Proceedings of the 3rd International Conference on Signal Processing and Machine Learning

DOI: 10.54254/2755-2721/6/20230799

Animal detection in farms using machine learning

S.Thylashri1, N.R.Rajalakshmi2, P. Karthikeyan3, L.Sahaya Senthamil4

1
Assistant Professor, Department of Computer Science And Engineering, Vel Tech
Rangarajan Dr Sagunthala R&D Institute of Science and Technology, Chennai, India.
2
Professor, Department of Computer Science And Engineering, Vel Tech Rangarajan
Dr Sagunthala R&D Institute of Science and Technology, Chennai, India
3
School of Information Technology and Engineering, Vellore Institute of Technology,
Vellore, Tamil Nadu, India.
4
Department of Electrical & Electronics Engineering, PSNA College of Engineering
and Technology, Dindigul.

[email protected]

Abstract. Agriculture is extremely important to a country’s development. Agricultural issues

have consistently stymied the country’s progress. Farmers face a slew of problems, including a
lack of water for irrigation, crops withering due to climatic changes, nutrient-depleted soils, and
crop damage from pests and wildlife. In recent decades, innovation has developed to provide
effective solutions to a large number of these difficulties. In any event, agricultural protection
from wild animals has not been fully addressed until now. The fields have been destroyed by
elephants, monkeys, and wild boars. The production of crops is diminished when wild animals
trample them or eat them. The presence of wild animals on farmland has always been a
contentious issue for farmers. Deer, wild boar, moles, elephants, and monkeys are just some of
the animals that might ruin a harvest. When the farmer isn't there, these animals may eat the
crops and wander the field, which is bad for the harvest. Because of this, yields might drop
significantly, and supplementary financial protection could be necessary to cope with the fallout.
This problem has to be solved as soon as possible, and a workable solution needs to be found
and put into action. Accordingly, the goal of our programme is to address this problem. Wireless
sensors might be used to set off an alarm in the event of an animal intrusion, communicating the
situation to the proprietor and forest authorities with an attached photo. This provides an early
warning so that measures may be taken according to the intruder type.

Keywords: Machine Learning, Deep Learning, Image Processing, OpenCV.

1. Introduction
Agriculture is the primary sector of the Indian economy, yet crop loss caused by wild animals has
recently become a significant social issue. There is currently no viable solution to this issue., which
necessitates careful thought. Building wire fences and electric fences are two existing options that are
ineffective. Electric fences have batteries that are charged by solar panels and shock animals that come
into touch with them. If plants or shrubs grow too close to the barrier, they can catch fire.
Electromagnetic interference occurs when the fence is not properly maintained, interfering with
telephone and radio signals. Electric fencing, although being the most extensively used farm protection

© 2023 The Authors. This is an open access article distributed under the terms of the Creative Commons Attribution License 4.0
(https://creativecommons.org/licenses/by/4.0/).

318
 Proceedings of the 3rd International Conference on Signal Processing and Machine Learning
DOI: 10.54254/2755-2721/6/20230799

device, is dangerous to both animals and humans [1]. A thorn fence, which is also extensively utilized,
has a similar effect to the previous strategy. This paper proposes a clever solution to this issue. When
an animal intrudes, a picture is collected, and the image is then categorized as domestic or wild using a
Convolution Neural Network and deep learning technique. This classification assists in informing the
farmer through SMS in the event of a wild animal encroachment. Crops are secure and protected thanks
to the smart farm protection system. There has been a rise in reports of dangerous wild animals venturing
into human settlements, especially in areas close to forests. The expenditures associated with animal
incursions, whether they be due to agricultural damage or attacks on livestock, may be quite high. One
of the key causes leading to the growth in human animal disputes is the growing human population,
which leads to decreasing forest cover. Erection of electric fences or posting of sentries to watch for
animals overnight are two methods now available for settling such problems [2]. Technology
advancements like computer vision are needed to solve the interconnected problems. The major goal of
this study is to explore the development of a computer vision system that can recognize and track wild
animals. The project’s main purpose is to keep an eye on the farming field for wild animals and to protect
them by pushing them away rather than killing them. The project also intends to protect human lives
from animal attacks. We’re using an integrative Deep Learning methodology to create a checking and
repelling architecture for crop insurance against monster attacks. First, automated systems were
developed to eliminate the brutes by exploiting their weaknesses (such as controls and sensors).
Camera's scheming with the boo pi module is a similar method. Images of wild animals are taken using
a camera and transmitted to the Jeer pi module. When the boo pi captures a picture, it is compared to
the images already in the database. If the wild animal is also spotted after comparing photos, the
GSM module is commanded. The estate's owner was contacted using a GSM gunshot [3]. Connect boo
pi to the speaker to hear the event unfold in audio form. Disadvantage There's also the fact that other
forms of assault element aren't very effective, and their conservation is difficult. Recently, efforts have
been made to find a solution to this issue by using AI. The World Congress of Hog Owners and Growers
was one such notion that had been offered already (Weighted- circumstance of Histogram-Acquainted
Slants). LI- BLINEAR (Library for large direct type) classifier is used in order to get better delicacy for
high dimensional data. The trials were conducted on two standard datasets called Wild- Beast and
Camara-Trap dataset. But it’s a com- Megaplex process. This point vector is supplied to the classifier
to descry the beast in a particular window. In sliding window fashion different sliding windows are used
in order to identify brutes of different sizes predicated upon drone position of the camera. In the CoHOG
(Co-occurrence of Histogram-Acquainted Slants) system, grade directions are used to calculate point
vector and the magnitude is ignored (2). One another approach was by relating the sound of brutes and
recovering it by concepts of signal processing and deep knowledge which involved more attack and as
well as complex process depending on the database.

