ResNet-CNN Model for Plant Disease Classification

for e-Agriculture Applications

Arava Vamsi Kumar
PG Scholar, Department of Computer S. V. N. Sreenivasu Kandi Venkata Narasimha Reddy
Science and Engineering, Professor, Department of Computer Assistant Professor, Department of
Narasaraopeta Engineering College Science & Engineering, Narasaraopeta Computer Science & Engineering,
Narasaraopet , Andhra Pradesh, India Engineering College Narasaraopeta Engineering College
arava.vamsi224@[Link] Narasaraopet , Andhra Pradesh, India Narasaraopet , Andhra Pradesh, India
drsvnsrinivasu@[Link] narasimhareddynec03@[Link]

Abstract—The development of an economy's ability to image processing-based tools are prevalent, researchers are
innovate is directly linked to the progress of its agriculture exploring other methodologies such as natural language-
sector. The main objective of this project is to incorporate deep guided, rule-based engines. Although most apps currently
learning models in the development of a Plant Disease Detection provide suggestions to farmers on managing invasive species,
and Classification Network (PDDC-Net). In addition, the manual techniques involving agronomy specialists are still
preprocessing stage was conducted to eliminate different types necessary to expand the knowledge base [5]. Integrating
of interference, hence standardizing the images contained in the diverse data sources remains challenging, but ontologies and
dataset. Furthermore, the PDDC-Net implements the operation semantic technologies show promise in integrating data from
by utilizing a ResNet-CNN, which is a convolutional neural
various application domains. Natural language processing
network that incorporates residual networks, for the specific
tasks of extracting features and performing classification. The
methods can automatically populate ontologies with data from
recommended PDDC-Net model achieved a satisfactory Spanish unstructured sources, enabling access to information
accuracy rate in detecting and classifying plant leaf diseases, as on plant diseases and pests [6]. Through these advancements,
evidenced by the test results. The proposed model tested on a unique method of presenting plant disease symptoms based
various crop diseased leaves such as potato, tomato, and pepper on observations and the synergy of plant parts has been
with an accuracy of 99.84% suggested. Additionally, a knowledge-based agricultural pest
recognizer capable of distinguishing between overlapping
Keywords— PDDC-Net, ResNet-CNN, Plant Disease pests has been developed. The proposed architecture is easily
Detection, agricultural specialists, plant pathologists expandable to include visual evidence and other types of data,
further enhancing its capabilities. Traditional approaches [7]
I. INTRODUCTION to pest monitoring and detection in agriculture, such as manual
On a worldwide scale, India is the second-largest food monitoring, sticky traps, and black light traps, require
producer, offering employment opportunities to a wide variety significant time and rely on trained experts to identify pests.
of households. Given the industry's significant market size, it Pathogens, which are agents that cause sickness, often
is imperative to increase output levels across the board. manifest on plant leaves or stems. Accurately identifying
Approximately 67 percent of the labor force is engaged in plants, leaves, and stems, as well as detecting pests and
agriculture, indicating that nearly two-thirds of the population diseases, the frequency of pest or disease occurrences, and the
relies on the agricultural sector for sustenance. However, signs of infestation are critical components of successful crop
current farming techniques lack proper irrigation infrastructure cultivation. Both living agents (biotic agents) and non-living
and services, resulting in negative environmental impacts. substances (abiotic agents) can harm or kill plants [8]. Living
These challenges, including inadequate irrigation, crop loss agents include insects, bacteria, fungi, and viruses, while non-
due to water shortage and floods, poor seed quality, and lack living agents encompass extreme temperatures, moisture
of cold storage, can lead to failed harvests and waste of abundance, inadequate light, nutrient deficiencies,
resources [1]. The term "e-agriculture," which refers to the use unfavourable soil pH, and air pollution [9]. Deep learning [10]
of information and communication technology in agriculture, has made significant advancements in digital image
is gaining global popularity [2]. E-agriculture, or electronic processing, surpassing traditional techniques. Researchers are
agriculture, encompasses the integration of technology in increasingly concerned with how to utilize deep learning
agricultural settings. Smart farming, a subset of precision technology for plant disease and pest detection, as it holds
agriculture, involves utilizing contemporary information great potential in this field.
technology (IT) to optimize productivity and efficiency. Using The remaining parts of the paper are organized as described
global positioning systems (GPS) and geographic information in the following: The thorough analysis of the literature survey
systems (GIS) for precision farming is one example of a use of with its limitations is presented in section 2. Section 3 gives
information technology in agriculture. Other examples include the detailed analysis of proposed ResNet-CNN feature
e-learning apps for smartphones and agroadvisory services, extraction. Section 4 gives the detailed simulation analysis of
and RFID technology for product monitoring [3]. These proposed method. Further, the conclusion of the proposed is
technologies improve various aspects of agriculture, such as presented in section 5.
weather monitoring, soil preparation, sowing schedules, farm
management, soil fertility, crop nutrition, insect control, II. LITERATURE SURVEY
irrigation, drainage, and research applications [4].In the
context of diagnosing plant diseases and pests, information and Harikrishna, et al. [11] proposed Recognition of Crop
communication technology (ICT) is increasingly used. While Disease and Pesticide Suggestion using Convolution Neural
Network. The primary persistence of this investigation is to

