INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,

MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)

ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XIV, Issue XI, November 2025

Animal Enchroachment Detection in Croplands using Machine

Learning Approaches

Anjaly Santhosh

Department of Information Technology, Jawaharlal Nehru Technological University Hyderabad,

GNITC Hyderabad

DOI : https://doi.org/10.51583/IJLTEMAS.2025.1411000074

Received: 30 November 2025; Accepted: 05 December 2025; Published: 11 December 2025

ABSTRACT

Animal Enchroachment is a major threat to the productivity of the crops, which affects food security and reduces

the farmer’s profit. Machine learning-based solutions are used to overcome this problem. Convolu- tional Neural

Network (CNN), ResNet-50, and Inception v3 are the three methods used to identify the animals. The proposed

model classifies the detected animals and alerts humans through a message to avoid animal intru- sions into

properties. Hence, minimising the dangerous consequences caused by the intrusion. The Inception v3 model

provides more accurate results compared to the other two models, and it is considered the main method for the

proposed model.

Keywords: Convolutional Neural Network, ResNet-50, Inception V3, Machine learning.

INTRODUCTION

Agriculture is the most important sector in India. Day by day, the production of crops is decreasing due to a lack

of interest in farming. The major problem faced in farming is animal intrusion. The animals cause damage to the

crop and also reduce the productivity of farmers. In this situation, we need a proper detection system to detect

the pres- ence of animals. By knowing which animal is most likely to come into cropland, farmers can use good

prevention methods to keep the animals away from croplands. Traditional methods are harmful to ani- mals and

humans, time-consuming, and lead to the need for more elaborate solutions. Machine learn- ing with

convolutional neural networks, ResNet 50, and Inception v3 models offers a promising approach to this problem.

Animal infestations are always a problem for farmers. Sheep, cows, ele- phants, monkeys, etc. roam the fields

without the consent of the farmer and destroy and eat the crops. By doing so, the yield could consequently expe-

rience a substantial loss, prompting the purchase of further financial insurance to pay for the harm. Every farmer

should be aware of the animals in the area who need to be protected from suffering while using their land to

grow food. Right away, this issue needs to be addressed, and a workable solution needs to be developed and

implemented.

RELATED WORKS

Mowen Xue et al.[1] Developed a technique for aerial animal surveillance. Mainly focuses on the difficulties of

identifying small animals from aerial images, where creatures may seem like small and far-off objects due to the

altitude of the aerial platform. This approach integrates super-resolution and altitude data exploitation directly

into deep animal detection pipelines for aerial survey appli- cations.

T. Sandeep et al.[2] They proposed a prototype that can be used as software that recognises ani- mals and

classifies themaccordingly. The software required can be developed using openCV and deep learning algorithms.

This can be embedded with an ultrasonic repellent hardware system that drives the animal away from the farm

and also informs the farmer about this. This is a low-cost project that aims to drive the animals away without

causing any lethal harm.

Davide Adami et al.[3] The proposed system is based on IoT platforms that provide a sat- isfactory

www.ijltemas.in

Page 805

INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,

MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)

ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XIV, Issue XI, November 2025

compromise between performance, cost and energy consumption. More specifically, in this work, we deployed

and evaluated various edge computing devices running real-time object detector (YOLO and Tiny-YOLO) with

custom- trained models to identify the most suitable animal recognition Hardware or Software platform to be

integrated with the ultrasound generator.

K Balakrishna et al.[4] The authors propose an integrated system that combines IoT and machine learning for

crop protection against animal intru-sion. The system consists of sensor nodes that are deployed in the field to

collect data on environ- mental parameters such as temperature, humidity, soil moisture, and animal presence.

These sensor nodes are connected to a central hub or gateway through wireless communication, forming an IoT

network.

Devsmit Ranparia et al.[5] Developed a system that uses audio data to detect the presence of wild animals

and trigger repelling actions. The system consists of microphones that capture audio data from the field, which

is then processed using machine learning algorithms to identify animal vocalisations or sounds associated with

animal presence.