2. Related Work
Working together, we developed an Internet of Things (IoT) software to deter animals from damaging
crops in the field. The authors coupled modern IoT gateways with wireless devices such as 6LoWPAN,
WiFi to harvest or monitor data from the agricultural field [4]. The small solar and LiPo batteries were
used to power an ultrasonic repellent that could function in low-light conditions. This gadget performs
frequency transmission and networking by broadcasting a small frame across a distance of 50 metres; a
PIR (Passive Infrared) sensor was employed to increase its efficiency. When an animal is identified, the
device uses the RIOT-OS software to communicate by making a 120dB noise. Its effectiveness drops to
below 90% after that. Produce from buffaloes, cows, goats, and birds, among others, were preserved in
the field with the help of a PIC microcontroller [5]. The device employs a motion sensor to identify the
presence of animals in a farming area, and then notifies the microcontroller so that the farmers may take
the necessary precautions. The microcontroller sends signals to the PIR-based motion detector to
identify the presence of animals, as well as the bell sounds to notify the farmers. Authors claim this
method saves farmers from blocking off an entire field for an entire day. Deep deep learning and the
Convolutional neural networks method [6] are employed by the author to safeguard crops against animal
damage. The goal of the author is to protect crops from animals without endangering either animals or
humans. Researchers developed a prototype to direct the animals toward the croplands. To use the neural

319
 Proceedings of the 3rd International Conference on Signal Processing and Machine Learning
DOI: 10.54254/2755-2721/6/20230799

network concept and computer vision method, a model is developed using a machine learning algorithm
to identify animals entering the agricultural field. In this approach, we periodically deploy a camera to
the field to check in on the crops' progress. The method uses a combination of libraries and neural
network concepts to deduce the animal's motion from the camera frames. The present study was focused
on a model to safeguard the crop from pests and rodents, and utilized Internet of Things (IoT) technology
to regulate irrigation [7]. The right irrigation and crop protection are managed by an Arduino UNO
microcontroller in the suggested design [8[. The level of moisture in a farm's soil is used to control when
water syphons are activated. Intentional sensors installed in the agricultural field protect the crops from
insects, animals, and other pests; the sensors measure the motion of insects and animals closer to the
crop and give the signal to the Arduino Uno microcontroller for distance calculations and other activities
[9]. The microcontroller permits high-frequency sound based on the estimated distance values [10]. The
collaborated to create a prototype that uses two new technologies, IoT and machine learning, to monitor
crops and alert wild animals [9]. PTZ (Pan- Tilt-Zoom) camera, GSM module, Sensors, and Arduino
UNO microcontroller are among the IoT components used[11]. Machine learning algorithms for animal
categorization include KNN (K-Nearest Neighbour) Algorithm, Logistic Regression, andSVM (Support
Vector Machine) Algorithm [7]. Elephants, horses, Zebras, and other animals are represented in 605
photos [12][13]. For the iterated regularization parameter of C=100, SVM outperforms the KNN and
Logistic Regression models withan accuracy of 89.6 [14][15]. Numerous of the above-mentioned state-
of-the-art prototypeand classifier systems function on IoT and recognize animals using machine learning
approaches [16]. This system, however, is still in its infancy for real-time use because it has yet to achieve
the needed performance [17][18]. As a result, we offer a better model based on IoT and machine learning
techniques for defending the farm field from intruders [19].

