Smart Spraying Unleashed: Exploring the Power of AI in

Pesticide Applications
Manan Shah
Department of Chemical Engineering, School of Energy Technology, Pandit Deendayal Energy
University, Gandhinagar, 382426, Gujarat, INDIA
Email: [Link]@[Link]
Harsh Shah
Department of Computer Engineering, School of Energy Technology, Pandit Deendayal Energy
University, Gandhinagar, 382426, Gujarat, INDIA
Email: harshdev624@[Link]
Kushagra Darji
Department of Computer Engineering, School of Energy Technology, Pandit Deendayal Energy
University, Gandhinagar, 382426, Gujarat, INDIA
Email: darjikushagra@[Link]
Karan Shah
Department of Computer Engineering, School of Energy Technology, Pandit Deendayal Energy
University, Gandhinagar, 382426, Gujarat, INDIA
Email: karanshah2100@[Link]
Adit Shah
Department of Computer Engineering, School of Energy Technology, Pandit Deendayal Energy
University, Gandhinagar, 382426, Gujarat, INDIA
Email: aditjignesh44@[Link]

Abstract
Artificial intelligence can be defined as the technique of adding human-like intelligence to the
machine, making it perform activities involving decision-making. IoT is a physical device with a
software application and electronic elements that provide them the ability to connect and
communicate over the internet. It makes agriculture spell serious health and environmental problems
due to excessive pesticide application. The chapter analyzes how Artificial Intelligence (AI) and
Internet of Things (IoT) technologies could be applied to optimize the pesticide spraying with the aim
of minimizing chemical footprints with the least effect on crop productivity. Combining AI algorithms
with IoT sensors, we could make advanced systems that monitor the condition of crops, detecting
threats from pests and diseases and applying pesticides only when and where needed. It will be
workable to improve agricultural sustainability with the least harm to the environment. This chapter
entails some of the AI techniques in smart pesticide spraying and discusses further the pros and cons
related to its application. Some issues concerning this technology are also about saving on pesticides,
improving quality crops, and reducing costs.

Keywords: Artificial Intelligence; Agriculture; Pesticides; Internet of Things; Smart spraying

Techniques
I. Introduction
Smart Agriculture
Smart agriculture is the use of information technology in agriculture to achieve precision in various
aspects of it. In this, agricultural machines and devices are equipped with various IoT based devices
which are used to collect data and make decisions according to it. The rise of smart agriculture is
backed by rise of various technologies like IoT, Big Data, Artificial Intelligence and Satellite
observations. Smart Agriculture is the amalgamation of various technologies used in the agriculture
sector to increase the yield of crops. (Taylor 2017)
Smart Agriculture is significant in the current scenario to keep the pace of increase in agricultural
production with the rising world population. The world has almost reached the peak of harvestable
land and that is also depleting due to excessive use of pesticides, climate change and increasing land
pollution. So, to increase the yield of crops there is a need to increase the yield per hectare of crops
which can be achieved using smart agriculture. Smart agriculture increases the efficiency of resources
with thoughtful use of it. It involves rotation of crops so that fertility of soil remains intact and use of
advanced methods like drip irrigation and greenhouse farming to save resources.
(Zinke-Wehlmann and Charvát 2021)

Artificial Intelligence (AI)

Artificial Intelligence deals with automation of tasks that are believed to require human intelligence.
These include recognizing faces of different people, colour, objects, playing mind games like chess,
etc. It tries to achieve taking decisions by computer with same precision as human which requires
human intelligence. As Artificial Intelligence can be used to automate any human decision, it has a
potential to contribute enormously to numerous fields. Though it is used in various fields, it is
regarded as domain of Computer Science because it is thoroughly used to formulate and solve the
problem. Artificial Intelligence is a huge domain as it includes other domains like Deep Learning and
Machine Learning in it. Machine Learning is used to analyse past data and draw predictions based
upon it. Deep Learning (a subset of Machine Learning) is used to analyse a particular set of data.
There are tremendous applications of AI in wide range of fields to automate tasks and reduce human
interference. It is used in self-driving cars so that they can drive safely on the road which involves
taking numerous humans like decisions at every point of time. (Honavar)

Artificial Intelligence in Smart Agriculture

AI serves as the technological facilitator in numerous industrial IoT scenarios, encompassing diverse
applications such as environmental monitoring, agriculture, smart grids, urban infrastructure,
buildings, homes, transportation, and healthcare. Within this context, AI employs various techniques
such as image enhancement, sensor data analysis, data transfer, and decision-making. These processes
necessitate swift data streams, minimal delays in communication, quick processing, and prompt
actions within the IoT system. Leveraging reinforcement learning models, the AI model operates
efficiently within the limited memory capacity of IoT devices. (Adli et al. 2023)
Smart farming heavily relies on machine learning (ML) to carry out crucial tasks like crop selection
and management. ML algorithms are utilized to assess and determine the most suitable crops for
specific soils. Hence, very critical and needed is that the farmers make land choices, thus; for those
goals as well, ML classification algorithms are also employed. Optimal water resource levels turning
up to match crops are also determined by using ML regression algorithms for this purpose. (Akkem et
al. 2023)
According to experts, although the feature is at its fastest-growing rate, smart agriculture technology
is still at a baby stage. IoT agricultural applications are yet to be fully populated with a lot of space
waiting to get filled regarding possible integration and its effects on agricultural growth, this
according to (Yousaf et al. 2023).
Smart spraying of pesticides
Smart spraying of pesticides is a very significant innovation in modern farming. Indeed, it enhances
the crop yield and quality while the damages on the environment due to pesticides are reduced.
Pesticides are required for agricultural production, but some monitoring on their usage will have to be
effected to avoid over-usage of pesticides that have some adverse effects on the environment. Proper
usage of pesticides could help save the environment from these effects (Chaitanya* et al., 2020;
Dhumale and Bhaskar, 2021). Overuse and blind spraying, a common act related to conventional
pesticide application methods, affects the natural environment negatively. Smart spraying
technologies adapt only a small amount of insecticide in an intelligent way under the leadership of
advanced technologies such as image processing, machine learning, and robotics by deciding which
crop needs treatments. On the other hand, reduction in pesticide use and avoidance in environment
damage means an increase in crop yield and quality. (Chaitanya* et al. 2020; Unal et al. 2020;
Murugan et al. 2020; Dhumale and Bhaskar 2021; Dhanoore 2022).
Other instances are, but not exclusive to the automated system for disease detection and pesticide
application by image processing techniques of (Dhumale and Bhaskar 2021); smart spraying robots
for pesticides of (Chaitanya et al. 2020); and an advanced automated pesticide spraying bot by
(Murugan et al. 2020). Furthermore, organic/inorganic composite material containing chitosan has
been used in the synthesis of pH-responsive pesticide carriers for controlled delivery under the
influence of UV light (Liu et al., 2022a). Other innovative approaches being now developed include
solar-operated grass cutters with in-built mechanisms for spraying pesticides (Dhanoore, 2022) and
UAVs that are well-configured for the smart spraying of pesticides over agricultural fields (Unal et al.,
2020).

Artificial Intelligence in Smart spraying of pesticide

Success in AI interferes with almost every walk of life, and agriculture is no exception. The use of
data and technology has undoubtedly led to better accuracy and efficiency in farming. AI continues to
progress and keeps on driving favorable changes across numerous divisions (Tadpatri et al.).
Using robots in farming lowers expenses and makes farming more precise. Precision agriculture
utilizes sensors to detect and spray plants with the correct number of pesticides. A robot with a colour
sensor effectively identifies weeds, making it suitable for commercial pesticide application in fields.
(Tahmasebi et al. 2022a)
Traditional farming techniques were not good enough for modern needs. Artificial Intelligence in
agriculture, especially 'Smart Spraying of Pesticides and Herbicides,' has revolutionized farming
methods. (Talaviya et al. 2020a)
Modern farming equipment uses Machine Vision along with Artificial Intelligence to improve
spraying techniques. One method involves a cost-effective smart sensing system for controlling
sprayers in tree crops, specifically citrus. This system was designed and tested to enhance efficiency.
(Partel et al. 2021a)
Scientists have created prototypes of sprayers that use Artificial Intelligence (AI) and Deep Learning
to identify specific weeds. These sprayers are connected to different nozzles that spray only on the
targeted weeds. They will be tested in real vegetable fields to compare their effectiveness and
chemical usage with traditional sprayers. (Partel et al. 2019a)
The old ways of spraying herbicides have harmed farmer’s health and cost them more work. To solve
these issues, the Smart Spraying System was created. It moves on its own in the field, making things
easier and safer for farmers. The system can be improved with more technologies for better
performance. (Tahmasebi et al. 2022a)
II. Advanced Technologies

