2023 9th International Conference on Advanced Computing and Communication Systems (ICACCS)

Smart Farming System using NPK Sensor

Bharadwaj Cheruvu Dr. S. Bhargavi Latha

Computer Science Department Computer Science Department
Gokaraju Rangaraju Institute of Engineering and Gokaraju Rangaraju Institute of Engineering and
Technology Technology
Hyderabad, India Hyderabad, India
cheruvubob@[Link] [Link]@[Link]

Mada Nikhil Hitesh Mahajan

Computer Science Department Computer Science Department
Gokaraju Rangaraju Institute of Engineering and Gokaraju Rangaraju Institute of Engineering and
Technology Technology
Hyderabad, India Hyderabad, India
madanikhil143@[Link] hiteshmahajan774@[Link]

Kongari Prashanth
Computer Science Department
Gokaraju Rangaraju Institute of Engineering and Technology
Hyderabad, India
kongariprashanth23@[Link]

Abstract—The human population is only growing, reserves; we face a hunger-stricken world in the
and certain steps are to be implemented to meet the years to come.
future requirements with respect to food. This paper
discusses the implementation of a smart farm using In India, an estimated 60 percent of the land is
Internet of Thing. IoT has led to a faster form of used for agricultural purposes with majority of the
gathering data and inferring from our surroundings. population considering it their primary source of
A farmer, with the help of this smart farm system, will livelihood[1]. It’s also believed that India will be the
be able to keep track of plant and soil vitals in real- most populous state in 2023 and will continue to see
time and use the recommendation system, based on a an increase in its numbers and simultaneously parts
model trained from a dataset, to suggest the best of the existing agricultural land will be converted
suitable crop based on sensor values. IoT-enabled into amenities and residencies for the growing
Smart Farming will enable growers and farmers to population. With other factors like increasing Global
enhance productivity and reduce the wastage of Warming, adverse climate change, excessive use of
resources. The proposed system is a more reliable fertilizers and chemicals, Natural Calamities, etc. it
concept and can be easily implemented as it consists of is becoming crucial to address the prime agricultural
sensors that can collect vital information about the concern: What will we do when the demand
environment from soil nutrients, temperature,
surpasses the supply? The advent of technological
humidity, and soil moisture regularly which is
advances in the agricultural field has given some
displayed on an easy-to-understand interface to be
interpreted by the growers and farmers to understand
spite.
the best conditions to give their plants. The term "Internet of Things" refers to a broad
network of hardware and software, including
Keywords—Internet of Things, smart farming, IoT sensors, microcontrollers, connections, user
sensors, NPK sensor, soil measurement, machine interfaces, and data processing capabilities, that
learning, python API, Thingspeak server
enables communication between hardware and the
I. INTRODUCTION cloud. IoT started off with modest beginning being
part of Coca Cola machines and a toaster connected
The United Nations Food and Agriculture to the Internet. With technological advent in the 21st
Organization estimates that an additional 75% of the century, IoT has been at the forefront of major
production of agricultural produce is needed to technological research in domains such as Smart
sustain the world by 2050. With shrinking Cities in [2], Smart Homes in [3], [4], Smart Energy
agricultural lands, slowing yield trends in major in [5], Autonomous Vehicles in [6], [7], Healthcare
crops, water shortages, and depleting mineral
 2023 9th International Conference on Advanced Computing and Communication Systems (ICACCS)