Kuei Chung Chang et al.[6] The authors propose an IoT-based system that utilises object detection and tracking

techniques to monitor and detect the presence of monkeys in agricultural fields. The proposed system consists

of IoT devices, such as cameras or sensors, that are deployed in the field to capture images or data related to the

presence of monkeys. These devices are connected to a central hub or gateway through wireless communication,

forming an IoT network.

METHODOLOGIES

Convolutional Neural Nework(CNN)

CNN models can be used for animal intrusion detection in cropland by training the model to recognise images

or video footage captured by cameras installed in the cropland. The CNN model can learn to differentiate between

different types of animals and identify their locations within the cropland. To develop an effective animal

intrusion detection system using CNN, a large dataset of labelled images or videos of animals in cropland would

need to be collected. The dataset can be used to train the CNN model, which can learn to recognise the features

of different animals and their behaviours. Once the CNN model is trained, it can be used for real-time animal

intrusion detection by analysing the video feed from the cameras installed in the cropland. The model can detect

when an animal enters the cropland and identify the type of animal and its location. This information can be

used to trigger an alarm or send alerts to farmers or farm managers, allowing them to take immediate action to

prevent crop damage.

Fig. 1. Training and validation accuracy graph of CNN

www.ijltemas.in

Page 806

INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,

MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)

ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XIV, Issue XI, November 2025

Fig. 2. Training and validation loss graph of CNN

ResNet 50

ResNet-50, a well-known deep neural network design, can be used as a machine learning method to identify

animal encroachment on agricultural lands. It has been shown that the deep convo- lutional neural network

ResNet-50 excels in a wide range of picture categorization tasks. It is a popular choice for many computer vision

applica- tions, including the detection of animal trespass in agriculture. A dataset of photos taken by cameras

installed in crops for the purpose of detecting animal intrusions can be used to train ResNet-50. The network

may be taught to classify photos as either including animals or not based on the visual traits of the animals in

the image.

Fig. 3. Training and validation accuracy graph of ResNet 50

Fig. 4. Training and validation loss graph of ResNet 50

www.ijltemas.in

Page 807

INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,

MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)

ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XIV, Issue XI, November 2025

Inception v3

Inception v3, which is a popular deep neural network architecture, can also be used for animal intrusion detection

in croplands using a machine learning approach. The Inception v3 model works by using multiple convolutional

layers with dif- ferent kernel sizes to extract features from in- put images. These features are then concatenated

and fed through additional convolutional and fully connected layers to make the final classification decision. To

use Inception v3 for animal intru- sion detection in cropland, the model is typically first trained on a large dataset

of images that includes animals. During training, the model learns to recognise the visual features of animals and

distinguish them from other objects. Once the Inception v3 model is trained, it can be used to classify new images

captured by cameras in the cropland. The input image is fed into the model, and the model outputs a probability

distribution over different classes, indicating the likelihood that the image contains animals. The Inception v3

model can be fine-tuned on a smaller dataset of images specifically related to animal intrusion detection in

cropland, which can help improve the accuracy of the system. Additionally, transfer learning techniques can

be used to further improve the performance of the model by adapting it to the specific characteristics of the

cropland envi- ronment.

Fig. 5. Training and validation accuracy graph of Inception v3

Fig. 6. Training and validation loss graph of Inception v3

PROPOSED METHOD

To address the problem of animal intrusion in cropland, propose a method that utilises CNN (convolutional

neural networks), ResNet 50, and Inception v3 models. This approach involves train- ing these models on a large

dataset of images con- taining various animals that are commonly found in crop fields. By fine-tuning the pre-

trained mod- els, they achieve high accuracy in detecting animal intrusion in real time. Once the model is trained,

it can be used to detect animals in real-time by processing images captured by cameras placed in the cropland.

