Jaipur Service Company

Dr. Nikhil GOVIL Department of Anukul Gangwar

Computer Engineering & Application B.Tech computer Science
specialization in AML

Raghav Mittal
Raghav Gupta B.Tech computer Science
B.Tech computer science specialization in AML
Specialization in AIML

Abstract: Today, people is move towards digital in on time with a minimum wages got an issue with an AC
world so they are used various E-commerce call and ear conditional expert who will fix your AC just a
application . our project is also a E-commerce few minutes near right by your location. with a minimum
service based application which has aim of provide cost a consumer has no problem regarding a finding a
on demand services to people available on smart worker to fix their applications. or their household
phone application and web based services . Our products similarly these apps help in fixing the problem of
aim is to design an application which helps a needy corporate house cleaning home Appliance and all other
people to find a service worker very easily in a cost household problems.
efficient manner. We will design a Jaipur Service
Company. It will provide a list of worker to a Keywords— On-demand services, consumer satisfaction,
particular selected category and use the will be sustainability, wages.
Workers are also available at their near locations.
and they can provide. if no worker is available near. I. NEED OF THE STUDY
the consumer do. finding worker radius or the area
is increased. gradually so that we will find a new Now people increasingly move toward the technology so
worker for the consumer help. Now a days Our they need a platform for their needs reaching. which includes
travel , eat , shop and even dating all has undergone a shopping, dining, even dating. Know the youth demand on
a tremendous change so we can design Interactive demand services platform, which provided. on demand
Android application. named as Jaipur Service services to the pupil according to their needs and help. in their
Company Love in the marketplace. for on demand common livelihood In this context, we are defining a Jaipur
service available. through a smartphone app. service. This company, which is easily accessible by all the
Today on demand solution have proven blessings to pupils, have efficiency to provide a service provision and it is
everyone. in all the environment condition . Our a localized problem. solution which focused on provided based
Jaipur based company provide the advanced solution to the human with the cost efficient way in their
livelihoods. The rapid evolution of technology makes a Jaipur
technology to the younger generation which fulfill
service company. a new term for adapting the human changing
the demands everything to happen instantly
demands behavior. and New South the customer experience
anywhere at anytime. Jaipur service company through which company has addressing all the demand and
provides everything Add their doorstep whether their needs for their own demand services platform named as
it’s in South Cleaning, or innovation, plummy Jaipur Service Company platform It’s a simple Pupil friendly
assistance, appliance repair or electric application platform in which user can. demand its basic needs like
of fitness coaches who come to them. method of plumbing, furnishing, carpentering, painting and Etc. services
providing services. is being changed as a result. in their local peoples.
Customer demand exchanges rapidly. Jaipur based
company involved in customer needs and technical
improvement on demand app development II. INTRODUCTION
emerged. The cost of a worker or an employee. is
checked. or calculated by our system dynamically Today is the world dependent on tech All we can say fully
after considering a different parameter, a final hi-tech world . As a group of the technologies, it increases day
causeway despite to the consumer for which the by day. So the concept of working is Exchanging and so is
cost of the worker or helper is minimized. The main the future of work and employment and production is also
idea behind this if two dynamic prices is applied on changes With this, we can introduce our project that is served.
the user order to overcome a left salary problems on service demand platform named as Jaipur Service
of the worker and also provided a cheap and the Company, which provided New trend in the market of on
best experience labor or helper to the consumer. demand application. The demand of home based services. is
increased day by day. to increase these demand. home service
As the name suggest, our company is a service based application will be rapidly the old age people are not going for
company, which provided a platform to the consumer for any worries anymore, The trusted cost efficient A home
which. a consumer can hire a professional for your service application with professional qualification and
household course at your fingertips. Jaipur service personal quality. and fixed everything. at your home in very
company is a local Jaipur based company which provided efficient way. Now the problem is with the growth of city and
a services in the Jaipur marketplace. Jaipur based urbanization. the problem of Abbudance of low cost worker
company provides on demand in build services. with all the and finding a worker for their used like painting.
essential features. For example, got a leakage issue at Carpentering. preparing. lumbering and Etcetera purposes. is
home. Higher a plumber near at your locality who will fix most difficult challenge in the urban or metropolitan city of
Global. World people are constantly in a hurry. and want to in such a way that it will help in. organizing ,Support
fix their program at any cost, immediately They do not have reasoning user friendly and behaviour of the system
time to find a labor for their work. They are busy with their
everyday calculus and their work life balance to maintain this IV.I. Architectural Design
problem. Jaipur service company is available in the Metro The architecture design of the Jaipur Service Company is a
City of Rajasthan. which provides a solution of this program. representation of the. working and its Organized way to To
enriched at their home or doorstep to help you with this. Jaipur make it user friendly for the people and work efficiently. with
Service company has aimed to implement a contract based that, the basic architecture system design a mobile app
model for the people in the Jaipur, the pink city of Rajasthan.
application server which is client side on Internet service
Jaipur service company has the ability of the on demand
provide For and which connect to the server in the database
services economy to make everyone happy, including service
providing services. providers and customers happiness to of the. of the architecture represent the design of the system
receive essential service at their convenience is a major factor and mobile application of the Jaipur service based company.
of its success. The home services connect the local services to
the consumer while maximizing the benefits of a technology
with the cost efficient and less complicate manner With the
cost efficient method. Now the problem of finding a local
worker to help in their problem is very efficient with the
Jaipur service based company. Jaipur service company is the
IOS and app based application which is Which is commonly
used in Iphone And Smart phone Providing efficiency to the
human. it is designed. by HTML, CSS Javascript, angular
framework, Ionic framework, and firebase for its backend
efficiency.

