VISVESVARAYA TECHNOLOGICAL UNIVERSITY

Jnana Sangama, Santhibastawad Road, Machhe, Belagavi - 590018, Karnataka, India

TECHNICAL SEMINAR (18CSS84) REPORT

ON
“Waste Classification and Segregation using Machine Learning and IOT”

Submitted in the partial fulfillment of requirements for the award of the degree of

BACHELOR OF ENGINEERING
IN
INFORMATION SCIENCE AND ENGINEERING
For the Academic Year 2023-2024
Submitted by

AKASH N 1JS20IS010

Under the Guidance of

Mr CHETAN R
Assistant Professor
Dept. of ISE, JSSATE

2023-2024
DEPARTMENT OF INFORMATION SCIENCE AND ENGINEERING
JSS ACADEMY OF TECHNICAL EDUCATION
JSS Campus, Dr. Vishnuvardhan Road, Bengaluru-560060
 JSS MAHAVIDYAPEETHA, MYSURU
JSS ACADEMY OF TECHNICAL EDUCATION
JSS Campus, Dr.Vishnuvardhan Road, Bengaluru-560060

DEPARTMENT OF INFORMATION SCIENCE & ENGINEERING

CERTIFICATE

This is to certify that Technical seminar (18CSS84) Report entitled “Waste

Classification and Segregation using Machine Learning and IOT” is a bonafide work
carried out by Akash N [1JS20IS010] in partial fulfillment for the award of degree of
Bachelor of Engineering in Information Science and Engineering of Visvesvaraya
Technological University Belagavi during the year 2023- 2024.

Signature of the Guide Signature of the HOD

Chetan R Dr Rekha PM
Assisstant Professor Professor and Head
Dept. of ISE Dept. of ISE
JSSATE, Bengaluru JSSATE, Bengaluru
 ACKNOWLEDGEMENT

The satisfaction and euphoria that accompany the successful completion of any
task would be incomplete without the mention of the people who made it possible.
So, with gratitude, we acknowledge all those whose guidance and encouragement
crowned my effort with success.

First and foremost, we would like to thank his Holiness Jagadguru Sri
Shivarathri Deshikendra Mahaswamiji and Dr Bhimasen Soragaon., Principal,
JSSATE, Bangalore for providing an opportunity to carry out the Technical
Seminar (18CSS86) as a part of our curriculum in the partial fulfillment of the
degree course.

It is our pleasant duty to place on record our deepest sense of gratitude to our
respected internal guide Mr Chetan R, Assistant Professor for the constant
encouragement, valuable help and assistance in every possible way.

We express our sincere gratitude for our beloved Professor & Head of the
department, Dr. Rekha P M, for her co-operation and encouragement at all the
moments of our approach. We are thankful to our department Technical Seminar
Coordinator Mr Chetan R , Assistant Professor, Dept. of ISE, for his continuous
cooperation and support.

We would like to thank all Teaching and non-teaching Faculty for providing us
with their valuable guidance and for being there at all stages of our work.

Akash N
1JS20IS010
 ABSTRACT

This abstract introduces a novel approach to waste management through the implementation of a fully
automated system for waste segregation. The proposed method leverages computer vision and deep
learning technologies integrated with an Internet of Things (IoT) framework to efficiently categorize
municipal waste into Organic and Recyclable components. By eliminating the need for manual segregation,
this system not only enhances operational efficiency but also mitigates health risks associated with manual
handling of waste by municipal workers.

Traditional waste management processes often entail manual sorting of waste, exposing workers to various
health hazards including the risk of contracting transmissible diseases. By automating the segregation
process, our system minimizes direct human contact with waste materials, thereby reducing the likelihood
of illness and enhancing overall worker safety.Moreover, automation brings about notable improvements in
the speed and cost-effectiveness of waste segregation. With the ability to rapidly and accurately classify
waste items, the proposed system streamlines the entire waste management process, leading to increased
operational efficiency and reduced labor costs.

This study is driven by the goal of fostering innovation and sustainability in waste management practices.
By envisioning and implementing cutting-edge solutions such as the automated waste segregation system
described herein, we aim to revolutionize traditional waste management methodologies, minimizing human
intervention while maximizing efficiency and safety. Through this research, we demonstrate the feasibility
and efficacy of automated waste management systems, highlighting the potential for widespread adoption
and integration into municipal waste management infrastructures. By embracing technological
advancements and innovative approaches, we can pave the way for a cleaner, safer, and more sustainable
future in waste management.

Dept of ISE, JSSATEB 4

 TABLE OF CONTENT

Acknowlegement i.
Abstract ii.
List of figures iii.
Chapter 1: Introduction 7-8
Chapter 2:Literature Survey 9-11
Chapter 3: Proposed Work
3.1 Problem Statement 12
3.2 Problem Solution 12
3.3 Problem Identify 12
3.4 Objectives 13
3.5 Scope of the project 13
Chapter 4: Methodolgy
4.1 Waste Classifier and Sorter 14
4.2 Model Pipe Lining 15
4.3 Importing Libraries 20
4.4 Data Enrichment 21
Chapter 5: Experimental Setup
5.1 Datasets 22
5.2 Classifier Performance 23
5.3 Classifier Function 25
5.4 Hardware Segregator 26
5.5 Comparison with other CNN Models 28
5.6 Results and Discussion 30
Chapter 6: Conclusion 31
References 32

Dept of ISE, JSSATEB 5

 LIST OF FIGURES

Figure 1 Organic and recyclable bins 8

Figure 4.1 Model Pipe Lining 15
Figure 4.2 Importing Libraries 15
Figure 4.3 Convolution Neural Networks 18

