WASTE SEGREGATION.

A CAPSTONE PROJECT REPORT

Submitted in partial fulfillment of the

requirement for the award of the
Degree of

BACHELOR OF TECHNOLOGY
IN
COMPUTER SCIENCE

by

UJJWAL REDDY K S (20BCI7203)

VISWATEJA CHITTURI (20BCE7237)
YOGESHWAR KOTHA (20BCN7091)

Under the Guidance of

DR. P KUPPUSAMY

SCHOOL OF COMPUTER SCIENCE AND ENGINEERING

VIT-AP UNIVERSITY
AMARAVATI- 522237

NOVEMBER 2023
 CERTIFICATE

This is to certify that the Capstone Project work titled “WASTE SEGREGATION”
that is being submitted by UJJWAL REDDY K S (20BCI7203),VISWATEJA

CHITTURI (20BCE7237), YOGESHWAR KOTHA (20BCN7091) is in partial

fulfillment of the requirements for the award of Bachelor of Technology, is a record

of bonafide work done under my guidance. The contents of this Project work, in full
or in parts, have neither been taken from any other source nor have been submitted
to any other Institute or University for award of any degree or diploma and the same
is certified.

Dr. P KUPPUSAMY
Guide

The thesis is satisfactory / unsatisfactory

Internal Examiner External Examiner

Approved by

PROGRAM CHAIR DEAN

B. Tech. SCOPE School Of Computer Science and Engineering
1
 ACKNOWLEDGEMENTS

We would like to express our heart full gratitude to Dr. P Kuppusamy for his
support and contribution to our project. Without his guidance we may not complete
this project. His guidance always gave us confidence. He gave true full reviews
when we explained to him regarding this project. He gave me confidence to
complete this project. Programme chair gave us the chance to do this project to
improve our knowledge.

2
 ABSTRACT

Garbage management is critical for environmental sustainability, and the development of

better computer vision models has opened up new opportunities for automating garbage sorting
operations. In this research, we give a thorough investigation into the use of the YOLOv8 model
for trash classification into three separate categories: biodegradable, non-biodegradable, and
recyclable. To achieve accurate and efficient categorization, we train the YOLOv8 model on a
varied collection of garbage photos. We achieved a stunning 90% classification accuracy,
indicating the model's resilience in discriminating between different categories of waste items.
The training method included fine-tuning the pre-trained YOLOv8 model, optimizing
hyperparameters, and tackling waste classification issues. In addition, we successfully deployed
the trained YOLOv8 model in real-time settings, effortlessly integrating it into both a website and
a mobile application. This implementation enables real-time trash categorization at the point of
disposal, allowing users to make more informed decisions about correct waste disposal methods.
The real-time implementation improves the usefulness and practicality of our trash categorization
system, which contributes to the larger objective of encouraging sustainable waste management
practices. Our findings indicate that using YOLOv8 for waste categorization not only produces
high accuracy but also provides a feasible option for real-time waste management applications.
This study adds to continuing efforts to develop intelligent systems that assist effective garbage
sorting and contribute to a more sustainable and environmentally conscious future.

3
 TABLE OF CONTENTS

S.No. Chapter Title Page Number

1. Acknowledgement 2
2. Abstract 3
3. List of Figures and Table 5
4. 1 Introduction 6

1.1 Objectives 7

1.2 Background and Literature Survey 7

Organization of the Report
1.3
8
5. 2 WASTE SEGREGATION 10

2.1 Proposed System 10

2.2 Working Methodology 11

2.3 Standards 11

2.4 System Details 12

2.4.1 Software Details 12

2.4.2 DataBase 13

2.5 Connecting Dataset 14

2.5.1 Connecting with ultralytic hub 15

2.5.2 Deployment 16

2.5.3 Website deployment 17

6. 3 Results and Discussion 19

8. 4 Conclusion & Future Works 21

9. 5 Appendix 22
10. 6 References 25

4
 List of Table

Figure Title Page No.

No.
1 Result 21

List of Figures

Figure No. Title Page No.