III. PROPOSED SYSTEM

Jaipur Service company is an integrated app base service
marketplace that connects a customer to the service
professional. The People or the person Which participate in
the whole process is. consist of worker and a client The IV.II Use-Case Diagram
administration of the Jaipur service based company have
access to manage a customer and the worker. Jaipur services Use Case diagram provide a high level overview of the
company is used in. both iphone and a smartphone app. First, functionality of the system. This. system which indicate
a user can download this Jaipur service company application how the user can interact with the system through various
use cases. to achieve its goal and target the use case
from the play store and the ios store. Then a login page can diagram also help in to identify clarify and the organize
open in which a user can log in through their mobile number. the system requirement by visualizing the interaction
The user can receive an OTP to confirm the identity of. its between users and the system. The use case diagram of a
number then it should be redirect on the login page. where Jaipur service company has four basic part which include
using can register itself. by a three simple steps. that is actor actor , usecase , system boundary and interaction. In
registered their name. the agenda and its current address. this diagram, a actor represents. a customer interacting
Jaipur service ensure that the address. which was entered by system. It’s an visual representation of the whole Jaipur
service company model.
the user. is genuine. so it can help it out. in very efficient way.
The administration has also. verify all the details. of the user.
The administration of a access. over the customer and the
worker system ported. in the Jaipur company. The major
categories of service is cleaning, painting and installation. A
user can select according to their needs. so that the worker
can help it down. at the minimum cost acquisition price for
that system. A user can select according to their need. What
he and she want to do. at their apartment Firstly, the odd
consultant team. come at the customer home. or address? to
respect the area. The whole charges for this inspection is
minimum. After the inspection. they tell the charges in the
appointment date. of the worker. A user can add. all the
services which he and she want to do at your home. user can
add all the services at the cart and book now with the
minimum wages. The description and the type features is also
available at the cart so a user can explain at efficient way to
the consultant people. after the user can buy the consultation
of the Jaipur Service Company A company will send a
consultant people to the user home . The whole process is
confirmed by the vacant team and send its consultant as soon
as possible.

