J.

Electrical Systems 20-10 s (2024): 5191-5200

Kiran Bhowmick1* Smart Waste Segregation System:

Chetashri Bhadane2*,
Param Shah3†, Proximity-Triggered Waste
Harsh Patel4†, Classification Using Convolutional
Chaitya Lakhani5†,
Rishab Pendam6† Neural Networks
† These authors contributed equally to this work.
* These authors jointly directed this study.

Abstract—The rapid increase in the amount of waste produced by human societies has led to the urgent need for proper waste
management so as to curtail risks to both health and the environment. The infrastructure facilities and public awareness are a
prerequisite for effective automation, still are often neglected. This paper introduces a Smart Waste Segregation System (SWS) aimed
at increasing efficiency in managing waste by automating its classification and disposal activities. The Arduino microcontroller is used
in this system along with an OV7670 camera and an ultrasonic sensor for accurate contactless classification and disposal of waste into
different categories. When waste is placed on a designated flap, the image is captured by an OV7670 camera. The image is subsequently
analysed by a deep learning model trained on a diverse dataset containing images of various category waste items. Based on the model’s
classification, the waste can be disposed in the appropriate bin.

Index Terms—Automated waste segregation, deep learning, Arduino microcontroller, OV7670 camera, ultrasonic sensor, image
classification, EfficientNetB0, CNN model.

I. INTRODUCTION
The rapidly growing volume of waste generated globally has stressed the urgent need for efficient and effective waste
management practices. Management of waste has become a critical challenge for municipal corporations and
environmental agencies. Further, improper waste management practices pose significant risks to public health and the
environment, and also serve as an obstacle to achieving sustainable development goals. Segregating waste into different
stacks, such as recyclables, organic garbage, and plastics, is a critical component of proper waste segregation. This process
not only reduces environmental pollution but also saves resources and minimizes health risks associated with improper
waste disposal. Due to these reasons, developing an innovative solution for waste segregation is imperative. Hence, SWS
is designed to automate and streamline the waste classification process using advanced technologies, such as deep learning
and sensor-based systems.
The initial approach was based on a hardware system relying on Arduino Uno microcontroller to coordinate and manage
the system’s functions. Ultrasonic sensors were integrated to measure dustbin levels, while a moisture sensor aided in
identifying and segregating waste as either dry or wet. A servomotor facilitated the tilting of a flap dedicated for waste
disposal.
However, this hardware-based approach had its limitations. For example, the categorization of wet waste was simply
based on moisture content, which led to inaccuracies, resulting in classification of non-biodegradable wet plastics as wet
waste. On the other hand, certain types of dry household waste, such as fruit and vegetable peels, were incorrectly disposed
as dry waste despite their biodegradable nature.
The proposed approach builds upon the initial hardware-based system and tries to overcome prior encountered limitations
by integrating advanced technologies for more accurate waste segregation. An Arduino Uno microcontroller continues to
manage the system, ensuring smooth operation. The ultrasonic sensor detects the user’s hand movement close to the bin,
triggering the waste classification process. One of the most noteworthy upgrades in this system is the incorporation of an
OV7670 camera for image capture. When a disposal signal is detected, the camera captures an image of the waste, which
is then transmitted and stored on a server. This image data is crucial for the waste classification process that follows.
This method is a huge advancement in the domain of waste segregation. Instead of relying only on hardware-based sensors
to classify waste, this system leverages the power of deep learning algorithms. The system incorporates EfficientNetB0,
a powerful Convolutional Neural Network (CNN) architecture, to analyse captured waste images. This integration of deep
learning model enables the system to address the flaws of the hardware-based approach, like inaccuracies in classifying
wet waste and misclassification of biodegradable dry waste.
1*Department of Computer Engineering Dwarkadas J. Sanghvi College of Engineering Mumbai, India
Email:- [email protected]
2Department of Computer Engineering Dwarkadas J. Sanghvi College of Engineering Mumbai, India