When an animal is detected, an alert can be sent to the farmer or a control system, triggering appropriate actions

to deter the animal and prevent crop damage. Compared to traditional methods like fence-based systems,

machine learn- ing approaches using CNN can be less expensive and more flexible, as they do not require

www.ijltemas.in

Page 808

INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,

MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)

ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XIV, Issue XI, November 2025

physical barriers to be installed. They can also provide more accurate and reliable detection by analysing multiple

factors, such as animal size, shape, and movement patterns.

Fig. 7. Architecture of Proposed Method

Dataset

The data used in this work was an animal’s image. Here, use six classes of animal images: chicken, cow,

elephant, monkey, scoiattolo (squir- rel), and sheep. After collecting the images of these six classes, divide the

data into training and testing data sets. Initially, five classes were used for modelling (chicken, cow, elephant,

scoiattolo (squirrel), and sheep). After selecting the final model, use the six classes of dataset. The collected

images of the above classes were used as a dataset for data training and testing.

Fig. 8. Dataset

Data preprocessing

The collected images need to be preprocessed to ensure that they are all of the same size and format. This can

involve resizing the images, converting them to grayscale or RGB, and normalising the pixel values. Splitting

the data into training, valida- tion, and test sets. Data augmentation techniques, such as rotation, flipping, and

zooming, can also be applied to increase the diversity of the data and improve the model’s generalization ability.

Modelling and Training

Convolutional Neural Network (CNN), ResNet 50, and Inception V3 are the models used for training the data

(preprocessed data). The model that provides the highest accuracy in testing is considered the final model for

animal intrusion detection in cropland. During training, the model learns to recognise features that distinguish

an- imals and also to distinguish between different types of animals.

www.ijltemas.in

Page 809

INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,

MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)

ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XIV, Issue XI, November 2025

Testing the model

Once the model is trained, it needs to be tested on a separate set of images that were not used during training.

This helps to ensure that the model can generalize to new images.

Evaluation

Here, consider the F1 score, test accuracy, val- idation, and train accuracy, validation, and train loss. The

Inception v3 model provides good accu- racy and less validation and train loss.

Graphical Visualization

Once the model is tested and validated, it can be saved. In this step, show the confusion matrix, ROC curve of

used models, validation accuracy, validation loss, training accuracy, and loss graphs. The model was uploaded

to Visual Studio Code to create the front end.

Requirements are

IMPLEMENTATION

Animal Intrusion Detection in cropland using Machine Learning approach used Inception v3 as a model because

Inception v3 provides higher accuracy compared with CNN, ResNet-50 models. First, collect a dataset of animals

that damage the croplands. After the data collection, import the libraries required for animal intrusion detection.

Inception v3 is used as a model to create a data generator. The training and testing datasets are added to this data

generator. Data generators are commonly used in animal intrusion detection sys- tems to generate labelled

training data for machine learning algorithms. These algorithms are trained to distinguish between normal animal

behaviour and anomalous behaviour that may indicate intru- sion or potential threats. Next, train the model (

Inception v3 ) and save the trained model, then do the graphical visualisation, which includes valida- tion and

train loss graphs and validation and train accuracy plotting. Create the classification report and predict the images

from the test data. Then check the confusion matrix, plot the ROC curve, accuracy, and f1-score, and test the

accuracy of the model. These are the evaluation procedures for animal intrusion detection in cropland. After

the evaluation process, use the Visual Studio code. The saved model is uploaded to the Visual Studio code to

create the front end. HTML ( HyperText Markup Language ), CSS ( Cascading Style Sheets), and JavaScript are

the programming languages used to create the front end and run the programme on the local host and get the

outputs. To receive alerts when an animal is located, connect a mo- bile device. This makes use of both the

Firebase cloud service provider and the WhatsApp API. Firebase Cloud Messaging (FCM) enables a secure and

power-efficient connection between the server and devices for the free delivery and receiving of messages

and notifications on iOS, Android, and the web. The Pyrebase library, which may use a configuration dictionary

to interact with the Firebase platform, is used to initialise a Firebase app, which is afterwards used to authenticate

a user by logging in with an email and password. After a successful sign-in, it receives the account information

for the authenticated user.

