Combating Agricultural Challenges with Secure Digital Farming

Cheikhou Akhmed Kane

1 a

and Pascal Francois Faye

2 b

Universit

e Rose Dieng France-S

egal, Dakar, Senegal

Department of Mathematics and Computer Science, Universit

e du Sine Saloum El Hadj Ibrahima NIASS, Kaolack, Senegal

Keywords:

Digital Farming, Machine Learning, Distributed Control.

Abstract:

This paper introduces Secure Digital Farming, a comprehensive approach to enhancing farm security and op-

timizing crop yield. SDF addresses critical challenges faced by modern agriculture, including climate change,

pest control, rural crime, and demographic pressures, all of which threaten agricultural perimeters and impact

yield. Our SDF-based solution leverages deep-learning algorithms to analyze sensor data and video streams

from security cameras, enabling intelligent access control, pest detection, and yield estimation. This paper

outlines the implementation framework for SDF, highlighting its feasibility for real-life testing and validation.

We plan to conduct ﬁeld tests on our educational farm in the peanut basin of Senegal to evaluate the efﬁcacy

and practicality of SDF in a real-world setting.

1 INTRODUCTION

Global agricultural yields are increasingly compro-

mised by climate change, input taxes, government

restrictions without farmer consensus, and various

security issues within farming environments. Fig-

ure 1 illustrates a comparison between a set of se-

curity issues and the corresponding security solu-

tions necessary to balance performance and social

welfare in farming. Additionally, farmers face ar-

duous labor, high equipment acquisition costs, risks

associated with pests (insects, wild animals, domes-

tic animals, etc.), rising rural crime rates, and demo-

graphic expansion that threatens agricultural perime-

ters. This balance is delicate, as attackers—including

humans, animals, insects, bacteria, nematodes, and

termites—consistently devise methods to circumvent

security measures.

Farmers typically secure their farms based on re-

current risks in their area using techniques familiar to

aggressors. Therefore, it is crucial to categorize all

potential risks, even the least likely ones, to develop

comprehensive solutions. However, relying on real-

time monitoring by guards or watchdogs has physio-

logical limitations such as sleep, hunger, and exhaus-

tion, which can hinder effective performance. A well-

trained distributed artiﬁcial intelligence (AI) system

can process large volumes of data and make better

https://orcid.org/0009-0000-9341-2466

https://orcid.org/0000-0002-2078-5891

decisions without experiencing fatigue or biased judg-

ment, offering a more reliable alternative for securing

farms.

Figure 1: Farm’s security solutions versus threats that af-

fects farm effectiveness.

Fig. 2 below shows different ﬁeld in order to en-

sure a secured digital farming. Despite technological

advances, the management of agricultural perimeters

remains a persistent challenge, particularly in regions

like the peanut basin of Senegal. In this study, we

advocate for the implementation of an artiﬁcial intel-

ligence (AI) system to address critical security issues

in agricultural ﬁelds. The integration of AI has proven

beneﬁcial and sustainable across various sectors, in-

cluding arboricultural farming.

An arboricultural farm is deﬁned by its capacity to

produce fruit, operate independently day-to-day, and

meet speciﬁed thresholds in both area and fruit pro-

duction. The suitability of soil types, such as clay-

sandy, airy, supple, fertile, and cool, varies across

376

Kane, C. and Faye, P.

Combating Agricultural Challenges with Secure Digital Farming.

DOI: 10.5220/0013011800003822

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 21st International Conference on Informatics in Control, Automation and Robotics (ICINCO 2024) - Volume 1, pages 376-383

ISBN: 978-989-758-717-7; ISSN: 2184-2809

Figure 2: The main components for secure agricultural

spaces with the contribution of digital.

the Senegalese peanut basin, inﬂuenced further by the

challenges posed by climate change. Each type of soil

presents unique advantages and challenges. Arbori-

cultural environments are complex due to the size of

trees, their foliage, and the diverse activities occur-

ring within them, such as pest monitoring and poten-

tial intrusions by animals or humans. Therefore, aid-

ing farmers in identifying these activities, situations,

or speciﬁc trees on their farms is crucial. This identi-

ﬁcation process spans various stages of tree growth

and fruit development, with particular emphasis on

pre-harvest planning. AI is particularly suitable for

this task due to its capability to analyze large volumes

of data efﬁciently, identify patterns, and provide pre-

dictive insights, thereby enabling farmers to optimize

resource allocation and anticipate yield outcomes ef-

fectively.