Email:- [email protected]
3Department of Computer Engineering Dwarkadas J. Sanghvi College of Engineering Mumbai, India

Email:- [email protected]
4Department of Computer Engineering Dwarkadas J. Sanghvi College of Engineering Mumbai, India

Email:- [email protected]
5Department of Computer Engineering Dwarkadas J. Sanghvi College of Engineering Mumbai, India

Email:- [email protected]
6Department of Computer Engineering Dwarkadas J. Sanghvi College of Engineering Mumbai, India

Email:- [email protected]

5191
 J. Electrical Systems 20-10 s (2024): 5191-5200

II. LITERATURE REVIEW

In [1], authors developed a smart waste bin and named it Automated Teller Dustbin. In this, an object recognition and
detection model for classifying garbage items was implemented using an AlexNET based on CNN by training it with a
dataset of 20 images for each of the 10 categorized objects. But the drawback is that training the CNN model on only 20
images per object category might be too small for high accuracy and generalizability since it may not capture appearance
variability of waste items in size, shape, and orientation.
Paper [2] introduces an automated waste segregation and management system for households, utilizing an Arduino
microcontroller and Raspberry Pi. The system includes sensors for moisture and metal detection, segregation bins, and a
camera for image analysis. However, the accuracy and reliability of sensors can be affected by environmental conditions,
and sensor quality. Moreover, moisture sensors and metal detectors may not always correctly identify waste types, leading
to incorrect segregation.
[3] involved creating a system that uses an RLC metal detector circuit to identify and separate metallic waste from non-
metallic garbage. Two compartments make up the waste bin, and the method of classification depends on how an object
affects the coil’s inductance when it gets close to it. A plate tilts toward one side of the bin partition when it comes into
contact with metallic waste, which is detected by large variations in inductance. An Arduino microcontroller with Wi-Fi
connectivity and a 9V battery was used to power this device. However, the system only segregates metallic waste from
non-metallic garbage.
In [4], the “Automatic Waste Segregator” system sorts waste into metal, dry, and wet categories using sensors and
mechanisms like flaps and blowers. Moisture and IR sensors are used to distinguish between dry and wet waste, the
system also detects and separates metallic waste, and alerts when garbage tanks are full. Hardware includes ARM
LPC2148, sensors, motors, and a GSM module, with software requirements in Embedded C and Kiel M vision.
The paper in [5] utilizes a PIC16F877A microcontroller for its versatility and flash memory technology. A GSM modem
enables wireless communication, while an IR sensor detects objects via infrared transmission. An LCD display shows
data output, and gas and rain sensors monitor hazardous gases and rainfall, respectively. A kit breaker sends alerts for
maintenance issues, ensuring operational efficiency.
The waste segregation process in [6] involves setting up a conveyor system with an integrated camera for image analysis.
Develop image processing software to identify waste types, control conveyor movement based on sensor signals, and
parameters of systems are tuned appropriately for high accuracy.
The paper [7] details the design and development of a smart trash bin prototype aimed at improving municipal solid waste
management. The proposed system integrates various sensors, including level detection sensors and GPS modules, to
monitor bin status and optimize waste collection routes. The smart bin features a communication interface that sends real
time data to a central management system, allowing for efficient scheduling and reducing unnecessary collection trips.
This approach seeks to make waste management more efficient and cut down on management costs. However, its success
depends heavily on how well the sensors and communication network function, which can affect how smoothly the system
runs, especially in different urban environments.
The paper [8] introduces to the realization of a smart waste management system through the use of IoT to enhance
efficiency in waste collection. This involves the placement of sensors in bins to monitor fill level by height and send this
information instantaneously to a central system. As bins get filled up, notifications are transmitted for proper timing in
collecting them as well as for route optimization, reducing unworthy traveling. The developed system was perceived to
help cut costs and increase the orderliness of cities and offer better-informed decisions in managing waste. However,
sensor reliability and network infrastructure may be significant problems that are more acute in different urban settings.

