EcoSort – Deep Learning approach for E-Waste Segregation.

Rujul Chaure
Siddhesh Dandekar
Computer Science & Engineering Computer Science & Engineering
Department G H Raisoni College of Department G H Raisoni College of
Engineering Engineering
Nagpur, India Nagpur, India
[Link]@[Link] [Link]@ghrc
Shreyash Sarve [Link]
Computer Science & Engineering Janhavi Bajpai
Department G H Raisoni College of Computer Science & Engineering
Engineering Department G H Raisoni College of
Nagpur, India Engineering
[Link]@[Link] Nagpur, India
[Link]@ghrce.
[Link]

Prof. Ashish Kumar Soni

Computer Science & Engineering
Department G H Raisoni College of Engineering
Nagpur, India
[Link]@[Link]

disposal activities, and deliver relevant news and tips about

Abstract
E-waste.
Introducing EcoSort, a pioneering application designed for
With a robust backend infrastructure developed on [Link]
the management of electronic waste (E-waste). Using the
and MongoDB, EcoSort ensures seamless communication
power of advanced technology, EcoSort employs innovative
between the frontend interface built in React and the
image processing techniques to scan and identify various
backend servers. This architecture guarantees smooth
types of E-waste through a smartphone camera. Powered by
performance and reliability, even during periods of high
PyTorch, the application utilizes sophisticated machine
user traffic.
learning models with support for ONNX and TensorFlow
files, ensuring accurate and real-time detection of E-waste
items.
EcoSort features aimed at enhancing user’s experience and Key Words: E-waste, Deep Learning,
promoting environmental sustainability. Through intuitive Convolutional Neural Networks (CNNs),
user profile management, users can customize their Object detection, PyTorch, YOLOv5,
experience and effortlessly track their E-waste disposal
YOLOv8, ONNX, TensorFlow.
activities which is extremely easy to handle. Additionally,
provide guides within the application to educate users on
proper waste disposal practices, empowering them to make 1 Introduction
informed decisions about their environmental impact. In today's digital age, the significance of e-waste
management cannot be overstated. Electronics
Moreover, EcoSort integrates a user-friendly chatbot contain precious resources such as gold and copper,
interface to address user queries and provide immediate whose extraction contributes to environmental
assistance. The geolocation feature enables users to easily degradation. By recycling e-waste, we not only
locate nearby recycling centres or drop-off points,
reduce the need for new raw materials but also curb
facilitating convenient disposal of E-waste. The help and
support functionality ensures that users receive timely
energy consumption and greenhouse gas emissions
assistance and guidance whenever needed by us, they don’t associated with manufacturing. Furthermore,
encounter any problem by using the EcoSort application. effective e-waste management helps alleviate the
To further engage users and promote accountability, strain on landfills, extends their lifespan, and ensures
EcoSort incorporates notification functionalities to remind compliance with legal regulations aimed at
users of upcoming recycling events, provide updates on their protecting public health and the environment.
 Investing in e-waste recycling infrastructure holds EcoSort fills a critical need in today's world by
the potential to create job opportunities and stimulate addressing the numerous challenges of electronic
economic growth. In essence, e-waste management waste (E-waste) management. With the proliferation
is integral to sustainable development, offering of electronic devices, the disposal of obsolete or
solutions to environmental, social, and economic damaged electronics has become a pressing
challenges in our increasingly digitalized world. So, environmental concern.