The objective of this work is to develop an ap-

plication employing various machine learning tech-

niques, such as deep learning and K-nearest neigh-

bors, to enhance farm management and security. This

application will be capable of detecting and recogniz-

ing activities, machinery, and agronomic equipment

on farms. Additionally, it will monitor the farm envi-

ronment, assist farmers in predicting yields, and help

anticipate attacks or intrusions. Our proposed ap-

proach focuses on securing access to different cate-

gories of crops and trees, thereby preserving yield.

2 RELATED WORK

(Sellam and Poovammal, 2010) investigates the im-

pact of environmental factors, such as annual rainfall

and cultivated area, on crop yield using multivariate

regression analysis over a 10-year period. The study

ﬁnds a strong correlation between these factors and

yield, as indicated by R

value. While the authors ac-

knowledge the potential inﬂuence of economic factors

(e.g., minimum support price, cost price index, etc..),

they do not consider crop security aspects like pest

or intrusion detection. (Zhang et al., 2010) compares

the accuracy of linear regression and a spatial au-

toregressive model for predicting corn yield in Iowa.

The study demonstrates the superiority of the autore-

gressive model, which accounts for spatial autocor-

relation not captured by OLS, leading to improved

predictions. While highlighting the importance of

NDVI (Normalized Difference Vegetation Index) and

precipitation as predictors, the study does not incor-

porate other factors like soil health or pest infesta-

tions. (Zingade et al., 2018) introduces an Android

application and website that leverage machine learn-

ing to recommend the most proﬁtable crops based on

current weather, soil, and environmental conditions.

This system aids farmers in crop selection for long-

term proﬁtability but overlooks crop security consid-

erations. (Sun et al., 2022b) focuses on a different

aspect of crop management, proposing an improved

density peak clustering algorithm for RGB images,

incorporating depth information to locate and rec-

ognize target fruit (green apples) for harvesting or

yield estimation. However, this approach also ne-

glects security factors. (Feng et al., 2022a) focuses on

disease identiﬁcation in crops using image process-

ing techniques, employing morphological operations,

contrast stretching, and image scaling for preprocess-

ing, followed by circle-ﬁtting-based segmentation of

leaf lesions. Classiﬁcation is then performed using

SVM and random forest models trained on LBP his-

togram features. While this approach addresses dis-

ease detection, it does not consider broader security

aspects. (Fu et al., 2022) introduces a rapeseed dataset

and a target-dependent neural architecture search for

analysis. Addressing the challenges of data loss and

misrepresentation in smart agriculture, (Cheng et al.,

2022) proposes an anomaly detection model for mul-

tidimensional time series data. (Uyeh et al., 2022)

employs a multi-objective machine learning approach

to optimize sensor placement in a protected cultiva-

tion system, utilizing a gradient boosting model with

observed and derived environmental variables. (Maia

et al., 2022) analyzes sensor data from Australian cot-

ton ﬁelds, revealing a correlation between soil matric

Combating Agricultural Challenges with Secure Digital Farming

377

potential and satellite-derived cumulative crop evapo-

transpiration, but does not provide yield comparisons.

(Feng et al., 2022b) focuses on disrupting the mat-

ing of brown planthoppers (BPH), a signiﬁcant rice

pest. They develop a system to record, monitor, and

playback BPH courtship vibrations, identifying key

frequencies for potential disruption strategies. (Sun

et al., 2022a) tackles the problem of monitoring mi-

gratory rice pests by developing an intelligent system.

This system utilizes a searchlight trap to capture in-

sects, which are then automatically identiﬁed using

computer vision, providing real-time monitoring data.

These studies highlight various aspects of smart agri-

culture, but a holistic approach that integrates farm

security considerations into yield estimation has not

been fully addressed. In this work, our goal is to use a

set of AI algorithms for the detection and the recogni-

tion of objects (person, animals, truck, car, etc.), trees,

and fruits in a farm for safety and harvest prediction

in a farm environment.