Figure 1: Advanced Technologies

A. IntelliSprayNet: Deep Learning-Enabled Smart Variable Rate Spraying System

Conventionally, pesticides are sprayed on all the crops at regular intervals of time. This spraying of
pesticides is uniform throughout the field. Uniform application (UA) of pesticides uses large amount
of chemicals which surges the crop cost and degrades the soil quality over a long period of time.
Variable rate application (VA) is a better approach to use pesticides on the crops which reduces the
use of chemicals and decreases the cost of the crop. It also helps to preserve soil quality by using
minimal amount of pesticide. Now, Deep Learning is used to enhance the functioning of variable rate
sprayer in which pesticides are sprayed on the crops based on their requirements. Pesticides are
sprayed only on the diseased plants and weed patches by the smart sprayer. Diseased plants and weed
patches are detected using various Deep Learning algorithms based on morphology and texture of
plants. All the diseased plants and weed patches are detected and pesticides are sprayed in real time.
Real time detection depends upon quality and quantity of images, available hardware resources like
GPU and algorithms like Deep Convolutional Neural Networks (DCNN). By spraying pesticides to
only diseased plants and patches of weeds around 40% use of pesticides can be reduced. Around 75%
of pesticides can be saved in the case of bushes and tree crops. This reduction in use of pesticide
considerably reduces the cost of crops and prevents the harmful environmental effects of excessive
pesticide present in the soil. (Hussain et al. 2020)
In a study conducted by (Liu et al. 2021), a Deep Learning-based variable rate spraying method was
employed to effectively manage weed control in strawberry crops. The aim of this paper was to
minimize the pesticides usage which significantly decrease the cost of crops and prevent soil pollution
caused by chemical pesticides. The dataset of this research consists of 12, 443 images containing
images of spotted spurge (4200), Shepherd’s purse (4265) and strawberry plants (4178). With this
dataset, three Deep Convolutional Neural Network (DCNN) models, namely AlexNet, GoogleNet and
VGG-16, were trained using the TensorFlow framework. All the DCNN models were trained using
70% of dataset images and then DCNN models were used to sort the weeds from the images taken by
camera in real-time. To assess the performance of the DCNN models precision, F1-score, recall and
overall accuracy were considered. Every model's performance was converted into matrices of binary
confusion which comprises of true positive (Tp), false positive (Fp), true negative (Tn), false negative
(Fn). Here, Tp are the images which are correctly classified as weeds, Tn are the images which are
correctly identified as strawberry plants not containing weeds, Fp are the images which are identified
as weeds but are not weeds and Fn are the images which are identified as strawberry plants but are
weeds. Precision is calculated as Tp / (Tp + Fp). In all parameters, VGG-16 model scored higher
values than GoogleNet and AlexNet. Also, VGG-16 model registered the highest number of sprayed
weed targets. Though VGG-16 model is most accurate but results of GoogleNet model were also
significant and AlexNet model is less accurate for the real-time classification of weeds. In the
laboratory experiment, VGG-16 model scored 93% in completely sprayed weed targets while
GoogleNet scored 90% and AlexNet stood at 87%. From the combined results of lab and real time
field experiments, VGG-16 model is very accurate in classifying weeds from the strawberry plants
and AlexNet model is not preferable for variable spraying of pesticide on weeds present among the
strawberry plants due to its low precision.
A study by (Reas as Alam et al. 2020) focused on variable rate spraying of pesticide on weeds present
among the crops using Deep Learning. Traditional spraying of pesticide may result in under or over-
dosing of chemicals. Under-dosing results in loss of nutrients to the weeds and over-dosing causes
serious environmental damage. The main purpose of this paper is to prevent under or over-dosing of
pesticide to get optimal amount of crop without harming the environment. The training dataset for this
study consists of 99 images of crop plants, 99 images of weeds and 99 irrelevant images. Also, 33
images are used as testing dataset for each segment. So, the total dataset contains 396 images. All the
images were captured over the span of many days at different times with varied lighting conditions
and at different angles to enhance the classifier ‘s accuracy. During image pre-processing, all the
undesired and noisy regions were removed. Techniques of Image enhancement were used to improve
the image’s visual appearance and pixel brightness value of the images were modified to improve its
quality. Now, thresholding-based segmentation technique is used to remove the background contents
of the image. Next step is to extract features from the image, and it is done using several global
features. Classification of weeds and crops can be done using colour, edges, textures and shapes as
both have slight differences in these features. After extracting features and labelling the training
dataset, a classification model based on machine learning is trained. During training, model is trained
using labelled data and it segregates each image. Using the training dataset, images having noise and
sunlight variation along with blurred images were classified by the model. Various learning models in
the research based on Scikit Learn includes K Nearest Neighbours, Support Vector Machine, Gaussian
Naive Bayes and Random Forest Logistic Regression. Random Forest, a supervised Machine
Learning algorithm, was used to solve both regression and classification problems. A forest is formed
by multiple decision trees constructed using randomly selected samples. A greater number of decision
trees ensures more accurate results. The accuracy of classifier of Random Forest is 95% using all
features of classification which drops when classification features are reduced. The algorithm's
performance is evaluated by measuring the time it takes for the algorithm to execute. The time
required by the whole algorithm to execute was calculated as 57.4ms at 17.4fps. This study introduces
an agricultural sprayer that utilizes real-time vision-based technology for weed and crop detection.
Based on several experiments, Random Forest algorithm works satisfactorily in real-time
environment. Variable rate sprayer used Pulse Width Modulation to limit the rate of flow of pesticide.
In a paper by (Tewari et al. 2020) , authors focused on implementing variable rate spraying system
which works in real-time to control disease in paddy crops based on image processing. The main
purpose of this paper is to develop a real-time image processing algorithm to assess the severity level
of plant disease and develop a real-time variable rate spraying system for optimizing spraying of
pesticide depending upon disease severity. This study considers one of the common illnesses in the
paddy crop white tip disease which has visually distinct symptoms. The leaf tip of plant suffering
from this disease becomes white for up to 5 cm. The RGB images of the crop were captured in each
row using web cameras with the resolution of 640 x 480 pixels. To reduce the effect of varying
environmental conditions of the field in the images, green cloth covering was used on image
acquisition device and image scene. Another advantage of green colour background is that it avoids
false detection of diseased plants and reduces noise in the image capturing area. To detect disease
region in the paddy crop, colour segmentation algorithm was used which is based on chromatic
aberration (CA). From the boxplot derived from original RGB images, it was derived that values of R
and G pixels are different for lesion and healthy region. To distinguish between lesion regions and
healthy regions as well as the background, the RGB colour space model can be employed. The CA
(colour attribute) values of the lesion region exhibit noticeable variations compared to the healthy
region and background, allowing for successful identification and extraction. Now, in thresholding
segmentation step, binary images were formed from RGB images by using threshold for CA values
(T=0). After segregating the infected plants, it is important to estimate the disease severity. Disease
severity can be estimated by finding the percent value of the area of plant leaf which is infected and
infected area of plant leaf can be known by white colour area on the leaf tip. The variable rate sprayer
sprays pesticides based on the severity of the disease of the paddy plant which is decided in real-time.
To make decision of the amount of pesticide that needs to be sprayed simple, the severity of the
disease is categorized into three levels: low, moderate, and high. Every category of disease severity is
assigned a reduction coefficient which determines how much less pesticide needs to be sprayed from
the recommended volume. This reduction coefficient is used to calculate the reduction in use of
pesticide for the whole paddy field. Reduction coefficient for low severity is 0.33, medium severity is
0.67 and high severity is 1. Use of pesticide was reduced by 33.88% to 40.47% from the
recommended volume per unit area. Further the accuracy of spraying pesticide on the infected region
of paddy plant can be improved using ultrasonic or infrared sensors.

B. Agronomatron: Advanced Colour-Based Vision System for Autonomous Pesticide

Spraying Robot
Self-regulating machines equipped with a visually focused sensory system that enables the precise
deployment of chemicals have emerged as a cutting-edge solution in the field of agriculture. The
traditional method of applying insecticides frequently results in haphazard application, excessive
chemical usage, ecological issues, and possible health concerns. However, the integration of a colour-
based vision system into autonomous sprayer robots aims to address these challenges by enabling
targeted and precise pesticide spraying. Color-based perception is one of the perceptions that can be
done automatically and perceptibly in real-time image processing. A robot will perceive a given color
in a target area of interest and take control of the domain for that region autonomously to treat it with
pesticides. In addition to that, the autonomous pesticide-spraying robot is equipped with a precision
mechanism for spraying, and the color of the target areas in the field would then be defined. These
robots are autonomous and make their way through the field by using GPS or other positioning
systems, and vary the parameters of the spraying, such as pressure, nozzle angle, and volume of
sprayed liquid, to hit the spot. Automated functions in a sprayer robot enable it to work by itself,
either following a path set before or adjusting according to real-time conditions in the environment.
This would, therefore, obviate manual intervention and thus guarantee uniformity and consistency in
the spread of sprays with pesticides in a field. The other aspects where the autonomous pesticide
sprayer robot bases its vision are color detection and hence has several advantages. This makes the
application of the pesticide very accurate and directed, hence reducing volumes used and lowering the
environmental impacts. This is because it only sprays the infested areas, hence avoiding non-target
zones in the process, saving beneficial insects and its soil health and water resources. More so, such
systems save operational cost through the time and labor savings. Large-scale coverage efficiencies of
autonomous robots equate to substantial labor reductions in manual sprays. It will also allow
chemicals to be applied when most effective with the least crop damage. A color-based vision system,
in combination, could offer real-time monitoring and information gathering on the prevalence of pests
or diseases within a field. That would inform, and could also help in making decisions at every
juncture pertaining to pest management strategies or regarding crop heath and yield optimization
(Tahmasebi et al. 2022b).
(Nasir et al. 2023) worked toward a robotic sprayer for precision agriculture and real-time tobacco
identification, meant for integration into a spraying system. The work aims to develop a vision-guided
robot platform that will selectively spray the tobacco plants with minimal agrochemical applications
while applying pesticide using the developed deep learning technologies. The experiment was
conducted on a real tobacco field in Pakistan, and a data set comprising 6500 images containing
tobacco plants and 2000 images containing weeds, each with different environmental conditions. In
this paper, deep learning algorithms—more precisely, the YOLO-based models—have been used for
defining or identifying tobacco plants in accordance with the preset objectives of the present study.
This paper deals with the experiments currently conducted in tobacco fields to evaluate the
performances of the six YOLO-based models for tobacco plant detection. The work proposes a new
method of control, with the name disturbance-based control, which generalizes the problem of
pressure fluctuation under the traditional bang-bang control. A qualitative and quantitative evaluation
of this new control method was carried out to measure its effectiveness in bringing improvements to
the spray's quality under three different scenarios of spraying: broadcast spraying, selective pressure
of 20 psi taking into account variations of pressure between 15 to 120 psi, and variable rate spraying.
The YOLOv5n model trained with field data results in an execution of 67 fps with an F1-score of
87.2%. This may be combined with an application strategy or method for selective and variable rate
spraying to result in approximately a 60% saving, effective against conventional blanket broadcasting.
The possibilities of reduced pesticide application through the proposed scheme for deep learning
algorithm-based tobacco plant selection for spraying will help reduce associated costs and
environmental damage. They finally reached the fact that appropriate spraying systems, designed with
deep learning for in-precision agricultural robots—real-time recognition of tobacco—can usefully
overcome current problems in tobacco-field overuse and underuse of agrochemicals. Also mentioned
in this study is the possibility of the extension of this technology developed for other crops, which
calls for future research in this area. Proof of concept in deep learning algorithms has been established
in their selective applications for tobacco spraying and it generates insight valuable for developing
similar systems for different crops with excess or deficiencies dealing with pesticide applications.
A paper by (Adamides et al. 2017) centres over the creation and implementation of a semi-
independent vineyard sprayer for agriculture, with a specific emphasis on the user interface for
teleoperation in the designated study area. The objective and aim of this paper are to present a
methodology for transforming a general-purpose mobile robotic platform into an agricultural robot
sprayer solving both user interface and hardware problems. This paper fails to specify any dataset
employed during the study. But it explains in detail the computer and software components that must
be added to the system with a special emphasis on the interaction between humans and robots and
explores the efficacy of the user interface in two on field studies and two laboratory experiments. The
researchers conducted heuristic evaluations on various user interfaces using HRI (Human-Robot
Interaction) heuristics grounded in research. The field studies provided evidence indicating enhanced
usability of the SAVSARv2 (final) system, which could lead to its widespread acceptance among end
users. Regarding target selection input devices, the Mouse and Digital Pen demonstrated superior
usability compared to the Wiimote. The researchers observed significant improvements in usability
but reduced efficiency when utilizing multiple views compared to a single view. Additionally, the PC
keyboard exhibited significantly better interaction efficiency and perceived usability compared to the
PS3 gamepad. The paper discusses the consequences of the results and their potential application in
other areas of mobile robotics. Finally, the paper asserts that the methodology used can be applied to
various distinct mobile robot applications and provides proof of the improved utility of the
SAVSARv2 system, which received positive feedback on the User Experience Questionnaire.
In their study, Sammons et al. (2005) introduce an engineering strategy to address the health risks
faced by humans when applying hazardous chemicals in the enclosed and humid conditions of a
glasshouse. The target is the commercial greenhouses where there is a critical issue with applying
pesticides in controlling pests and diseases. This paper presents the design and development of an
autonomous mobile robot that can travel through the rows in a greenhouse to spray the required
quantity of pesticides on the plants efficiently and uniformly while alleviating the concerns of human
health related to the use of pesticides in greenhouses. That is, the results of this study were numerical
and determined by data of three observers, different tests in two cucumber greenhouses, and the
outcome was even application of pesticides to all plants by the autonomous robot. This paper
elaborates the design and development procedures for the robot, the safety procedures regarding the
robot, the software used for operating the robot and evaluates its cost-effectiveness in terms of
automation under which this Greenhouses. This means that machine functions are designed for
pesticide spraying, fruit collection, and care of infected plants within the greenhouse. The autonomous
pesticide spraying robot successfully navigated down rows of a greenhouse while efficiently covering
all the plants equally with pesticide spray in the prescribed dosages. Employing an autonomous robot
in a greenhouse offers several advantages, such as lowering labour expenses, minimizing pesticide
wastage expenses, and notably mitigating human health risks. The study concludes that the benefits of
utilizing an independent robot in a greenhouse far exceed the costs associated with the manual process
of spraying of pesticide, both economically and socially.
The robot is intentionally designed with a high degree of modularity, allowing for the development or
customization of each specific task independently. The efficacy of the robot in a fake greenhouse
environment indicates no potential for further development of automation systems in greenhouses.