XXX-X-XXXX-XXXX-X/XX/$XX.00 ©20XX IEEE

apply deep learning models to the process of constructing Plant information on the crop that was gathered may be included in
Disease Detection. The authors Marwaha et al. [12] suggested the data set for an experiment. After that, the Relief algorithm
that the use of artificial intelligence and machine learning may is used in order to choose the characteristics that will be
be beneficial to the sector of agriculture. The authors discuss employed in the subsequent stages. This occurs after the
the many applications of AI and ML that are now available, as previous step.
well as how these methods have been used at different phases
of the production cycle in the agricultural industry. Singh et al. III. PROPOSED METHOD
[13] devised a technique for tracking pests that makes use of Agriculture is one of the most crucial industries as it
wireless sensor networks as its foundation. Simply detecting supplies food to the world's population and plays a significant
insect activity via the use of a multitude of sensors is all that is role in the country's economy. Additionally, it provides food
required for this approach. The support vector machine (SVM) for animals, making it essential for their survival. However,
is a technique for the speedy and accurate identification and plant diseases pose a significant challenge as they can lead to
categorization of insects that is based on the top five crops and
a decrease in both the quantity and quality of agricultural
the insect pests that are associated with those crops. The
products. Modern farmers face various obstacles that hinder
approach was based on the five primary crops and the insect
pests that are associated with those crops. Lanjewar et al. [14] their ability to increase crop yields, such as fluctuating
proposed the implementation of a real-time disease prediction temperatures and a large population of insects. To boost yields,
system. This system would make use of a CNN that would be reducing the number of destructive insects in the environment
stored in the cloud using a Platform-as-a-Service (PaaS) is vital. Significant investments are made globally to ensure
infrastructure. When the CNN model was put into action to crop safety, agricultural product quality, and robust output.
generate a prediction about the sickness that affects tea leaves, Protecting crops from bio-aggressors like pests and insects is
it achieved a remarkable accuracy of one hundred percent crucial to avoid substantial crop loss and damage. In India, pest
across the training, validation, and test datasets. This was an infestations cause substantial agricultural productivity losses,
impressive accomplishment. estimated at ninety quadrillion rupees and accounting for
around eighteen percent of the country's total agricultural
Shaikh et al. [15] suggested using machine learning and
output.
artificial intelligence to get optimal results. using information
and communication technologies in one's daily life. Deep learning models, which can learn complicated
Specifically, to conventional methods of farming is the subject characteristics from the pictures that are fed into the system,
of this study. The logistic regression classifier (LRC) are responsible for the high accuracy rate that is achieved by
addresses the advantages and disadvantages of using these the system. It is possible for agricultural experts to diagnose
technologies, as well as the possibilities and challenges that and cure plant illnesses before they cause substantial crop
might arise from putting them into practice. Kumar et al. [16] losses by using the PDDC-Net system, which has the
proposed the use of the TensorFlow machine learning capability of correctly and effectively detecting plant
technique for the purpose of cotton disease detection. The diseases. Figure 1 shows the proposed ResNet-CNN block
prediction model is built with the help of the decision tree diagram. The crop disease datasets are then pre-processed
classifier (DTC) algorithm, which is accessed using before being input into ResNet-CNN for the aim of feature
TensorFlow's Keras application programming interface (API). extraction. This step is performed in order to improve
This model is then employed in the production of mobile apps accuracy. Photos of leaves, on the other hand, go through the
that aid farmers in recognizing cotton illnesses and offer same pre-processing step before being uploaded to ResNet
advice about the pesticides that may be used to battle such CNN for analysis. The leaf images, the crop disease datasets,
diseases. These programs may be downloaded from the and the learnt features that have previously been trained with
Google Play Store or the Apple App Store. the plant diseases are compared with one another and
Shaikh et al. [17] presented their idea for how machine compared to the leaf images. Both the accuracy and the
learning may be used in intelligent agriculture and precision calculation of the recovered attributes are reduced to some
farming. They focused on the possible applications of degree because of this process. The comparison graph has the
information and communication technologies in traditional capacity to provide predictions about the many kinds of plant
agricultural processes. It also discusses the issues that may diseases. The PDDC-Net uses a residual network based
occur as a result of adopting these technologies. A convolutional neural network, also known as a ResNet-CNN,
representation of an agricultural knowledge mining strategy to aid in the process of feature extraction and categorization.
that pulls information from many ontologies was shown by According to the findings of the research, the PDDC-Net
Murali et al. [18]. This technique mines information for model was successful in achieving a high degree of accuracy
agricultural purposes. As one of its many capabilities, The when it came to the detection and categorization of plant leaf
purpose of this system is to provide a graphical representation diseases. This was proved by the fact that the model was
of a number of different knowledge management systems.
successful in meeting the goals set forth by the trials.
Data imagining is a relatively new method that has emerged in
A. Image Preprocessing
recent years as an approach to the representation of data.
Yadav and his colleagues [19] presented a proposal titled It is essential to pre-process pictures and standardize
"Disruptive innovations in smart farming: an extended features, such as image size and colour, which may result in
perspective using sentiment analysis." The provision of an increase in the total amount of datasets. Image size and
instruments for various agricultural and farming operations colour are only two examples. This is done to ensure that the
under the general heading of "Smart Farming" has the criteria of model training, such as picture size, are satisfied,
overarching objective of increasing productivity while and to reduce noise throughout the process of gathering
simultaneously cutting expenses, eliminating waste, and images. Parameters, such as image size, must meet these
limiting the amount of work that is required. Gupta et al. [20] requirements. The bulk of the time, image pre-processing
developed an approach for the application of machine learning processes such as resizing, padding resizing, and letterboxing
to be applied to the forecasting of agricultural output. The are some of the procedures that are used. Other procedures
include picture vector normalization and data augmentation maps can be computed from them. Each neuron in the layer
approaches. Image denoising, flipping, blur operations, color that is now active in the feature map is linked to the adjacent
enhancement, edge detection enhancement, logarithmic neurons in the layer that is active.
transformation, and image denoising are some of the most
popular methods that are used in the process of altering
images. Other techniques that may be utilized include edge
detection enhancement and logarithmic transformation.

Fig. 2. ResNet-CNN architechure.