IV. MPLEMENTATION
The implementation of a Jaipur service company design
is a user friendly. so that a customer and a worker can
use according to. their work. the whole formal
description and the representation of system is designed
These encounters do not only affect the suffering of people and
wildlife who are constantly affected by conflict. It can also
have a global reach, as groups akin to sustainable
development agencies and businesses are feeling the residual
goods. The problem compass is important and truly global,
but it cannot be addressed on the scale needed [2].

A. Why is human-animal conflict on the rise?

As the human population and demand for space continue to

grow, humans and wildlife will increasingly interact and
compete for resources, potentially increasing human-wildlife
conflict [2].
Along with other threats, human-wildlife conflicts are
causing the decline of many species and pushing others
to the brink of extinction. These conflicts can arise from many
reasons such as;
• Deforestation due to agricultural expansion and
human settlement.
• Increase in livestock and an increase in poaching
which serves as a good source of prey for wild
animals such as Tigers, Leopards, and Bears.
• Climate Changes also affect human-wildlife conflict
via droughts, floods, forest fires, and conflict over
food and space among the predators.

B. Who is affected by human-wildlife conflict and in what

ways?

Wildlife and nearby communities are most directly

affected by human-wildlife conflicts. Such conflicts may
lead to the loss or potential extinction of species in this
region or parts of the region, but communities will suffer
economic loss and loss to health and safety, livelihoods,
food security, and property. “The burden of coexisting
with wildlife is not distributed equally, as certain
communities often encounter systemic obstacles such as
low incomes and limited economic opportunities. In
contrast, healthy wildlife populations and ecosystems are
crucial for the survival and sustainability of global
 communities. This disparity places an unjust and Furthermore, the model integrates the CCTV camera
disproportionate burden on communities that are already using RTSP and delivers the detections to the rest API in the
marginalized and struggling to make ends meet” [2]. JSON file format or telegram port, which makes the process
highly configurable and easy to use. The resulting video can
The problem of human-wildlife conflict is a critical be configured as per the user's needs, and the frequency of
issue in many regions, and it requires effective solutions to different categories of animals can be counted to provide
mitigate its impacts. One potential solution is the use of more insights.
machine learning and deep learning algorithms to detect
and monitor wildlife before they can cause harm to humans Several key points should be highlighted, including the use
or farmlands. This paper aims to implement such models of various machine learning and deep learning
for detecting wildlife, with a focus on elephants and tigers. algorithms, the implementation of the model on different
types of wildlife, and the use of RTSP for CCTV camera
The research paper proposes using deep learning integration. The integration with the telegram port or API
algorithms that can automatically pick up images from server to send encoded results in the JSON file format
the dataset in order to accomplish the goal. The You for clarification of animal presence is also critical.
Only Look Once (YOLO) algorithm is used for object
recognition; YOLOv5 is the most recent version, and Overall, this study offers a potential method for
it has produced desired results. The Proposed model employing deep learning algorithms to identify wildlife,
includes multiple processes, including class selection, which can lessen conflict between people and animals by
dataset gathering from open sources, image annotation, identifying animals early and enabling prompt intervention to
image preprocessing, data augmentation, model building, avoid unwanted interactions. Farmers who will use it can also
IOU implementation, and configuration of the code to protect their crops and avoid financial damages. Digital fence
run on any GPU accelerator device. can also lessen the need for physical barriers and animal
relocation, which can be expensive and logistically difficult.
A. Key Points
The rest of the paper is organized as follows: Section II
1. Learn about various Machine Learning and Deep discusses the literature review followed by Section III, which
learning algorithms. explains the proposed methodology; Section IV discusses the
2. To implement the model on various types of results and analysis. Finally, section V concludes the
wildlife, especially Elephants and Tigers. research paper and provides direction for future work.