Figure 4.4 Dataset for Organic Waste 18

Figure 4.5 Class Diagram 19

Figure 4.6 Use Case Diagram 20

Figure 5.1 Waste Images Dataset 20

Figure 5.2 accuracy v/s epoch 22
Figure 5.3 Classifier Function 22
Figure 5.4 Comparison With Other CNN Models 24
Figure 5.5 Epoch Levels 26
Figure 5.6 Proposed System complete model setup. 26
Figure 5.7 Propsed System Hardware Model 27

Dept of ISE, JSSATEB 6

CHAPTER 1
INTRODUCTION

In the contemporary world, one of the most pressing challenges confronting human society is the effective
management and disposal of waste, exacerbated by the exponential growth of the global population.
Reports from the World Bank alarmingly indicate that a staggering 4 billion tonnes of waste are generated
annually, equivalent to a daunting 9000 tonnes each day. This crisis demands urgent solutions that not only
address the escalating demand for new landfills but also prioritize the swift processing and segregation of
waste, all while minimizing the margin for human error in these critical tasks.

Recognizing the imperative for innovation, our project presents a pioneering approach to waste
management through the integration of cutting-edge technologies. Leveraging Convolutional Neural
Networks (CNNs) and image classification techniques alongside Internet of Things (IoT) devices, our
solution seeks to automate and optimize the waste management process.

Solid Waste Management (SWM) represents the systematic collection and treatment of Municipal Solid
Waste (MSW) before its safe disposal. Yet, it remains a formidable challenge, compounded by factors such
as rapid population growth, inadequate infrastructure, and insufficient financial planning. Regrettably,
inefficient waste management systems prevail, particularly in densely populated urban areas of
underdeveloped or developing nations. The sheer magnitude of waste generation, as highlighted by the
World Bank, underscores the urgency for transformative approaches. Notably, a significant portion of this
waste, approximately 50%, comprises organic matter that could be decomposed, while 30% is recyclable,
and the remaining 20% is non-recyclable, destined for landfills.

Central to our proposal is the vision of an ideal waste management cycle, characterized by effective
segregation, followed by recycling or proper disposal. However, traditional methods, particularly manual
segregation, prove time-consuming and often ineffectual, especially when confronted with vast quantities
of Municipal waste. Herein lies the essence of our endeavor: harnessing the power of computer vision and
Deep Learning in tandem with advanced hardware devices to create intelligent systems capable of
emulating human behavior. Such systems promise to revolutionize repetitive tasks, typically reliant on
human judgment, by operating at speeds far surpassing those achievable by manual labor, all while
minimizing errors. Hence , we present a comprehensive waste management system poised to address the
global waste crisis head-on. Through innovative integration of technology and methodology, our solution
aims to usher in a new era of efficiency and sustainability in waste management practices worldwide.
Machine Learning (ML), specifically harnessed through Convolutional Neural Networks (CNNs), plays a
Dept of ISE, JSSATEB 7
pivotal role in waste classification within the proposed system. Leveraging intelligent image classification,
ML algorithms, particularly CNNs, distinguish between various types of waste, including bio and non-bio
waste. The CNN model excels in learning intricate patterns and features from waste images, ensuring
precise categorization. This optimized sorting process, driven by ML algorithms, guarantees that materials
are effectively separated for recycling or proper disposal. The system continually refines its classification
accuracy over time, adapting to evolving waste compositions and enhancing overall performance.

The Internet of Things (IoT) contributes to real-time data monitoring, a cornerstone of the smart waste
management system. Through sensors like ultrasonic and load measurement sensors, IoT enables the
collection of data on waste levels in the smart trash bin, offering instantaneous updates on fill levels and
waste types. Remote accessibility is benefit facilitated by IoT, allowing users to monitor the waste from
anywhere. This interconnected system not only empowers users to make timely interventions based on real
time data but also enables data-driven decision-making for waste management strategies.

In essence, our project represents a paradigm shift in waste management, transcending the limitations of
traditional methodologies to embrace the transformative potential of machine learning and IoT. By
harnessing the power of data-driven decision-making and real-time connectivity, we envision a future
where waste management transcends the realm of necessity to become a cornerstone of environmental
stewardship and sustainable development. This introduction encapsulates the essence of your project,
emphasizing the fusion of machine learning and IoT to address the challenges of waste segregation and
classification. Let me know if you need further refinement or additional details!

Figure 1: Organic and recyclable bins

Dept of ISE, JSSATEB 8

CHAPTER 2
LITERATURE SURVEY

Title 1: Waste Detection and Segregation Using Machine Learning and YOLO
Algorithm
This research study, conducted by Aishwarya, Parth Wadhwa, and Owais, introduces an innovative
application of image processing powered by machine learning and the YOLO (You Only Look Once)
algorithm. The aim is to detect non-biodegradable waste within bins for efficient separation. The study
involved collecting and labeling 450-500 images each of glass, metal, and plastic waste to train the
machine learning model. However, while the accuracy for detecting metals reached 75%, distinguishing
between plastic and glass proved challenging, resulting in lower accuracy. Nonetheless, the project
underscores the potential of technology to automate waste segregation processes and mitigate health risks
associated with manual sorting.

Title 2: Development of Biodegradable Partitioning Materials

Authored by Dumindu Dassanayaka and Dilshan Hedigalla, this research proposes a novel solution to
address the environmental impact of conventional construction practices. By manufacturing biodegradable
partitioning materials using banana yarn reinforcement and Poly-lactic acid matrix via compression
molding, the study aims to reduce reliance on non-biodegradable materials like PVC and asbestos. Flexural
and impact strength tests were conducted on different composite variations, offering insights into optimal
manufacturing parameters. This innovative approach holds promise for sustainable construction practices,
although further optimization may be required for widespread adoption.