This process is referred to as the receptive field of the

neuron, which is the phrase that is used to describe it.
Additionally, the input is jumbled up by mixing the element-
wise nonlinear activation function with the learned kernel in
order to build a new feature map. This is done in order to
produce a new feature map. In order to generate a new feature
vector, this step has to be taken. This step must be completed
before continuing on to the next stage, which will build a new
Fig. 1. Block diagram of proposed system.
feature map. In order to fulfill the objective of producing a new
The aforementioned procedures were used in this study to feature map, this step has to be completed in its entirety. It is
change the gathered images in order to increase the number of essential to infer a variety of distinct kernels in order to provide
datasets, prevent the model from being overfit, guarantee the comprehensive feature maps. This is required. For example,
availability of the datasets, and speed up the process of training the feature value that corresponds to the coordinates (I, J) on
the model. The target's coordinate position is the co-ordinate feature map K may be figured out by applying Equation (3) to
information that is kept in the label data. This information is that layer of the map. This allows for the map to be broken
𝐿
represented by the numbers [𝑥𝑚𝑖𝑛 , 𝑥𝑚𝑎𝑥 , 𝑦𝑚𝑖𝑛 , 𝑦𝑚𝑎𝑥 ], which down into its constituent parts. The symbol 𝑆𝐼,𝐽,𝐾 may be used
stand for the X and Y coordinate values of the target frame's to refer to the feature map that was produced as a consequence.
upper left corner and lower right corner, respectively. The
𝑇
following information is included in the data for the YOLO 𝐿
𝑆𝐼,𝐽,𝐾 = 𝑊𝐾𝐿 × 𝐹𝐼,𝐽
𝐿
+ 𝐵𝑎𝑖𝑠𝐾𝐿 ()
label: [𝑐𝑙𝑎𝑠𝑠, 𝑋𝑐𝑒𝑛𝑡𝑒𝑟 , 𝑌𝑐𝑒𝑛𝑡𝑒𝑟 , 𝑤𝑖𝑑𝑡ℎ, ℎ𝑒𝑖𝑔ℎ𝑡], which
represent the target box's category, the centre point's X and Y In this context, the weight vector 𝑊𝐾𝐿 𝑇 and the bias vector
coordinates, as well as the target frame's width and height, 𝐵𝑎𝑖𝑠𝐾𝐿 relate to the feature value that resides at the location
respectively. The equation that describes the process of (I, J) in layer L of feature map 𝐾. This feature value may be
converting from the VOC format to the YOLO format is in accessed by clicking on the 𝐵𝑎𝑖𝑠𝐾𝐿 node. The letter L is used to
Equation (1). indicate both of these specific sites. In addition, the feature that
𝐿
was extracted is referred to as 𝐹𝐼,𝐽 , and it is this feature that is
𝑥 + 𝑥𝑚𝑖𝑛 = 2𝑥__𝑐𝑒𝑛𝑡𝑒𝑟
{ 𝑚𝑖𝑛 () input into CNN at the location (I, J) in the 𝐿𝑡ℎ layer. This layer
𝑥𝑚𝑖𝑛 − 𝑥𝑚𝑖𝑛 = 𝜔 ∗ 𝑤𝑖𝑑𝑡ℎ
is the one that corresponds to the Kth map in the training set.
In addition to this, the weight function W is fine-tuned with the
After everything has been completed, it is possible to find
help of a new CSUBW model. This is done with the intention
the second equation:
of lowering the loss (error) while simultaneously recognizing
1 whether or not the subject is sick. The component of the CNN
𝑥𝑚𝑖𝑛 = 𝑥𝑐𝑒𝑛𝑡𝑒𝑟 − ∗ 𝜔 ∗ 𝑤𝑖𝑑𝑡ℎ known as the "activation function" is the one that is in charge
{ 2
1 () of bringing the nonlinearity into the process. The nonlinear
𝑥𝑚𝑖𝑛 = 𝑥𝑐𝑒𝑛𝑡𝑒𝑟 + ∗ 𝜔 ∗ 𝑤𝑖𝑑𝑡ℎ
2 activation function is represented by the symbol (. ) , which is
𝐿
a period. When calculating the activation function 𝐴𝐹𝐼,𝐽,𝐾 for
where 𝑥𝑚𝑖𝑛 and 𝑥𝑚𝑎𝑥 are the X-coordinate values that 𝐿
correspond, respectively, with respect to the target frame's top the convolutional feature 𝑆𝐼,𝐽,𝐾 , Equation (4) is the formula
left and lower right corners. These values may be found by that is used.
using the variable x.
𝐿
𝐴𝐹𝐼,𝐽,𝐾 𝐿
= 𝐴𝐹(𝑆𝐼,𝐽,𝐾 ) ()
B. ResNet-CNN Classifier
The collected features are fed into a CNN that has been The three activation functions that see the most common