3. Use of Deep Learning Algorithms to automatically II. LITERATURE REVIEW

learn the features of the images from the dataset.
This literature review discusses the growing human-
4. Use RTSP for CCTV camera integration with the animal conflicts and the demand for monitoring and
python based deep learning model for live safeguarding animal welfare. It suggests using deep learning
detection. and machine learning algorithms to identify wildlife and stop
possible threats to people and farms. The assessment
5. Integration with the telegram port or API server to
emphasizes the factors that lead to conflicts between people
send encoded results in the JSON file format for
and wildlife, including deforestation, an increase in
clarification of the animal presence.
livestock, poaching, and climate change, as well as the
6. Configure the resulting video as per the needs and effects of these factors on animals and communities;
count the frequency of the different categories of including damage to economies, safety, livelihoods and food
the animals. security. It is dangerous and may violate animal welfare
rights to monitor and control animals using conventional or
coercive ways. Consequently, using sensors, cameras, GPS,
B. Proposed model for the problem drones, unmanned aerial vehicles, and ground vehicles can
The steps for creating this model include Class Selection improve physical security, human property protection, and
of the interested animals, Image Collection from open monitoring of animal welfare. The review addresses several
sources, Image Annotations, image preprocessing, data techniques for tracking animals, including RFID, GPS, and
augmentation, model building, IOU implementation and, drones with cameras, and it focuses on their drawbacks and
finally, the configuration of the project code to be executed on difficulties. In order to detect wildlife and avert potential
any GPU accelerator device. The collected images were danger to people and farmlands, the review's conclusion
annotated in the YAML format and fed into the YOLO v5 for suggests using machine learning and deep learning
training purposes. The finished model has been algorithms, specifically the YOLO object detection
configured to be used in GPU accelerators that can work technique. The suggested model includes picking animal
with Python along with the integration of the camera using classes, gathering image datasets, annotating them, data
RTSP and delivering the detections to the rest API in the preprocessing, enhancing the data, developing models and
JSON file format or telegram port as required by the user. setting up the code to run on any GPU accelerator device.
Depending on the user's preferences, the completed model
can be coupled with a camera and provide detections to the After collecting and preparing the dataset as discussed
REST API in the JSON file format or telegram port. earlier, we passed the dataset images into the YOLO v5
custom object detection model. “YOLO v5 is the latest
Using Wireless Sensor Networks (WSNs), digital version of the YOLO (You Only Look Once) algorithm,
fencing is a novel approach to resolving conflicts between which is a popular and widely used deep learning algorithm
people and wildlife. Devi and Madheswaran created the for object detection tasks.”
WSN-based technology known as "Smart Fencing" to
monitor and manage animal invasions. The system detects YOLO v5 is a family of compound-scale object detection
the presence of animals and alerts the control unit using models that have been trained on the COCO (Common
a combination of infrared and ultrasonic sensors. A deterrent Objects in Context) dataset. It includes simple functionality for
device, such as an alarm or sprinkler, is then turned on by the Test Time Augmentation (TTA), which helps to improve the
control unit to prevent animals from accessing the accuracy of object detection by generating multiple
protected area [3]. versions of the input images during testing. Additionally, it
also includes functionality for model assembling,
Digital fencing was proposed by Kumar and Sivakumar as hyperparameter evolution, and export to ONNX, CoreML,
a long-term approach to wildlife management. In order to and TFLite, which makes it a very versatile and flexible
encourage cohabitation between people and animals, the deep learning model for object detection tasks.
study emphasized the significance of striking a balance
between human activity and conservation efforts. The The YOLO v5 model was trained on our custom dataset,
authors addressed the various digital fencing technologies which consisted of images of different categories of
and their uses in managing wildlife, including Radio- animals. The model was trained using transfer learning,
Frequency IDentification (RFID), Global Positioning which is a popular technique in deep learning that involves
systems (GPS), and WSNs [4]. using a pre-trained model as a starting point and fine-tuning it
on a new dataset. This allowed us to leverage the
Comita and Robbins reviewed the most recent knowledge and features learned by the pre-trained model
technological advancements and how they were used in and adapt it to our specific animal detection task. Fig 1,
wildlife management. The study explored the possible Shares the proposed model of the paper.
advantages and disadvantages of digital fencing technology,
including potential effects on animal behavior, decreased
human-wildlife conflict, and better safety for both people
and wildlife [5].