C. CropSentinel: Intelligent Drone Pesticide Spraying Technology

Currently, conventional methods are predominantly used worldwide for pesticide application.
The manual mechanical sprayer is widely utilized in traditional pesticide application, yet it poses
significant risks to human health, potentially resulting in various ailments such as asthma,
hypersensitivity, cancer and related disorders. (Koç 2017). Moreover, regular methods have various
limitations, including excessive chemical use, labour shortages, uneven spray distribution,
environmental pollution, and limited coverage area. These drawbacks result in higher pesticide
application costs and less effective pest and disease control. To address these issues, the use of drone-
mounted sprayers has been introduced. Employing drone-mounted sprayers in the field has brought
about several improvements. They made it possible to cover great areas with greater effectiveness in
applying the chemicals and improved the process of spraying, making it easier and taking the least
possible time. Recent drones are made with agrochemical tanks holding 40 liters in order to follow the
route for applying pesticides by the need's preset. Such drones turn out to be highly useful when
reaching those places that prove to be very hard to reach for tractors and aircraft. Figure 1: (SINHA
2020). Besides the actual act of carrying out agricultural activities, drones in agriculture help expand
the level of mastery gained over pests and other organisms that are harmful while causing only
negligible threats to humans and nature. Typically, a general process of coming up with a pesticide-
spraying drone starts with payload estimations. Depending on the payload needs, additional parts of
the drone could be decided upon. The battery type is chosen based on the voltage and current
requirement according to the modules incorporated into the drone. Finally, the drone frame is
designed with parameters like the number of arms and payload capacity. There has been much
research to enhance its stability, control, and spray mechanism of an effective pesticide spray. These
pave the way for easy and efficient access to drone-based pest management over small plots and
difficult terrains.
The work of (Kumar Shaw) reported here is on the development of a tailored work-drone system for
the spraying of various agrochemicals within the agricultural sector. The intended objectives of the
study are to surmount the problems posed by conventional manual spraying methods, which are
normally laborious, involving much outlay, and sometimes health risks. The objective of the
researchers was to revolutionize farming practice by introducing an automated solution to bring
precision, efficiency, and crop health improvement while simultaneously reducing the demand for
labor. In effectuation of this approach, the researchers went on an exhaustive exercise of design and
development in building a drone system meant exclusively for use in agricultural spraying. Advanced
technologies in the design of the drone included GPS, computer vision, and precision-spraying
systems. It had a payload tank to hold the chemicals and a spraying mechanism that could precisely
dispense its contents. The major highlights of the drone system were autonomous navigation using
GPS over large agricultural areas; it has also reduced the human involvement factor as less intensive
labor requirement, computer vision algorithms to detect crop boundaries for optimized spraying routes
and accurate, even application. The drone's spraying mechanisms were supposed to use those patterns
and volumes of spray characteristics for avoidance of wastage and burden on the environment.
Extensive performance assessment of the drone system was made through field tests across a range of
agro-ecological zones. In this regard, the study results on efficiency in spraying, reduction of cost in
labor, and improvement in crop coverage are important. Since compared to the manual spraying,
automatic spraying is accurate and uniform; hence one can only expect healthier crops with higher
yield. In summary, the potential of the drone technology as a game-changer in the agricultural
practices had been outlined by the research from Shaw et al. The developed drone system effectively
overcomes problems associated with traditional manual spraying, increases precision and efficiency
rendering healthy crops and finally ensures high levels of productivity. Advanced technologies in GPS
guidance, computer vision, and precision spraying devices are therefore merged to bring out very
important advantages like reduced manpower and less environmental impact with full crop coverage.
(Spoorhti et al. 2017) introduced Freyr drone, which is an innovative solution for uniform spraying of
agrochemicals over the field in agriculture. Freyr drone integrated many advanced options that
enhance the performances of Freyr drone. Then to make this easier than ever before, researchers
developed an intuitive Android application having the Wi-Fi interface, through which a farmer could
command their Freyr drone straight from their mobile devices. The entire control system was
essentially provided in an app-based way, acting as an effective control center for communication and
operation. An Arduino Mega-2560 acted as the smart controller board at the heart of the Freyr drone
system, which formed the core element in controlling the whole process. Onboard, this robust board
provided precision control over the drone's flight and spraying systems, ensuring an accurate hit and
application. The prime benefits of the drone were location overpass and covering any awkward shape
of a field area, particularly where crops are farmed at low level height with crops near to the ground.
In contrast, the Android app based itself on a farmer being very technical in order to fully exploit it
and consequently help that farmer maximize the usage of the capacity of the drone in improving their
farming process. The drone by Spoorthi et al. developed in 2017 presented very high stimulus to the
agricultural sector with the advanced solution brought in terms of uniform spraying, user-friendly app,
smart control board, and flexibility in attaining different shapes of fields; hence, high potential for
improvement in the practice of agriculture can be realized.
(Borikar et al. 2022) give a critical review on the applicability of drone systems used in modern
agriculture to apply pesticide sprays. The authors perform a literature review by comparing views of
the research articles, technical reports, and industrial publications on the benefits, challenges, and
technical development with the use of drones for pesticide sprays. The benefits of this review are
increased accuracy, reduced labor, and reduced drifting of chemicals during spraying. Drones have
made it possible to pinpoint spray areas in an increase of efficiency with a low cost of application.
Research finds there has been a serious push technologically in developing drone systems: accuracy
improvements for better spraying; data analytics integration; navigation improvements; obstacle
avoidance. Prognosis over the long term: Drones could be game-changers in the application of
pesticides, hence for sustainable agriculture. This adds a call for heavier efforts to be channeled into
developing and working out the problems currently associated with drone systems, so as to achieve
much better spray performances, and develop regulatory frameworks and training programs to assure
proper and safe use of drones in agricultural applications. This emphasizes the great advantage of
drone systems for pesticide spraying in advanced agriculture. Emphasis is given to the many
advantages and benefits related to drone-based spraying, though pointing out challenges and the very
possibility of drones changing the way pesticides are applied. The research could add knowledge
about agricultural drone technology and would also provide more input into the future research and
development of this kind.

D. SpecieSeek: Adaptive Target Detection and Species Classification

While smart pesticide sprinkling on the crops involves two key components of Target Detection,
application is done through Artificial Intelligence and Machine Vision and further through Robotics
and Drones. For Target Detection, the identification of weeds or unwanted plants is done through the
use of Machine Vision. It has two methods. In the first method, first of all, all plants are detected and
the weeds among them are classified. The second method involves identifying all plants, then
specifically detecting the crop plants and classifying any remaining plants as weeds or unwanted
vegetation. Detection of unwanted plants is dependent on numerous parameters like texture, colour,
shape and size of the leaves of the plant. Machine Vision is a very efficient method to detect the
physical parameters of the plant which are used for detection. So, Machine Vision is widely used in
target detection, species classification and identifying diseased plants. Using Machine Vision targeted
plants are classified and then pesticides are sprayed only on the targeted plants. Instead of spraying
pesticides on all plants this method is used to reduce the cost of crops and pesticides content in the
soil. Pesticides content in the soil affects the fertility of the soil so this technique is also useful to
maintain the fertility of the soil. (Song et al. 2015)
A study by (Yu et al. 2019) focuses on weed detection for spraying of herbicides in ryegrass using
DCNN (Deep Convolutional Neural Network). The objective of this paper is to decrease the use of
herbicides by spraying them only on the weeds present in the ryegrass which reduces the cost of weed
detection and control in turfgrass. A and B training datasets consist of three types of weed species (E.
maculate, T. officinale and G. hederacea) along with perennial ryegrass. Balanced Dataset A consists
of 19, 500 positive images (images containing targeted weeds along with ryegrass where 6, 500
images for each weed species) and 19, 500 negative images (images containing only ryegrass).
Unbalanced Dataset B consists of 17, 600 positive images (E. maculate (6, 500), T. officinale (6, 500)
and G. hederacea (4, 600)) and 15, 486 negative images. For validation, 900 positive and 900 negative
(300 for each species) images were used. Three models based on DCNN VGGNet, GoogleNet and
AlexNet were trained on both datasets. Results obtained from validation and testing were converted
into binary confusion matrices consisting of Tp, Tn, Fp, Fn. In this context, Tp represents the images
of target weeds that are accurately identified, Tn represents the images of ryegrass without target
weeds that are accurately identified, Fp refers to the images inaccurately identified as weeds, and Fn
refers to the images inaccurately identified as ryegrass. Precision, F1-score, recall and Mathew's
correlation coefficient (MCC) have been calculated for all models. Results using validation dataset is
that VGGNet showed excellent performance in recognizing E. maculata and G. hederacea and
achieved high recall (≥0.9952) and F1-score (≥0.9278) for both datasets. For identification of T.
officinale, VGGNet showed high recall and F1-score when trained with dataset B than A. MCC values
were highest using VGGNet and lowest using GoogleNet. AlexNet also showed significant
performance, but GoogleNet was ineffective in recognizing weed species due to less accuracy.
VGGNet and AlexNet can be used to identify multiple weed species, but GoogleNet is not acceptable
due to very little accuracy.
In a paper by (Zhao et al. 2022a) , authors focused on precise and efficient detection of weeds in
potato fields using improved YOLOv4 model. The aim of this paper is to reduce the amount of
pesticides used in the potato fields and propose an efficient method of controlling weeds by detecting
weeds present in the potato fields using the improved YOLOv4 model. This paper can be used as
reference to detect weeds in other crops using YOLOv4 model. The dataset of this research consists of
2000 images which consists of 654 images of only weeds, 490 images of only potatoes and 856
images of weeds and potatoes. Images were captured in various lighting and environmental conditions
at various growth stages of plant from fixed height. To enhance the size of the dataset, data
enhancement techniques were applied, including random horizontal flipping, variations in brightness,
the addition of Gaussian noise, pretzel noise, and scatter noise with suitable magnitudes. From the
initial 2000 images were randomly selected of which 550 images were enhanced so enhanced dataset
contains 2550 images. From the dataset of 2550 images, 1785 images were utilized for training, 510
images for validation and 255 images for testing. Target detection of weeds can be done using two
methods: one stage and two stage target detection algorithms. Target detection algorithms with two
stages have high detection accuracy but slow detection. Target detection algorithms with one stage
consists of the SSD series, YOLO series, Retia-Net, etc. have low detection accuracy but fast
detection. Mosaic data enhancement method is used to enrich the background of the target and
improves the model efficiency. The main cause for low accuracy of the YOLOv4 model is the choice
of anchor box. In the initial YOLOv4 model, an anchor box was derived from the COCO dataset
using K-means clustering algorithm. To improve the accuracy, an improved YOLOv4 model (known
as MC-YOLOv4) was used in which K-means++ clustering algorithm was used rather than K-means
clustering algorithm. To evaluate the results, single image detection speed, average precision (AP),
mean average precision (mAP) and other model parameters were utilized. The mAP of MC-YOLOv4
is highest among the different models of YOLO series that is 98.52% with average precision of weed
at 98.63% and average precision of potatoes at 98.42%. The MC-YOLOv4 model proposed in this
paper detects weed targets with great accuracy along with fast detection matching the real time field
requirements. The MC-YOLOv4 model requires very less dataset to achieve high accuracy so it can
be very useful for detection of weeds in potato fields, and it can also be used in other crops.
A paper by (Zhang et al. 2022) addressed the difficulty of detection of weeds in the peanut fields due
to cross growth of peanuts and wide variety of weeds using EM-YOLOv4-Tiny model which is an
improved version of YOLOv4-Tiny model. The original training dataset contains 700 images of 6
types of weeds and peanut plants. To increase the richness of the training dataset, flip horizontally, flip
vertically, brightness increase, brightness decrease, Gauss noise enhancements techniques were used
and 500 images per technique were added. So, the total training dataset contains 3200 images with
enhancement. The test dataset contains 155 images without enhancement. The YOLOv4-Tiny model
consists of four elements: an input layer, a backbone network, a FPN and an output prediction layer.
To achieve feature fusion, features extracted from CSPDarkNet53-Tiny were sent to FPN. The
CSPDarkNet53-Tiny mainly contains CBL and CSP (cross-stage partial) module.
The CBL (Controlled Feature Binding and Sampling) module, serving as the smallest component
within the network, is responsible for feature control splicing and sampling. On the other hand, the
CSP (Cross Stage Partial) module is an enhanced version of the residual network structure that
enables the division of the input map into two segments. Three modules of CSP: CSP1, CSP2, CSP3
are present in CSPDarkNet53-Tiny. As the output map dimensions are reduced within the CSP
module, the spatial information becomes less precise, resulting in a scarcity of detailed information
and decreased ability to detect small objects. To address these challenges, a path connected to the
CSP2 module is introduced in the FPN (Feature Pyramid Network) layer. Additionally, the ECA
(Efficient Channel Attention) module is applied iteratively to process the effective features present in
the FPN module, thereby enhancing accuracy. The Soft-NMS (Non-Maximum Suppression) algorithm
is utilized to filter out redundant prediction boxes surrounding the targeted weeds in the YOLOv4-
Tiny output. The performance evaluation of the model is conducted using precision, F1-value, recall
rate and mean average precision (mAP). The YOLOv4-Tiny model mis detect smaller targeted weeds
(resolution lower than 32 x 32). So, mAP value for small targets using YOLOv4-Tiny model is low
(79.53%). This problem is overcome in the improved EM-YOLOv4-Tiny model as its mAP value for
small targets is 89.65%. Overall mAP value for YOLOv4-Tiny model is 87.71% and EM-YOLOv4-
Tiny model is 94.54%. Among the various models of YOLO series EM-YOLOv4-Tiny has the highest
precision of 96.7%. Small target detection was always a burning problem in identifying weeds among
the peanut plants which is to some extent is addressed by EM-YOLOv4-Tiny model. Still, there is
further scope of research as all the dataset in this research only includes peanut plants in their seeding
stage. Also, the precision of the new YOLO model is higher than the previous ones, but it increases
the size of the model. The proposed EM-YOLOv4-Tiny model has higher recognition accuracy than
previously used models of the YOLO series.