trained to determine whether or not diseases that affect mango use are the sigmoid, the tanh, and the ReLU. It is feasible to
leaf are present. This CNN has been tweaked to make this achieve shift-invariance in the maps even if the resolution of
determination. The “convolutional,” “pooling,” and “fully the feature maps is decreased. The poling layer is denoted by
connected layers” make form the foundational framework of the symbol of a dot (.). The output of CNN is denoted by the
CNN as shown in Figure 2. The feature representations for the notation 𝑂(𝐼,𝐽,𝐾)𝐿 , and its mathematical representation can be
inputs are taught to be learned in order to be learned in the found in Equation (5). In this instance, 𝑅𝑖,𝑗 determines the
convolutional layer, which is made of numerous convolution neighborhood that is located close by and is centered on the
kernels. This is done in order for the feature representations to location (I, J). The output layer is the very last layer of the
be learned. Within the neural network, this layer may be found CNN, and it is immediately followed by additional layers that
in the center of the structure. These convolution kernels are are fully connected to one another.
applied to the data several times before the various feature
 𝐿
𝑂𝐼,𝐽,𝐾 𝐿
= 𝑝𝑜𝑜𝑙(𝐴𝐹𝑜,𝑝,𝐾 ), ∀(𝑜, 𝑝) ∈ ℛ𝐼,𝐽 () convolutions on the input image, extracting features through
filters or kernels. The spatial dimensions of the feature maps
There is a count of N for the number of inputs–output are shrunk thanks to the MAXPOOLING layer, but the
essential details are preserved in their entirety. The FLATTEN
relations (𝐹𝑛 , 𝑂𝑛 ; 𝑛 = [1, . . . , 𝑁]) . In this case, the letter N
layer reformats the feature maps such that they are in the form
stands for the nth group of input data, while the letter o
of a one-dimensional vector. This gets them ready for input
indicates the desired label (the presence or absence of a into a layer that is completely linked. Because every neuron in
disease in a mango leaf). The output of CNN is represented the preceding layer is linked to every neuron in the current
by the symbol 𝑦𝑛 . Equation (6) is used in CNN to compute the layer through the DENSE layer, which is also referred to as a
loss function, which must be reduced as much as possible. In completely connected layer, it is possible to learn intricate non-
the current body of work, the loss function is made as small linear correlations. Following convolutional and fully
as possible by adjusting the parameters of CNN via the use of connected layers is a frequent place to implement the ReLU
a recently established hybrid optimization model known as activation function. It does this by introducing non-linearity
CSUBW. Equation (7) illustrates this point further. Figure 4 into the network, which gives it the ability to learn complicated
depicts the output of the solution that was provided to the patterns and boosts the performance of the model.
CSUBW model as an input.
Figure 5 shows the predicted outcomes of different plant
1
leaf images. Here, the plant disease name is indicated by green
𝐿𝑜𝑠𝑠 = ∑𝑁 𝑛 𝑛
𝑛=1(𝜁, 𝑂 , 𝑦 ) () colour, and pesticide suggestion is highlighted by red colour.
𝑛
Figure 6 shows the accuracy and loss comparison graph of
𝑂𝑏𝑗 = min(𝐿𝑜𝑠𝑠) () PDDC-Net for 10 epochs. The increasing accuracy suggests
that the model is becoming more accurate in its predictions,
while the decreasing loss indicates that the model's predictions
IV. RESULTS AND DISCUSSION are getting closer to the true values.
This section gives the detailed analysis of simulation
results, which are compared to existing methods using same
dataset.