1 Waste Segregation Categories 7

2 System Block Diagram 11

3 Google colab 14

4 Dataset 15

5 Connecting dataset 15

6 Connecting ultralyticHub 16

7 Ultralytic Hub 16

8 Ultralytic Hub MobileApplication 17

9 Ultralytics Hub Website 18

10 Result 19

11 Metrix plot 19

12 Training plot 20

13 Validation plot 20

14 Result Plots 21

CHAPTER 1
INTRODUCTION

5
 Rapid urbanization and population expansion have resulted in an unprecedented increase
in worldwide garbage creation, causing severe environmental issues. As the globe grapples with
the implications of inappropriate garbage disposal, creative solutions to limit environmental
effects are critical. The merging of artificial intelligence (AI) and computer vision technology
appears to be a viable route for transforming waste management techniques in this regard.

Figure 1 Waste Segregation Categories

Manual labor is typically used in traditional garbage sorting procedures, which is both
time-consuming and prone to mistakes. The introduction of deep learning and powerful computer
vision models has created a game-changing potential to automate and improve trash
categorization operations. The application of the YOLOv8 (You Only Look Once version 8)
paradigm to the critical issue of garbage categorization is the topic of this study, with a particular
emphasis on classifying waste into three separate classes: biodegradable, non-biodegradable, and
recyclable. Figure 1 depicts the various forms of waste segregation, laying the groundwork for the
forthcoming investigation of our technique and outcomes.

The YOLOv8 model, which is well-known for its speed and accuracy in object
identification tasks, provides a solid framework for real-time trash categorization. The goal of this
project is to use YOLOv8's capabilities to achieve high-precision waste classification, allowing

6
for a more efficient and sustainable waste management system. Our research intends to contribute
to the increasing corpus of research that uses artificial intelligence for environmental applications,
with an emphasis on practical implementation and real-world deployment.
In the following sections, we will look at the methodology used to train the YOLOv8
model, the composition of our diverse waste dataset, the optimization processes used to achieve
high accuracy, and the model's seamless integration into real-time waste management scenarios
via a website and a mobile app. We hope to illustrate the potential of YOLOv8 in tackling modern
waste management difficulties, opening the road for smarter, more automated, and
environmentally conscientious trash disposal procedures through this research.

1.1 Objectives
The following are the objectives of this project:
● Model Training: Create and train a YOLOv8 model specifically designed for exact trash
classification, assuring excellent performance in discriminating between biodegradable,
non-biodegradable, and recyclable materials.
● Dataset Compilation: Create a diversified and comprehensive dataset that includes a wide
range of waste items to improve the model's capacity to generalize and properly categorize
numerous real-world circumstances.
● Systematically fine-tune the model's hyperparameters to obtain maximum accuracy while
balancing efficiency and reliability in waste classification activities.
● Deployment in Real Time: Integrate the trained YOLOv8 model easily into a website and
mobile application to provide real-time garbage categorization at the disposal point,
allowing for immediate and informed decision-making.
● Sustainability Impact: Validate the model's performance in real-world waste
management contexts, with the overriding objective of developing environmental
consciousness among end-users and contributing to sustainable waste disposal practices.

7
1.2 Background and Literature Survey

The growing issues of global waste management have spurred academics to investigate
novel solutions based on advances in artificial intelligence (AI) and computer vision. Effective
waste segregation is a vital component of contemporary waste management, since it increases
recycling rates while minimizing environmental damage. This section summarizes the available
research, highlighting major contributions and insights from recent investigations. Al Rakib et al.
presented a "Dry and Wet Waste Segregation and Management System" [1], highlighting the
significance of categorizing garbage. By addressing the requirement for effective waste sorting
systems, their work provides the groundwork for contemporary study. Similarly, Gupta et al. [2]
presented a "Automatic Waste Segregation" system, defining waste management instrumentation
and control engineering concepts. This research provides important insights into the automation of
waste segregation operations.
Susanth et al. presented "Garbage Waste Segregation Using Deep Learning Techniques"
[3] in the field of deep learning techniques, illustrating the application of deep learning for waste
categorization. Their research looks at the possibilities of modern technology to improve trash
segregation accuracy. Furthermore, in "A Novel YOLOv3 Algorithm-Based Deep Learning
Approach for Waste Segregation" [4], Kumar et al. developed a YOLOv3 algorithm-based
strategy, demonstrating the usefulness of YOLOv3 in the context of smart waste management.
Narayanswamy et al. investigate the junction of computer vision and trash management in
"Development of Computer Vision Algorithms for Multi-class Waste Segregation and Their
Analysis" [5]. This work adds to our understanding of multi-class waste segregation by shining
light on the problems and possibilities in this rapidly expanding industry.
We want to use the YOLOv8 model to enhance our existing understanding of trash
classification as we continue our study, building on these basic findings. Our method not only
coincides with prior works' objectives, but also aims to improve real-time waste management
through the seamless integration of our model into practical applications such as websites and
mobile apps.