5192
 J. Electrical Systems 20-10 s (2024): 5191-5200

The authors in [9] propose a smart waste management system to realize effective and environmentally friendly waste
segregation and collection by automatically categorizing the waste into groups such as biodegradable, recyclable, and
nonrecyclable through sensors using machine learning and IoT. If it utilizes image processing and deep learning
technology to attain very high levels of accuracy in garbage sorting, then this is what makes proper waste stream
management possible. Such real-time data from the system leads to optimal collection routes where unnecessary trips are
reduced thus cutting on the costs. The authors pay attention to how this solution can improve urban sanitation on the
grounds of sustainability principles but at the same have raised issues related to technological solutions regarding
absorption capacity toward them being highly advanced and also questioned scalability for megacities.
The paper [10] discusses an IoT-based smart dustbin management system for reducing manual work and to maintain
cleanliness in cities. The system uses an IoT based sensor in the bins to detect the garbage level. The information collected
by the systems is passed on to optimization algorithms of waste collection routes so that time and efficiency are observed
with a reduction in unnecessary trip collections as well. These sensors provide real-time information regarding fill levels
from bins across a centralized platform or system. It also applies various machine learning models for estimating the trend
in accumulating waste so that actions could be taken much earlier based on prediction rather than waiting until when it
happens. The review portrayed great potential for improving productivity, lowering expenses, and decreasing
environmental footprints connected with such technologies but mentioned that issues like network reliability, as well as
sensor accuracy, were valid constraints felt under different conditions for which performance systems strongly influence
them.
The paper [11] presents an ANN-based integrated feature data of automated waste sorting and recycling classification
system for urban management to enhance productivity. Many data features, such as visual and sensor inputs, are integrated
by the system to increase the accuracy of waste classification. The combination supports enhanced ways of sorting and
recycling waste so as to meet the circular economy goals in urban areas. As digital technologies help optimize these
processes, they also decrease operational costs and improve sustainability within smart cities. The authors explore its
application from a practical standpoint for achieving efficient waste recycling and resource recovery.
The paper [12] explains deep learning methods for object detection. It examines how Convolutional Neural Networks
(CNNs) and Vision Transformers have been used in this context. The authors discuss the evolution of these techniques,
comparing the ability of CNN to extract features with Vision Transformers ability to communicate globally through self-
attention mechanisms. They also highlight the progress that has been made in designs that combine both CNNs and
Transformer architectures to leverage on their respective strengths. It should be noted that the importance of Vision
Transformers in improving detection accuracy and efficiency is increasing.
The paper [13] presents an advanced deep learning-based object detection system designed to improve accuracy and
efficiency in image analysis. The system aims to enhance accuracy and efficiency in image analysis by optimizing feature
extraction and neural network architectures. Their approach incorporates advanced techniques to address challenges such
as detecting small objects and handling various backgrounds. The system illustrates notable improvements in detecting
speed and precision compared to existing methods, making it suitable for real-time surveillance and industrial quality
control applications.
The literature review highlights the important developments as well as the continuous challenges faced by automated
waste segregation systems. Various approaches, from microcontroller-based systems to machine learning models, have
demonstrated the potential for efficient waste management. However, limitations such as the inability to handle certain
waste types and the need for frequent maintenance underscore the necessity for further innovation. The project uses
ultrasonic sensors for contactless disposal, OV7670 cameras for image capture, Arduino microcontrollers for control, and

5193
 J. Electrical Systems 20-10 s (2024): 5191-5200

a deep learning model (EfficientNetB0 CNN) for precise waste classification to overcome these difficulties. This
integrated strategy seeks to improve on the shortcomings of previous systems in waste management.