Kahuga and Muraya evaluated the efficiency of digital

fencing in minimizing the conflict between people and
wildlife in Kenya. The authors investigated on how digital
fencing affected elephant behavior and evaluated how well the
device functioned to keep elephants out of agricultural
fields. The financial benefits of digital fencing as a low-cost
strategy to lessen conflict between people and wildlife were
also addressed in the paper [6].

Overall, digital fencing has the potential to be a successful and

long-lasting method of resolving conflicts between
people and wildlife. The use of physical barriers can be
diminished thanks to technology, which can also allow real-
time monitoring and management of animal behavior.
However, more study is required to assess the effectiveness
over the long run and the potential adverse effects of digital
fencing on animal behavior and ecosystem dynamics.

Fig. 1. Flow Chart of the model.

III. PROPOSED METHODOLOGY
The methodology for digital fencing is divided into six
phases namely data collection phase, data augmentation
A. Data Collection
phase, Image Annotation and Acquisition phase, Model
training and detection processing phase, IOU
implementation phase, and Integration with the Camera The dataset preparation for animal detection is a crucial part of
phase. Fig. (1) depicts the research approach, which we any deep-learning project. In our project, we made sure to use
followed for digital fencing. a well-balanced dataset of images for training, validation
and testing purposes. As mentioned earlier, there are various
ways to collect data and create a dataset. We
chose to collect our dataset from open-source images, video RoboFlow being a web-based tool of annotation supports
frame extractions and available datasets. various formats for annotation such as YOLOv5,
TensorFlow, and PyTorch. It also provides several other
Our dataset had approximately 10,000 images of seven useful features such as data augmentation, dataset splitting,
different categories of animals that we focused upon, and export functionality. This tool being designed to
including bear, wild boar, tiger, leopard, elephant, deer, and streamline the annotation process for their dataset helps the
macaque. We wanted to ensure that the dataset was well data scientists and developers train their deep learning
balanced, so we included approximately 1300 images of models more quickly and more efficiently.
each category of animal. The dataset was divided into three D. Model Training and Detection Processing
parts - the training set, validation set, and testing set - in a ratio The next step was to pass the processed dataset to the
of 7:1:2 respectively. This allowed us to train our models YOLOv5 for the custom training of the model. The next
effectively and also evaluate the performance of our models challenge was to create a debugging video that required
on new and unseen data. counting of detections. The detection alarms and notification
cannot be based on a single frame. So, we used multiple
To simulate the night vision of the cameras, we also used frame check, in this the frames are counted for the detection of
grayscale images in our dataset. This was important as we the same animal and after the threshold is met, the
wanted to make sure that our models could work effectively in
notification or the alarm goes on to the data-sharing
low-light conditions as well. The dataset was required to be platforms.
balanced and unbiased. The number of day and night
images were kept of necessary amount to make sure that our One of the biggest challenges we faced during the data-
model works perfect in both day as well as night when the
sharing process was how to transmit the detection image to a
animal attacks are prominent.
centralized checking system. We could not send the
The balancing of the dataset was a very important task as it is detection image directly as it would require large amounts
mandatory to train the model equally for all the selected of bandwidth and storage. To overcome this challenge, we
classes. There should be generalization and equality in
came up with the idea of converting the detection image into a
images for an unbiased classification. This was necessary