E. AgroIntellect: AI-Enabled IoT Herbicide Spraying Solution for Precision

Agriculture
In recent times, there has been excessive technological advancements for spraying herbicides using AI
and IoT in the agricultural industry. The conventional approach to herbicide application frequently
results in excessive chemical usage, elevated expenses, and ecological concerns. The integration of AI
and IoT technology intends to address these challenges by enabling targeted and precise herbicide
spraying. These autonomous systems consist of artificial intelligence algorithms, Internet of Things
devices, and automated sprayers. The fusion of Artificial Intelligence and Internet of Things
technologies in AgriHaze will alleviate the problems that face the traditional method of herbicide
application, such as too much pesticide use, high expenses, and environmental impacts. The aim of
the paper is the precise determination of weed groups from crop vegetation and the proper dosage of
herbicides for an effective weed control process. It has therefore proposed new thoughts: it now has
autonomous systems applying herbicides, powered by artificial intelligence and the Internet of
Things. The corresponding information is in turn wirelessly sent to a central artificial intelligence
system that allows constant monitoring and assessment. These AI algorithms use this information
alongside historical factors and environmental parameters to come up with the exact
recommendations on spraying. The AI system stipulates directions to the robotic sprayer for
independent navigation within the agricultural field. The sprayer throttles patterns of spraying while
adjusting dosage and coverage through the instantaneous feedback from the IoT devices to assure
accuracy in the application of effectiveness in herbicides. All these systems hold a number of
advantages to the sector. First, they offer precision and, as a result, efficiency in weed infestations and
herbicide applications, thereby providing the opportunity for chemical optimization, cost saving, and
environmental footprint reduction. Second, it impacts the environment by not only reducing overspray
but also avoiding sprayed non-target areas, hence protecting valued flora, soil quality, and water
resources. In addition, the systems save money for agricultural producers through more efficient
applications of herbicides and improved allocation of labor resources. This would also mean that big
areas would be covered within a short duration as compared to manual spraying methods; hence,
time-saving. (Katiyar 1AD; Liakos et al. 2018; Tadpatri et al.)
(B et al., 2021) study the viability of an autonomous, self-sufficient spraying system on herbicide
crops using future-driven technologies like IoT and AI. As traditional methodologies of spraying
herbicides affect human health and plant growth, this paper is written with the goal of diminishing the
negative environmental impact while simultaneously attaining precision and efficiency in agriculture.
It is a herbicide-spraying automated system, capable of applying herbicides and other chemicals
effectively without posing any risk to plants, of course. All this basically does is portray how immense
the power technological advances have in that it can only know precisely its route through integration
with the system of GPS, magnetometer, and liquid level sensors for herbicide levels and from this
make decisions regarding its heading. Furthermore, the weeding robot estimates and identifies
unwanted plants by analyzing the location, color, shape, and texture of the plants. Heading error and
radial distance are calcualated by mathematical equations for ensuring precise navigation. The
Haversine rule is determined to quantify distances on a three-dimensional surface, such as Earth's
surface. The recommended technique has many advantages compared to typical traditional
methodologies. These advantages comprise of lowered labour costs, heightened efficiency, and better
accuracy. Furthermore, the application of weedkillers is reduced, which decreases ecological impact
and advocates for sustainable farming methods. The combination of Artificial Intelligence and the
Internet of Things in an independent system for administering herbicides is unlikely to enhance
effectiveness and precision in agriculture while also decreasing environmental impact. The system's
inability to navigate autonomously, detect weeds, and measure herbicide levels in real-time makes it a
useless tool for farmers and agricultural practitioners.

(Qazi et al. 2022a) contributed a comprehensive overview of the existing literature and recent
advancements in Smart Agriculture Systems (SAS). The paper explores the application of IoT
technologies and AI methods in the agricultural field. The main objectives of the paper are to provide
a detailed tutorial on the advancements in SAS through IoT and AI, critically evaluate these
technologies, discuss challenges in their widespread adoption, and examine future trends in the global
acceptance of SAS by farmers. Although the specific dataset used is not mentioned, the paper explains
how information can be gathered from various IoT sensors and UAV images in smart agricultural
fields located in different geographic areas. This data can be utilized to make more accurate decisions
regarding pest detection, plant diseases, smart irrigation, and the limited use of harmful substances
like herbicides. Additionally, it discusses recent advancements in AI algorithms based on Deep
Learning (DL) that can leverage data collected from diverse sources. The document explores the
current state-of-the-art technologies in SAS, the challenges associated with their implementation, and
the expected future developments. It emphasizes the significance of smart agriculture practices given
the increasing gap between global food demand and current production, the scarcity of arable land,
stricter regulations on toxic pesticides/herbicides by international organizations, and the global water
resource shortage for irrigation. The paper concludes that IoT and AI technologies have revolutionized
traditional agricultural practices and have the potential to optimize resource management, such as
minimizing water requirements for irrigation and reducing the use of toxic pesticides. However,
barriers still exist in the widespread adoption of these technologies, and future research should focus
on addressing these challenges to ensure global acceptance of SAS by farmers.

F. OptiMist: High-Precision Low-Volume Pesticide Spray Technology

The agricultural industry incurs huge losses due to pests, which decreases the overall yield of the
fields. A very effective way of removing pests from the fields is by spraying chemical pesticides on
the crops. Spraying chemical pesticides on all the crops regularly increases the cost of crops
significantly and deteriorates the quality of the soil. With increasing content of chemical pesticides in
the soil, the fertility of soil decreases. To maintain the fertility of the soil, more chemicals are being
sprayed which starts the vicious cycle of using more and more chemicals for regular yields. This
vicious cycle can be countered by using sustainable amount of pesticides which maintains the soil
quality and removes pests also. This equilibrium amount of pesticides can be achieved by spraying
low volume pesticides precisely to the pest affected plants at the right time. If pesticides are sprayed
at the initial stage of the pest attack where only a few plants are affected, then less pesticides needs to
be sprayed. If this technique is combined with targeted spraying of pesticides, then use of pesticides
can be decreased significantly. The nozzle of the sprayer changes the air flow and spray volume to
achieve optimal effects of pesticides. Also, the pesticides are sprayed from ideal distance to ensure
that only affected area of plant receives the pesticides. Ultrasonic Detection, LiDAR (Light Detection
and Ranging) and Imaging are used to identify plants affected by pests. (Nan et al. 2023)
(Mahmud et al. 2021) address the critical need to reduce the risks associated with pesticide use in tree
fruit agriculture. They conducted an in-depth investigation into the potential of precision spraying as a
solution. The study focuses on identifying the factors influencing spraying parameters, with the aim of
reducing excessive use of agrochemicals. The authors’ primary goal was to systematically summarize
the current technologies for precision spraying in tree fruits, emphasizing their potential to minimize
unnecessary chemical applications. To achieve this, the researchers reviewed existing precision
spraying technologies and analyzed the factors that impact spraying efficiency. Although they
provided a thorough overview of these technologies, some specific datasets were not included in the
paper. Despite this, the study offers practical solutions to help curb agrochemical overuse in the
industry. One of the study’s significant findings is that successful droplet deposition during spraying
depends heavily on accurate sensing technologies that detect the target effectively. The authors also
highlighted the importance of factors such as air jet velocity, travel speed, wind speed and direction,
droplet size, and canopy characteristics in achieving effective droplet deposition. By integrating these
variables into precision spraying systems, the over-application of agrochemicals can be minimized.
Ultimately, the study concludes that precision spraying holds promise as a strategy to mitigate the
risks associated with pesticide use in tree fruits. The research underscores the potential of precision
spraying to reduce excessive chemical applications and offers valuable insights into the factors that
influence spraying outcomes. The authors are optimistic that their work will contribute to the ongoing
development of more sustainable and environmentally-friendly agricultural practices.
(Liu et al. 2022b) explored advanced spraying techniques for orchards, focusing on efficiency and
accuracy. They introduced an innovative method utilizing a single 3D LiDAR sensor to guide a
spraying robot in real-time. The study aimed to demonstrate the effectiveness of this technology in
achieving accurate variable-rate spraying, minimizing chemical use, and enhancing spraying
efficiency in orchard management. The authors developed a precision spraying robot equipped with
the LiDAR sensor, enabling the robot to autonomously navigate orchards and detect tree canopies.
The high-resolution point cloud data generated by the sensor allowed the robot to precisely determine
the size and shape of tree canopies. By analyzing the LiDAR data, algorithms calculated the required
spray volume for each tree based on canopy dimensions, enabling the robot to adjust its spraying rate
for accurate variable-rate spraying. The study included experiments conducted in real orchard
environments, comparing the performance of the variable-rate spraying system with traditional fixed-
rate methods. The results showed that the LiDAR-based system provided better spray coverage and
reduced chemical usage compared to fixed-rate spraying. The authors highlighted the benefits of this
approach, such as enhanced spray efficiency, lower environmental impact, and cost savings. They
concluded that this precision spraying technology has significant potential for improving orchard
management and precision agriculture, offering a more efficient and eco-friendly alternative to
conventional spraying methods. The study also suggested that this technology could be applied in
other agricultural settings where precise spray delivery is essential.
(Zhao et al. 2022b) designed a highly accurate and resilient control system for targeted spraying in
challenging natural environments. The system is engineered to precisely identify cabbage plants and
different types of weeds. It features an active light source along with a targeted spraying delay model
to minimize the effects of changing natural light conditions on target detection. The identification and
positioning of targets are conducted using video streaming, with encoder-based ranging determining
the positioning distance for precise control. While the specific dataset is not mentioned, the authors
highlight that real-time identification of cabbage targets is accomplished using a Support Vector
Machine (SVM). The study introduces the use of skeleton point-to-line ratio and ring structure
characteristics as distinguishing features for the SVM model, which is used to classify and test
different combinations of these features. The methodology involves an SVM-based cabbage
identification model that identifies the optimal hyperplane to separate cabbage from weed samples.
The paper also describes a communication protocol for the control system, where video streaming
aids in target identification and positioning, and encoder ranging is used to calculate positioning
distance for accurate control during targeted spraying.
The research findings reveal that utilizing the combination of the point-to-line ratio, maximum
inscribed circle radius, and fitted curve coefficient as the characteristic vector yields the highest
identification accuracy, reaching 95.7%. The system's response time is 33ms, making it viable for
real-world production use. Field trials confirmed the accuracy of target identification and control of
the spraying machine. The system achieved a 95.0% precision rate in identifying cabbage samples,
while the precision for various weed identifications ranged from 92.2% to 96.2%, averaging 93.5%. In
the discussion, the paper notes that although the machine typically differentiates between cabbage and
weeds effectively, there are cases where crops are mistakenly identified as weeds due to growth
patterns or partial overlap. Additionally, some cabbage samples are not sprayed because of uneven
growth or localized reflections, leading to incorrect identification. To address these issues, the
proposed targeted pesticide spraying control system incorporates an active light source and a targeted
spraying delay model, achieving a high identification accuracy of 95.7% with the specified
characteristic vector.