in [8], Logistics in [9] and Smart Agriculture in [10] 1) NodeMCU, Figure 1, is an open-sourced
to name a few. software and hardware component built on the
In the next section, we highlight the recent System-on-chip system called ESP8266.
advances with respect to IoT applications in the ESP8266, created by Espressif Systems, includes
sector of agriculture. central processing unit (CPU), network
capabilities, random access memory (RAM), an
II. LITERATURE STUDY operating system (OS) and system development
There have been several projects and research kit (SDK). It comes with 16 GPIOs and 1 analog
done using various technologies and Internet of input pin. However only 10 pins can be utilized
Things has been one promising technology. IoT for digital input and output. Its operating voltage
based solutions focused on reduces human is 3.3V with an input voltage ranging from 4.5V
involvement. IoT based agricultural solutions are
to 10V.
classifies into Monitoring, Tracking and tracing, and
Smart Precision Farming, Livestock Management 2) The NPK sensor, depicted in Figure 2 is a sensor
and Greenhouse Production solutions as mentioned used to measure the content of Nitrogen,
in [10], [11]. Phosphorous and Potassium in milligram per
liter. This has been an area of study in
Some of the studies so far have dealt with Smart agricultural growth techniques recently as
Farm mentioned in [12] provides the concept of
mentioned in [17], [18]. The measurements of
traceability systems by collecting data from sensors
and to monitor these in real-time, capabilities of such precision sensors will give an
Wireless Sensor Network (WSN) in [13], [14], a understanding on the fertility of the soil to the
monitoring system for transmitting sensor data using user. The sensor is resistant to long-term
the Wi-Fi technology to a database [15]. An android electrolysis, corrosion resistance and
application to suggest crops based on a trained waterproof. It runs on a power supply of 5-30V
model was explained in [16]. DC. Its protection grade is 68 and has a
Farmers tend to rely on their basic observations resolution of 1mg/L. The measurement is done by
and conventional approaches when it comes to detecting the conductivity transformation caused
making decisions on what crop to plant without by different nitrogen, phosphorous and
considering the nutrients or environment effect on potassium concentrations. Since the output
the crop. A machine learning based recommendation signal of NPK sensor is RS485 and this signal
system will help the farmer make an informed cannot be interpreted by ESP8266, we must
decision and the real-time monitoring system will make use of the MAX485 convertor module.
help to keep a track of the crop conditions.
There is a need to build a user-friendly interface
to communicate the model result based on the sensor
values and to display the values in real-time for
monitoring. The inclusion of a precision IoT
component as the Nitrogen-Phosphorous-Potassium
(NPK) sensor to measure the primary nutrients in the
soil has been done as Nitrogen, Phosphorous and
Potassium are the three primary components Figure. 1. NodeMCU ESP8266 Wi-Fi Module
required for good crop growth. A fairly accurate
model has to be trained to recommend the suitable
crop depending on the dataset and its attributes.
In this research, we studied the major measurable
factors deciding plant growth and how it can be
monitored by the farmer in real-time to take an
appropriate action. Our research involves a Plant
recommender based on sensor values and a
continuous monitoring system that can be accessible Figure. [Link] Sensors
from the Python interface.
III. PROPOSED APPROACH
A. Components
In the IoT setup, we have used Nitrogen-
Phosphorous-Potassium sensor, Digital
Temperature and Humidity (DHT22) sensor, soil
moisture sensor, ESP NodeMCU Wi-Fi module and
MAX485 TTL to RS485 Module. Figure. 3. MAX485 TTL to RS485 Convertor
 2023 9th International Conference on Advanced Computing and Communication Systems (ICACCS)

technique known as Random Forest Classifier. It

uses multiple predictions from models and hence is
an ensemble technique. The feature importance is a
computed factor that signifies how well a certain
selection of attributes affects the model.
Consider the importance of node j as explained
Figure. 4. DHT22 Sensor in (1).

3) The MAX485 module, Figure 3, converts the TTL imk = wt I - wtlt k Ilt k - wtrt(k) Irt(k) (1)
signals from NPK sensor to RS485 readings that
can be interpreted by the ESP Module. It imk = importance of node k.
converts transistor-transistor logic (TTL) into wt k = weighted samples number that reach node k.
RS485 to transmit high data rate prone
differential communication as mentioned in [19]. Ik = impurity value for k node.
4) The DHT22 Sensor is a low-cost sensor having a lt(k) = child node on left split of node k.
capacitive humidity sensor and a thermistor.
DHT22, Figure 4, has a range 0-100% humidity rt(k) = child node on right split of node k.
with 2-5% accuracy and negative 40-80 ºC with Whereas the importance for an individual feature
an accuracy of ±0.5ºC[20]. It has 4 pins is calculated from (2).
including ground, data, VCC and NC.
∑n:node j splits on feature k imj
B. Dataset fik = ∑m∈all nodes imm
(2)
To train a ML model, we should have sufficient
entries for each class and only those parameters that fik = importance of the feature k.
truly have a significant effect on the plants condition = importance of the node j.
are taken into consideration before training. We
have trained our model on the After which the data is scaled between 0 and 1
Crop_Recommendation dataset available on Kaggle using (3).
which is based on the report from Indian Council of
Agricultural Research. Most crops require 3 fik
Nfik = (3)
essential nutrients for growth, namely, Nitrogen, ∑j ∈ all features fij
Phosphorous and Potassium. Hence, we have used a
dataset having these parameters. The final feature importance is the average over
all the trees represented by (4).
It has 2200 entries of 22 different plant species
with parameters of Nitrogen content, Potassium ∑j ∈all trees Nfij
Fk = (4)
content, Phosphorous content, temperature, T
humidity, pH of the soil and rainfall as depicted by Fk = importance of feature k computed from all
Table 1. We ensure that the dataset does not have trees.
any null values or missing data that can hinder the
training of the model. The dataset is split into a 75:25 Nfij = normalized feature importance for k in tree j.
split, where 75 percent of the data is part of training
T = number of trees generated.
samples and the rest for testing.
The reason for using Random Forest i.e., the
TABLE I. DATASET STRUCTURE accuracy comparison is expressed in Table 2. It is
also advantageous due to its diversity in each model
Parameter Count Data type
and stability due to majority voting.
Nitrogen
2200 Int64
content TABLE II. COMPARISON BETWEEN CLASSIFIERS
Phosphorous
2200 Int64
content Classifier Accuracy (%)
Potassium
2200 Int64 Logistic Regression 88.000
content
Temperature 2200 Float64 K Neighbours 91.090