8
1.3 Organization of the Report

The remaining chapters of the project report are described as follows:

● Chapter 2 contains the proposed system, methodology, hardware and software details.
● Chapter 3 discusses the results obtained after the project was implemented.
● Chapter 4 concludes the report.
● Chapter 5 consists of codes.
● Chapter 6 gives references.

9
 CHAPTER 2

WASTE SEGREGATION
This Chapter describes the proposed system, working methodology, software and hardware
details.

2.1 Proposed System

The following block diagram (figure 2) shows the system architecture of this project.

Figure 2 System Block Diagram

10
2.2 Working Methodology
The study utilized the YOLOv8 model for waste classification using a Colab notebook and
GPU T4 runtime. The training dataset consisted of 2040 images divided into training, validation,
and test sets. The model was fine-tuned to accurately classify waste into three categories:
Biodegradable, Non-Biodegradable, and Recyclable. The division of the dataset allowed for
robust model evaluation and generalizability beyond the training data. To streamline the
workflow, the Ultralytics hub beta version was used to deploy and execute the YOLOv8 model on
the notebook. The model's computational power was harnessed through the Ultralytics hub,
achieving high accuracy in waste classification.
The trained YOLOv8 model was successfully deployed on a website and mobile
application, demonstrating its practicality for real-time waste management scenarios. This
approach not only ensures accurate waste classification but also highlights the potential for wider
adoption of AI-driven solutions in promoting environmental sustainability. The methodology
highlights the effectiveness of advanced computer vision techniques in addressing contemporary
waste management challenges.

2.3 Standards
Various standards used in this project are:
Image Annotation Standards:
● To maintain consistency and interoperability in labeling the dataset, use
standardized picture annotation formats such as COCO (Common Objects in
Context) or Pascal VOC (Visual Object Classes).
Model Training Best Practices:
● Following known procedures for training machine learning models, taking into
account TensorFlow and PyTorch framework concepts, and following to
community-accepted hyperparameter tuning standards.
Accuracy Assessment Standards:
● Using common measures like precision, recall, and F1 score to assess the
accuracy and effectiveness of the YOLOv8 garbage categorization model.

11
Environmental Impact Considerations:
● Adhering to rules for reducing the environmental effect of computing resources,
particularly during the training phase, by optimizing code, using energy-efficient
hardware, and considering the model's carbon footprint.
Deployment and API Standards:
● Using RESTful API standards for interaction between the trash categorization
model and the website or mobile application, following industry norms for
deploying machine learning models.
Documentation Standards:
● Following clear and thorough documentation standards to ensure that the
project's codebase, model architecture, and deployment methods are
well-documented for future reference and collaboration.

2.4 System Details

For efficient processing, the trash categorization model was trained on a Colab laptop with GPU
T4 runtime. The features of the system for training the trash categorization model include a
synergistic combination of Colab, the YOLOv8 architecture, and the Ultralytics hub. This mix of
tools and frameworks, aided by GPU acceleration, demonstrates our approach's efficiency and
scalability. The next sections dig into the complexities of model optimization, assessment, and
deployment, demonstrating a comprehensive system designed to handle waste management
difficulties using cutting-edge technology.