III. PROPOSED METHODOLOGY

Fig. 1. Proposed Methodology

A. Proximity Detection
1. Ultrasonic sensor HC-SRO4
The HC-SR04 ultrasonic sensor module is a widely used distance-measuring device that operates by utilizing the
ultrasonic waves’ echo time. It is made up of a receiver and a transmitter. The sensor emits ultrasonic waves that bounce
off nearby objects and then return to the receiver. On measuring the time, it takes for the waves to return, the sensor is
equipped to calculate the distance to the object. When a user approaches, the HC-SR04 ultrasonic sensor senses their
presence and activates the camera module to capture an image for contactless waste disposal. This ensures convenience
and hygiene. Additionally, the bin’s fill levels may also be tracked using this module.

Fig. 2. Ultrasonic sensor HC-SRO4 connection

2. Arduino Uno Microcontroller

Based on the ATmega328P microcontroller, the Arduino Uno is a popular microcontroller board. It has six analog inputs,
a 16 MHz quartz crystal, 14 digital input/output pins (six of which can be used as PWM outputs), a USB port, a power
jack, an ICSP header, and a reset button. It is renowned for being straightforward and simple to use, which makes it an
ideal choice for beginners as well as professionals working on electronics projects.
In SWS, signals from the OV7670 camera and the ultrasonic sensor are received by the Arduino Uno, which is then
responsible for controlling the system. It processes these signals to manage the waste segregation tasks efficiently.
B. Image Capture
1. OV7670 Camera
The camera module can be easily connected to other hardware devices due to its low cost. It is suitable for image

5194
 J. Electrical Systems 20-10 s (2024): 5191-5200

Fig. 3. OV7670 camera connection

processing applications and also interfaces well with microcontrollers. The module supports the connection to other
hardware devices through its multiple pins. The camera module can successfully link with other hardware components
via its several pins for power, ground, data transfer, and control signals. The primary pins include power (VCC), ground
(GND), and several data and control pins (SIOC, SIOD, VSYNC, HREF, PCLK, XCLK, and D0-D7) for communication
and synchronization.
The OV7670 camera captures images by connecting to an Arduino-based system. The camera’s internal settings are
configured using setup routines, which also determine important details for taking pictures, like color format and image
resolution. After the setup is complete, the camera begins taking pictures, changing light into electric signals and then
into digital image data.
The Arduino setup initializes the OV7670 camera and any linked displays, and it also sets the system’s clock rates if
necessary. The processFrame() function manages the primary image capture procedure inside the program loop. The way
this function behaves depends on the setup parameters, such as pixel-by-pixel processing or buffered processing. When
the camera is set to buffer mode, the code reads lines from the camera and displays them on the screen during blank lines.
Once the camera is set to pixel mode, data is processed pixel by pixel, with each pixel reading loop delivering data to the
screen. Through the collaborative approach, hardware and software components are seamlessly integrated, enabling
continuous image capture and presentation on connected output devices.
2. Arduino Uno Microcontroller
Besides managing proximity detection, the Arduino Uno regulates the OV7670 camera, ensuring timely and precise image
capture. The Arduino transmits the image data to the image repository after capturing the image.
3. Image Repository
The captured images are stored in a designated image repository, where they are temporarily held before being processed
for waste classification. The repository ensures that the latest image is readily available for analysis by the backend server.
4. Backend Server
The backend server gets signals from the Arduino Uno, fetches the most recent image from the repository, and processes
it with a pre-trained Convolutional Neural Network model. The server categorizes the waste based on the image data, and
the classification results are forwarded for further processing.

C. Waste Classification
1. Image Preprocessing
5195
 J. Electrical Systems 20-10 s (2024): 5191-5200

Images are preprocessed before being fed into the CNN model. During this stage, they are resized to a standard resolution
of 180x180 pixels, which enhances efficiency and consistency in classification. Additionally, normalizing pixel values
through re-scaling ensures that the image data falls within the suitable range for the model.
2. Underlying CNN Model
The processed images are given to a CNN (Convolutional Neural Network) model, specifically EfficientNetB0, which
has been optimized for this classification task. This model utilizes pre-trained weights from ImageNet to categorize the
images into one of nine waste types. Leveraging its deep learning architecture, the CNN effectively determines the type
of waste by extracting and analyzing details from the images.