G. SentiGrow: AI-Optimized Tree Crop Sprayer leveraging Sensor Fusion for

Unmatched Precision
The research findings reveal that the combination of the point-to-line ratio, maximum inscribed circle
radius, and fitted curve coefficient as the characteristic vector achieves a top identification rate of
95.7%. The testing process is also efficient, taking only 33 milliseconds, which makes it viable for
practical production applications. Field tests were conducted to confirm the accuracy of the target.
The efficient use of chemicals in agriculture is vital for reducing costs, minimizing waste, and
optimizing labor. To achieve this goal, advanced agricultural machinery equipped with artificial
intelligence (AI) and machine vision technologies can be utilized. These technologies enable precise
and variable-rate spraying applications tailored to the specific needs of each plant. In a recent study
focusing on citrus crops, researchers developed a cost-effective smart sensing system to control air
blast sprayers. The smart sensing system included several components, such as Light Detection and
Ranging (LiDAR), machine vision, GPS, flow meters, sensor fusion, and AI. This system was
designed to scan citrus trees and collect important information such as tree height, tree classification,
and fruit count. By analyzing this data, the system could optimize the spraying process by adjusting
the spraying nozzles accordingly. To process and control the collected data, specialized software was
developed and implemented on an embedded computer called Nvidia Jetson Xavier NX. This
software, written in C++, utilized advanced AI techniques and data fusion methods. The study results
demonstrated the effectiveness of the smart sensing system, achieving accurate estimates of tree
height with a relatively low average error rate. Additionally, a convolutional neural network (CNN)
was used for tree classification, accurately identifying mature, young, and dead trees, as well as
distinguishing them from non-tree objects. The system also employed a CNN-based fruit count
module, which accurately counted mature and immature citrus fruits.
By implementing this innovative sensing technology, farmers can substantially decrease the amount of
chemicals used for spraying compared to traditional methods where nozzles remain continuously
open. This reduction in chemical usage not only minimizes waste but also enhances the efficiency of
resource utilization in agriculture. Originally developed for citrus orchards, the smart sensing system
has the potential to be adapted for other tree crops like peaches, apples, and pecans. It provides an
effective approach to optimizing spray applications, reducing agrochemical waste, and generating
valuable data that can aid in making informed decisions, such as forecasting crop yield. To improve
the system's functionality and test its resilience, further research will be conducted in commercial
orchards, addressing challenges like tall weeds and other environmental factors. The ongoing studies
aim to refine data collection methods, enhance the integration of camera and LiDAR data, and
broaden the system’s use across various tree crops. In summary, this smart sensing system marks a
significant leap forward in agricultural technology, offering a cost-effective and efficient solution for
optimizing spraying practices, reducing waste, and bolstering decision-making processes. (Partel et al.
2021b)
(Zhang et al. 2018) highlighted the significance of Precision variable-rate spraying, a key focus in
modern agricultural practices, which tailors pesticide application based on the size of crop canopies.
This approach leads to both cost savings and reduced environmental impact. The chapter reviews
various sensing technologies employed for real-time detection and characterization of tree plantations,
including infrared, ultrasonic, Light Detection and Ranging (LiDAR), and stereo vision sensors. Laser
scanners and stereo vision systems are noted for their ability to generate 3D images and maps of
plants and canopies, with LiDAR sensors being particularly valued for their speed and accuracy in
creating detailed 3D representations. However, the practical application of LiDAR in real-time
variable-rate spraying, particularly in liner production fields, is hampered by its limited resolution and
high costs. Researchers have attempted to use LiDAR to measure canopy volume, achieving results
comparable to manual methods, yet the narrow row spacing in liner fields presents a significant
challenge, and the expense of multiple LiDAR systems hinders broader adoption. The development of
alternative variable-rate spraying technologies aims to achieve precise, low-dosage pesticide
applications to reduce environmental impact. Various methods for detecting tree canopies have been
proposed, utilizing ultrasonic, imaging, and optical systems. Infrared and ultrasonic sensors are
especially effective in assessing average canopy characteristics and tree density, thereby easing the
operator's workload and providing essential data for sprayer dosage adjustments. Among the detection
systems reviewed, laser scanners and stereo vision are identified as the most promising for capturing
detailed 3D plant and canopy information, while infrared and ultrasonic sensors continue to be useful
for specific purposes. The integration of multiple sensors holds great promise for future
advancements, with Flash LiDAR sensors, which offer real-time 3D measurements at a moderate cost,
expected to drive progress in this domain.
Camera sensors have been explored to measure plantation density, but there is still a need for
advancements to accurately capture geometric properties and adjust spraying rates accordingly. The
focus should be on developing cost-effective and user-friendly detection technologies, like LiDAR
and stereo vision, to achieve precise estimations and support sustainable pesticide application.
LiDAR, in particular, is anticipated to be instrumental in the development of trailer-mounted and self-
propelled sprayers. Progress in sensors specifically designed for geometric characterization will be
key to optimizing variable-rate spraying in agriculture, ultimately enhancing productivity and training
systems. Technological and commercial issues that have to be addressed include detection systems,
advanced software, cost reduction, and large-scale deployment. Further efforts need to be made in
creating low-cost reliable and accurate sensing systems that can support precision and sustainable
agriculture. The drawn benefits are so huge from this variable-rate spraying that to increase the
process of spraying and support the agricultural practice becomes of paramount importance to the
society and environment.
(Stajnko et al. 2012) proposed an automated orchard sprayer prototype to allow for the realization of
selective application of pesticides, with respect to certain characteristics of the target. The targets were
basically the reduction of costs and environmental load plus effectiveness in applying pesticides. The
system used ultrasound sensors, operated by microcontrollers, to measure apple tree canopy density.
The ultrasound signals would then be processed by an embedded computer for the real-time actuation
of electromagnetic valves, opening the spraying nozzles according to the canopy structure. This was
directed to happen very precisely with clear and defined ultrasound thresholds for spraying. In an
apple orchard, testing involved the use of tartrazine dye on apple leaves to measure the effectiveness
of the system. Programmable microcontrollers and electromagnetic valves reduced the number of
spray applications by up to 48.15%. Deposits on some crown positions were able to reduce the
tartrazine dye deposition by 37.7%, and in between trees by up to 65.1%. There is a possibility of
reducing usage and the environmental impacts while having similar or better leaf coverages using an
automated sprayer. Further, the study observed that ultrasound sensitivity should be optimized by
adding both cumulative and maximal response thresholds such that the value of the thresholds can be
adjusted based on different factors like orchard type, fruit species, tree age, and humidity. The
addition to this system suggested is fuzzy logic control and pulse-width modulation to further
optimize the system response for ultrasound signals and hence avoid over-spraying with increased tree
crowns. This study has successfully devised and field-tested an automated sprayer to operate in
orchards using the selective application of pesticides targeted for characteristics, thus fundamentally
reducing spray use and deposits of tartrazine dye, hence promising savings on both cost and
environmental stewardship. Further future development using advanced control techniques will
greatly enhance performance over a variety of orchard conditions.
(Y. Chen et al. 2012) applied developed advanced hyper precision spray equipment and techniques to
minimize pesticide use with tree crops. They developed an experimental variable rate air assisted
sprayer, which with the use of data acquired from a high speed laser scanning sensor, determined
volume, and the foliage density present in the tree canopy. Such a system would require a control
system designed to time-vary the output of each nozzle based on the characteristics of a target tree: its
occurrence, height, width, and foliage density. An innovative algorithm was developed that processed
the sensor data and coordinated the nozzle flow rates accordingly. The introduction of timing in the
deployment of nozzle response, following canopy interception, was captured using high-speed video
cameras to determine the coherence of spray deposit in the canopies.
Results indicated that there were no significant statistical differences regarding spray coverage, but
structures, volumes of the canopy, and foliage density differed significantly among canopies of the
two different tree types. Measurements of volume and foliage density by the laser sensor showed very
little variation due to speed. The lag time of the spraying system was determined to be 0.119 seconds,
which was deemed sufficient for timely action. However, it was observed that as the travel speed
increased from 3.2 to 6.4 km/h, spray volume, coverage, and uniformity inside canopies decreased.
The required spray outputs exceeded the nozzle capacity at higher speeds. The conclusion drawn from
the study was that the experimental air-assisted variable-rate sprayer, integrated with high-speed laser
scanning technology and a specially designed controller, demonstrated the ability to achieve variable
spray rates based on canopy characteristics. The design criteria of the sensor-controlled system were
deemed acceptable for variable-rate application, offering potential for up to 48.19% spray volume
reduction and minimizing environmental impact. However, to address the decrease in spray volume
and coverage at higher speeds, larger-capacity nozzles should be considered for future improvements.
Future research will focus on validating the algorithms for tree volume and foliage density and
evaluating the sprayer's performance for different tree crops with varying structures. This will involve
assessing spray deposition uniformity, achieving further spray volume reduction, and minimizing off-
target losses to optimize the precision spraying system.