3 SYSTEM DESCRIPTION

Knowledge on farm security and its effects on the

various sectors of the national economy is a major

challenge for the country’s policymakers. Various ini-

tiatives are therefore being developed to better iden-

tify the implications of risks in the agricultural sec-

tor. However, while a causal relationship has clearly

been established between the vulnerability of the agri-

cultural sectors and a set of risks on crops, livestock,

machinery, etc. Supporting farmers to better manage

the risks associated with security is a major necessity.

All economic activities which promote food security

and suitable agriculture must incorporate the risks of

farm management into their planning. The aims of

this work are proposing :

1. An intrusion detection system that automates the

intrusion detection process.

2. An intrusion prevention system that can detect

and also attempt to stop possible incidents.

3. An network topology for a secure digital farming

This by using a full mesh sensor network, a set of

IOT card for local and remote control. In order to

improve decision making we store information in a

local database server and on a ThingSpeak platform.

In this way, a farmer can make decisions based on a

real-time view of his or her farm. This work combines

several Machine learning algorithms (ML) like:

1. Deep Learning (see Fig 3) ;

2. K-Nearest Neighbours (see Fig 4).

This combination delivers a solution that addresses

well the dynamism and uncertainty challenges tar-

geted in this work.

3.1 Preliminaries

In this part, we will give a set of analytic view for un-

derstanding of our solution. This work is based on AI

algorithms like Deep Learning and K-Nearest Neigh-

bours. The Deep Learning is a subcategory of neural

networks. Indeed, it is a set of tools and methods of

machine learning based on the use of neural networks.

We use the words Deep in reference to the number of

layers of neurons that make up these networks: the

greater the number of layers the deeper the network

and the more it allows to treat complex learning prob-

lems, but the harder it is to train.

Figure 3: Visual neural network with 4 layers.

In these mathematical architectures, each neuron

performs simple calculations but the input data passes

through several layers of computation before produc-

ing an output. The results of the ﬁrst layer of neurons

serve as input to the calculation of the next layer and

so on. It is possible to switch on different parameters

of the network architecture, like: the number of lay-

ers, the type of each layer, the number of neurons that

make up each layer. A Deep Learning algorithm can

be summarised by the steps:

• Forward propagation

• Cost function

• Backward propagation

• Gradient descendent

Further information about Deep Learning can be

found in (Schmidhuber, 2015).

The K-Nearest Neighbours (kNN) algorithm

is an ML algorithm that belongs to the class of

ICINCO 2024 - 21st International Conference on Informatics in Control, Automation and Robotics

378

unsupervised learning algorithms that can be used

to solve classiﬁcation and regression problems. In

unsupervised learning, an algorithm receives a data

set that is tagged with corresponding output values on

which it will be able to train and deﬁne a prediction

model. This algorithm can subsequently be used on

new data in order to predict their output values.

Figure 4: Visual cluster assignment.

The main steps of the KNN algorithm are:

• Step 1: Find the appropriate number of K, by us-

ing Elbow method (Umargono et al., 2020);

• Step 2: Select the number K of neighbours;

• Step 3: Calculate

Euclidian distance

∑

i=1

−Y

| (1)

or Manhattan distance;

∑

i=1

−Y

)

(2)

• Step 4: Take the K-Nearest Neighbours according

to the calculated distance.

• Step 5: Count among these K neighbours, the

number of points belonging to each cluster.

• Step 6: Assign the new point to the category most

present among these K neighbours.

• Step 7: Our model will be ready when all points

are in a cluster

3.2 Main Steps of Our Mechanism

In this section, we present the operational architec-

ture and working principles of our mechanism as il-

lustrated in Figures 5 and 6, respectively.

Figure 5: Visual illustration of our mechanism.

Figure 6: Main steps when a human is detected.