JSON file format along with other properties of the
for the proper classification of animals that look similar in detection.
person and also that boosted our precision and recall value.
B. Dataset Augmentation The JSON file included information such as the class
The open source images, camera video clips of animals and detection coordinates, camera IP, and an image of the
openly available images of the animals were the major detection. This format allowed us to transmit the
source for our model’s training, testing, and validation. The information quickly and efficiently through various
major problem with these images was their raw nature and platforms, including REST APIs.
unclear motive. Most of these images were not fit for the E. IOU Implementation
training purpose or were of too low quality. The images
We could not afford multiple notifications of the same
were then to be scrapped out. The quality images were left to
detection. Hence, we used the concept of IOU that marks the
around 10,000, 1300 for each class. The images were than
animal as the same till it is in the frame so that it does not
annotated according to the required format of annotations
awakes the system again. IOU if calculated on each
for YOLOv5. The annotation part was done through the
successful detection to the previous frame detections do that
Roboflow software. This was the most crucial, time
the warning signals are legit and non-repetitive. The
consuming and important part of the whole project as the
threshold value of the IOU can be modified according to the
quality of dataset determines the working of the model in requirement.
the long run. Open sources could only give us raw images but
without data refining and image annotation the training was
The code to count the frames of the detection was self-
not possible. So, the dataset was then converted to the
developed to ensure specificity in the program and to
annotation format of the algorithm used. prevent any side cases. This ensured that the animal was
C. Image Annotation and Acquisition really present at that place for a certain amount of time. This
Even after the refining of the images, without the proper was important to check before sending out a warning signal.
tool for annotation, it is practically very difficult to annotate a One can also set the no. of frames to be checked before the
whole dataset of 10,000 images. Thus, we used Roboflow for waring signal.
the annotation process. “Roboflow is a Computer Vision
developer framework for better data collection to Overall, the multiple warning problem or the single frame
preprocessing, and model training techniques.” Annotation warning problems were addressed by the above techniques
includes labelling of the selected object in a bounding box used. There will be no warning on the basis of a single
that gets a class label. This label is the actual way that the deep frame detection and if the animal occurs in multiple frames, it
learning learns and distinguish the classes. Using roboflow will be noted as only one warning.
was a great advantage for us and it helped us manage the F. Integration with the Camera
dataset easily.
RTSP protocol is a protocol that is used mainly for the
connection of video surveillance systems to the computer
device, as live video stream is transmitted directly to the equal to 0.434. A further increase in the accuracy of the
computer for live detection or other purposes. User are able to results can be gathered by the refinement of the dataset and
access the cameras remotely from anywhere and any increasing the size and quality of the dataset.
devices, being fixed or mobile.
After obtaining the IP address of the camera for the To further improve the accuracy of our proposed model, we
connection purpose to our systems, we connected the can use transfer learning. Transfer learning is a machine
camera using RTSP so as to avail all the available use cases of learning technique where a pre-trained model is used as
the above-mentioned method of connection. The next step a starting point for a new task. It allows us to take advantage
was port setup and creating username and password for the of the pre-existing knowledge gained by the model from a
camera device. different task and apply it to a new problem.