Figure 1. Proposed system Architecture

3. Proposed Framework
The object sensor will be fixed in the cropland that covers a specific area in our system architecture. The
object sensor detects and transmits a signal to the Arduino microcontroller whenever any object enters

320
 Proceedings of the 3rd International Conference on Signal Processing and Machine Learning
DOI: 10.54254/2755-2721/6/20230799

the cropland. The user interface is then notified that an object has entered the farmland via this controller.
We employed a MatLab software interface with a trained data set as the user interface. When we upload
an input image, it goes through deep learning classification and is compared to the training data set. If it
matches, the categorized result will be communicated to the microcontroller, which will issue a buzzer
alarm, scaring the animals away from the farmland and reducing crop loss. If the classified image does
not match the trained data set, it will deliver an unclassified result message and sound a buzzer to inform
the user. Figure 1 depicts the proposed system architecture and Figure 2 depicts the working model
flowchart.

Figure 2. Flow chart of the working model

We’ll need to collect some data in order to construct this system. Images of wild animals from eight
distinct classes are included in our datasets. After gathering the data, it will be subjected to picture pre-
processing, commonly known as annotation, before being transformed into a training dataset. The class
file, photos, and txt files for those images are all included in this dataset, which is generated automatically
following successful annotation. We can now begin to train our system, and we will do so using Google
Colab. While the dataset is being trained, weights are generated that can be used for testing later. Although
the training method can take 12-15 hours, testing can be done quickly with the help of those weights that
have been trained.

4. Experimental Results
Python language is used for wild animal detection. this initiative has a huge social impact since it will
assist farmers in safeguarding their farms, Protect them from significant financial losses and the wasteful
operations that are required to sustain their fields. This could also help with preventing frequent human-
animal confrontations and the loss of human life, as well as avoiding any substantial injury to people.
Figure 3 represents the image classification like; cat or dog.

321
 Proceedings of the 3rd International Conference on Signal Processing and Machine Learning
DOI: 10.54254/2755-2721/6/20230799

(a) (b)
Figure 3. Image Classified (a) as dog; (b) as cow

5. Conclusion
Farmers are facing a growing problem due to the destruction of their crops by wild animals. An timely
response and a viable solution are essential to resolving this urgent problem. To address the issue facing
the farmer, we employed IoT to develop a protective and proactive system capable of early detection.
The primary objective is to lessen the number of human deaths and to protect agriculturally productive
land from destructive wild animals. When animals are seen on the forest's edge, it's best to take
precautions as soon as possible. Therefore, our scientific method will aid farmers in protecting their
farms, saving them money and relieving them of the effort and time needed to defend their lands. Future
improvements in sensor efficiency will allow us to spread our wings to a broad variety of locations. The
proposed system would only provide service to a small area around a few villages that are located close
to a forest. It is possible that the range of uses will grow as more efficient sensors are included in the
future. In light of the persistence of the aforementioned problems regardless of intervention, we resolved
the issue by using computer vision to mechanically scare away the animals. CCTV footage is used as
an input in the model we propose (Closed Circuit Television). The system analyses and anticipates the
images from the camera, then plays a repellent sound to frighten away the detected animal. This
programme has far-reaching societal effects since it helps farmers prevent agricultural failures, saves
them money, and frees them from time-consuming and inefficient maintenance tasks. The frequent
encounters between humans and animals, the resulting human loss of life, and the severe harm that often
follows may all be reduced with this measure.

References
[1] Rajalaxmi, R. R., Saradha, M., Fathima, S. K., Sathish Kumar, V. E., Sandeep kumar, M., &
Prabhu, J. (2022). An Improved MangoNet Architecture Using Harris Hawks Optimization
for Fruit Classification with Uncertainty Estimation. Journal of Uncertain Systems.
[2] Sathishkumar, V. E., Rahman, A. B. M., Park, J., Shin, C., & Cho, Y. (2020, April). Using
machine learning algorithms for fruit disease classification. In Basic & clinical pharmacology
& toxicology (Vol. 126, pp. 253-253). 111 RIVER ST, HOBOKEN 07030-5774, NJ USA:
WILEY.
[3] Babu, J. C., Kumar, M. S., Jayagopal, P., Sathishkumar, V. E., Rajendran, S., Kumar, S., ... &
Mahseena, A. M. (2022). IoT-based intelligent system for internal crack detection in building
blocks. Journal of Nanomaterials, 2022.
[4] Sathishkumar, V. E., Venkatesan, S., Park, J., Shin, C., Kim, Y., & Cho, Y. (2020, April). Nutrient
water supply prediction for fruit production in greenhouse environment using artificial neural
networks. In Basic & Clinical Pharmacology & Toxicology (Vol. 126, pp. 257-258). 111
RIVER ST, HOBOKEN 07030-5774, NJ USA: WILEY.
[5] Krishnamoorthy, N., Prasad, L. N., Kumar, C. P., Subedi, B., Abraha, H. B., & Sathishkumar, V.