Title 3: IoT-Based Waste Segregator for Efficient Recycling

Jeberson Retna Raj, B.Infant Philo Rajula, R.Tamilbharathi, and Senduru Srinivasulu present an IoT-
enabled waste segregator designed to streamline municipal waste management processes. Equipped with IR,
moisture, and metal sensors, the system effectively segregates metal, wet, and dry waste, respectively.
While the system successfully segregates waste and stores sensor data in the cloud for further analysis,
there are some drawbacks. Notably, the system's reliance on a conveyor belt for waste transport may pose
operational challenges, and further refinement may be necessary to optimize efficiency and reliability.
Nonetheless, this IoT-based solution represents a significant step towards sustainable waste management
practices in urban environments.

Dept of ISE, JSSATEB 9

Title 4: Polyethylene Detection through CNN
Bobulski and Kubanek (2020) concentrated on waste classification, particularly the detection of
polyethylene. Their experiments included the identification of various waste types such as terephthalate,
high-density polyethylene, polypropylene, and polystyrene, showcasing the capabilities of Convolutional
Neural Networks (CNN) in targeted waste detection. Title 7: Waste Segregation with CNN Algorithm In
the study by R S Sandhya Devi, Vijaykumar VR, and M Muthumeena, Convolutional Neural Networks
(CNN) were applied for waste segregation. Utilizing object recognition, the algorithm classified objects
into bio and non-biodegradable waste in real-time through a webcam. The study, leveraging Python
libraries like OpenCV, acknowledges challenges related to training diverse objects with varying
dimensions and shapes.

Title 5: IoT-Based Smart Waste Management with LoRa and TensorFlow

Teoh Ji Sheng, Mohammad Shahidul Islam, and Norbahiah Misran introduced an IoT-based smart waste
management system. Using LoRa communication and TensorFlow deep learning, the system detected and
classified objects, providing a relative match percentage. The Raspberry Pi 3 Model B+ camera served as
the primary sensor, highlighting the integration of IoT and deep learning technologies.

Title 6: IoT Technology in Waste Management

Hong et al. (2017) presented a smart trash can with IoT and Raspberry Pi, while Bai et al. (2018)
implemented an IoT-based smart garbage system to reduce food waste. The latter achieved a notable 33%
reduction through mesh technology, router integration, and server analysis.Muthugala et al. (2020)
showcased a waste collection robot with 95% accuracy in deep learning- based waste detection.

Title 7: Capsule-Net for Solid Waste Management

Sreelakshmi et al. (2019) utilized Capsule Neural Network (Capsule-Net) for solid waste management,
achieving impressive accuracy rates of 96.3% and 95.7%. The system distinguished between plastic and
non-plastic materials, demonstrating the effectiveness of Capsule-Net in waste classification on various
hardware devices.

Dept of ISE, JSSATEB 10

The deleterious impact of waste on the environment and human health has long been recognized,
prompting concerted efforts to mitigate its adverse effects . Over the years, technological interventions
such as RFID (Radio Frequency Identification) devices have been deployed to streamline waste
classification and minimize environmental footprint Previous studies have explored methodologies
including landfill diversion, recycling initiatives, organic waste management, employing (CNNs) to
categorize waste effectively. Inspired by these endeavors, our approach similarly leverages CNN
architectures, augmented with IoT (Internet of Things) technologies, to facilitate accurate waste
segregation.

The evolution of CNN structures since 2015 has revolutionized the landscape of waste classification,
enabling robust and efficient solutions to complex categorization problems. Notable contributions include
the integration of characteristics from SIFT (Scale-Invariant Feature Transform) and shape analysis within
Bayesian computational frameworks. Additionally, innovative solutions like the Auto Trash system
developed by Zaman exemplify the fusion of hardware (e.g., Raspberry Pi) with machine learning
frameworks such as Google TensorFlow, underscoring the versatility and scalability of modern waste
management solutions.

Recent literature highlights the detrimental impact of plastic waste on ecosystems and human health,
emphasizing the urgency of sustainable waste management practices. Furthermore, advancements in waste
recycling technologies, such as the review on solar panel recycling by Xu et al. , underscore the
multifaceted nature of waste management challenges. Contemporary trends underscore the prominence of
CNNs and neural networks in waste classification, underscoring the pivotal role of machine learning
algorithms in addressing complex segmentation tasks.

Traditionally, waste segregation has relied on manual sorting methods, but recent advancements in image-
based classification have gained traction due to their efficiency and scalability. Leveraging publicly
available datasets such as ImageNet from platforms like Kaggle, researchers have developed robust CNN
models capable of accurately classifying waste materials. Our study builds upon this foundation, employing
CNNs trained on diverse datasets to achieve precise waste classification.

By integrating CNN-based classification with IoT-enabled hardware, our approach offers a comprehensive
solution to waste management challenges. Arduino microcontrollers, guided by CNN predictions,
orchestrate the precise segregation of waste materials into designated bins through servo arm mechanisms.
This symbiotic fusion of machine learning and hardware automation promises to streamline waste
management processes, mitigating environmental impacts and enhancing operational efficiency.
Dept of ISE, JSSATEB 11
CHAPTER 3
PROPOSED WORK

3.1 PROBLEM STATEMENT

Inefficient waste management practices contribute to environmental pollution by leading to the
accumulation of waste in landfills and open areas. Improper disposal methods posing serious threats to
ecosystems and public health. The lack of proper waste segregation and disposal can also lead to the
production of greenhouse gases, soil degradation, water contamination, and the proliferation of disease
vectors.To address this issue, this project proposes a smart waste management system integrating deep
learning and IoT technologies. The solution aims to enhance waste classification precision and provide
real-time monitoring, promoting sustainable waste disposal practices.