Fig. 4. ResNet-CNN layers.

Fig. 3. Sample dataset with healthy and plant diseases.

A. Dataset
The Plant Disease Image Dataset (PDID) is a specialized
dataset that primarily focuses on leaf diseases as shown in (a)
Figure 3. It contains a collection of high-quality images
showcasing both healthy and diseased leaves caused by
various pathogens. The PDID serves as a valuable resource for
researchers and practitioners interested in studying rice
diseases and developing models for disease detection in plants.
By utilizing this dataset, scientists can enhance their
understanding of diseases and explore innovative methods for
identifying and managing them.
B. Performance evaluation
Figure 4 shows the ResNet-CNN architecture layers pro.
The ResNet-CNN architecture typically consists of several
repeating blocks, each containing convolutional layers, (b)
activation functions (such as ReLU), and shortcut connections.
These shortcut connections allow the network to learn residual
mappings, enabling better gradient flow and alleviating the
vanishing gradient problem. The CONV2D layer performs
 proposed PDDC-Net method surpasses DTC [16] with a
substantial improvement of 8.15%. The F1-Measure of the
proposed PDDC-Net method demonstrates a significant
improvement of 2.36% compared to DTC [16].
Table 1. Performance evaluation
Method Accuracy Precision Recall F1-Measure
SVM [13] 80.0 92.77 85.76 88.86
LRC [15] 86.66 95.00 89.61 93.41
DTC [16] 96.66 93.33 90.0 96.42