Traditional methods of sorting and managing E- PyTorch, with support for ONNX and TensorFlow
waste are often labor-intensive, time-consuming, files, ensuring high precision in E-waste identification.
prone to error and more over thrown away rather than The application's frontend, developed using React,
recycling them. By leveraging cutting-edge delivers a seamless user experience, complemented by
technology, EcoSort offers a transformative solution a robust MongoDB backend that facilitates efficient
to this problem. Its ability to scan and detect E-waste communication between the frontend and backend
items through a smartphone camera streamlines the servers by [Link].
identification process, making it more efficient and In short, EcoSort represents a holistic approach to E-
accurate. Moreover, EcoSort goes beyond simple waste management, combining technological
detection. Its comprehensive set of features, innovation with user-centric design to promote
including user profile management, interactive responsible waste disposal and contribute to a cleaner,
guides, chatbot assistance, geolocation services, and more sustainable environment.
notifications, ensures a seamless and user-friendly
experience.
2. Literature Survey:
By empowering users with knowledge and resources,
EcoSort encourages responsible waste disposal Aatmaj Amol Salunke from Manipal University Jaipur
practices and promotes environmental stewardship. introduces WasteSegNet, a designed Convolutional
Neural Network (CNN) aimed at automating waste
In essence, EcoSort is not just an application; it's a segregation into Organic and Recyclable categories.
vital tool for addressing the heavenly increasing The model is worked on a dataset of 22,564 waste
challenge of E-waste management in a sustainable and images and achieves an impressive 87.5% accuracy on
effective manner. It represents a proactive approach to a test dataset of 2,513 samples. The WasteSegNet
mitigating environmental impact, safeguarding natural suggests highlights the significance of efficient waste
resources, and creating a healthier planet for current segregation for sustainable urban waste management
and future generations. EcoSort has been an amazing and emphasizes the potential of WasteSegNet in
application poised to redefine E-waste management. optimizing waste sorting processes, reducing landfill
Our innovative solution leverages cutting-edge waste, combating environmental pollution, and
technology to revolutionize how electronic waste is fostering sustainable waste management practices in
identified and managed. Through the seamless smart cities.
integration of advanced image processing algorithms,
EcoSort empowers users to scan E-waste items using In the "Smart e-waste management system utilizing
their smartphone cameras anytime-anywhere, Internet of Things and Deep Learning approaches,"
facilitating swift and accurate detection of nonuse able the authors address the pressing global concern of
E-waste. Beyond its core functionality, EcoSort offers electronic waste (e-waste) by proposing an innovative
various
system based on the Internet of Things (IoT) and Deep
features designed to enhance user experience and
promote environmental responsibility among Learning (DL) for more efficient e-waste
individuals. From interactive guides that educate users management. The study focuses on developing a
on proper disposal practices to robust user profile smart e-waste management system that integrates
management, our application prioritizes user object detection models YOLOv8s achieved the
engagement and empowerment. Additionally, a user- highest mAP@50 of 72% and mAP@50-95 of 52%,
friendly chatbot interface provides immediate making it the most effective model for e-waste object
assistance, while geolocation capabilities streamline
detection. The proposed system also incorporates
the process of locating nearby recycling centers and
drop-off points. At the heart of EcoSort lies a monitoring sensors to detect and prevent fires, provide
sophisticated machine learning model trained using real-time information, and store data on the
ThingSpeak cloud platform. The study emphasizes offers insights into the use of deep learning techniques.
the potential of this innovative system promote
sustainability in e-waste management. The article "Application of deep learning object
classifier to improve e-waste collection planning"
discusses the application of a deep learning object
The document presents a research study on a novel
classifier to identify and classify waste electrical and
deep learning-based small object detection and
electronic equipment from photos. The proposed
classification model for garbage waste management in
system uses a deep learning convolutional neural
smart cities and IoT environment. The performance of
network (CNN) to classify the type of e-waste for a
the DLSODC-GWM technique is validated using
faster region-based convolutional neural network deep
benchmark datasets, and the experimental results
learning model and a (R-CNN) to detect the category
and sizes of the waste equipment in the dataset of E -
demonstrate its promising performance with a
waste images. The accuracy of the CNN and R-CNN
maximum accuracy of 98.61%. The study highlights
for the 8e-waste categories ranged from 90 to 97%,
the potential of the proposed model in addressing the
enabling waste collection companies
challenges of waste management in smart cities and

3. System Architecture:
waste (E-waste) management. CNNs are advanced deep
learning models specially designed for working task like
image classification and object detection. In our application,
CNNs analyze images captured through a smartphone
camera, enabling EcoSort to accurately identify various
types of E-waste in real-time. This technology forms the
backbone of our image recognition system, allowing
EcoSort to efficiently categorize and process E-waste items
with precision and speed.