Humidity 2200 Float64 Decision tree 94.090

Label 2200 Object Gradient Bossting 94.545

C. Random Forest Classifier Random Forest 95.090

Regression and classification problems are
addressed by the supervised machine learning
 2023 9th International Conference on Advanced Computing and Communication Systems (ICACCS)

D. Hardware Organization Using the Thingspeak Application Programming

The sensors that are implemented are NPK Interface (API) keys, we can transfer readings from
sensor for measuring nutrient content, DHT22 for the sensors to their respective fields on the
temperature and humidity and a soil moisture sensor. Thingspeak platform[24], [25] as depicted in figures
7-12. Data on the Thingspeak Server is represented
The sensors will connect to the NodeMCU in the form of graphs with suitable limits and units.
ESP8266 microcontroller Wi-Fi module. The block The data is updated as and when a new value is read
and schematic diagram are given in Figure 5 and by the sensor. Each field indicates a specific sensor
Figure 6 respectively. Block Diagram shows the value that it visualizes, and the field numbers are set
general arrangement of principal parts or functions in the Arduino Sketch.
to show the relationship between them. The system
will require MAX485 module to integrate the NPK The Python based user interface allows the
with NodeMCU whereas the other sensors can be farmer to continuously monitor and recommend the
directly connected. suitable crop based on the conditions at any time.
The interface was created using FLASK, that was
used for similar cause in [26], Thingspeak API keys
to transfer the data and the trained classifier on the
dataset. FLASK is a community-maintained micro
web framework. The user interface is depicted in
Figure 13.

Figure. 5. Block diagram for system. Figure. 7. Thingspeak - Nitrogen content graph

Figure. 8. Thingspeak - Phosphorous content graph.

Figure. 9. Thingspeak - Potassium content graph.

Figure. 6. Block diagram showcasing connectivity and
functionality of components.

E. Data Analytical Tool and User Interface

Initially the data is transmitted to IoT analytical
platform called Thingspeak. MQ Telemetry
Transport (MQTT), a publish/subscribe
communication method that utilizes TCP/IP sockets
or WebSockets, is used to update sensor values.
explained in [21]–[23]. Figure. 10. Thingspeak - Moisture content graph.
 2023 9th International Conference on Advanced Computing and Communication Systems (ICACCS)

The number of correctly predicted positive

outcomes is called precision.
TP
Precision= (4)
TP+FP

Recall or sensitivity is the proportion of correctly

predicted positive cases among all the positive
examples in the data that the classifier produced.
TP
Figure. 11. Thingspeak - Temperature graph. Recall= (5)
TP+FN

The F1-Score is measured as the harmonic mean

of the two. The following shows its calculation.
Precision × Recall
F1 Score=2 × (6)
Precision + Recall