H. EnviroWings: Unmanned Aerial Vehicles

Unmanned Aerial Vehicles (UAVs), also known as drones, sensors, sprayers, payload delivery
systems, Sample collectors, Navigation and obstacle avoidance systems are being increasingly used in
agriculture and environmental applications. In many developing countries, farmers rely on traditional
knowledge and outdated practices, leading to low productivity and resource degradation. Finally, PA
depends on the use of appropriate technology to apply treatments at the right time and in precisely the
correct place on-farm. More importantly, drones can capture high-resolution data and images. From
these images, detailed information may be garnered concerning soil and crop conditions—therefore
helping the farmer to make better decisions. Through the assessment of vegetation indices—leaf area,
for example—the health status of crops is established, which otherwise would have been stressed by
such factors as nutrient deficiency, low soil moisture, or pests. drones help farmers to quickly identify
areas under stress and apply interventions focused, much like localized fertilization and pest control.
This improves farm management, increases productivity, and reduces impacts on the environment.
Beyond this, drones have diverse applications in such fields as transport, surveillance and security,
rescue missions, and entertainment. Fitted with various sensors, drones collect complementary data
that complements data obtained from ground sensors and satellites, hence giving more detailed
analyses. According to (Avtar and Watanabe, 2020), UAVs are major tools used in agriculture and
environmental research, covering issues such as precision agriculture, crop monitoring, health of
plants, and topographic surveys. Their book represents case studies showing how UAVs can be
applied in practice and offers valuable insights to students and professionals working on UAV-related
applications. While it is not on the subject of how to design a UAV, it does offer practical advice when
applying drones in agriculture and the environment. In general, the promise that UAVs hold for
farming practices and environmental management is realized in their capacity to support data
collection, crop monitoring, and further informed decision-making, therefore leading to more efficient
and sustainable agricultural practices while furthering environmental understanding. (Raj et al. 2019)

Figure 2: Pesticide Spraying Drone (Sane et al. 2021)

(Psirofonia et al. 2017) investigated the use of drones into the agriculture, having their study site on
the island of Crete-Greece in the field of crop protection. The scope of the present study was
threefold: identifying pests and diseases on olive trees, palm trees mapping, and integration into
sensitively spraying. In the original, they tried to find a faster, more efficient method to detect the
bacterium Xlyella fastidiosa, claiming to cause the Olive Quick Decline Syndrome in olives. It was
too time-consuming to inspect the wide range of olive trees on foot, so drones, equipped with cameras
are used to draw the telltale vector signs, such as discoloration, annotating it from the rest easily and
quickly. This way, they can cordon off infected trees speedily so the plants can promptly be tested and
then uprooted to prevent further spread of the disease. In the second case, the study involved mapping
palm trees and the detection of red palm weevil infestations, a dangerous pest that often brings about
severe damage and even the death of palm plantations. They used drones to capture images of palm
trees in the early detection of infestation by weevils. Early detection allowed for targeted
interventions, like cutting down infested trees and interception with insecticides on at-risk trees,
therefore lowering general damage. In the last case, drones were integrated with electronic traps on
research for pest surveillance and targeted spraying. Pesticide application activity was based on field
populations of pests recorded through the use of electronic traps, thereby engaging drones in
precision-spraying activity. In such a manner, the use of drones in combination with electronic traps is
a source of accurate and targeting means of getting rid of the pests. For the research, they flew DJI
Phantom 3 Advanced drones equipped with high-definition cameras. In addition, the researchers used
several types of software, such as DJI Go, Pix4Dcapture for carrying out the flight mission, and data
processing. Drone Kit Python and Drone Kit-SITL were used for simulations and communicating with
drones. In a large degree of potential, drone applications in agriculture—especially in crop protection
—were pointed out in their research as well. Drones effectively helped in early pest and disease
detection and plantation mapping, which eventually led to reduced damage in targeted spraying. The
practical insights and implementation guidelines, as brought out in this study, can be useful for the
integration of drone technology in agriculture by farmers and other stakeholders. The research aimed
to extend the drones' use within agriculture and crop protection, focusing more on the olive and palm
tree farming challenges. The study was able to demonstrate the benefits and potential cost savings in
the adoption of drones in farming by integrating them with electronic traps, thus making the
technology even more accessible to smaller agricultural stakeholders.
 Figure 3: Discoloration of Image (Zhou et al. 2022)

(Kim et al. 2019) outline the application of unmanned aerial vehicles in agriculture, research on the
same applications, and emerging trends in control technology, with prospects for future development.
It tries to present an analysis of the state of UAVs in agriculture in regard to hardware setting, control
system, and practical applications. The objective of this study is to enlighten interested parties about
the limitations, applications, and technological development regarding agricultural UAVs, with a view
to further helping in guiding future research and market advancement. This study does not point to
any specific dataset, but this is entirely based on a critical review of existing literature and research
works in the area. Accordingly, the methodology adopted incorporates a literature review in quest of
an answer to a number of aspects relating to agricultural UAVs with respect to their hardware, control
mechanisms, and practical uses. The research also explores the constraints, current uses, and the
emerging trends of UAVs. The findings underline the potential and versatility that UAVs have for
agriculture, detailing their uses associated with mapping, spraying, sowing, and monitoring tasks. The
study classifies these applications and provides further valuable insight into the state of UAV
technology within the agricultural sector.
The study offers an insight into the development process of agricultural UAVs, noting their
accomplishments, challenges, and future prospects. In this study, constraints facing UAVs are
presented, areas of application—which include harvesting and precision mapping, among others—
looked into, and trends in the emerging technology analyzed. This will be very useful to researchers,
industry professionals, and stakeholders in agricultural UAVs. This paper discusses the constraints,
challenges, and potential solutions pertaining to the battery life of a UAV, management of multiple
UAVs, user interfaces, and integration of multi-robot systems in smart farming. The objective is to
draw a comprehensive picture of the current status of agricultural UAVs and stimulate discussions
about future developments. Concluding, the study underlines that, on one hand, agricultural UAVs
hold huge potential; on the other hand, they are still facing numerous challenges at their early stage. It
is the most prospective area of agriculture in terms of data collection, performance of precision tasks,
and cost-effectiveness compared to traditional methods. The contribution of this work helps to better
understand agricultural UAVs and would further guide future research, market trends, and practical
applications related to this field.
The paper by (Shamshiri et al. 2018) continues by discussing the use of UAV technologies in
precision agriculture in an oil palm plantation through the use of advanced sensing devices. This is to
realize issues concerning UAV reliability, accuracy, and timeliness. There are further challenges for
UAVs compared with ground-based systems about their payload capacity, flight duration and stability,
autonomy, and outside disturbances. This study investigates how UAVs can improve smart farming
and precision agriculture regarding the assessment of plant health; disease detection; pests
monitoring; crop yield estimation; creation of virtual plantations; and generation of dynamic maps.
The authors, in that respect, consider the improvement regarding the stability of UAV while capturing
high-resolution images at lower altitudes to have a better mapping. To this end, the researchers
developed and tested three different controllers: PID, LQR with full state feedback, and LQR with an
observer. These were designed to enhance the automatic flight missions of a fixed-wing Osprey drone
configured for crop surveillance. The results were supposed to stabilize the drone system when a
velocity step input was introduced. It was found from the simulation that the PID controller had
suppressed the effects of the step changes on the pitch of the drone but was being negatively impacted
by the pitch rate sensor noise. The controllers using LQR control resulted in less actuator dynamics,
and, in turn swerved out of control with some oscillation upon introduction of noise. Overall, the LQR
controller performed more effectively than the PID controller in noisy conditions. The study
highlighted the challenges UAVs face in precision agriculture, including platform stability, flight
dynamics, environmental factors, and regulatory issues. It highlighted the importance of addressing
the coupling between different variables during control design. The controllers' relative performance
was evaluated, emphasizing the robustness and effectiveness of the LQR controller for multi-input,
multi-output (MIMO) control.
The study emphasized the potential applications of UAVs in precision agriculture for oil palm
plantations. It acknowledged the challenges faced by UAVs and presented the design and performance
of different controllers. There are more techniques for smart spraying which are listed in Table 1.

Table 1: Advanced Techniques

Techniques Algorithm Objective Accuracy References

IntelliSprayNet Convolutional The purpose of this paper is to Recognition rate Gao et al.
Neural Network suggest a method for identifying exceeding 96%. The
(CNN) the location of a spray target to accuracy of object
2020
algorithm diminish the environmental location and
impact and waste associated with processing speed
chemical control used to manage takes less than 150 ms
American Hyphantria cunea per image.
larvae by utilizing a convolutional
neural network (CNN).
SpecieSeek YOLOv3, The primary aim is to establish a The best performing Partel et al.
Faster R-CNN, clever and autonomous herbicide network = Resnet50.
and Resnet50, sprayer mechanism that can No specific accuracy.
2019b
and lacks recognize various weed categories
specifications in real-time and use the
on the algorithm appropriate herbicide only in
utilized for certain regions, to achieve better
training and real-time target detection
testing the performance.
models.
SpecieSeek Maximum The study aims to improve ResNet-50 based Asad and
Likelihood precision farming and reduce SegNet model has
Classification herbicide impacts by developing a shown the best results
Bais 2020
and Deep weed detection system and with mean
Convolution proposing a variable-rate intersection over
Neural herbicide application based on union value of 0.8288
Networks, weed densities. and frequency
ResNet-50 weighted intersection
based SegNet over union value of
 Model 0.9869.
AgroIntellect System The paper aims to create a SMCS SMCS could save Chen et al.
dynamics and for greenhouse cultivation to 66.8% of water and
machine regulate the micro-environment energy (electricity)
2022
learning and increase agricultural
techniques, resilience while promoting
Back resource preservation, leading to
Propagation carbon-emission mitigation and
Neural Network sustainable water-energy-food
(BPNN) nexus.
Agronomatron Convolutional The paper suggests a technical Robot works 98% Kotkar 2021
Neural Network solution utilizing machine accurately
(CNN), Image learning algorithms for the
Processing identification of crop diseases and
the automated spraying of
pesticides to assist farmers in
enhancing agricultural production
by addressing the issue of pest
infestations in crops.
OptiMist Random Forest The paper aims to explore the No specific accuracy Et. al. 2021
algorithm for utilization of artificial intelligence
crop protection and language processing to
anticipate herbicides and manure
in farming, while providing an
outline of the existing state of the
art and suggesting future research
domains.

IntelliSprayNet SSD The analysis aims to create a It assumes 100% Sanchez and
MobileNetV1 computerized algorithm using precision of CNN
for developing CNN to detect crops in model for the purpose
Zhang 2022
the vision agriculture, considering factors of calculating the
module, CNN like speed and camera settings to plant detection rate.
model improve efficacy.

OptiMist No Specific The motive of this analysis is to Crop cost reduction = Zanin et al.
Algorithm Used measure the decrease in pesticide 56.16%.System based
usage on soybean and maize on real-time sensors
2022
crops by implementing a site- was accurate in
specific spraying technique that is targeting the areas
dependent on real-time sensors, that required pesticide
having the goal of maintaining application
crop yield and profitability.

SentiGrow No Specific According to the findings of this 50-80% Blackshaw et

Algorithm Used study, the weed-sensing sprayer is
most effective in managing
al. 1998
patchy weed infestations,
specifically perennial weeds
examined in the study, as well as
occasional weed growth observed
during dry fallow periods in
prairies.

CropSentinel Sensors(LiDAR The objective of this study is to ±0.05 - ±0.1(Unit=m) Hentschke et

,LADAR,RAD analyze and contrast various
AR,IMU) range finders used for altitude
al. 2018
sensing on a rotating wings UAV.

CropSentinel Drone mounted This technology reduces pesticide Around 60% Yallappa et
sprayer costs, minimizes pollution, and Efficiency on
technique improves application Mentioned Crops
al. 2017
effectiveness for small-scale
 farming communities

EnviroWings Genetic The main goal of this system is to Convergence rate= Faiçal et al.
Algorithm provide pilots with a higher level 80%- 100%
of accuracy in determining the
2017
geographical position of the
aircraft compared to other
alternatives like the Global
Positioning System (GPS).

EnviroWings No Specific The project introduces an No specific accuracy P and

Algorithm Used affordable unmanned aerial
vehicle (UAV) designed for
Gorantla
pesticide spraying in agricultural 2019
fields.