RTSP protocol, unabled easy integration of the camera with We can also experiment with different hyperparameters,
the system. We used the OpenCV library in Python to read the such as learning rates, batch sizes, and optimization
video stream from the camera and passed the frames to our algorithms to further improve the accuracy of our model.
object detection model for real-time detection of Another approach to improve accuracy is by using ensemble
animals. methods where multiple models are combined to give a
more accurate prediction. One of the hyperparameters is the
By using the RTSP protocol, we were able to seamlessly confidence interval of the given prediction. Fig. 2 shows the
integrate a wide variety of cameras into our system, making it different accuracy of the prediction cum detection results
highly scalable and adaptable to different environments. according to the different values of the confidence interval.
This also allowed us to remotely monitor and control the
cameras, enabling us to quickly respond to any potential
threats or incidents.

IV. RESULTS AND ANALYSIS

In this proposed paper, we have considered around 1500

images of each animal species. We divided the whole
dataset into various sections, some data for the training
phase and some for the testing phase. Various Machine
Learning and Deep Learning models were applied to the
datasets such as Yolo v5.
Table 1 shows the training results after 50 epochs.

Class Images Instances P R mAP50 mAP50-

95
All 1986 2489 0.872 0.858 0.904 0.584
Bear 1986 326 0.874 0.837 0.905 0.589
Deer 1986 248 0.86 0.895 0.91 0.689
Elephant 1986 477 0.826 0.883 0.909 0.671
Leopard 1986 372 0.938 0.919 0.961 0.568
Monkey 1986 316 0.867 0.794 0.869 0.496
Tiger 1986 333 0.942 0.871 0.932 0.645

Wild 1986 417 0.8 0.804 0.845 0.434

Boar

As shown in Table 1, we got fair results, the combined all

classes' precision was 0.872 with a recall value of 0.858
making our mAP over confidence value 0.5 equal to 0.904 and
that with the confidence interval of 0.5:0.95 equal to
0.584. The class with the highest mAP at the confidence
interval of 0.5 is the Leopard class having an mAP equal to
0.961. The class with the highest mAP at the confidence
interval of 0.5:0.95 is the Deer class having an mAP equal to
Fig. 2 Change of the F1 score per the confidence value. The
0.689. The class with the least mAP at the confidence
interval of 0.5 is the wild boar class having an mAP equal to
0.845. The class with the highest mAP at the confidence F1 score of the model increases rapidly till a confidence
interval 0.5:0.95 is also the wild boar class having an mAP value of 0.15 and then reaches the max of 0.86 at 0.395 and
then degrades down rapidly to almost 0 with a
confidence
value of 1. There are two factors of the f1 score i.e.,
precision and recall. Let’s now study them uniquely. Down
below fig.3 and Fig. 4 show the relation of the confidence
value with the precision and recall values respectively.

Fig. 4 Relation between the recall and the confidence value

of the model.