2.4.1 Software Details

i) Google Colab
Google Colab allows anybody to write and execute arbitrary python code through the browser,
and is especially well suited to machine learning, data analysis and education.
Google Colab is simply an online representation of Jupyter Notebook. While Jupyter Notebook needs
installation on a computer and can only use local machine resources, Colab is a full-fledged cloud app
for Python coding. You can write Python codes using Colab on your Google Chrome or Mozilla
Firefox web browsers.

12
 Figure 3 Google Colab

ii)DataBase
The dataset used to train our trash categorization algorithm consists of a selected collection of
2040 photos depicting various garbage objects in various circumstances. The dataset is divided into three
major categories: biodegradable, non-biodegradable, and recyclable. The dataset is deliberately partitioned
to allow successful model training, with 87.3% dedicated to the training set (1781 photos), 8.5% devoted to
the validation set (173 images), and 4.2% allocated to the test set (86 images). Furthermore, a tiny portion
of the dataset (0.1%) is purposefully left unlabeled to expose the model to instances in which not all
trash items are marked. Each trash item is meticulously labelled using technologies such as LabelImg
and VGG Image Annotator (VIA), assuring uniformity and accuracy.The emphasis is on ethical
issues, privacy rules, and avoiding the inclusion of sensitive information in the photos. Regular quality
checks during the annotation process ensure the dataset's integrity, which is used to train the YOLOv8
model to properly categorise waste items in real-world waste management settings.

13
 Figure 4 Dataset

Connecting with Dataset

• After adding the screens, we will add buttons and create an authentication key.
• After appropriate authentication it is linked to the Google Firebase.

Figure 5 Connecting Dataset

14
Connecting with Ultralytic Hub

Figure 6 Connecting Ultralytics HUB

The Ultralytics HUB interface streamlines the process of uploading data, training models, and
deploying them to real-world applications. It removes the requirement for coding experience and
makes it simple to begin with object detection jobs. HUB also provides pre-trained models and
bespoke model training. It enables the deployment of trained models to web servers, mobile devices,
and other platforms. HUB provides real-time performance, scalability, and community support. All of
these capabilities combine to create HUB a strong and adaptable machine learning platform for users
who wish to harness the power of object identification and picture recognition without the hassles of
coding.

15
 Figure 7 Ultralytics HUB

iii) Deployment
Mobile Application Deployment

For on-the-go waste classification, our mobile application offers a user-friendly interface with
seamless integration of the YOLOv8 model. Users can capture images of waste using their mobile
devices, and the application sends these images to the YOLOv8 model through the Ultralytics Hub
API. The model processes the images in real-time, and the results are promptly displayed on the user's
mobile device. The mobile application is designed to enhance the accessibility and convenience of our
waste classification system. Whether individuals are at home, in a public space, or managing waste in
a business setting, the app provides a practical solution for immediate waste categorization.

16
 Figure 8 Ultralytics HUB Mobile Application

Website Deployment
The YOLOv8 trash categorization model has been integrated into a dedicated area of our website.
Users can use this function to immediately upload photographs of waste materials using the website's
interface. The Ultralytics Hub API enables the model to process these photos in real-time,
categorizing garbage as Biodegradable, Non-Biodegradable, or Recyclable. The findings are then
given to users, allowing them to make more educated waste disposal decisions. The internet interface
is user-friendly, allowing people, communities, and corporations to easily use the waste classification
capabilities. The incorporation of the YOLOv8 model into the website broadens the scope of our
waste management solution by offering a user-friendly platform accessible from a variety of devices.

17
Figure 9 Ultralytics HUB Website

18
 CHAPTER 3

RESULTS AND DISCUSSIONS

The garbage categorization system, which used the YOLOv8 model and was
implemented via a website and mobile app, achieved a noteworthy accuracy of almost 90% on
a diversified sample of 2040 waste photos. The ability to do real-time inference on
user-uploaded garbage photos was a significant accomplishment. The system's accuracy and
user-friendly interfaces have received excellent comments from users. The generalizability of
the model was secured by various dataset segmentation, and the integration of the Ultralytics
Hub API maintained real-time processing performance. Continuous enhancement efforts are
ongoing, including fine-tuning with more datasets and improving the system based on user
feedback. Finally, the accuracy, real-time processing, and positive user feedback verify the
system's usefulness in encouraging sustainable trash disposal practices.

