ILCNN: An Improved Lightweight Convolutional Neural Network

Based Recycling Garbage Classification Strategy with Image

Processing Technique

S. G. Balakrishnan, S. Abinaya, M. Harinishree, M. Jansitha and S. Kalaivani

Department of Computer Science and Engineering, Mahendra Engineering College, Tamil Nadu, India

Keywords: Lightweight Network, Garbage Recycling, CNN, Garbage Classification, Image Processing, Convolutional

Neural Network, ILCNN.

Abstract: Domestic garbage has grown at an alarming rate in recent years, making the use of intelligent waste sorting

technology an absolute necessity. Unfortunately, embedded garbage classification devices aren't a good fit for

current garbage classification algorithms due to their high parameter counts and bad real-time performance.

More and more people are using traditional garbage cans, which mean there's a growing need for effective

segmentation and identification algorithms. Modern computer systems' increased processing power and more

effective picture recognition technologies are in line with this desire. By utilizing an image processing logic

known as Improved Lightweight Convolutional Neural Network (ILCNN), a new garbage classification

procedure is established, which decreases the time and expenses associated with waste segregation. This helps

to solve these challenges against test how well the suggested model works, it is compared against a standard

deep learning model known as a Convolutional Neural Network (CNN). Reducing the need for human

involvement and increasing efficiency in garbage segregation are the goals of automating the process. Using

a publicly accessible dataset that included pictures of different kinds of garbage gathered from different

places, we ran the various state-of-the-art deep learning models. On this dataset, we fine-tuned pre-trained

ILCNN models using image augmentation approaches and transfer learning techniques. With its proposed

ILCNN model, the network can classify and recognize garbage with great accuracy while using very little

energy.

1 INTRODUCTION

Sorting garbage, sometimes known as garbage

categorization, is the process of dividing waste into

distinct groups according to its characteristics

(Shanshan Meng, et al., 2020). Simplifying disposal

is the primary goal of the project, which will lead to

improved recycling, less environmental effect, and

support for sustainability. (Sri Kruthika M, et al.,

2024). It is common practice to classify waste into

many types, such as: 1. Organic waste, which

includes things like discarded food, yard trimmings,

and paper goods. 2. Items that may be recycled

include paper, cardboard, glass, specific types of

plastic, aluminum, and metals. 3. Items that are

considered hazardous garbage include old batteries,

electronics, paints, solvents, pesticides, and certain

chemicals. 4. Mixed garbage that cannot be recycled,

including plastic bags, some forms of packaging, used

diapers, and sanitary goods. 5. Industrial, medical,

and construction-related garbage are examples of

specialized waste streams. In most cases,

communities and individuals need to be educated on

the need of waste segregation, separate methods for

collecting and labeling garbage, and facilities for

proper disposal and recycling in order to sort garbage

properly. (Yu Song, et al., 2023).

Governments, local authorities, corporations, and

individuals all have a role to play in promoting

efficient waste management systems and decreasing

the environmental impact of waste. Extensive

research has been conducted in the crucial and

demanding area of object detection within computer

vision. The objective of object detection is to identify

and categorize all items. Its applications are

practically endless: driverless vehicles, medical

imaging, robot vision, intelligent video surveillance,

remote sensing pictures, etc. (Yang Shen, et al.,

2023).

Balakrishnan, S. G., Abinaya, S., Harinishree, M., Jansitha, M. and Kalaivani, S.

ILCNN: An Improved Lightweight Convolutional Neural Network Based Recycling Garbage Classiﬁcation Strategy with Image Processing Technique.

DOI: 10.5220/0013876900004919

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 1st International Conference on Research and Development in Information, Communication, and Computing Technologies (ICRDICCT‘25 2025) - Volume 2, pages

59-69

ISBN: 978-989-758-777-1

The quantity of domestic garbage has grown

substantially in recent years due to rising urbanization

and rising living standards. Incineration is the current

gold standard for waste disposal; however, sorting

waste before burning can help make better use of

available resources. Conventional waste sorting

involved transporting household waste to a treatment

facility, where workers would stand on either side of

a conveyor belt and manually sort the material by

hand or with implements. Nevertheless, this method

of garbage sorting requires a lot of workers, doesn't

get the job done very well, and the pollution and smell

coming from the treatment plant might be harmful to

the workers' health (He Bai, et al., 2021).