AgroIntellect LSTM (Long This paper aims to showcase the No specific accuracy Dahane et al.
Short-Term efficacy of AI techniques in
Memory facilitating efficient irrigation
2020
Recurrent decisions and optimizing water
Networks) and usage in smart farming practices.
GRU based
models.

Agronomatron Method used This paper outlines algorithms for Detection Accuracy= Nakao et al.
here is Laser path planning and control of a ± 40 mm
Range pesticide spraying robot
Finder(LRF). specifically designed for use in
greenhouses.

SentiGrow R-CNN, The paper emphasizes the need precision: 97.31%; Mahmud et
YOLO-V4 for advanced technologies in the recall: 95.70%
Dense Model nursery crop industry, stating that
al. 2023
network, their adoption benefits sustainable
YOLO-V3- production of ornamental nursery
dense crops.

Challenges
There are various challenges related to smart spraying of pesticides using Artificial Intelligence
related to security of data, privacy and trust issues. One of the challenges is the lack of advancement
in the use of communication protocols as current communication protocols can transfer data over
short range only. So, they cannot be used on large farms. Node devices used in farms transmit data
collected from surroundings operate on battery which needs to be replaced frequently. These devices
stop working when their battery is exhausted due to which farmer cannot collect data about the farm
remotely and urgent replacement is required. Another challenge in spraying pesticides is early
detection of diseased plants before disease is spread across many plants. If disease is not detected at
its initial stage, then it can reduce the harvest of the field to a great extent. (Idoje et al. 2021)
Smart spraying of pesticides reduces use of pesticides but even trace amounts of chemicals present in
crops can be harmful for health. This leads to an increase in production and consumption of organic
crops which do not require any chemicals. The EU (European Union) is also making policy favouring
imports of organic products and no tolerance of certain chemicals even in trace amounts. All these
factors combined decrease the research funding required for the improvement of smart farming. Due
to climate change, the area of agriculture fields is decreasing rapidly which eventually decreases crop
production. Now, this problem is combined with acute water shortage in many areas of globe. All
these factors combined decrease the viability of farming in open fields where surrounding
environmental factors like temperature, humidity, sunlight, etc. cannot be controlled. This leads to
vertical farming in controlled environments like greenhouses in which environmental factors
necessary for plant growth can be controlled. Another problem associated with smart spraying of
pesticides is the cost of all the equipment and technology required for implementing this. Many small
farmers having very small area of field cannot afford to buy all the equipment. Also, the advantage of
this technology is not substantial in small farms where very few crops are grown as compared to its
cost. So, it’s not a viable economic decision for the small farmers to use this technology. Another
major threat related to this technology is data security and integrity. Hacking attacks on the computers
which make decisions for spraying pesticides can lead to huge losses and all the data stored about the
farm can be exposed. Malware attacks on these computers can lead to loss of all the data stored till
date about the farm gained from various node sensors. All these security threats to the smart farming
system need to be addressed to minimize losses of the farmers. (Qazi et al. 2022b)

Future Scope
Despite the immense potential of AI in agriculture, there is still a lack of familiarity with advanced
machine learning solutions among farmers worldwide. Farmers face challenges due to external factors
such as temperature, soil conditions, and susceptibility to pest attacks. These parameters can affect the
appropriateness of crop-raising programs designed at the beginning of the season. Additionally, AI
systems need a substantial amount of data to generate accurate projections or estimates. While spatial
data can be obtained relatively quickly in rapidly growing regions, gathering temporal data proves to
be more difficult. Developing a stable AI machine learning model takes time, as the software needs to
mature. As such, AI is primarily used in agronomic products such as seeds, fertilizers, and pesticides,
rather than as precise field solutions. The future sustainability of agriculture heavily relies on the
implementation of cognitive approaches. While there are ongoing efforts to address the challenges and
demands faced by farmers using AI decision-making tools and quantitative approaches, AI
implementation in farming is still in its early stages. In order to fully harness the vast potential of AI,
agricultural frameworks need to be more reliable. (Bhardwaj et al. 1AD)
AI technology holds great potential in predicting weather patterns and various agricultural conditions
such as land quality, groundwater levels, crop cycles, and pest attacks. By providing accurate
projections and predictions, AI can address many concerns faced by farmers. The integration of AI-
driven sensors in agriculture allows for the extraction of valuable data, which can contribute to
enhancing production. While technology, including the use of pesticides, fertilizers, and machinery,
has played a role in agricultural advancements, much of the progress has been achieved through the
expansion of cultivated land, often at the expense of forests and water resources. However, in the
present scenario, a more resourceful approach is needed. AI is expected to revolutionize agriculture
and transform the market in the coming years. Drones can assist farmers in effectively spraying
pesticides and herbicides in their fields, while plant monitoring is no longer a burdensome task with
the help of AI. These advancements in technology enable farmers to optimize their pest control and
crop management practices, leading to improved agricultural outcomes. (Talaviya et al. 2020b)
The pest control industry is currently experiencing a period of turbulence characterized by declining
pest populations, changing priorities in terms of target pests, advancements in application techniques,
and the introduction of new chemical solutions. As a result, there is a pressing need for rapid
development cycles in the creation of new equipment. However, catering to the specific requirements
of the global pest control market is a challenging task, as the demand from any individual country is
insufficient to justify the costs associated with equipment development. Moreover, countries are at
various stages of pest control evolution, making it impractical to design equipment that
accommodates the unique needs of each nation. In the future, one significant change in equipment
development for pest control is expected to be the establishment of confidential agreements between
chemical companies and equipment manufacturers for collaborative efforts. This collaborative
 approach will ensure that equipment advancements keep pace with developments in chemical
solutions. It aims to prevent revolutionary chemistry from remaining confined to laboratories due to
inefficient application methods.

Conclusion and prospectives

This chapter comprehensively explored the integration of AI and IoT technologies for optimizing
pesticide spraying in agriculture. By analysing various AI algorithms and their applications through
case studies, we demonstrated the potential of these technologies to significantly reduce pesticide
usage while maintaining crop productivity. The implementation of AI-driven smart spraying systems
can contribute to a more sustainable and environmentally friendly agricultural practices. However,
challenges such as data quality, algorithm complexity, and infrastructure requirements need to be
addressed for widespread adoption. Future research should focus on developing more robust and
adaptable AI models, exploring new sensor technologies, and conducting large-scale field trials to
validate the effectiveness of these systems. By addressing these challenges and capitalizing on
emerging opportunities, we can unlock the full potential of AI and IoT in revolutionizing pesticide
management and ensuring a sustainable future for agriculture.

References

Adamides G, Katsanos C, Constantinou I, Christou G, Xenos M, Hadzilacos T, Edan Y (2017) Design and
development of a semi-autonomous agricultural vineyard sprayer: Human–robot interaction aspects.
J Field Robot 34(8):1407–1426. [Link]

Adli HK, Remli MA, Wan Salihin Wong KNS, Ismail NA, González-Briones A, Corchado JM, Mohamad
MS (2023) Recent Advancements and Challenges of AIoT Application in Smart Agriculture: A
Review. Sensors 2023, Vol 23, Page 3752 23(7):3752. [Link]

Akkem Y, Biswas SK, Varanasi A (2023) Smart farming using artificial intelligence: A review. Eng Appl
Artif Intell 120:105899. [Link]

Alam M, Alam MS, Roman M, Tufail M, Khan MU, Khan MT (2020) Real-Time Machine-Learning Based
Crop/Weed Detection and Classification for Variable-Rate Spraying in Precision Agriculture. 2020 7th
International Conference on Electrical and Electronics Engineering, ICEEE 2020 :273–280.
[Link]

Asad MH, Bais A (2020) Weed detection in canola fields using maximum likelihood classification and deep
convolutional neural network. Information Processing in Agriculture 7(4):535–545.
[Link]

Avtar R, Watanabe T (eds) (2020) Unmanned Aerial Vehicle: Applications in Agriculture and Environment.
Springer International Publishing, Cham
B P, Shridhar R, Tadpatri AS, Ugargol PP, M AHB (2021) Autonomous Herbicide Spraying System using
AI and IoT. International Journal of Engineering Research & Technology 10(9).
[Link]

Bhardwaj H, Tomar P, Sakalle A, Sharma U (1AD) Artificial Intelligence and Its Applications in
Agriculture With the Future of Smart Agriculture Techniques.
[Link] :25–
39. [Link]

Blackshaw RE, Molnar LJ, Lindwall CW (1998) Merits of a Weed-Sensing Sprayer to Control Weeds in
Conservation Fallow and Cropping Systems. Weed Sci 46(1):120–126

Borikar GP, Gharat C, Deshmukh SR (2022) Application of Drone Systems for Spraying Pesticides in
Advanced Agriculture: A Review. IOP Conf Ser Mater Sci Eng 1259(1):012015.
[Link]

Chaitanya* P, Kotte D, Srinath DrA, Kalyan KB (2020) Development of Smart Pesticide Spraying Robot.
International Journal of Recent Technology and Engineering (IJRTE) 8(5):2193–2202.
[Link]

Chen T-H, Lee M-H, Hsia I-W, Hsu C-H, Yao M-H, Chang F-J (2022) Develop a Smart Microclimate
Control System for Greenhouses through System Dynamics and Machine Learning Techniques. Water
(Basel) 14(23):3941. [Link]

Dahane A, Benameur R, Kechar B, Benyamina A (2020) An IoT Based Smart Farming System Using
Machine Learning. In: 2020 International Symposium on Networks, Computers and Communications
(ISNCC). IEEE, pp 1–6

Dhanoore RT (2022) DESIGN AND MANUFACTURING OF SMART SOLAR GRASS CUTTER AND
PESTICIDE SPRAY MACHINE. INTERANTIONAL JOURNAL OF SCIENTIFIC RESEARCH IN
ENGINEERING AND MANAGEMENT 06(05). [Link]

Dhumale NR, Bhaskar PC (2021) Smart Agricultural Robot for Spraying Pesticide with Image Processing
based Disease Classification Technique. 2021 International Conference on Emerging Smart
Computing and Informatics (ESCI) :604–609. [Link]

Et. al. PP (2021) A Survey on Machine Learning and Text Processing for Pesticides and Fertilizer
Prediction. Turkish Journal of Computer and Mathematics Education (TURCOMAT) 12(2).
[Link]
Faiçal BS, Freitas H, Gomes PH, Mano LY, Pessin G, de Carvalho ACPLF, Krishnamachari B, Ueyama J
(2017) An adaptive approach for UAV-based pesticide spraying in dynamic environments. Comput
Electron Agric 138:210–223. [Link]

Gao Y, Zhao Y, Ji Y, Zhao D, Wang C, Sun Q (2020) A Screen Location Method for Treating American
Hyphantria cunea Larvae Using Convolutional Neural Network. Math Probl Eng 2020:1–11.
[Link]

Hentschke M, Pignaton de Freitas E, Hennig C, Girardi da Veiga I (2018) Evaluation of Altitude Sensors
for a Crop Spraying Drone. Drones 2(3):25. [Link]

Honavar V Artificial Intelligence: An Overview *

Hussain N, Farooque AA, Schumann AW, McKenzie-Gopsill A, Esau T, Abbas F, Acharya B, Zaman Q
(2020) Design and development of a smart variable rate sprayer using deep learning. Remote Sens
(Basel) 12(24):1–17. [Link]

Idoje G, Dagiuklas T, Iqbal M (2021) Survey for smart farming technologies: Challenges and issues.
Computers & Electrical Engineering 92:107104.
[Link]

Katiyar S (1AD) The Use of Pesticide Management Using Artificial Intelligence. [Link]
[Link]/resolvedoi/[Link]?doi=104018/978-1-6684-5141-0.ch005 :74–94.
[Link]