3.2 PROPOSED SOLUTION

The project introduces an architecture that combines deep learning, specifically a Convolutional Neural
Network (CNN), for waste classification, and IoT for real-time data monitoring. The system includes a
smart trash bin with a microcontroller and multiple sensors, offering an innovative approach to sorting
digestible and indigestible waste. IoT enables remote real-time data control.The proposed CNN model
achieves a good classification accuracy and the System Usability Scale (SUS) score is effective. This
adaptive smart system can effectively monitor household waste activities in real time.

3.3 PROBLEM IDENTIFY

The existing waste management practices often lack precision and real-time monitoring capabilities,
leading to environmental pollution ,inefficient resource utilization and suffer from imprecise waste
categorization.our new approach integrates deep learning and the Internet of Things (IoT) to create a smart
waste management system going to make all the corrections of its limitation.

Dept of ISE, JSSATEB 12

3.4 Objectives
 Develop a fully automated waste management system integrating computer vision, deep learning, and
IoT technologies.
 Create a system capable of accurately classifying municipal waste into organic and recyclable
categories to facilitate efficient segregation.
 Reduce health risks to municipal workers by minimizing manual labor involvement in waste
segregation processes.
 Improve operational efficiency in waste management by automating waste segregation, thereby
reducing processing time and costs.
 Implement a scalable solution that can be deployed in various settings, including urban and industrial
environments, to address diverse waste management challenges.

3.5 Scope of the Project:

 The project focuses on the development and implementation of a prototype waste management system,
encompassing both hardware and software components.
 It involves the utilization of computer vision techniques for image processing and deep learning
algorithms for waste classification.
 The scope includes the integration of IoT devices, such as Arduino controllers and servo motors, to
automate waste segregation based on classification results.
 The project encompasses the training of convolutional neural networks (CNNs) using a suitable dataset
to achieve high accuracy in waste classification.
 The proposed system targets the segregation of organic and recyclable waste types, addressing a
significant portion of municipal waste management challenges.
 While the project aims to provide a proof-of-concept demonstration, it also lays the groundwork for
potential scalability and future enhancements, including the addition of more waste categories and the
optimization of hardware components for industrial-scale deployment.

Dept of ISE, JSSATEB 13

CHAPTER 4
METHODOLGY
The proposed methodology integrates two essential components: the utilization of convolutional
neural networks for waste classification and the design of intelligent trash boxes with real-time data
monitoring capabilities using IoT. This fusion of two structural models aims to achieve outstanding
outcomes in the field of waste management. The systematic classification of wastes into appropriate
categories facilitates the identification of reusable materials, allowing utilization without deterioration.

Systems design is the process of defining the architecture, components, modules, interfaces, and data
for a system to satisfy specified requirements. Systems design could see it as the application of
systems theory to product development. System design is one of the most important phases of
software development process. The purpose of the design is to plan the solution of a problem
specified by the requirement documentation. In other words, the first step in the solution to the
problem is the design of the project. The design of the system is perhaps the most critical factor
affecting the quality of the software. The objective of the design phase is to produce overall design of
the software. It aims to figure out the modules that should be in the system to fulfil all the system
requirements in an efficient manner.

WASTE CLASSIFIER AND SORTER

In urban and household settings, waste bins often accumulate mixed garbage, leading to
environmental hazards and health risks. To address this issue, we propose a neural network-based
solution utilizing layers to classify waste into respective bins. A servo-arm system transports the
waste to designated bins based on classification. Initially, waste images are captured and fed into the
computer's memory for classification as recyclable or organic. The model then instructs the servo-arm
accordingly. Importantly, this entire process is automated, eliminating the need for human
intervention—a significant advantage in modern automation techniques.

Dept of ISE, JSSATEB 14

4.1 Model Pipe Lining

Figure 4.1 Model Pipe Lining

Step 1: Downloading and Extraction of Images:

Data Set Selection: We began by acquiring a dataset from Kaggle, which served as the foundation
for our project. This dataset contained a substantial collection of waste images essential for training
our model.

Image Extraction: Upon obtaining the dataset, extraction of the images, totaling 20,000, to prepare
them for further processing.

Organizing Images: Next systematically organized the extracted images into distinct folders. These
folders were categorized based on the type of waste, with separate subfolders designated for organic
and recyclable waste.

Dept of ISE, JSSATEB 15

Step 2: Model Training:

Utilization of Dataset: With the organized dataset in hand, we proceeded to train our model. This
involved utilizing the input images from the dataset to train the model as an image classifier.

Objective of Training: The primary objective of the training phase was to equip the model with the
capability to accurately classify waste images as either organic or recyclable. This classification task
served as the cornerstone of our waste management system.

Step 3: Image Classification:

Capturing and Classifying Images: Once the model was trained, it was ready to classify waste
images in real-time. By capturing images from a webcam, the model provided predictions regarding
the type of waste, facilitating automated sorting without human intervention.

Significance of Real-time Classification: Real-time image classification plays a pivotal role in

streamlining waste management processes by enabling automated sorting. This not only reduced
manual effort but also enhanced efficiency and accuracy in waste segregation tasks.

4.2 Importing Libraries: It leverage Keras, an open-source library renowned for its Python
interface facilitating artificial neural network development. Specifically, we utilize Keras version 2.3,
chosen for its efficiency in numerical computations and rapid experimentation. This library supports
multiple back-end neural network computations, offering versatility and performance optimization.
By importing essential libraries and layers from Keras, we lay the foundation for constructing our
neural network model, ensuring streamlined development and implementation.