322
 Proceedings of the 3rd International Conference on Signal Processing and Machine Learning
DOI: 10.54254/2755-2721/6/20230799

E. (2021). Rice leaf diseases prediction using deep neural networks with transfer learning.
Environmental Research, 198, 111275.
[6] VE, S., & Cho, Y. (2020). A rule-based model for Seoul Bike sharing demand prediction using
weather data. European Journal of Remote Sensing, 53(sup1), 166-183.
[7] VE, S., Shin, C., & Cho, Y. (2021).Efficient energy consumption prediction model for a data
analytic-enabled industry building in a smart city. Building Research & Information,
49(1),127-143.
[8] Sathishkumar, V. E., Park, J., & Cho, Y. (2020). Using data mining techniques for bike sharing
demand prediction in metropolitan city. Computer Communications, 153, 353-366.
[9] Subramanian, M., Lv, N. P., & VE, S. (2022). Hyperparameter optimization for transfer learning
of VGG16 for disease identification in corn leaves using Bayesian optimization. Big Data,
10(3), 215-229.
[10] VE, S., Park, J., & Cho, Y. (2020). Seoul bike trip duration prediction using data mining
techniques. IET Intelligent Transport Systems, 14(11), 1465-1474.
[11] Easwaramoorthy, S., Thamburasa, S., Samy, G., Bhushan, S. B., & Aravind, K. (2016, April).
Digital forensic evidence collection of cloud storage data for investigation. In 2016
International Conference on Recent Trends in Information Technology (ICRTIT) (pp. 1-6).
IEEE.
[12] Easwaramoorthy, S., Sophia, F., & Prathik, A. (2016, February). Biometric Authentication using
finger nails. In 2016 international conference on emerging trends in engineering, technology
and science (ICETETS) (pp. 1-6). IEEE.
[13] Easwaramoorthy, S., Thamburasa, S., Aravind, K., Bhushan, S. B., & Rajadurai, H. (2016,
August). Heterogeneous classifier model for e-mail spam classification using FSO feature
selection method. In 2016 International Conference on Inventive Computation Technologies
(ICICT) (Vol. 1, pp. 1-6). IEEE.
[14] Subedi, B., Sathishkumar, V. E., Maheshwari, V., Kumar, M. S., Jayagopal, P., & Allayear, S. M.
(2022). Feature learning-based generative adversarial network data augmentation for class-
based few-shot learning. Mathematical Problems in Engineering, 2022.
[15] Rajalaxmi, R. R., Narasimha Prasad, L. V., Janakiramaiah, B., Pavankumar, C. S., Neelima, N.,
& Sathishkumar, V. E. (2022). Optimizing Hyperparameters and Performance Analysis of
LSTM Model in Detecting Fake News on Social media. Transactions on Asian and Low-
Resource Language Information Processing.
[16] Zhongshan, C., Xinning, F., Manickam, A., & Sathishkumar, V. E. (2021). Facial landmark
detection using artificial intelligence techniques. Annals of Operations Research, 1-19.
[17] Shanthi, N., VE, S., Upendra Babu, K., Karthikeyan, P., Rajendran, S., & Allayear, S. M. (2022).
Analysis on the Bus Arrival Time Prediction Model for Human-Centric Services Using Data
Mining Techniques. Computational Intelligence & Neuroscience.
[18] Subramanian, M., Sathishkumar, V. E., Ramya, C., Kogilavani, S. V., & Deepti, R. (2022, May).
A Lightweight Depthwise Separable Convolution Neural Network for Screening Covid-19
Infection from Chest CT and X-ray Images. In 2022 18th International Conference on
Distributed Computing in Sensor Systems (DCOSS) (pp. 410-413). IEEE.
[19] Shanmugavadivel, K., Sathishkumar, V. E., Kumar, M. S., Maheshwari, V., Prabhu, J., &
Allayear, S. M. (2022). Investigation of Applying Machine Learning and Hyperparameter
Tuned Deep Learning Approaches for Arrhythmia Detection in ECG Images. Computational
& Mathematical Methods in Medicine.

323