4 IMPLEMENTATION

In this section, we explain the steps we follow for im-

plementation and tests in a real-life environment. Our

environment is one hectare with sides of the same

size. We started with the virtual grid to identify the

number of optimal access points we will need to cover

our ﬁeld. This, to ensure that, there is no white

space (without Wi-Fi coverage) that can prevent sen-

sors from transmitting their data. We have used access

points with 20-60 meters of reach for a mesh network

that seamlessly connects machinery, sensors, etc. en-

suring uninterrupted harvesting and other agriculture

operations by establishing a self-conﬁguring network.

Thus, the maximum diameter is D = 120 meters as

shown in Fig 8.

• Access point ray coverage

20 meters < r < 60 meters (3)

Combating Agricultural Challenges with Secure Digital Farming

379

Figure 7: Access point : 6W 4G solar router;WiFi re-

peater;4G router solar powered all in one;IP66 Water-

proof;Customized according to different regions.

Figure 8: Virtual grid to identify the required access point

required for 1 hectare. This to set up a sensor network for

data collection. This proposition for 1 hectare is a sampling

in order to show how to cover a farm.

• We can have a inside square

d =

+ r

= r

√

2 (4)

• On one hectare we will have

X =

1 hectare

(5)

In this work, we assume that, due to trees and other

farm’s component which block the signal, the bet-

ter coverage distance we can have is r = 35 meters.

we need X =

1 hectare

= 4 access points in our farm’s

mesh network with an overlapping signals range (see

Fig. 9) to prevent a sensor from being without a Wi-

Fi signal. This implementation produce a set of data

Figure 9: Overlapping signals range to avoid white space.

collected through an IOT network which can be rep-

resented by the Fig 10.

The data are collected via:

• sensors : ultrasonic, infra-red, thermal, pressure,

RGB, UV, rain, temperature, accelerometer, PIR

motion, ultrasonic transducer, obstacle detector,

vibration detectors.

• cards : ESP 8266, ESP CAM, ARDUINO UNO,

servomotors to control equipment (access, sensor,

fence, pump, tractor, irrigation, etc.), to have vi-

sual data, sound data and create a mesh network

Figure 10: Data network of our Secure Digital Farming En-

vironments.

of ﬁeld sensors that will serve as a medium for

transmitting data from sensors but also user con-

trols.

– The different cards are connected to the access

point networks to transmit their data. The data

is sent simultaneously on the ThingSpeak plat-

form and on our local server via https (Hyper-

text Transfer Protocol Secure) and SQL (Struc-

tured Query Language) requests.

In our local server, we have a web applica-

tion that implements our SDF algorithm and

creates data visualization to support decision

making.

ThingSpeak (MathWorks, 2024) is a cloud

IoT analytics platform service that allows to

aggregate, visualize, and analyze live data

streams. We send data to ThingSpeak from

our devices, create instant visualization of live

data, and send alerts. We could set up more

features such as turning on a motor when the

water level in the water tank drops below a

speciﬁed limit er even remotely control de-

vices, such as battery-operated door locks.

· Send sensor data privately to our cloud ac-

count.

· Analyze and visualize data with MATLAB.

· Trigger a reaction after a threshold value or a

queue up command for a device to execute.

• Viewing data or launching commands can be done

via smart-phone or computer.

ICINCO 2024 - 21st International Conference on Informatics in Control, Automation and Robotics

380

5 RESULTS

In this section, we give a set of theorems to prove the

reliability and feasibility of our system. As we use a

set of IOT device, we deﬁne a sensor σ ∈ Σ which

is constrained by the parameters: {Rσ

,ϑ

Net

Rσ

is its resource(s). A view ϑ

is the set of

sensors in its neighborhood with whom it can directly

communicate at time t. L

Net

deﬁnes the dependence

level between the received data in a given sensors

network (Net).

Corollary 1. For each Access Point (AP) coverage, it

exists an inside square.

Proof. Let AP be an access point centered at A =

(x,y) with ray coverage r. Consider a square with

center A and side length d = r. The distance from A

to any vertex is

√

. Since

√

< r, the vertices of the

square lie within the circle.

This proves our corollary.

Theorem 1. ∀ S as a farm’ size, it exists a ﬁnite

number X of access points (AP) for sensor data

transmission.