The burden of post-recycling treatment can be

substantially alleviated if sorting can be finished at

the time of waste collection and recycled in separate

bins according to distinct categories. There has been

a rise in the visibility of intelligent garbage cans,

recycling bins, and machine vision-based intelligent

garbage categorization systems owing to

advancements in AI technology (Yujin Chen, et

al.,2023).

The goal of this project is to create software that

can analyze gathered photographs and identify cases

of abandoned garbage. Create a computer vision

model that can sort waste by material, including

paper, metal, glass, cardboard, and rubbish. To make

recycling the waste materials as easy as possible and

the worker's health may be harmed by the physical

labor (Zhichao Chen, et al.,2022).

As a result, this study uses a self-constructed

rubbish dataset for rubbish detection and

classification for a small, efficient classification

model. To enhance the spatial feature perception of

the network model, we apply the Improved

Lightweight Convolutional Neural Network

(ILCNN) attention mechanism. Highly efficient and

effective models across a range of tasks are produced

by EfficientNet through the application of a

compound scaling approach, which balances the

depth, breadth, and resolution of the models.

Efficiency in terms of parameters and computing cost

allows it to attain extremely excellent performance

(Kishan PS, et al., 2021).

• In order to enhance the recognition performance

of the pre-trained MobileNetV3 model in waste

classification tasks, a dataset was created that

featured four typical forms of waste from

homes.

• Instead of using SE-Net, the model is equipped

with CBAM (Convolutional Block Attention

Module) to improve its spatial perception of

features. This allows it to adaptively emphasize

or suppress distinct feature information based on

the distribution of feature maps.

• The convolution layer employs the Mish

activation function to enhance deep networks'

information representation ability and

generalization performance.

• In order to decrease the model's parameter,

count and prevent overfitting, the classifier opts

for global average pooling rather than a

complete connection layer.

Waste categorization using deep learning

algorithms, especially ILCNN can assist overcomes

some of the problems that come with conventional

machine learning methods. The following are some

examples of how deep learning could help with these

problems: Feature extraction, from unstructured data,

deep learning systems may automatically learn

feature hierarchies. Because of this, diverse machine

learning models no longer require feature engineering

that is done by hand and improving the model's

capacity to distinguish between various forms of

waste, ILCNN use convolutional layers to extract

useful characteristics from rubbish images (Shoufeng

Jin, et al., 2023).

Reduced requirement for massive volumes of

labeled data is a result of size-transfer learning, which

permits fine-tuning of pre-trained ILCNN models

(e.g., trained on ImageNet) on smaller garbage

classification datasets. The ability to generalize to

new contexts and unknown data is strength of pre-

trained models on big and varied datasets.

Additionally, domain adaptation and data

augmentation are two methods that can take model

generalization to the next level. Considerations such

as model complexity, processing resources, and

classification accuracy determine the relative merits

of the many deep learning models available for waste

image categorization (Wei Liu, et al., 2023).

2 RELATED WORKS

Effective environmental sustainability and garbage

management depend on correct garbage classification

(Kirit Rathod, et al.,2024). Typical methods of waste

classification rely on hand sorting. This might be a

somewhat demanding and error-prone procedure that

leads the government to carry out insufficient policy.

In this research, we offer a garbage categorization

method that uses deep learning and GCDN to

automate and enhance the accuracy of the process.

Shoe, green-glass, paper, cardboard, battery,

biological, plastic, metal, brown-glass, white-glass,

and waste are among the twelve types of trash that our

ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,

COMMUNICATION, AND COMPUTING TECHNOLOGIES

system uses through an extra layer of convolutional

neural networks (CNNs). Utilizing a publicly

accessible dataset that comprises images of diverse

waste materials gathered from numerous sites, we

have educated many cutting-edge deep learning

models. We applied picture enhancing methods

before using transfer learning methods on pre-trained

CNN models on this dataset. We obtained a training

phase classification accuracy of 98.64% and a

validation phase accuracy of 93.23% according to our

results analysis. Experimental findings reveal that

our approach effectively identifies garbage in

challenging environments with different backgrounds

and illumination levels. We also discuss some of the

uses for our technology in smart garbage cans and

recycling facilities to simplify environmental and

consumer sorting of waste.