The Keras Conv2D layer is pivotal in our neural network, specializing in image convolution and
feature extraction. Through learnable filters and activation functions like 'Relu', it identifies intricate
patterns within images, crucial for subsequent classification tasks. This layer's parameters, including
filter size and activation functions, optimize feature extraction and representation learning. By
applying convolution operations to input data, it enhances the network's ability to discern meaningful
information. Its output tensors serve as rich representations of input images, facilitating accurate
classification. Utilizing Keras 2.3, we harness the efficiency and flexibility of this layer for numerical
computations. Conv2D's role in identifying edges, textures, and shapes within images contributes
significantly to the network's predictive capabilities. Its integration into our model architecture is
essential for robust and accurate image classification.

Dept of ISE, JSSATEB 16

Conv2D Layer:
The Conv2D layer creates a neural network kernel for image convolution, essential for feature
extraction in classification tasks. By applying learnable filters and activation functions like 'Relu', it
identifies edges, textures, and shapes within images. This layer enhances the network's ability to
discern meaningful information from input images, contributing to accurate classification. With
specified parameters such as filter size, Conv2D optimizes feature extraction and representation
learning. Its output tensors serve as rich representations of input data, facilitating subsequent layers'
decision-making processes.

MaxPooling2D Layer:
MaxPooling2D downsamples feature maps, summarizing key information while reducing
computational complexity. By selecting maximum values within specified patches, it enhances the
efficiency of subsequent layers. Utilizing a pooling size of (2,2) optimizes information retention and
computational efficiency. This layer promotes spatial invariance, enabling the network to generalize
better to variations in input data. It enhances the network's ability to identify important features and
patterns within images.

Flattening Layer:
Flattening reshapes tensors into one-dimensional vectors, ensuring compatibility with dense layers. It
aligns extracted features for seamless integration with subsequent layers, facilitating accurate
classification. This transformation bridges convolutional and fully connected layers, streamlining
information flow. By condensing hierarchical representations into a format conducive to classification,
Flattening promotes efficient utilization of extracted features. Its role is crucial for transitioning from
feature extraction to decision-making in the network.

Dense Layer:
Dense layers add fully connected neural network layers, consolidating extracted features for
classification. With experimentally chosen parameters like 128 nodes in a hidden layer, they introduce
non-linearity and capture complex relationships within the data. These layers serve as the final stage
in the network's decision-making process, providing predictions based on learned features. Dense
layers optimize classification accuracy by aggregating and processing features extracted by preceding
layers.

Dept of ISE, JSSATEB 17

4.3 Convolutional Neural Network (CNN):
CNNs excel in analyzing imagery through hierarchical feature extraction, inspired by the human
brain's visual processing. Comprising convolutional, pooling, and fully connected layers, they
effectively identify and classify features within images. Each layer contributes uniquely to the
network's ability to discern patterns and make accurate predictions, culminating in a robust model for
image classification tasks. The hierarchical architecture of CNNs allows for efficient feature
extraction and representation learning, enabling high-performance image classification.

Figure 4.2 Importing Libraries

Figure 4.3 Convolutional Neural Network

Dept of ISE, JSSATEB 18

4.4 Data Enrichment
In the data enrichment process, we captured images in low-light conditions and applied image
inversion techniques to minimize errors in classification, especially for dimly lit environments. By
varying camera positions, we obtained a diverse dataset comprising a total of 20,000 images,
consisting of 502 organic and 1502 recyclable waste images. Each image was standardized to a
resolution of 64x64 pixels before being fed into the classifier. To augment our dataset and enhance
model robustness, we employed Keras' image data generator. This tool facilitated the generation of
numerous transformed images through operations such as zooming, flipping, inversion, and horizontal
and vertical adjustments. These enriched images enabled our classifier to learn from a more
comprehensive range of environmental conditions, improving its accuracy and performance in real-
world scenarios.

Figure 4.4 Dataset for Organic Waste

4.5 CLASS DIAGRAM

The figure 4.5 shows a class diagram, it is the main building block of object-oriented modeling. It is used
for general conceptual modeling of the structure of the application, and for detailed modeling translating
the models into programming code. Class diagrams can also be used for data modeling. The classes in a
class diagram represent both the main elements, interactions in the application, and the classes to be
programmed

Dept of ISE, JSSATEB 19

 Figure 4.5 Class Diagram

4.6 Use Case Diagram

A Use Case Diagram is a lot of situations that reflect a client-frame relationship. A use case chart shows
the entertainer-to-use relationship. Usage cases and on-screen characters are the two main elements of an
usage case diagram. An on-screen character refers to an user or other person connected with the
demonstrated process. A use case chart in figure is an out-of - the-box perspective that speaks to some
activity each module will perform to complete an errand.