Kim J, Kim S, Ju C, Son H Il (2019) Unmanned aerial vehicles in agriculture: A review of perspective of
platform, control, and applications. IEEE Access 7:105100–105115.
[Link]

Koç C (2017) Design and Development of a Low-cost UAV for Pesticide Applications

Kotkar VA (2021) An automatic pesticide sprayer to detect the crop disease using machine learning
algorithms and spraying pesticide on affected crops

Kumar Shaw K Design and Development of a Drone for Spraying Pesticides, Fertilizers and Disinfectants
Liakos KG, Busato P, Moshou D, Pearson S, Bochtis D (2018) Machine Learning in Agriculture: A Review.
Sensors 2018, Vol 18, Page 2674 18(8):2674. [Link]

Liu B, Chen C, Teng G, Tian G, Zhang G, Gao Y, Zhang L, Wu Z, Zhang J (2022a) Chitosan-based
organic/inorganic composite engineered for UV light-controlled smart pH-responsive pesticide
through in situ photo-induced generation of acid. Pest Manag Sci 78(6):2299–2308.
[Link]

Liu J, Abbas I, Noor RS (2021) Development of deep learning-based variable rate agrochemical spraying
system for targeted weeds control in strawberry crop. Agronomy 11(8).
[Link]

Liu L, Liu Y, He X, Liu W (2022b) Precision Variable-Rate Spraying Robot by Using Single 3D LIDAR in
Orchards. Agronomy 2022, Vol 12, Page 2509 12(10):2509.
[Link]

Mahmud MS, Zahid A, Das AK (2023) Sensing and Automation Technologies for Ornamental Nursery
Crop Production: Current Status and Future Prospects. Sensors 23(4):1818.
[Link]

Mahmud MS, Zahid A, He L, Martin P (2021) Opportunities and Possibilities of Developing an Advanced
Precision Spraying System for Tree Fruits. Sensors (Basel) 21(9). [Link]

Murugan K, Shankar BJ, Sumanth A, Sudharshan CV, Reddy GV (2020) Smart Automated Pesticide
Spraying Bot. 2020 3rd International Conference on Intelligent Sustainable Systems (ICISS) :864–
868. [Link]

Nakao N, Suzuki H, Kitajima T, Kuwahara A, Yasuno T Path Planning and Traveling Control for Pesticide-
Spraying Robot in Greenhouse

Nan Y, Zhang H, Zheng J, Yang K, Ge Y (2023) Low-volume precision spray for plant pest control using
profile variable rate spraying and ultrasonic detection. Front Plant Sci 13.
[Link]

Nasir FE, Tufail M, Haris M, Iqbal J, Khan S, Khan MT (2023) Precision agricultural robotic sprayer with
real-time Tobacco recognition and spraying system based on deep learning. PLoS One 18(3
MARCH). [Link]
P VSR, Gorantla SR (2019) Design and Modelling of anAffordable UAV Based Pesticide Sprayer in
Agriculture Applications. In: 2019 Fifth International Conference on Electrical Energy Systems
(ICEES). IEEE, pp 1–4

Partel V, Charan Kakarla S, Ampatzidis Y (2019a) Development and evaluation of a low-cost and smart
technology for precision weed management utilizing artificial intelligence. Comput Electron Agric
157:339–350. [Link]

Partel V, Costa L, Ampatzidis Y (2021a) Smart tree crop sprayer utilizing sensor fusion and artificial
intelligence. Comput Electron Agric 191:106556. [Link]

Partel V, Costa L, Ampatzidis Y (2021b) Smart tree crop sprayer utilizing sensor fusion and artificial
intelligence. Comput Electron Agric 191:106556. [Link]

Partel V, Kim J, Costa L, Pardalos P, Ampatzidis Y (2019b) Smart Sprayer for Precision Weed Control
Using Artificial Intelligence: Comparison of Deep Learning Frameworks

Psirofonia P, Samaritakis V, Eliopoulos P, Potamitis I (2017) Use of Unmanned Aerial Vehicles for
Agricultural Applications with Emphasis on Crop Protection: Three Novel Case - studies.
International Journal of Agricultural Science and Technology 5(1):30–39.
[Link]

Qazi S, Khawaja BA, Farooq QU (2022a) IoT-Equipped and AI-Enabled Next Generation Smart
Agriculture: A Critical Review, Current Challenges and Future Trends. IEEE Access 10:21219–21235.
[Link]

Qazi S, Khawaja BA, Farooq QU (2022b) IoT-Equipped and AI-Enabled Next Generation Smart
Agriculture: A Critical Review, Current Challenges and Future Trends. IEEE Access 10:21219–21235.
[Link]

Raj R, Kar S, Nandan R, Jagarlapudi A (2019) Precision Agriculture and Unmanned Aerial Vehicles
(UAVs). Unmanned Aerial Vehicle: Applications in Agriculture and Environment :7–23.
[Link]

Sammons PJ, Furukawa T, Bulgin A (2005) Autonomous Pesticide Spraying Robot for use in a Greenhouse

Sanchez PR, Zhang H (2022) Simulation-Aided Development of a CNN-Based Vision Module for Plant
Detection: Effect of Travel Velocity, Inferencing Speed, and Camera Configurations. Applied Sciences
12(3):1260. [Link]
Shamshiri RR, Hameed IA, Balasundram SK, Ahmad D, Weltzien C, Yamin M, Shamshiri RR, Hameed IA,
Balasundram SK, Ahmad D, Weltzien C, Yamin M (2018) Fundamental Research on Unmanned
Aerial Vehicles to Support Precision Agriculture in Oil Palm Plantations. Agricultural Robots -
Fundamentals and Applications. [Link]

SINHA JP (2020) Aerial robot for smart farming and enhancing farmers’ net benefit. The Indian Journal of
Agricultural Sciences 90(2):258–267. [Link]

Song Y, Sun H, Li M, Zhang Q (2015) Technology Application of Smart Spray in Agriculture: A Review.
Changed publisher: TSI Press 21(3):319–333. [Link]

Spoorthi S, Shadaksharappa B, Suraj S, Manasa VK (2017) Freyr drone: Pesticide/fertilizers spraying drone
- An agricultural approach. Proceedings of the 2017 2nd International Conference on Computing and
Communications Technologies, ICCCT 2017 :252–255.
[Link]

Stajnko D, Berk P, Lešnik M, Jejčič V, Lakota M, Štrancar A, Hočevar M, Rakun J (2012) Programmable
Ultrasonic Sensing System for Targeted Spraying in Orchards. Sensors (Basel) 12(11):15500.
[Link]

Tadpatri AS, Shridhar R, B M AH, Ugargol PP Autonomous Herbicide Spraying System using AI and IoT

Tadpatri AS, Shridhar R, B M AH, Ugargol PP Autonomous Herbicide Spraying System using AI and IoT

Tahmasebi M, Gohari M, Emami A (2022a) An Autonomous Pesticide Sprayer Robot with a Color-based
Vision System. International Journal of Robotics and Control Systems 2(1):115–123.
[Link]

Tahmasebi M, Gohari M, Emami A (2022b) An Autonomous Pesticide Sprayer Robot with a Color-based
Vision System. International Journal of Robotics and Control Systems 2(1):115–123.
[Link]

Talaviya T, Shah D, Patel N, Yagnik H, Shah M (2020a) Implementation of artificial intelligence in

agriculture for optimisation of irrigation and application of pesticides and herbicides. Artificial
Intelligence in Agriculture 4:58–73. [Link]
Talaviya T, Shah D, Patel N, Yagnik H, Shah M (2020b) Implementation of artificial intelligence in
agriculture for optimisation of irrigation and application of pesticides and herbicides. Artificial
Intelligence in Agriculture 4:58–73. [Link]

Taylor M (2017) Climate-smart agriculture: what is it good for?

[Link] 45(1):89–107.
[Link]

Tewari VK, Pareek CM, Lal G, Dhruw LK, Singh N (2020) Image processing based real-time variable-rate
chemical spraying system for disease control in paddy crop. Artificial Intelligence in Agriculture
4:21–30. [Link]

Unal B, Savas T, Yazar I (2020) Design of a Pesticide Spraying Quadcopter. International Journal of
Aviation Science and Technology vm01(is01):9–13. [Link]

Y. Chen, H. Zhu, H. E. Ozkan (2012) Development of a Variable-Rate Sprayer with Laser Scanning Sensor
to Synchronize Spray Outputs to Tree Structures. Trans ASABE 55(3):773–781.
[Link]

Yallappa D, Veerangouda M, Maski D, Palled V, Bheemanna M (2017) Development and evaluation of

drone mounted sprayer for pesticide applications to crops. In: 2017 IEEE Global Humanitarian
Technology Conference (GHTC). IEEE, pp 1–7

Yousaf A, Kayvanfar V, Mazzoni A, Elomri A (2023) Artificial intelligence-based decision support systems
in smart agriculture: Bibliometric analysis for operational insights and future directions. Front Sustain
Food Syst 6:1053921. [Link]

Yu J, Schumann AW, Cao Z, Sharpe SM, Boyd NS (2019) Weed Detection in Perennial Ryegrass With
Deep Learning Convolutional Neural Network. Front Plant Sci 10:483304.
[Link]

Zanin ARA, Neves DC, Teodoro LPR, da Silva Júnior CA, da Silva SP, Teodoro PE, Baio FHR (2022)
Reduction of pesticide application via real-time precision spraying. Sci Rep 12(1).
[Link]

Zhang H, Wang Z, Guo Y, Ma Y, Cao W, Chen D, Yang S, Gao R (2022) Weed Detection in Peanut Fields
Based on Machine Vision. Agriculture 2022, Vol 12, Page 1541 12(10):1541.
[Link]
Zhang Z, Wang X, Lai Q, Zhang Z, Zhang Z, Wang X, Lai Q, Zhang Z (2018) Review of Variable-Rate
Sprayer Applications Based on Real- Time Sensor Technologies. Automation in Agriculture -
Securing Food Supplies for Future Generations. [Link]

Zhao J, Tian G, Qiu C, Gu B, Zheng K, Liu Q (2022a) Weed Detection in Potato Fields Based on Improved
YOLOv4: Optimal Speed and Accuracy of Weed Detection in Potato Fields. Electronics 2022, Vol 11,
Page 3709 11(22):3709. [Link]

Zhao X, Wang X, Li C, Fu H, Yang S, Zhai C (2022b) Cabbage and Weed Identification Based on Machine
Learning and Target Spraying System Design. Front Plant Sci 13:924973.
[Link]

Zhou H, Yuan X, Zhou H, Shen H, Ma L, Sun L, Fang G, Sun H (2022) Surveillance of pine wilt disease by
high resolution satellite. J For Res (Harbin) 33(4):1401–1408. [Link]
01423-8

Zinke-Wehlmann C, Charvát K (2021) Introduction of Smart Agriculture. Big Data in Bioeconomy :187–
190. [Link]

(PDF) Review on Application of Drone in Spraying Pesticides and Fertilizers.

[Link]
g_Pesticides_and_Fertilizers. Accessed 31 Jul 2024

Index Terms
Artificial Intelligence Random forest algorithm
Computer Vision Rate spraying
Convolutional Neural Network Real time
Crop Robot
Deep Learning Smart agriculture
Drone/Drones Smart Spraying Pesticides
Farming Spray Volume
Herbicide Target Detection
Image processing Technology
Internet of Things Tiny model
Machine learning Unmanned Aerial Vehicles
Machine vision Variable Rate
Model Variable Rate Spraying
Pest control Weeds
Pesticides Yolo V4 tiny