Proof. Let r an access point (AP) ray coverage and

S a farm’ size. Thus, ∀ AP coverage ∃ (d ∗d) as a

square include in S = πr

((d ∗d) ⊂ S/d is the side of

the square (cf. corollary 1).

This proves our theorem, ∀S,∃X =

∈N

Lemma 1. ∀ S as a farm’ size, a sensor belongs at

least in one AP’ signal range.

Proof. Let σ

∈ Σ be a sensor in this network de-

ployed in a farm of size S. Due to our overlapping

AP topology signal range, it exists at least one AP

which can allow data transmission. In addition, ∀ AP

it transmits data with MIMO (Multiple Input Multiple

Output) strategy and may allows ﬁfty (50) simultane-

ous sensor data transmission.

Theorem 2. ∀ S as a farm’ size, if an intrusion

happen, it exists at least one sensor which can detect

the event.

Proof. Let σ

∈ Σ a sensor in this network deployed

in a farm of size S. Due to our overlapping AP topol-

ogy signal range and MIMO strategy, there exists at

least one AP which can allow data transmission for

Cap (cf. Lemma 1). Therefore, if an intrusion hap-

pens anywhere within the farm, it will be within the

detection range of at least one sensor. Thus, for any

farm size S, if an intrusion happens, there exists at

least one sensor which can detect the event.

Lemma 2. The message costs are bounded and poly-

nomial for each sensor σ until the convergence of net-

work.

Proof. Let

Msg be the message cost of our sensor net-

work and σ

∈Σ be a sensor in this network. For each

∈ Σ, if it probes its neighborhood, it sends 2 mes-

sages (one for the probe and another for the acknowl-

edgment from the receiver). Additionally, σ

may

receive messages from its neighboring sensors. The

number of these neighbors is denoted by Card(ϑ

where ϑ

represents the set of neighbors at time t.

Thus, 2 ≤

Msg ≤ Card(ϑ

). For each σ

∈ Σ, if a

network layer L

Net

exists, σ

sends 1 message to its

non-zero neighbor set ϑ

̸= 0. Additionally, σ

must

send 1 response message after receiving input from its

neighbor sensors in ϑ

. Thus, initially,

Msg = 2 for

the probe and acknowledgment. If additional infor-

mation is received after detection, an extra message is

needed, making

Msg = 3. Since the number of sen-

sors is limited in any neighborhood ϑ

, the value of

Card(ϑ

) is bounded. Therefore, the message cost

Msg is not exponential but polynomial in the number

of sensors. This proves our lemma.

Lemma 3. ∀ sensor σ

, the energy cost is optimized.

Proof. Let

Msg be the message cost of our sensor net-

work and σ

∈ Σ be a sensor in this network.

Msg is

bounded and polynomial (cf. Lemma 2). σ

∈ Σ con-

sumes energy during data transmission, i.e the energy

cost is proportional to communication, the number of

messages sent. Therefore, it is also bounded and poly-

nomial. This ensures controlled and optimized energy

consumption. This proves our lemma.

Theorem 3. For each decision-making, it is the

better and rational we can do without human inter-

actions.

Proof. Let σ

be a sensor in the network. Our over-

lapping AP topology and MIMO strategy ensure reli-

able data transmission. All intrusions are detected by

at least one sensor σ

(cf. Theorem 2). This guaran-

tees that sensors have access to complete and accurate

data for decision-making. Since the network’s com-

munication framework ensures data completeness and

reliability, sensors can make well-informed and ratio-

nal decisions autonomously. The bounded and poly-

nomial nature of the message cost

Msg and optimized

energy use further support this capability by ensuring

Combating Agricultural Challenges with Secure Digital Farming

381

efﬁcient communication. Thus, the decision-making

process within the network is both optimized and ra-

tional, operating effectively without human interac-

tions. This proves our theorem.

Lemma 4. The near and most reliable sensors of the

neighbourhood ϑ

of the sensor σ

which initiate the

network have always more ability to be selected as a

data relay.