Figure 4.6 Use case Diagram

Dept of ISE, JSSATEB 20

CHAPTER 5

EXPERIMENTAL SETUP
5.1 Dataset
The model evaluation is conducted on a subset of a larger dataset, comprising a total of 22,564 images
categorized into two classes: 'Organic' and 'Recyclable', with an equal distribution of 2513 images each.
These images encompass a diverse range of objects captured from different camera angles and under
various lighting conditions. Prior to training, all images are resized to a uniform resolution of 64 by 64
pixels to standardize the input data. The original dataset, totaling approximately 2.44 gigabytes, is
partitioned into separate training and test sets for each class to ensure comprehensive model evaluation. For
model training, a training step size of 706 and a validation step size of 2000 are employed over 15 epochs.
The classifier's performance metrics are subsequently computed, with the model achieving an accuracy of
95.04% and a loss value of 0.4508 upon completion of the training process. These metrics serve as
indicators of the model's effectiveness in accurately classifying waste images into their respective
categories, showcasing its robustness and reliability for real-world application in waste management

Figure 5.1 Waste images dataset

5.2 The Classifier Performance:

In Figure 10: Accuracy vs. Epoch . Epochs are crucial parameters in our model, representing the number of
passes through the training dataset. With a defined epoch of 4 and 706 steps per epoch, our model's
performance is evaluated. Additionally, 2000 validation steps are undertaken to ensure robust validation.

Dept of ISE, JSSATEB 21

In the first pass, the model achieves an accuracy of 80.41% and a loss of 0.4362. Subsequently, in the
second pass, the accuracy improves to 83.41% with a reduced loss of 0.37. Continuing to the third pass, the
accuracy further increases to 85.41%. By the fourth pass, the model's accuracy approaches 90%.

Figure 5.2 accuracy v/s epoch

5.3 Classifier Function

In figure 5.3 The graph illustrates the classifier function, showcasing the relationship between epoch levels
and model efficiency. It suggests that increasing the number of epochs could potentially enhance model
performance to over 95% accuracy. Remarkably, our model achieves nearly 90% accuracy within just 4
epochs, demonstrating rapid convergence and efficiency. This efficiency is particularly notable as the
training process completes within 2 minutes, indicating the model's effectiveness and computational
efficiency. In conclusion model exhibits promising performance metrics, with accuracy steadily increasing
across epochs. While further epochs may potentially improve accuracy, our model already achieves
commendable results with minimal training time. These findings underscore the model's effectiveness in
waste classification tasks and its potential for further optimization and real-world deployment.

Figure 5.3 Classifier Function

Dept of ISE, JSSATEB 22

5.4 Comparison with our CNN models
Comparing our CNN model's performance with AlexNet and ResNet models, we noted that our CNN
classifier, despite its smaller size, demonstrated competitive accuracy levels. While increasing the
number of epochs beyond 4 could potentially yield accuracy exceeding 90%, our results already
showcased significant enhancement in model performance. This indicates that our proposed CNN
classifier, despite its simplicity with only four layers, exhibits robust performance in image
classification tasks.

In this paper conduction of a comparative analysis of our proposed four-layer CNN method against
well-established models such as AlexNet and various ResNet architectures including ResNet34,
ResNet50, and ResNet101. The objective was to assess the impact of model size on both efficiency
and accuracy metrics. Our findings revealed a direct correlation between the number of epochs and
the accuracy of the CNN classifier. Beginning with epoch=1 and progressing to epoch=4, we
observed a notable increase in accuracy from 80% to 90% using the Kaggle dataset.

1.AlexNet:
AlexNet's architecture, with its deep layers and convolutional operations, can provide powerful
feature extraction capabilities, enhancing the model's ability to discern intricate patterns in waste
images. Its success in large-scale visual recognition tasks suggests its potential to accurately classify
diverse waste items, even in cluttered or complex environments. However, AlexNet's deeper
architecture may require more computational resources compared to simpler models, potentially
impacting efficiency in real-time waste classification scenarios.

2.ResNet34, ResNet50, and ResNet101:

ResNet architectures, particularly ResNet50 and ResNet101, are known for their depth and ability to
learn complex hierarchical features. The introduction of residual connections in ResNet models helps
alleviate the vanishing gradient problem, allowing for more effective training of deeper networks.
This can lead to improved accuracy in waste classification tasks, especially with large and varied
datasets. ResNet34, although less deep, still offers superior performance compared to shallower
architectures like AlexNet. It strikes a balance between complexity and efficiency, making it suitable
for tasks requiring moderate levels of feature abstraction. While ResNet50 and ResNet101 offer
higher accuracy potential, they may require more computational resources and longer training times.
However, their superior performance in handling intricate features could justify these trade-offs in
waste classification scenarios where accuracy is paramount.
Dept of ISE, JSSATEB 23
 Figure 5.4 Comparison with other CNN models

Figure 5.5 : epoch levels

5.5 Hardware Segregator

The hardware setup comprises an Arduino controller and a camera module, pivotal for automated waste
segregation. Organic waste, designated as '1', prompts clockwise servo arm rotation, while recyclable waste,
labeled '0', induces anticlockwise movement. The ICL293D servo motor driver ensures precise arm control,
enhancing sorting accuracy. This integrated system efficiently translates classification results into physical
actions for effective waste management. By interfacing with the deep learning model, the Arduino
orchestrates seamless operation based on real-time classification data. LED indicators in bins further
enhance the system's functionality, signaling successful waste segregation. Overall, this hardware setup
embodies a sophisticated yet user-friendly solution for automating waste sorting processes.

Dept of ISE, JSSATEB 24

The operational sequence of the hardware setup delineates a systematic process for
automated waste segregation:

1. Waste Placement: This initial step involves strategically positioning waste items within the observation
range of the camera module. Placing the waste appropriately ensures that all items are captured effectively
for subsequent analysis.
2. Image Capture and Transmission: Once the waste is in position, the camera module captures images
of the waste items. These images are then transmitted to the deep learning model for analysis. It's crucial to
ensure clear and high-quality image capture to facilitate accurate classification.
3. Classification: The heart of the process lies in the classification phase, where the deep learning model
employs convolutional neural networks (CNNs) to meticulously analyze the captured images. The model
leverages its trained parameters to distinguish between organic and recyclable waste items based on visual
features and patterns.
4. Servo Arm Movement: Upon classification, the results are relayed to the Arduino controller, which
serves as the central control unit. The Arduino interprets the classification outcomes and directs the servo
arm to move in the appropriate direction. If the waste item is classified as organic, the servo arm rotates
clockwise, guiding the item towards the organic waste bin. Conversely, if the item is deemed recyclable,
the servo arm rotates anticlockwise, guiding it towards the recyclable waste bin.