Figure 10 Sample Result

Figure 11 Metrics Plot

19
Figure 12 Training Plot

Figure 13 Validation Plot

20
 train/ train/ train/ metrics/p metrics metrics/ metrics/m val/b val/c val/d lr/pg lr/pg lr/pg
epoch box_l cls_lo dfl_lo recision(B /recall( mAP50( AP50-95( ox_lo ls_lo fl_lo 0 1 2
oss ss ss ) B) B) B) ss ss ss

0.622 1.915 1.090 0.42509 0.4484 0.38476 0.14193 2.60 2.75 4.35 0.00 0.00 0.00
1 54 7 4 9 53 99 6 0472 0472 0472
08 08 08

0.552 1.264 1.022 0.35276 0.3044 0.26064 0.08532 2.53 3.21 3.96 0.00 0.00 0.00
2 71 9 1 7 99 04 9 0854 0854 0854
52 52 52

0.548 1.08 1.007 0.41515 0.3819 0.33013 0.12042 2.29 2.72 3.36 0.00 0.00 0.00
3 36 9 5 9 47 74 1142 1142 1142
6 6 6

0.496 0.927 0.979 0.36436 0.5959 0.50215 0.24206 2.17 2.29 2.56 0.00 0.00 0.00
4 77 72 64 3 18 05 95 1004 1004 1004
6 6 6

0.460 0.818 0.964 0.60286 0.5671 0.55791 0.24794 2.38 2.21 3.04 0.00 0.00 0.00
5 2 4 39 1 48 92 06 1004 1004 1004
6 6 6

0.418 0.721 0.949 0.36724 0.5305 0.41585 0.17378 2.14 2.25 2.65 0.00 0.00 0.00
6 03 2 82 7 4 57 55 0863 0863 0863
12 12 12

0.368 0.628 0.908 0.51877 0.5939 0.58804 0.28491 2.00 1.97 2.42 0.00 0.00 0.00
7 49 2 85 3 74 1 8 0721 0721 0721
64 64 64

0.332 0.545 0.904 0.67498 0.6219 0.66438 0.3561 1.90 1.84 2.22 0.00 0.00 0.00
8 72 93 91 26 53 0580 0580 0580
17 17 17

0.309 0.490 0.885 0.61574 0.7443 0.73187 0.41674 1.91 1.74 2.30 0.00 0.00 0.00
9 88 53 75 3 71 43 66 0438 0438 0438
7 7 7

0.278 0.423 0.882 0.83368 0.6756 0.79891 0.47213 1.80 1.61 2.21 0.00 0.00 0.00
10 12 69 52 69 91 01 0297 0297 0297
23 23 23

Table 1 Result

Figure 14 Result Plots

21
 CHAPTER 4

CONCLUSION AND FUTURE WORK

Finally, our trash categorization system, built on the YOLOv8 model and delivered via a
website and mobile app, has proven to be a reliable and effective solution for real-time waste
management. The system displays the YOLOv8 model's skill in differentiating between
Biodegradable, Non-Biodegradable, and Recyclable waste categories with an accuracy of roughly
90%. The inclusion of the Ultralytics Hub API guaranteed real-time processing capabilities, increasing
the system's applicability in a variety of waste management scenarios. The website and mobile
application's user-friendly interfaces, together with favorable user comments, highlight the system's
accessibility and usefulness. Our solution helps people, communities, and organizations to make
educated trash disposal decisions by giving rapid results, therefore contributing to more sustainable
and ecologically responsible behaviors. The trash classification system will be continuously improved
and expanded in the future. To increase its flexibility to different waste types and settings, the
YOLOv8 model will be fine-tuned with more diverse datasets. To improve accuracy and efficiency,
various architectural changes to the YOLOv8 model will be investigated. User feedback will be critical
in enhancing the system's interfaces and fixing any usability issues. Efforts will also be made to
improve the system's capabilities, which might include features like automatic garbage sorting or
connectivity with waste management databases. Continued collaboration with environmental
organizations, waste management entities, and community stakeholders will be pursued to ensure the
system remains aligned with evolving waste management needs. The goal is to establish the waste
classification system as a valuable and scalable tool for promoting sustainable waste disposal
practices on a broader scale. Overall, future work will aim to push the boundaries of innovation in
waste management technology while addressing emerging challenges in this critical domain.