Proof. Let σ

be the initiating sensor in the network,

and let ϑ

denote its neighborhood view. Consider

two random sensors σ

and σ

from ϑ

. Let d(σ

,σ

)

and d(σ

,σ

) be the distances from σ

to σ

and σ

respectively. Let ρ(σ

) and ρ(σ

) represent the re-

liabilities of σ

and σ

, respectively. The reliability

ρ(σ

) of a sensor σ

is computed using the Poisson

Law (Yates and Goodman, 2005). Suppose the num-

ber of times a sensor fails to transmit data follows a

Poisson distribution with an average rate of failure oc-

currences λ per unit time. The number of failures N

in a time interval t is given by:

P(N

= k) =

(λ)

−λ

where λ is the expected number of failures in time t.

The reliability ρ(σ

) is then deﬁned as:

ρ(σ

) = 1 −

Number of failures

Total number of

opportunities for transmission

which, in terms of the Poisson parameter λ , be-

comes:

ρ(σ

) = 1 −

Total number of opportunities

Assume d(σ

,σ

) < d(σ

,σ

), indicating σ

closer to σ

than σ

, and ρ(σ

) ≥ ρ(σ

), implying σ

has equal or higher reliability compared to σ

Since the reliability ρ(σ

) is inversely related to

the number of failures, sensors with a lower rate of

failures (higher ρ) are more reliable. The probability

of a sensor being selected as a data relay is positively

correlated with its reliability and effective communi-

cation role. Thus, sensors that are both closer to σ

and have a higher reliability ρ are more likely to be

selected as data relays compared to those further away

or with lower reliability. Therefore, sensors closer to

and with higher ρ are more likely to be selected as

data relays.

This proves our lemma.

Theorem 4. The protocol SDF, is auto-stabilizing if

a full connected network is reached.

Proof. Let Σ be the set of all sensors in the network,

and SDF be our secure digital farming protocol. We

aim to show that SDF is auto-stabilizing, meaning it

can recover from any transient faults and eventually

reach a stable state where all sensors are correctly

functioning in a fully connected network.

In a fully connected network, every sensor σ

∈

Σ can directly communicate with every other sensor

∈ Σ. This ensures that each sensor has complete

and accurate data about the network.

The message costs

Msg are bounded and polyno-

mial for each sensor σ

until the network converges

(cf. Lemma 2). This ensures that communication

overhead remains manageable and efﬁcient.

From our earlier results, for any farm size S, if an

intrusion occurs, there exists at least one sensor that

can detect the event. This implies that the protocol

can maintain security by ensuring that intrusions are

always detected.

Lemma 4 shows that the nearest and most reliable

sensors in the neighborhood of any given sensor σ

have a higher probability of being selected as data re-

lays. Thus, sensors with higher reliability are more

likely to be selected, which contributes to the stability

of the network.

Given that SDF ensures reliable communication

and intrusion detection in a fully connected network,

the protocol can recover from any transient faults.

Even if some sensors temporarily fail or produce in-

correct data, the protocol’s design ensures that the net-

work will eventually stabilize. The bounded message

costs and efﬁcient selection of reliable sensors con-

tribute to this auto-stabilization. Thus, the protocol

SDF is auto-stabilizing in a fully connected network.

This proves our theorem.

6 CONCLUSIONS

This paper presents a comprehensive approach to

enhancing farm security and optimizing crop yield,

introducing the Secure Digital Farming protocol.

We have deﬁned the feasibility of our SDF-based

solution, which addresses critical challenges faced

by modern agriculture, including climate change,

pest control, rural crime, and demographic pressures.

SDF leverages deep-learning algorithms to analyze

sensor data and video streams from security cameras,

enabling intelligent access control, pest detection,

and yield estimation. We present an implementation

framework ready for real-life testing and validation.

To further validate our solution’s efﬁcacy and

practicality, we will conduct ﬁeld tests on our educa-

ICINCO 2024 - 21st International Conference on Informatics in Control, Automation and Robotics

382

tional farm. This controlled environment will allow us

to rigorously evaluate its performance and ﬁne-tune

its components. Future work will focus on assessing

the scalability of our SDF-based approach for appli-

cation in diverse farming contexts. We will also ex-

plore potential collaborations to reﬁne and implement

the solution on a larger scale and evaluate the social

impact of our technology, particularly its potential to

improve farmer livelihoods and promote sustainable

agricultural practices.