As we can see that the precision of the model initially

increases rapidly till the confidence value reaches 0.17 and
Fig. 3 Relation between the precision and the confidence then increases gradually till the end besides a gradual dip in
value of the model. the three classes of the animals, namely, Deer, Monkey and
Wild Boar. All the classes' precision reaches 1 at a
confidence value of 0.923. In Fig. 4, the recall is at a
maximum value of 0.96 at a confidence value of 0.00. Then it
starts to decline at a slow rate initially and then rapidly. The
recall value reaches almost 0 at the confidence value of
approximately 1. Let’s now study the correlation between
precision and recall. Fig. 5 shows the precision-recall curve of
the model.
 Fig. 6 Confusion Metrix - Correlation between the animal
classes.
According to the above confusion matrix, we see that there is
merely any confusion between the different classes
whereas we can see major instances where the background
is detected as elephant and wild boar classes more often.
This can be due to the camouflage of the animals and can be
reduced by refining the images.
Fig. 5 Precision-Recall curve of the model at confidence
value 0.5. Furthermore, we can also consider augmenting the dataset
with more images of different angles and lighting
The above curve clearly shows that precision-recall shares conditions, which will make our model more robust to
an inverse relationship with each other. As soon as the recall variations in the real-world scenario.
is increased, the precision of the results got hampered. The
precision of the results can be compromised till the mAP of all In addition, we can also explore other object detection
classes reaches 0.904 at the edge of the curve and then again models such as EfficientDet, RetinaNet, and Mask R-CNN to
starts to reduce by the increase in the recall value. We now see if they give better results for our use case.
will study the correlation cum similarities in the
appearances of the classes that we worked upon. Down
below fig. 6 shows the apparent similarity between the
classes.
 Overall, by using a combination of transfer learning,
hyperparameter tuning, ensemble methods, dataset
augmentation, and exploring different object detection
models, we can further improve the accuracy of our
proposed model for wildlife conservation.
V. Conclusion and Future Scope
n this research paper, we proposed a digital
I fencing echanism using a Deep Learning algorithm to
mtackle the allenges of physical fencing such as being
chtime- nsuming, expensive, unmaintainable, etc. To
co be articular, we designed a ready-to-use version of yoloV5
pas object detection technique for the classification of the
animals that came in a certain region covered by the
anmeras. Thus, our approach did not affect animal rights as ey
ca were not being continuously tracked like the GPS and ther
th methods of prevention unless they appeared in a
o estricted area. To implement this work, we created our own
rataset of 10,000 images belonging to seven different
d tegories of animals. As a result, we achieved classification
ca ith a precision of 0.872, recall of 0.858, [email protected] of
w.904, and [email protected]:0.95 of 0.584.
0 ture work in this area would involve designing a more
Fuvanced version of the YOLO algorithm with an even large
adataset containing more no. of real-time quality images that ay
dbe provided by the Forest Department of India itself.
m
REFERENCES
1] A. Herlin, E. Brunberg, J. Hultgren, N. Högberg, A. Rydberg, and A.
Skarin, “Animal welfare implications of digital tools for monitoring
[
and management of cattle and sheep on pasture,” Animals, vol.
11, no. 3. MDPI AG, pp. 1–20, Mar. 01, 2021. doi:
10.3390/ani11030829.
2] “What is human-wildlife conflict and why is it more than just a
[
conservation concern?” (no date) World Wildlife Fund. Available at:
crossref.
3] S. Nirmala Devi and M. Madheswaran, "Smart Fencing: A Wireless
[
Sensor Network for Monitoring and Control of Animal Intrusion," in
IEEE Sensors Journal, vol. 18, no. 2, pp. 755-761, Jan.15, 2018, doi:
10.1109/JSEN.2017.2771695.
4] A. Kumar and S. S. Sivakumar, "Digital Fencing: An Innovative
[ Approach for Sustainable Wildlife Management," in IEEE Potentials,
vol. 37, no. 1, pp. 16-21, Jan.-Feb. 2018, doi:
Fig. 7 Shows different important graphs of the model 10.1109/MPOT.2017.2766163.
training. [5] L. S. Comita and P. A. Robbins, "Digital Fencing: A Review of
Current Technologies and Their Applications in Wildlife
Management," in IEEE Technology and Society Magazine, vol. 37,
The first graph shows the box loss reduction upon the no. 2, pp. 48-57, Summer 2018, doi: 10.1109/MTS.2018.2830623.
training dataset after simultaneous epochs with the [6] H. M. Kahuga and M. M. Muraya, "Evaluating the effectiveness
minimum box loss resulting to be 0.03 at the approximate of digital fencing in reducing human-wildlife conflict in Kenya,"
last epoch. The second graph shows the declining nature of the in IEEE Access, vol. 7, pp. 106780-106788, 2019, doi:
10.1109/ACCESS.2019.2931216.
object loss upon training the dataset that reached the [7] Leuven, K.U. (no date) “Automated Farming: Virtual Fencing,”
minimum of 0.016 at last. Graph 3 shows a similar behavior of Future Learn. Available at:
the class loss upon the training data set with the minimum https://www.futurelearn.com/info/courses/revolutionising-the-food-
reaching the value of 0.01. The next graph shows the chain/0/steps/170931
increasing precision of the model with the epochs followed by [8] R. A. Søraa and J. Vik, “Boundaryless boundary-objects: Digital
fencing of the CyborGoat in rural Norway,” J Rural Stud, vol. 87, pp.
the increasing recall graph of the model training. The 23–31, Oct. 2021, doi: 10.1016/j.jrurstud.2021.08.015.
lower row of the graphs in Fig. 7 shows similar three graphs [9] A. Muminov, D. Na, C. Lee, H. K. Kang, and H. S. Jeon, “Modern
of the box loss, Obj_loss, and the cls_loss with the Virtual Fencing Application: Monitoring and Controlling Behavior of
validation dataset. The last two graphs show the increasing Goats Using GPS Collars and Warning Signals,” Sensors (Basel), vol.
19, no. 7, Apr. 2019, doi: 10.3390/S19071598.
mAP scores with confidence intervals of 0.5 and 0.5:0.95
respectively.