TABLE III. CLASSIFICATION REPORT

Figure. 10. Thingspeak - Humidity graph.
Label Precision Recall F1-Score

Apple 1.00 1.00 1.00

Banana 1.00 1.00 1.00

Black gram 0.81 0.85 0.83

Chickpea 1.00 1.00 1.00

Cocunut 1.00 1.00 1.00

Coffee 1.00 1.00 1.00

Cotton 1.00 1.00 1.00

Figure. 13. User Interface for the monitoring and
Grapes 1.00 1.00 1.00
recommendation system created by Python and Flask.
Jute 0.81 0.89 0.85
IV. RESULTS
The distribution of data samples is made evenly Kidney beans 1.00 1.00 1.00
to train the model better. 25 percent of each labels
data sample is used in validation of the model and Lentils 0.72 0.76 0.74
the results for each label along with its precision,
Maize 1.00 1.00 1.00
recall, and f1-score are given in the classification
report expressed in table 3. The overall classification Mango 1.00 1.00 1.00
report is given in table 4 using weighted parameter.
The metrics used to evaluate the model are explained Moth bean 0.93 0.90 0.91
below based on the confusion matrix in Figure 14. A
confusion matrix will result in 4 terms. Mung bean 1.00 1.00 1.00
True Positives (TP) is the number of positive
cases correctly predicted. True Negatives (TN) is the Muskmelon 0.96 1.00 0.98
number of negative cases correctly predicted. False
Positives (FP) is the number of negative cases that Orange 1.00 1.00 1.00
are misclassified. False Negatives (FN) is the
Papaya 1.00 1.00 1.00
number of positive cases that are misclassified. The
mentioned equations are explained in [27]. Pigeon peas 0.88 0.85 0.87
The percentage of accurate results among all
Pomegranate 1.00 1.00 1.00
predictions is known as accuracy.
Rice 0.88 0.79 0.84
TP+TN
Accuracy= (3)
TP+TN+FP+FN
Watermelon 1.00 0.96 0.98
 2023 9th International Conference on Advanced Computing and Communication Systems (ICACCS)

TABLE IV. MODEL CLASSIFICATION REPORT parameters can be considered that have an impact on
the growth of plant. Microcontrollers allow easy
Parameter Percentage (%) addition of sensors and simple codes can implement
them.
Accuray 95.090
One can even implement a Global system for
mobile communication (GSM) module to improve
Precision 95.593 connectivity better. Additional improvements will
make the system more versatile and improve
efficiency.
Recall 95.454
VI. CONCLUSION
This paper has presented an IoT-based
recommendation and continuous monitoring system.
The different hardware components, the integration
between the components, dataset used for training,
model training, model accuracy and implementation
of a API using Thingspeak and Flask were
discussed. The Wi-Fi module results in lesser delay
and easier connectivity. Advancements like these
will help farmers by recommending, monitoring,
and thereby helping them grow better and lead to
sustainable agriculture.

Acknowledgment
This research has been conducted in the
supervision of faculty and aid from staff of Gokaraju
Rangaraju Institute of Engineering and Technology.