REFERENCES

Cheng, W., andXiaoting Wang, T. M., and Wang,

G. (2022). Anomaly detection for internet of

things time series data using generative adversar-

ial networks with attention mechanism in smart

agriculture. Frontiers in Plant Science, page

https://doi.org/10.3389/fpls.2022.890563.

Feng, Q., Wang, S., Wang, H., Qin, Z., and Wang,

H. (2022a). Circle ﬁtting based image segmen-

tation and multi-scale block local binary pattern

based distinction of ring rot and anthracnose on

apple fruits. Frontiers in Plant Science, 13:doi:

10.3389/fpls.2022.884891.

Feng, Z., Wei, Q., Ye, Z., Yang, B., Gao, Y., Lv, J.,

Dai, Y., Bao, J., and Yao, Q. (2022b). Vibrational

courtship disruption of nilaparvata lugens using arti-

ﬁcial disruptive signals. Frontiers in Plant Science,

13:https://doi.org/10.3389/fpls.2022.897475.

Fu, L., Li, S., Rao, Y., Liang, J., Teng, J., and He, Q. (2022).

A novel heuristic target-dependent neural architecture

search method with small samples. Frontiers in Plant

Science, page 10.3389/fpls.2022.897883.

Maia, R. F., Lurbe, C. B., and Hornbuckle, J. (2022).

Machine learning approach to estimate soil matric

potential in the plant root zone based on remote

sensing data. Frontiers in Plant Science, page

https://doi.org/10.3389/fpls.2022.931491.

MathWorks, I. (2024). Data collection in the cloud with

advanced data analysis using matlab.

Schmidhuber, J. (2015). Deep learning in neural networks:

An overview. neural networks. Ofﬁcial Journal

of the International Neural Network Society, 61,

https://doi.org/10.1016/j.neunet.2014.09.003:pages

85–117.

Sellam, V. and Poovammal, E. (2010). Prediction of crop

yield using regression analysis. IEEE Trans. Knowl.

Data Eng., pages vol. 23, no. 10, pp. 1498–1512.

Sun, G., Liu, S., Luo, H., Feng, Z., Yang, B., Luo, J., Tang,

J., Yao, Q., and Xu, J. (2022a). Intelligent monitor-

ing system of migratory pests based on searchlight

trap and machine vision. Frontiers in Plant Science,

13:https://doi.org/10.3389/fpls.2022.897739.

Sun, M., Xu1, L., Luo, R., Lu, Y., and Jia, W. (2022b).

Fast location and recognition of green apple based

on rgb-d image. Frontiers in Plant Science, 13:doi:

10.3389/fpls.2022.864458.

Umargono, E., Suseno, J. E., and K., V. G. S. (2020).

K-means clustering optimization using the elbow

method and early centroid determination based-on

mean and median. International Conferences on In-

formation System and Technology (CONRIST 2019,

DOI: 10.5220/0009908402340240:pages 234–240.

Uyeh, D. D., Iyiola, O., Mallipeddi, R., Asem-Hiablie,

S., Amaizu, M., Ha, Y., and Park, T. (2022). Grid

search for lowest root mean squared error in pre-

dicting optimal sensor location in protected culti-

vation systems. Frontiers in Plant Science, page

https://doi.org/10.3389/fpls.2022.920284.

Yates, R. D. and Goodman, D. J. (2005). Probability

and Stochastic Processes: A Friendly Introduction for

Electrical and Computer Engineers. John Wiley and

Sons, INC, Rutgers, The State University of New Jer-

sey.

Zhang, L., Lei, L., and Yan, D. (2010). Comparison of

two regression models for predicting crop yield. IEEE

International Symposium on Geoscience and Remote

Sensing (IGARSS).

Zingade, P. D. S., Buchade, O., Mehta, N., Ghodekar, S.,

and Mehta, C. (2018). Machine learning-based crop

prediction system using multi-linear regression. Inter-

national Journal of Emerging Technology and Com-

puter Science (IJETCS), pages Vol 3, Issue 2.

Combating Agricultural Challenges with Secure Digital Farming

383