Efficient Automation: The orchestrated sequence of operations highlights the efficiency and accuracy of
the hardware setup in automating waste segregation. By seamlessly integrating components and
sophisticated algorithmic processing, the system streamlines waste management processes. This
automation reduces reliance on manual sorting, minimizes human intervention, and enhances overall
operational efficiency. Moreover, it contributes to environmental sustainability by promoting proper waste
disposal and resource utilization.

The segregation process is facilitated by a servo arm controlled via an IC-l293d driver, coordinated by an
Arduino microcontroller. Through Arduino, instructions are relayed sequentially to the IC-L293D,
dictating motor movement for waste segregation. With two digital pins allocated on Arduino, motor
direction is efficiently managed. The ENA pin of PWM PIN-2 regulates motor speed, set at 600
Milliampere with a voltage of 4.5 V. Additionally, the Atmega 328-P micro-controller enhances control
and coordination, ensuring precise and effective waste segregation into designated bins.

Dept of ISE, JSSATEB 25

Components

1. Arduino Microcontroller: Serving as the central control unit, the Arduino microcontroller orchestrates
the operation by transmitting instructions to other components based on the classification results.
2. Servo Motor: The servo motor executes physical movements to segregate waste items into the
designated bins. Its precise control ensures accurate sorting based on the classification outcomes.
3. IC-l293d Driver: This driver module plays a crucial role in controlling the servo motor's direction and
movement, providing the necessary interface between the Arduino microcontroller and the motor.
4. Web Camera Module: Integrated into the system, the web camera module captures high-quality images
of waste items, which are then analyzed by the deep learning model for classification purposes.
5. LED Indicators: Strategically connected to each waste bin, LED indicators serve as visual feedback
mechanisms, illuminating upon waste entry into the respective bins, enhancing the segregation process's
transparency and efficiency.
6. Atmega 328-P Microcontroller: Enhancing the system's control and coordination, the Atmega 328-P
microcontroller complements the Arduino, ensuring smooth operation and precise execution of instructions
for effective waste segregation.

Figure 5.6 : Proposed System complete model setup.

Dept of ISE, JSSATEB 26

 Figure 5.7 Proposed System hardware model.

5.6 Results and Discussions

Results and Discussion:

The proposed waste management system, integrating computer vision, deep learning, and Internet of
Things (IoT) technologies, demonstrated significant advancements in waste segregation automation.
Through the elimination of manual segregation processes, this system offers considerable benefits,
including reduced health risks for municipal workers and increased operational efficiency and cost-
effectiveness.

1. Automation Benefits: By leveraging computer vision and deep learning algorithms, coupled with IoT
systems, the project successfully automated waste segregation processes. This automation not only
minimizes manual labor but also enhances the speed and accuracy of waste segregation, leading to
improved waste management practices.

2. Impact on Waste Management: Manual waste segregation, prevalent in many regions, often results in
inefficiencies and environmental hazards. With the introduction of automated systems, such as the one
proposed in this study, waste management practices can be revolutionized. The system's ability to classify
organic and recyclable waste with high accuracy mitigates environmental pollution and reduces the burden
on landfill sites.

Dept of ISE, JSSATEB 27

3. Technological Advancements: The study highlights the utilization of convolutional neural networks
(CNNs) and IoT devices to address waste management challenges. By employing CNNs for image
classification and IoT devices for waste segregation, the system achieves a seamless integration of cutting-
edge technologies to tackle real-world problems.

4. Experimental Evaluation: The experimental setup involved the evaluation of the proposed system's
performance using a subset of a larger dataset. With 22564 images belonging to two classes (Organic and
Recyclable), the system achieved significant accuracy, with an accuracy rate of 95.04% and a loss rate of
0.4508 after 15 epochs of training.

5. Comparative Analysis: A comparative analysis was conducted to assess the performance of the
proposed CNN method against established models like AlexNet and ResNet. The results indicate that
increasing the number of epochs enhances the accuracy of the CNN classifier, with accuracy levels
reaching above 90% with just four epochs of training.

6. Hardware Implementation: The hardware segment of the system, comprising Arduino, servo motor,
IC-l293d driver, and camera module, effectively facilitated waste segregation based on classification
results. The servo arm's precise movements, controlled by Arduino and IC-l293d driver, ensured accurate
sorting of waste items into designated bins, thus enhancing the overall efficiency of the waste management
process.

7. Future Implications: The study underscores the potential for future advancements in waste
management systems using CNNs and IoT technologies. With further research and development, the
integration of high-level GPU processing and advanced classification techniques could revolutionize waste
segregation processes at an industrial or societal level, contributing to a cleaner environment and improved
public health.

In conclusion, the proposed waste management system showcases the transformative potential of
integrating advanced technologies to address complex societal challenges. By automating waste
segregation processes and leveraging the capabilities of CNNs and IoT, the system offers a sustainable
solution for efficient waste management and environmental preservation.

Dept of ISE, JSSATEB 28

PYTHON

Python is an interpreted, object-oriented, high-level programming language with dynamic semantics

developed by Guido van Rossum. Python is used for server-side web development, software development,
mathematics, and system scripting, and is popular for Rapid Application Development and as a scripting or
glue language to tie existing components because of its highlevel, built-in data structures, dynamic typing,
and dynamic binding. Program maintenance costs are reduced with Python due to the easily learned syntax
and emphasis on readability.