References
Figure. 14. Confusion matrix of model. [1] M. Pyingkodi et al., “Sensor Based Smart Agriculture
with IoT Technologies: A Review,” in 2022
V. DISCUSSION International Conference on Computer Communication
and Informatics (ICCCI), Jan. 2022, pp. 1–7. doi:
Random Forest classifier worked well with the 10.1109/ICCCI54379.2022.9741001.
dataset giving us a highly accurate model for over 22 [2] N. Mishra, P. Singhal, and S. Kundu, “Application of
different plant species. Random forest Classifier IoT products in smart cities of India,” in Proceedings of
decides from a set of decision trees making it the 2020 9th International Conference on System
Modeling and Advancement in Research Trends,
efficient. The web interface can be easily accessed SMART 2020, Dec. 2020, pp. 155–157. doi:
and rather simple to use. It offers a custom 10.1109/SMART50582.2020.9337150.
recommender and continuous monitoring option [3] D. Shin, K. Yun, J. Kim, P. V. Astillo, J. N. Kim, and
with recommendation of plant species to grow based I. You, “A Security Protocol for Route Optimization in
on last available data point that is obtained from the DMM-Based Smart Home IoT Networks,” IEEE
sensors. As it is connected to the Internet, the Access, vol. 7, pp. 142531–142550, 2019, doi:
10.1109/ACCESS.2019.2943929.
transfer of data to the analytical tool, Thingspeak [4] Institute of Electrical and Electronics Engineers, 2019
and user interface happens within a fraction of a 4th International Conference on Internet of Things:
second. The recommendation system is not only Smart Innovation and Usages (IoT-SIU).
accurate but quickly responsive too. [5] Universiti Tunku Abdul Rahman and Institute of
Electrical and Electronics Engineers, ICSGCE 2018 :
The IoT-enabled smart farm will help the farmer 2018 International Conference on Smart Grid and
and grower to make decisions based on soil Clean Energy Technologies : 29 May - 1 June, 2018,
conditions quickly and accurately as compared to Tunku Abdul Rahman, Sungai Long Campus, Kajang,
Malaysia.
that of a soil testing. The continuous monitoring will [6] S. Kuutti, S. Fallah, K. Katsaros, M. Dianati, F.
keep the farmer aware of the crop’s condition and Mccullough, and A. Mouzakitis, “A Survey of the
can aid in timely response. State-of-the-Art Localization Techniques and Their
Potentials for Autonomous Vehicle Applications,”
In the future, more data samples can be included IEEE Internet of Things Journal, vol. 5, no. 2. Institute
to the dataset which currently has 2200 data sample of Electrical and Electronics Engineers Inc., pp. 829–
points for 22 plant species. Similarly other plant 846, Apr. 01, 2018. doi: 10.1109/JIOT.2018.2812300.
species can be included to widen the scope of the [7] Amity University, Institute of Electrical and
Electronics Engineers. Uttar Pradesh Section, Institute
system. of Electrical and Electronics Engineers, and Amity
With time, next generation sensors can replace University. School of Engineering and Technology,
2019 6th International Conference on Signal
existing sensors to make the system more accurate Processing and Integrated Networks (SPIN) : 7-8
and sensitive to the parameters. Additional
 2023 9th International Conference on Advanced Computing and Communication Systems (ICACCS)