D. Output
The waste classification model organizes waste into nine specific categories: Aluminium, Carton, Glass, Organic Waste,
Plastics, Paper and Cardboard, Other Plastic, Textiles, and Wood. These categories are further consolidated into three
main groups: Plastic, Organic, and Recyclable. The strategy includes utilizing a servomotor to align the correct bin
beneath the flap, ensuring proper waste disposal. This approach seeks to enhance the efficiency and precision of waste
management, resulting in a more automated disposal process.

IV. EXPERIMENTATION
1. Description of Dataset
The dataset [14] used for this project includes images sorted into nine categories: Aluminium, Carton, Glass, Organic
Waste, Plastics, Paper and Cardboard, Other Plastic, Textiles, and Wood. To maintain consistency in input dimensions for
the model, each image is resized to 180x180 pixels. Figure 4 offers a visual representation of some images from the
dataset, showcasing the variety of waste types included.

Fig. 4. Training dataset

2. Data Pre-processing
In the data preprocessing step, the images are rescaled to normalize their pixel values from the original range of 0-255 to
a range of 0-1. Normalizing the input data ensures that the neural network’s learning process is not adversely affected by
any single input during training, thus allowing the network to gain more convergence during training. Additionally,
TensorFlow’s data augmentation techniques, such as random flip, random rotation, random shear, and random zoom, are
applied to enhance the dataset’s quality. The image augmentation techniques create different versions of the training data.

5196
 J. Electrical Systems 20-10 s (2024): 5191-5200

This expansion of the dataset diversity helps improve the model’s robustness and its capacity to perform well on
previously unseen data. By exposing the model to various scenarios that it may encounter in the real world, this technique
leads to more accurate and reliable results.
3. CNN Model Development
EfficientNetB0 is a type of convolutional neural network (CNN) that has gained popularity due to its efficiency in terms
of model size and computation. Its efficiency is achieved through a method called compound scaling, which balances
depth, width, and resolution scaling to improve performance without significantly increasing computational costs. To
manage the size of the model, it introduces a compound coefficient called phi (Φ), with larger and more powerful models
being associated with higher values of Φ. The architecture incorporates MobileNetV2 blocks, depth-wise separable
convolutions, and squeeze-and-excitation blocks to efficiently extract and represent features.
This is loaded with pre-trained weights from ImageNet, a large-scale dataset commonly used for image classification
tasks. The input shape is set to (180, 180, 3) to match the size of the images in the dataset. The top layer of the model is
dropped to exclude the top classification layers of the pretrained model, enabling the addition of custom layers tailored
to the classification task.
Further the layers of the pre-trained EfficientNetB0 model are frozen so that they are not updated during training. This is
often done to leverage the knowledge learned from ImageNet while fine-tuning the model for a specific task.
The Sequential model is then defined, consisting of the pre-trained EfficientNetB0 model, a GlobalAveragePooling2D
layer to reduce spatial dimensions, Dense layers with a swish activation function, Dropout layers to reduce overfitting,
and finally, an output layer with 9 units (corresponding to each category in the dataset) and a softmax activation function
for multi-class classification.
After training, the model achieved a training accuracy of 98.65% and a validation accuracy of 91.65%. This indicates
strong performance in classifying the images into their respective categories. Figure 5 depicts the classification report for
model performance insights.

Fig. 5. Classification Report - Training dataset

Finally, a backend server is developed for real-time waste classification functions by leveraging a Flask framework in
Python and the trained model for image classification. Upon receiving a signal indicating the user’s proximity through
the COM port, the server retrieves the latest image from the image repository where the camera stores captured images.
This image is then analysed using the trained model for waste classification. The model predicts the category of waste in
the image, which is mapped to one of the predefined classes such as organic, plastic, or recyclable.