(c) Proposed PDDC-Net 99.84 99.3 98.15 99.001

V. CONCLUSION
In conclusion, the ResNet-CNN architecture combined
with crop disease identification that the PDDC-Net presented
achieves remarkable performance in terms of accuracy,
precision, recall, and F1-Measure. Compared to the baseline
methods SVM, LRC, and DTC, PDDC-Net demonstrates
significant improvements in all metrics. It outperforms SVM
in accuracy by 19.8%, LRC in accuracy by 13.18%, and DTC
in accuracy by 3.34%. The precision, recall, and F1-Measure
of PDDC-Net also exhibit notable improvements compared to
(d) the baseline methods. These results highlight the effectiveness
Fig. 5. Sanple prediction on test data. of the PDDC-Net approach in accurately detecting and
recognizing crop diseases, making it a promising model for
crop disease identification and management. Integrating the
PDDC-Net model with real-time monitoring systems and
deploying it in smart agriculture settings. This would enable
farmers to receive instant notifications and take prompt actions
to mitigate crop diseases, thereby minimizing yield losses.
REFERENCES
[1] El Mane, Adil, et al. "Agriculture Supply Chain Management Based on
Blockchain Architecture and Smart Contracts." Applied Computational
Intelligence and Soft Computing 2022 (2022).
[2] Chen, Ximing, and Jie Shang. "Analysis of Farmers’ Willingness to Use
Blockchain and Influencing Factors Based on the Binary Logit Model."
Wireless Communications and Mobile Computing 2022 (2022).
[3] Tsolakis, Naoum, et al. "Sustainability in the digital farming era: A
cyber-physical analysis approach for drone applications in agriculture
Fig. 6. Accuracy and loss comparision graph. 4.0." Information and Communication Technologies for Agriculture—
Theme IV: Actions. Cham: Springer International Publishing, 2022. 29-
Table 1 compares the performance evaluation of various 53.
plant disease detection classifiers. The proposed PDDC-Net [4] Rathod, Meenakshi L., et al. "Cloud Computing and Networking for
SmartFarm AgriTech." Journal of Nanomaterials 2022 (2022).
method demonstrates a significant improvement of 19.8% in
[5] Maraveas, C., et al. "Applications of IoT for optimized greenhouse
accuracy compared to SVM [13]. The precision of the environment and resources management." Computers and Electronics
proposed PDDC-Net method outperforms SVM [13] with an in Agriculture 198 (2022): 106993.
improvement of 6.53%. The recall achieved by the proposed [6] Debauche, Olivier, et al. "Cloud and distributed architectures for data
PDDC-Net method shows a remarkable improvement of management in agriculture 4.0: Review and future trends." Journal of
14.24% compared to SVM [13]. The F1-Measure of the King Saud University-Computer and Information Sciences 34.9 (2022):
7494-7514.
proposed PDDC-Net method exhibits a substantial
[7] Habeeb, Rimsha, et al. "A Proposed Comparative Algorithm for
improvement of 10.19% compared to SVM [13]. Regional Crop Yield Assessment: An Application of Characteristic
Objects Method." Mathematical Problems in Engineering 2022 (2022).
Compared to LRC [15], the proposed PDDC-Net method
[8] Dahane, Amine, Rabaie Benameur, and Bouabdellah Kechar. "An IoT