March 2019, Amity School of Engineering and [18] M. Masrie, A. Z. M. Rosli, R. Sam, Z. Janin, and M. K.
Technology, Noida, India. Nordin, “Integrated optical sensor for NPK Nutrient of
[8] M. N. Bhuiyan, M. M. Rahman, M. M. Billah, and D. Soil detection,” in 2018 IEEE 5th International
Saha, “Internet of Things (IoT): A Review of Its Conference on Smart Instrumentation, Measurement
Enabling Technologies in Healthcare Applications, and Application (ICSIMA), Nov. 2018, pp. 1–4. doi:
Standards Protocols, Security, and Market 10.1109/ICSIMA.2018.8688794.
Opportunities,” IEEE Internet of Things Journal, vol. 8, [19] Z. Xing, D. Li, Y. Shan, Y. Jin, and J. Shao, “The
no. 13. Institute of Electrical and Electronics Engineers Design and Implementation of a C8051F-Based
Inc., pp. 10474–10498, Jul. 01, 2021. doi: Embedded Image Acquisition System,” in 2010 2nd
10.1109/JIOT.2021.3062630. International Workshop on Intelligent Systems and
[9] Y. Song, F. R. Yu, L. Zhou, X. Yang, and Z. He, Applications, May 2010, pp. 1–3. doi:
“Applications of the Internet of Things (IoT) in Smart 10.1109/IWISA.2010.5473493.
Logistics: A Comprehensive Survey,” IEEE Internet of [20] Y. A. Ahmad, T. Surya Gunawan, H. Mansor, B. A.
Things Journal, vol. 8, no. 6. Institute of Electrical and Hamida, A. Fikri Hishamudin, and F. Arifin, “On the
Electronics Engineers Inc., pp. 4250–4274, Mar. 15, Evaluation of DHT22 Temperature Sensor for IoT
2021. doi: 10.1109/JIOT.2020.3034385. Application,” in 2021 8th International Conference on
[10] M. S. Farooq, S. Riaz, A. Abid, K. Abid, and M. A. Computer and Communication Engineering (ICCCE),
Naeem, “A Survey on the Role of IoT in Agriculture Jun. 2021, pp. 131–134. doi:
for the Implementation of Smart Farming,” IEEE 10.1109/ICCCE50029.2021.9467147.
Access, vol. 7. Institute of Electrical and Electronics [21] D. Parida, A. Behera, J. K. Naik, S. Pattanaik, and R. S.
Engineers Inc., pp. 156237–156271, 2019. doi: Nanda, “Real-time Environment Monitoring System
10.1109/ACCESS.2019.2949703. using ESP8266 and ThingSpeak on Internet of Things
[11] V. K. Quy et al., “IoT-Enabled Smart Agriculture: Platform,” 2019 International Conference on Intelligent
Architecture, Applications, and Challenges,” Applied Computing and Control Systems (ICCS), pp. 225–229,
Sciences (Switzerland), vol. 12, no. 7. MDPI, Apr. 01, May 2019, doi: 10.1109/ICCS45141.2019.9065451.
2022. doi: 10.3390/app12073396. [22] H. Benyezza, M. Bouhedda, K. Djellout, and A. Saidi,
[12] K. Wongpatikaseree, P. Kanka, and A. Ratikan, “Smart Irrigation System Based Thingspeak and
“Developing Smart Farm and Traceability System for Arduino,” in 2018 International Conference on Applied
Agricultural Products using IoT Technology,” in 2018 Smart Systems (ICASS), Nov. 2018, pp. 1–4. doi:
IEEE/ACIS 17th International Conference on 10.1109/ICASS.2018.8651993.
Computer and Information Science (ICIS), Jun. 2018, [23] L. Mhatre and N. Rai, “Integration between wireless
pp. 180–184. doi: 10.1109/ICIS.2018.8466479. sensor and cloud,” in 2017 International Conference on
[13] N. Ahmed, D. De, and I. Hussain, “Internet of Things I-SMAC (IoT in Social, Mobile, Analytics and Cloud)
(IoT) for Smart Precision Agriculture and Farming in (I-SMAC), Feb. 2017, pp. 779–782. doi: 10.1109/I-
Rural Areas,” IEEE Internet Things J, vol. 5, no. 6, pp. SMAC.2017.8058285.
4890–4899, Dec. 2018, doi: [24] A. Akhtar, T. Ahmad, N. Sabahat, and S. Minhas, “IoT
10.1109/JIOT.2018.2879579. Based Home Automation System Using ThingSpeak,”
[14] T. M. Bandara, W. Mudiyanselage, and M. Raza, in 2019 International Conference on Computing,
“Smart farm and monitoring system for measuring the Electronics & Communications Engineering
Environmental condition using wireless sensor network (iCCECE), Aug. 2019, pp. 163–168. doi:
- IOT Technology in farming,” in 2020 5th 10.1109/iCCECE46942.2019.8941737.
International Conference on Innovative Technologies [25] L. Mhatre and N. Rai, “Integration between wireless
in Intelligent Systems and Industrial Applications sensor and cloud,” in 2017 International Conference on
(CITISIA), Nov. 2020, pp. 1–7. doi: I-SMAC (IoT in Social, Mobile, Analytics and Cloud)
10.1109/CITISIA50690.2020.9371830. (I-SMAC), Feb. 2017, pp. 779–782. doi: 10.1109/I-
[15] W.-L. Chen et al., “AgriTalk: IoT for Precision Soil SMAC.2017.8058285.
Farming of Turmeric Cultivation,” IEEE Internet [26] A. Singh, R. Akash, and G. R. V, “Flower Classifier
Things J, vol. 6, no. 3, pp. 5209–5223, Jun. 2019, doi: Web App Using Ml & Flask Web Framework,” in
10.1109/JIOT.2019.2899128. 2022 2nd International Conference on Advance
[16] A. Sharma, M. Bhargava, and A. V. Khanna, “AI-Farm: Computing and Innovative Technologies in
A crop recommendation system,” in 2021 International Engineering (ICACITE), Apr. 2022, pp. 974–977. doi:
Conference on Advances in Computing and 10.1109/ICACITE53722.2022.9823577.
Communications (ICACC), Oct. 2021, pp. 1–7. doi: [27] J. Almulla and T. Alam, “Machine Learning Models
10.1109/ICACC-202152719.2021.9708104. Reveal Key Performance Metrics of Football Players to
[17] W. Fan, K. A. Kam, H. Zhao, P. J. Culligan, and I. Win Matches in Qatar Stars League,” IEEE Access,
Kymissis, “An Optical Soil Sensor for NPK Nutrient vol. 8, pp. 213695–213705, 2020, doi:
Detection in Smart Cities,” in 2022 18th International 10.1109/ACCESS.2020.3038601.
Conference on Intelligent Environments (IE), Jun.
2022, pp. 1–4. doi: 10.1109/IE54923.2022.9826759.