1. Machine Learning and Deep Learning: Python libraries such as TensorFlow, Keras, and Scikit-learn
are used for training and deploying machine learning and deep learning models. TensorFlow and Keras
provide high-level APIs for building and training neural networks, while Scikit-learn offers tools for data
preprocessing and model evaluation.

2. Image Processing and Computer Vision: Python's OpenCV library is extensively used for image
processing tasks, including image capture, resizing, preprocessing, and feature extraction. OpenCV
provides a wide range of functions and algorithms for manipulating and analyzing images, making it
suitable for tasks such as object detection and classification.

3. Data Manipulation and Analysis: Python's NumPy and Pandas libraries are employed for data
manipulation and analysis tasks. NumPy provides support for mathematical operations on arrays and
matrices, while Pandas offers data structures and functions for data manipulation and analysis, including
reading and writing datasets, data cleaning, and transformation.

4. IoT Integration: Python scripts are used to interface with IoT devices such as Arduino controllers and
servo motors. Libraries such as PySerial enable communication with serial ports, allowing Python code to
send instructions to IoT devices and receive data from them.

5. Model Evaluation and Optimization: Python scripts are utilized to evaluate the performance of
machine learning and deep learning models trained on the dataset. Metrics such as accuracy, precision,
recall, and F1-score are calculated to assess the model's performance. Additionally, techniques for model
optimization and hyperparameter tuning may also be implemented using Python libraries such as Hyperopt
or Optuna.

Dept of ISE, JSSATEB 29

CHAPTER 6
CONCLUSION

The comparison between our proposed four-layer CNN method and established models like AlexNet and
various ResNet architectures, including ResNet34, ResNet50, and ResNet101, provided valuable insights
into the performance of different model sizes. Through our evaluation, we observed a notable trend: as the
number of epochs increased, so did the accuracy of our CNN classifier. Beginning with epoch=1 and
progressing to epoch=4, we witnessed a steady improvement in accuracy, achieving results ranging from
80% to over 90%. This demonstrated the effectiveness of our proposed CNN classifier, which
outperformed other models in terms of accuracy metrics.

Furthermore, our study highlighted the critical importance of leveraging advanced technologies, such as
deep learning and computer vision, to address pressing real-world challenges like waste segregation. By
harnessing the power of these cutting-edge tools, we were able to develop a robust model capable of
accurately classifying waste materials, thereby offering a promising solution to the global waste
management crisis. Moreover, our findings underscored the scalability and potential impact of such
technologies, paving the way for future advancements in waste management practices.

Looking ahead, there is immense potential to further enhance the capabilities of our proposed CNN
classifier and extend its applicability to broader contexts, such as industrial-scale waste management
systems. By leveraging high-performance computing resources, such as GPUs, we can scale up our model
to handle larger datasets and more complex classification tasks with minimal human intervention.
Additionally, future research endeavors could explore additional classes of waste materials and incorporate
user databases to refine classification accuracy and optimize system performance.

In conclusion, our study underscores the transformative potential of leveraging state-of-the-art technologies
like CNNs for waste management applications. By harnessing the power of deep learning and computer
vision, we can pave the way for more efficient, sustainable, and environmentally conscious waste
management practices, ultimately contributing to a cleaner and healthier planet for future generations.

Dept of ISE, JSSATEB 30

 REFERENCES

1. Xiao, Yi, et al. Object Detection Based on Faster R-CNN Algorithm with Skip Pooling and Fusion of
Contextual Information. Sensors, vol. 20, no. 19, 2020, p. 5490.,
2. Automatic Waste Segregator for Future World. International Journal of Pharmaceutical Research, vol.
12, no. 02, 2 Oct. 2020, 10.31838/ijpr/2020.12.02.231. Accessed 5 Mar. 2021.
3.What Is Incremental Learning | IGI Global. Www.igi-Global.com,Accessed 5 Mar. 2021
www.igiglobal.com/dictionary/incremental-approach-to-classification-learning/14075.
4.The Raspberry Pi Foundation. Camera Module V2. Raspberrypi.Org,
https://www.raspberrypi.org/products/camera-module-v2/. Accessed 28 Feb. 2021.
5.Histogram of Oriented Gradients. Wikipedia, Wikimedia Foundation, 1 Feb. 2021,
en.wikipedia.org/wiki/Histogram_of_oriented_gradients.
6.Huiyu, L., O, O. G., & Kim, S. H, 2019. December. Automatic Classifications and Recognition for
Recycled Garbage by Utilizing Deep Learning Technology. In Proceedings of the 2019 7th International
Conference on Information Technology: IoT and Smart City, pp. 1–4.
7.Nowakowski, P., Pamuła, T., 2020. Application of deep learning object classifier to improve e- waste
collection planning. Waste Management 109, 1–9.
8.Singh, A., Aggarwal, P., Arora, R., 2016. In: September). IoT based waste collection system using
infrared sensors. IEEE, pp. 505–509.
9.ALFoudery, A., Alkandari, A. A., & Almutairi, N. M, 2018. Trash Basket Sensor Notification Using
Arduino with Android Application. Indonesian Journal of Electrical Engineering and Computer Science
(IJEECS) 10 (1), 120–128.
10.ConvoWaste: An Automatic Waste Segregation Machine Using Deep Learning Md. Shahariar
Nafiz,Department of Electrical and Electronic Engineering,Bangladesh.

Dept of ISE, JSSATEB 31