5197
 J. Electrical Systems 20-10 s (2024): 5191-5200

V. RESULTS

TABLE I
CLASSIFICATION RESULTS FOR TEST CASES
Image True Classified Misclassified
Category As / (%) As / (%)

Recyclable Plastic
Recyclable
(90.00%) (4.00%)

Organic Recyclable
Organic
(89.29%) (5.36%)

Plastic Organic
Plastic
(95.74%) (2.13%)

Plastic Recyclable
Plastic
(95.74%) (4.26%)

Recyclable Plastic
Recyclable
(90.00%) (4.00%)

Organic Recyclable
Organic
(89.29%) (5.36%)

To guarantee the SWS’ correctness and resilience, a series of comprehensive test cases were created and run. These test
cases were created to evaluate key features of the system, such as its overall reliability, input responsiveness, and accuracy
in classifying different kinds of waste products. A wide range of parameters were included in the test scenarios, such as
the composition of different waste materials, lighting conditions, and distances between the waste and the sensors. The
project team carried out very many tests for finding possible trouble spots to weed out inefficiency in SWS.

Fig. 6. Confusion Matrix - Test cases

5198
 J. Electrical Systems 20-10 s (2024): 5191-5200

The confusion matrix in Figure 6 helps to understand where this classification model does well or poorly by providing
insights into specific areas that it does well or struggles with. It included various test cases classifying waste into three
categories: Plastic waste, Recyclable waste, and Organic waste. Some examples of these test cases include Paper, Banana,
Organic waste in plastic, Plastic wrappers, Medicinal waste, and Organic waste. The matrix shows that ‘Plastic waste’
was correctly predicted 45 times but had minor misclassifications into ‘Recyclable waste’ (2 times) and ‘Organic waste’
(1 times). ‘Recyclable waste’ was correctly predicted 45 times, with 2 instances misclassified as ‘Plastic waste’ and 1 as
‘Organic waste’. ‘Organic waste’ was correctly predicted 50 times but was misclassified as ‘Plastic waste’ (4 times) and
‘Recyclable waste’ (3 times).
This confusion matrix explains the capability of the model to differentiate various types of waste and would provide
where it mostly errs giving insights into specific instances where misclassifications occur and guiding further
improvements.
The confusion matrix above also illustrates that the waste products are misclassified as organic waste with a very low
frequency. This is crucial as maintaining the purity of organic waste is a primary goal of our system, ensuring that it
remains uncontaminated and suitable for composting or other organic recycling processes.

VI. CONCLUSION
SWS has managed to present a system where solid waste management can be done using advanced hardware and software
technologies. The implementation is with the OV7670 camera, HC-SR04 ultrasonic sensor, and robust backend server in
real-time classification of the waste, with particular attention to identifying organic waste— which is very important for
effective segregation of waste and environmental sustainability. SWS applies pre-trained EfficientNetB0 and data
augmentation techniques for high accuracy in classifying the waste into organic, plastic, and recyclable materials; this
ensures their proper disposal and helps to enhance general waste management.
The proper sorting of organic waste enhances not only the management of solid waste but also paves the way for
prospective use, such as composting or generating biogas. This feature shows how SWS can be instrumental in helping
to keep our environment healthy. It is an excellent demonstration of the possibility of using smart and effective solutions
to solve common problems. The technology creates a precedent that calls for subsequent scientific work, which may give
rise to more sophisticated means of waste management, applicable on a broad basis in homes as well as industries.

VII. FUTURE SCOPE

The future development of the SWS system aims to improve its efficiency and scalability by incorporating a camera, a
servo motor-driven partitioner board, and a conveyor belt with a periodic nozzle for automated sorting. This setup enables
the camera to identify and sort waste as it moves along the conveyor belt, with the partitioner board directing it into the
appropriate compartment. This automated process increases both the speed and accuracy of waste segregation, making
the system scalable for larger facilities like industrial sites and recycling centres. Additionally, the system will integrate
advanced sensors for improved classification and smart bins with IoT capabilities for real-time waste management data.
By expanding the range of waste categories the system can handle, including more complex materials, SWS is intended
to contribute to a more sustainable waste management ecosystem by reducing manual labour, lowering operational costs,
and enhancing overall efficiency.