22
CHAPTER 5

APPENDIX

23
24
Reference Link:

Code Link:
https://colab.research.google.com/drive/1sw0--DMdRx5lCVnH3D0TI9aNYrxg-G1
s?usp=sharing

DataSet Link:
https://drive.google.com/file/d/16ZYfyvcHo5vhJl4cQDAfP16rUB8b1cjz/view?usp=share
_link
25
 REFERENCES

[1] Md Abdullah Al Rakib, Md. Sohel Rana, Md. Moklesur Rahman, Fysol Ibna Abbas.
"Dry and Wet Waste Segregation and Management System."

[2] Nimisha S Gupta et al. "Automatic Waste Segregation."

[3] G Sai Susanth, L M Jenila Livingston, L G X Agnel Livingston. "Garbage Waste

Segregation Using Deep Learning Techniques."

[4] Saurav Kumar, Drishti Yadav, Himanshu Gupta, Om Prakash Verma, Irshad Ahmad
Ansari, Chang Wook Ahn. "A Novel YOLOv3 Algorithm-Based Deep Learning Approach
for Waste Segregation: Towards Smart Waste Management."

[5] Neeraja Narayanswamy, A. R. Abdul Rajak, Shazia Hasan. "Development of

Computer Vision Algorithms for Multi-class Waste Segregation and Their Analysis."

[6] Ajay, V. P., Kishanth, A., Kumar, Vaishnavi, Devi, R. Santhiya, Rengarajan,
Amirtharajan, Thenmozhi, K., Praveenkumar, Padmapriya, et al. (2020). Automatic waste
segregation and management. In 2020 International Conference on Computer
Communication and Informatics (ICCCI) (pp. 1–5). IEEE.

[7] Agarwal, R., Chaudhary, M., & Singh, J. (2015). Waste management initiatives in
India for human well-being. European Scientific Journal. European Scientific Journal.

[8] Susanth, G. Sai, Livingston, L. M. Jenila, & Livingston, L. G. X. Agnel. (2021).

Garbage waste segregation using deep learning techniques. In IOP Conference Series:
Materials Science and Engineering, 1012(1), 012040. IOP Publishing.

[9] Wadehra, Shivani, & Mishra, Arabinda. (Title: Impact of Interventions on households’
waste disposal behavior: Field Evidence from Delhi)

[10] Sahiledengle, Biniyam. (2019). Self-reported healthcare waste segregation practice

and its correlate among healthcare workers in hospitals of Southeast Ethiopia. BMC
Health Services Research, 19(1), 1-11. BioMed Central.

[11] Srinilta, Chutimet, & Kanharattanachai, Sivakorn. (2019). Municipal solid waste
segregation with CNN. In 2019 5th International Conference on Engineering, Applied
Sciences and Technology (ICEAST) (pp. 1-4). IEEE.

[12] Flores, Myra G., & Tan, Jose. (2019). Literature review of automated waste
segregation system using machine learning: A comprehensive analysis. International
Journal of Simulation: Systems, Science, and Technology, 11.

26
 BIODATA

Name : UJJWAL REDDY K S

Mobile Number : 9845831793
E-mail : [email protected]
Permanent Address : Bangalore, KA

Name : ViswaTeja Chitturi

Mobile Number : 7780545410
E-mail : [email protected]
Permanent Address : Tanuku, AP

Name : YOGESHWAR KOTHA

Mobile Number : 93460 29420
E-mail : [email protected]
Permanent Address : Gosaveedu, AP

27