2) YOLOv8 (You Only Look Once version 8):

In EcoSort, we integrate YOLOv8 (You Only Look Once

version 8) as a key component of our object detection
system. In our application, YOLOv8 enhances EcoSort's
Figure 1: System Architecture ability to detect and classify various types of electronic
waste (E-waste) captured through a smartphone camera. By
leveraging YOLOv8, EcoSort ensures precise identification
Here, a roadmap for designing and developing complex, and classification of E-waste items in real-time, contributing
guiding decisions regarding the organization of modules, to the application's overall performance and user experience.
data flow, and system behavior of EcoSort. This
architectural design encompasses both functional and non-
functional aspects, including how the EcoSort will fulfil user 3) PyTorch:
requirements and achieve performance, scalability, and
reliability goals. System architecture defines the In EcoSort, PyTorch serves as a fundamental tool for
arrangement of components, their relationships, and the developing and training our neural network models,
communication protocols between them, ensuring seamless including Convolutional Neural Networks (CNNs) and
integration and interoperability. In essence, system YOLOv8. PyTorch is a well-known deep learning
architecture provides a high-level overview of how a framework known among machine learning developers for
software will work efficiently. It is a critical aspect of its flexibility, ease of use, and extensive community support.
software engineering that lays the groundwork for building With PyTorch, we can efficiently build and fine-tune
robust, scalable, and maintainable software solutions. complex neural network architectures tailored to the specific
requirements of EcoSort. This enables us to train our models
We used five components for our Machine Learning with high precision and accuracy, ensuring reliable
Project i) CNN ii) YOLOv8, iii) PyTorch, iv) ONNX, v) performance in identifying and classifying electronic waste
TensorFlow (E-waste) items captured through a smartphone camera. By
leveraging PyTorch, EcoSort delivers a robust and effective
1) Convolutional Neural Networks (CNNs): solution for E-waste management.

In EcoSort, Convolutional Neural Networks (CNNs) play a

pivotal character in enhancing the accuracy of electronic
 4) Open Neural Network Exchange (ONNX):
suitable format and quality for further analysis or modeling.
In EcoSort, we utilize the Open Neural Network Exchange Additionally, feature selection or extraction may be
(ONNX) format to ensure compatibility and interoperability performed to reduce dimensionality or enhance predictive
of our trained models across different platforms and power. Overall, dataset preprocessing aims to improve the
frameworks. By converting our models to the ONNX format, quality and usability of the data for subsequent tasks.
we ensure that they can be efficiently executed on different
devices and frameworks without the need for extensive Table 1: Datasets
modifications. This allows EcoSort to deliver a consistent
and reliable experience for users, regardless of the devices E-waste Image No. of Images
or platforms they use, contributing to the application's
overall accessibility and usability. Camera 250