5199
 J. Electrical Systems 20-10 s (2024): 5191-5200

REFERENCES
[1] S. H. Sunny, D. R. Dipta, S. Hossain, H. M. R. Faruque, and E. Hossain. Design of a convolutional neural network
based smart waste disposal system. In 2019 1st International Conference on Advanced Science and Engineering
Research and Technology (ICASERT), pages 1–5, 2019.
[2] B. S. Sasikanth, L. Naga Yoshita, G. N. Reddy, and M. P.V. An efficient smart waste management system. In 2021
International Conference on Computational Intelligence and Computing Applications (ICCICA), pages 1–6, 2021.
[3] K. C. Saranya, V. Sujan, B. Abivishaq, and K. N. Kanna. Smart bin with automated metal segregation and optimal
distribution of the bins. In Emerging Technologies for Agriculture and Environment, Lecture Notes on
Multidisciplinary Industrial Engineering, pages 115–125. Springer Singapore, 2020.
[4] V. Kesthara, N. Khan, S. P. Praveen, C. Mahesha, and N. Murali. Sensor based smart dustbin for waste segregation
and status alert. International Journal of Latest Technology in Engineering, Management and Applied Science,
VII(IV):171–173, 2018.
[5] R. Manikandan, S. Jamunadevi, A. Ajeyanthi, M. Divya, and D. Keerthana. An analysis of garbage
mechanism for smart cities. International Research Journal of Engineering and Technology, 6(1):1709– 1713, 2019.
[6] Lochan Shruthi, Shereya PV, Pavithra SV, and V Murali Bhaskaran. Automated waste segregation into biodegradable
and non-biodegradable fractions utilizing conveyor belt and image processing with design thinking, 2023. Student,
Department of Computer Science and Business Systems Engineering, Rajalakshmi Engineering College, Chennai,
Tamil Nadu, India; Professor, Dean Academics, Rajalakshmi Engineering College, Chennai, Tamil Nadu, India.
[7] Feisal Maulana, Theo Widyanto, Yudi , and Kusprasapta Mutijarsa. Design and development of smart trash bin
prototype for municipal solid waste management. pages 1–6, 10 2018.
[8] Tejashree Kadus, Pawankumar Nirmal, and Kartikee Kulkarni. Smart waste management system using iot.
International Journal of Engineering Research and, V9, 05 2020.
[9] Sehrish Munawar Cheema, Abdul Hannan, and Ivan Miguel Pires. Smart waste management and classification
systems using cutting edge approach. Sustainability, 14(16), 2022.
[10] Mohammed M. Ahmed, Ehab Hassanien, and Aboul Ella Hassanien. Iot-based intelligent waste management system.
International Journal of Environmental Science and Technology, 2023. doi: 10.1007/s13762023-04567-2.
[11] Mohammed MA, Abdulhasan MJ, Kumar NM, Abdulkareem KH, Mostafa SA, Maashi MS, Khalid LS, Abdulaali
HS, and Chopra SS. Automated waste-sorting and recycling classification using artificial neural network and
features fusion: a digital-enabled circular economy vision for smart cities. Multimedia Tools and Applications,
81(1):1–16, July 2022. Epub ahead of print.
[12] Ayoub Benali Amjoud and Mustapha Amrouch. Object detection using deep learning, cnns and vision transformers:
A review. IEEE Access, 11:35479–35516, 2023.
[13] Satya Prakash Yadav, Muskan Jindal, Preeti Rani, Victor Hugo C. de Albuquerque, Caio dos Santos Nascimento,
and Manoj Kumar. An improved deep learning-based optimal object detection system from images. Multimedia
Tools and Applications, 83(10):30045–30072, March 2024.
[14] Vijay Gadre. Waste classification using resnet152: Wasteimagesdataset, 2024. Accessed: 2024-08-21.

5200