RESULTS

The saved model is uploaded to Visual Studio Code to create the front end. The Inception v3 model was selected

as the final model because it provides better accuracy compared to the other two models used here. HTML

(HyperText Markup Language), CSS (Cascading Style Sheets), and JavaScript are the programming languages

used to create the front end, run the programme on the local host, and get the outputs. The output is given below.

After the prediction, the system provides an alert to the user via mobile phone.

www.ijltemas.in

Page 810

INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,

MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)

ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XIV, Issue XI, November 2025

Fig. 9. Select an image

Fig. 10. Upload an image

Fig. 11. Predict the image

Fig. 12. Prediction result

www.ijltemas.in

Page 811

INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,

MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)

ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XIV, Issue XI, November 2025

Fig. 13. Training and validation accuracy graph of final model

Fig. 14. Training and validation loss graph of final model

Fig. 15. Alert

CONCLUSION

The proposed system can help farmers monitor their fields and protect their crops from damage caused by

animals such as chickens, cows, sheeps, elephants, and monkeys and squirrels. By using machine learning

algorithms to analyse images captured by cameras installed in the cropland, the system can learn to recognise

the features of animals and distinguish them from other objects in the images. This approach can improve the

accu- racy and efficiency of animal detection, reduce the need for manual monitoring, and enable farmers to

respond quickly to potential threats to their crops. Ultimately, the goal is to help farmers reduce crop losses and

increase yields, leading to improved food security and economic benefits.

www.ijltemas.in

Page 812

INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,

MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)

ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XIV, Issue XI, November 2025

REFERENCES

1. Mowen Xue, Theo Greenslade, Majid Mirmehdi, Tilo Burghardt, “Small or Far Away? Exploiting

Deep Super- Resolution and Altitude Data for Aerial Animal Surveillance”, IEEE 2022

2. T.Sandeep, B.Manushree, S.Rahul, T.Bharath,“ Crop Protec- tion using CNN”, International Research

Journal of Modern- ization in Engineering Technology and Science, March 2022.

3. Davide Adami, Mike O Ojo, Stefano Glordano,“ Design, De- velopment and Evaluation of an

Intelligent Animal Repelling System for Crop Protection Based on Embedded Edge-AI ”, IEEE 2021.

4. Balakrishna K, Fazil Mohammed, C.R. Ullas, C.M. Hema,

5. S.K. Sonakshi, “Application of IOT and machine learning in crop protection against animal intrusion ”,

Global Transitions Proceedings 2, no. 2 (2021): 169-174.

6. Devsmit Ranparia, Gunjeet Singh, Anmol Rattan, Dr Harpreet Singh, Dr Nitin Auluck, “ Machine

learning based Acoustic Repellent System for Protecting Crops against Wild Animal Attacks ”, IEEE

2020.

7. K.S.P. Premarathna, R.M.K.T. Rathnayaka and J. Charles,“ An elephant detection system to prevent

human-elephant conflict and tracking of elephant using deep learning”, In 2020 5th International

Conference on Information Technology Research(ICITR), 1-6.IEEE 2020.

8. N S Gogul Dev, K S Sreenesh,P K Binu, “ IoT Based Automated Crop Protection System ”, IEEE

2019.

9. Kuei-Chung Chang,Zi-Wen Guo, “ The Monkeys are Coming Design of Agricultural Damage Warning

System by IoT- Based Objects Detection and Tracking ”, 2018 IEEE Interna- tional Conference on

Consumer Electronics-Taiwan (ICCE- TW).

10. P. Badre, S. Bandiwadekar, P. Chandanshive, A. Chaudhari, and M.S. Jadhav, “ Automatically

Identifying Animals Using Deep Learning”, vol. 6, no. 4, pp. 194–197, 2018.

www.ijltemas.in

Page 813