5) TensorFlow: Keyboards 200

In EcoSort, TensorFlow acts a very important role in the Laptops 200

development and deployment of our deep learning models,
including Convolutional Neural Networks (CNNs) and Microwave 250
YOLOv8. TensorFlow is a widely-used open-source
machine learning framework among machine learning Mobile 200
developer known for its flexibility, scalability, and
comprehensive set of tools and libraries. By leveraging Mouses 200
TensorFlow, we can efficiently build, train, and deploy our
models for E-waste detection and classification tasks. Smartwatch 250
TensorFlow's robust ecosystem and extensive
documentation enable us to implement state-of-the-art TV 250
algorithms and techniques, ensuring the high performance
and accuracy of EcoSort in identifying and categorizing
electronic waste (E-waste) items captured through a
smartphone camera. Through TensorFlow, EcoSort delivers
c) Deep Learning Models:
a reliable and effective solution for E-waste management
and sustainability.
In EcoSort, deep learning models play a pivotal role in the
4. Methodology application's functionality. These models, like
Convolutional Neural Networks (CNNs) and YOLOv8, are
Following are the undertaken steps for EcoSort application: tailored to analyze images captured through a smartphone
camera. Their primary function is to accurately detect and
a) Collecting Datasets:
categorize various types of electronic waste (E-waste) in
When collecting datasets, the process typically involves real-time. By harnessing deep learning techniques, EcoSort
several key steps. First, the research objectives were defined streamlines the process of E-waste identification, providing
and identification of the specific data was done. users with a seamless and efficient solution for responsible
Determinization suitable sources for the data and assess their waste management.
quality and reliability. Obtain any necessary permissions or
consent for data collection, ensuring compliance with legal
and ethical standards. Choose appropriate sampling d) Model Training:
techniques and design data collection instruments tailored to Model training is a crucial phase in the development of
the research objectives. Finally, implement the data machine learning and deep learning models. During this
collection process and ensure consistency and accuracy in process, the model learns from a labeled dataset to make
gathering the required datasets. accurate predictions. It involves iteratively adjusting the
model's parameters to minimize prediction errors. Typically,
b) Datasets preprocessing:
When preprocessing datasets, the focus is on preparing the the dataset is divided into training and validation sets, and a
data for analysis or modeling. Initially, it involves loss function is defined to measure the model's performance.
understanding the data and its quality, identifying missing Optimization algorithms like gradient descent are then used
values, outliers, and inconsistencies. Then, techniques like to update the model's parameters and minimize the loss. The
data cleaning, normalization, and transformation are applied training process continues until the model achieves
to address these issues.
satisfactory performance on the validation set
 e) Evaluate the Model
Model training is a crucial phase in the development of 4. Set Up Backend: Started by setting up the backend of our
machine learning and deep learning models. During this application. This involved creating a [Link] project,
installing dependencies, and setting up any necessary
process, the model learns from a labeled dataset to make
configurations.
accurate predictions. It involves iteratively adjusting the
model's parameters to minimize prediction errors. Typically, 5. Develop Backend Logic: Backend logic was made,
the dataset is divided into training and validation sets, and a including API endpoints, data models and Images. This
loss function is defined to measure the model's performance. involved integrating with databases or external APIs with
Optimization algorithms like gradient descent are then used MongoDB.
to update the model's parameters and minimize the loss. The
training process continues until the model achieves 6. Set Up Frontend: Next, setting up the frontend of our
satisfactory performance on the validation set. application. Initializing a React project, install dependencies,
and configure for the project settings.
f) Set Up API:
7. Develop Frontend Components: Developing the frontend
Setting up the environment for the application involves components of our application, included user interfaces,
several steps: navigation, and interactive elements like Chatbot. Connect
1. Install Required Software: For building the software tools these components to the backend as needed.
and dependencies were installed. This may include
programming languages (e.g., Python), frameworks (e.g., 8. Integrate Machine Learning Models: EcoSort has
React, [Link]), and libraries (e.g., PyTorch, TensorFlow). Machine components like PyTorch which were converted to
ONNX to TensorFlow the later integrate them into the
2. Create Project Directory: Setting up a dedicated directory backend via [Link].
for our application project. This helped in organize our files
and resources. 9. Test the Application: Once the backend and frontend are
developed, testing of our application to ensure that it
3. Initialize Version Control: We initialize a version control functions correctly. This involved in unit testing, integration
system GitHub in our project directory to track changes and testing.
collaborate with all our team members.

5. Result:

1) UI of EcoSort:

Figure 2: Welcome Page Figure 3: Home Page

After Login, User came into the Homepage of the

Users are welcomed with the display of the image and the
application where they can view user profile and chatbot
name if the application following the two boarding screens
for assistance. Below that they can view their score then
followed by creating a User Login Page and Create and
location of nearest dumpling yard and Navigation for
Account Page.
extra feature where users can locate easily.
 After, Login User came at Home Page where their Profile is
mentioned with Chatbot for assistance and their Score and
Location of the Map and the Navigation Bar can be seen.

Figure 2: AR Scan

After Navigating from Navigation Bar Users Can

Visit to AR Scan where they can Scan the E-waste
for the procedure. Figure 6: Test Annotated Images with Labels

2) Machine learning:

Testing annotated images with labels is a crucial step in

evaluating the performance of object detection models, such
as YOLOv8 in the EcoSort application. During this process,
annotated images, which include both the original images
and corresponding ground truth labels, are fed into the
model for evaluation

In essence, the testing of annotated images with labels is not

merely a routine evaluation step; it is a meticulous
examination that underscores the EcoSort application's
dedication to delivering accurate, reliable, and impactful
Figure 3: Result Display of YOLOv8 over the Image’s object detection
solutions for E-waste management.