achieves a notable improvement of 13.18% in accuracy. The low-cost smart farming for enhancing irrigation efficiency of
precision of the proposed PDDC-Net method surpasses LRC smallholders farmers." Wireless Personal Communications (2022): 1-
[15] with an improvement of 4.3%. The recall achieved by the 38.
proposed PDDC-Net method outperforms LRC [15] with a [9] Yang, He, et al. "Deep Learning Agricultural Information Classification
significant improvement of 8.54%. The F1-Measure of the Combined with Internet of Things Technology in Agricultural
Production and Economic Management." IEEE Access 10 (2022):
proposed PDDC-Net method demonstrates a substantial 54713-54719.
improvement of 5.59% compared to LRC [15]. The proposed [10] Thomas, Rosemary J., Gregory O'Hare, and David Coyle.
PDDC-Net method exhibits an impressive improvement of "Understanding technology acceptance in smart agriculture: A
3.34% in accuracy compared to DTC [16]. Compared to DTC systematic review of empirical research in crop production."
Technological Forecasting and Social Change 189 (2023): 122374.
[16], the precision of the proposed PDDC-Net method shows
[11] Harikrishna, A., et al. "Recognition of Crop Disease and Pesticide
a notable improvement of 5.97%. The recall achieved by the Suggestion using Convolution Neural Network." Turkish Journal of
 Computer and Mathematics Education (TURCOMAT) 14.2 (2023): [16] Kumar, Sandeep, Rajeev Ratan, and J. V. Desai. "Cotton Disease
394-403. Detection Using TensorFlow Machine Learning Technique." Advances
[12] Marwaha, Sudeep, et al. "Application of Artificial Intelligence and in Multimedia 2022 (2022).
Machine Learning in Agriculture." Translating Physiological Tools to [17] Shaikh, Tawseef Ayoub, et al. "Machine Learning for Smart
Augment Crop Breeding. Singapore: Springer Nature Singapore, 2023. Agriculture and Precision Farming: Towards Making the Fields Talk."
441-457. Archives of Computational Methods in Engineering 29.7 (2022): 4557-
[13] Singh, Kamred Udham, et al. "An artificial neural network-based pest 4597.
identification and control in smart agriculture using wireless sensor [18] Murali, E., and S. Margret Anouncia. "Visualization of Multiple
networks." Journal of Food Quality 2022 (2022). Ontology Agro Knowledge Mining Model." International Journal of
[14] Lanjewar, Madhusudan G., and Kamini G. Panchbhai. "Convolutional Reliability, Quality and Safety Engineering 29.05 (2022): 2241001.
neural network based tea leaf disease prediction system on smart phone [19] Yadav, Sargam, et al. "Disruptive technologies in smart farming: an
using paas cloud." Neural Computing and Applications 35.3 (2023): expanded view with sentiment analysis." AgriEngineering 4.2 (2022):
2755-2771. 424-460.
[15] Shaikh, Tawseef Ayoub, Tabasum Rasool, and Faisal Rasheed Lone. [20] Gupta, Sandeep, et al. "Machine learning-and feature selection-enabled
"Towards leveraging the role of machine learning and artificial framework for accurate crop yield prediction." Journal of Food Quality
intelligence in precision agriculture and smart farming." Computers and 2022 (2022): 1-7.
Electronics in Agriculture 198 (2022): 107119.