In the EcoSort application, the result display of YOLOv8

In the EcoSort application, the result display of YOLOv8 over the video's object detection showcases the effective
over the image object detection showcases the efficient identification and classification of electronic waste (E-
identification and categorization of electronic waste (E- waste) items in real-time.
waste) items. Through the integration of YOLOv8, users can Through the real-time display of YOLOv8's object
capture images of E-waste using their smartphone cameras, detection results, users gain valuable insights into the
and the application swiftly and accurately detects various presence and classification of E-waste items as they record
types of E-waste present in the image. This real-time video content.
detection capability enables users to receive immediate
feedback on the classification of E-waste items, empowering By examining the confusion matrix, users and developers
them to make informed decisions regarding responsible can gain valuable insights into the insights of YOLOv8's
waste management object detection features used in our model. This analysis
enables them to refine the model and enhance its accuracy
The confusion matrix of YOLOv8 over the image object over time, ultimately improving the effectiveness of the
detection in the EcoSort application provides a EcoSort application in facilitating responsible E-waste
comprehensive overview of the model's performance in management practices.
identifying electronic waste (E-waste) items. This matrix
categorizes the predictions made by YOLOv8 into four
distinct categories: true positives, false positives, true
negatives, and false negatives.
 3) Database Monitorization Database monitoring serves as a cornerstone in maintaining the
health and efficiency of the EcoSort application's data storage
infrastructure. By systematically analysing database metrics
and trends over time, EcoSort gains valuable insights into the
performance of its data storage systems. This continuous
monitoring enables EcoSort to make data-driven decisions
pertaining to various aspects of its infrastructure, including
capacity planning, resource allocation, and system
optimization.

One of the key benefits of database monitoring is its ability to

identify patterns and trends in database performance. By
closely monitoring metrics such as response times, query
throughput, and resource utilization, EcoSort can detect any
deviations from normal behaviour. These deviations may
Figure 7: Database Monitorization indicate potential issues or bottlenecks within the database
infrastructure, allowing EcoSort to address them proactively
Database monitoring in the EcoSort application is a before they escalate into more significant problems.
critical component that ensures the seamless operation and
performance of the system's data storage infrastructure can In summary, database monitoring is essential for EcoSort to
handle the further process promised by the application and maintain a robust and efficient data storage infrastructure that
the features satisfy the users requirements. By supports its mission of facilitating responsible E-waste
implementing robust monitoring mechanisms offered by management practices. By continuously monitoring and
our excellent frontend and the developing machine analysing database metrics, EcoSort can ensure the reliability,
learning application, EcoSort can continuously track and security, and performance of its data storage systems,
analyse various metrics related to its databases, ensuring ultimately contributing to the success of its E-waste
optimal functionality and reliability as state above. management initiatives.

Through database monitoring, EcoSort can monitor key

performance indicators (KPIs) such as database uptime,
response times, query throughput, and resource utilization
and all issues that makes our user frustrated to log out of
an application but not ours. This allows the application to
identify and address any potential issues or bottlenecks
proactively, minimizing downtime and optimizing
performance.

Figure 8: Monitoring Information

Additionally, database monitoring enables EcoSort to

detect anomalies or unusual patterns in database
activity, such as unauthorized access attempts or
abnormal spikes in traffic. By promptly identifying
and responding to these anomalies, EcoSort can
enhance its security posture and protect sensitive data
from potential threats.
6. Reference
[1] WasteSegNet: Deep Learning Approach
for Smart Waste Segregation in Urban
Environments (2023)
p. 90. doi:10.3390/ecsa-7-08224.

[2] Real-time Object Detection Using Deep

Learning
Sustainability 14 (2023), 11704.
doi:10.3390/su141811704.

[3] Smart e-waste management system

utilizing Internet of Things and Deep Learning
approaches
, International Journal of Integrated
Engineering 13 (2023), 1–9.

[4] Deep Consensus Network for Recycling

Waste Detection in Smart Cities (2023).

[5] ConvoWaste: An Automatic Waste

Segregation Machine Using Deep Learning
Proceedings of the IEEE/CVF Conference on
Computer Vision and Pattern Recognition,
2023, pp. 4650–4654.

[6] Image Based Trash Classification using

Machine Learning Algorithms for
Recyclability Status
(2020)

[7] Deep Learning-Based Small Object

Detection and Classification Model for
Garbage Waste Management in Smart Cities
and IoT Environment
Appl 82 (2020), 9243–9275.

[8] Efficient E-waste Classification for

Sustainable Recycling
IEEE, Cambridge, United Kingdom, 2010, pp.
72–77. doi:10.1109/NGMAST. 2015.71.

[9] M. Farjana, A.B. Fahad, S.E. Alam and

M.M. Islam, An IoT- and cloud-based e-waste
management system for resource reclamation
with a data-driven decision-making process,
IoT 4 (2020), 202–220.
doi:10.3390/iot4030011.

[10] E-waste Management for Environmental

Sustainability: An Exploratory Study (2020)