The appropriate categorization of garbage is

crucial for efficient waste management and the long-

term viability of the ecosystem. Using Convolutional

Neural Networks (CNNs), this study (Al Mahmud Al

Mamun, et al., 2024) thoroughly explores garbage

categorization. Our goal is to develop a deep

learning-based system that can classify waste with

remarkable accuracy. Proving its efficacy in waste

classification, the suggested CNN model attains a

staggering 98.45% accuracy (Wenbo Liu, et al.,

2024). The study covers all aspect, including data

gathering and preprocessing, model creation, training

techniques, and assessment. Results show that CNNs

can transform waste management for the better and

start a more environmentally friendly age.

Incorrect garbage classification may lead to

environmental pollution by means of recycling.

Designed on a convolutional neural network (CNN),

an enhanced garbage sorting and recycling system is

provided to effectively handle this problem. This

article offers an approach based on a well-designed

deep network topology to maximize the model

parameters. With an aim of increasing the processing

speed and accuracy in classification, it then employs

the graphics processing unit (GPU) to achieve parallel

processing and batch processing. With an accuracy

of at least 94% and as high as 99%, the CNN-based

garbage categorization system showed rather good

performance according to the research. Direct

outcome of the system design is better accuracy in

garbage classification. Using autonomous feature

learning and deep network architecture helps the

system to more identify significant components in

waste images. This guarantees efficient garbage

recycling and helps to increase the accuracy of

categorization results.

The destruction of past waste disposal methods is

a logical result of the always faster rate of junk

creation, which forces an unavoidable choice of

garbage classification (Qingqiang Chen, et al., 2020).

Also, much of interest is the accuracy of identification

and the multi-category classification of waste. The

present garbage classification systems lack diversity,

low accuracy, and a single category emphasis that

results in with an average accuracy of 64% and 92

frames per second, the paper proposes to detect 15

objects across 3 categories using the upgraded

YOLOV4 network framework. The upgraded

YOLOV4 fits well with embedded devices as they

increase its capacity to recognize different waste

kinds.

As the world population keeps growing today,

pollution levels are also rising (Volkan Kaya, 2023).

One main cause of environmental contamination is

harmful compounds found in garbage. From

incorrect waste management, damage to ecosystems

and human health is considerable. Particularly as

technology develops, the recyclability of the raw

materials used to create waste products affects both

national demands for resources and energy savings.

As such, recycling facilities do a lot of traditional

chores to re-use recyclable waste across different

countries. These operations start with pre-processing

and physical waste collecting, which depend on

human labor. Apart from endangering people, this

habit damages the surroundings as well.

This fact generates the need for an intelligent

system competent of independently recognizing and

classifying waste items. This work automatically

sorted waste by material using powerful deep learning

algorithms like Xception, InceptionResNetV2,

MobileNet, DenseNet121, and EfficientNetV2S.

Based on transfer learning methods, it also suggested

two further algorithms:

Xception_CutLayer and InceptionResNetV2_CutLa

yer. Using a dataset comprising six different kinds of

garbage, we trained and assessed the suggested

methodologies and deep learning techniques

grounded on artificial intelligence. Utilizing the

suggested Xception_CutLayer technique and 85.77%

utilizing the InceptionResNetV2_CutLayer method,

the study revealed that these methods exceeded the

others in terms of classification success rate 89.72%.

3 METHODOLOGY

In order to manage waste and keep the environment

sustainable, garbage categorization is crucial. The

government's policies may be inadequate since

ILCNN: An Improved Lightweight Convolutional Neural Network Based Recycling Garbage Classiﬁcation Strategy with Image Processing

Technique

traditional waste categorization systems sometimes

rely on manual sorting, which is both labor-intensive

and susceptible to human mistake. Unfortunately,

embedded garbage classification devices are not a

good fit for current garbage classification algorithms

due to their high parameter counts and bad real-time

performance. The volume of waste is causing

cleanup to be quite a hassle. Waste separation, one of

the desirable recycling activities, is by far the most

important step in achieving low-cost recycling.

When thinking about smart and automated cities, one

of the efficiency challenges of waste management as

it is currently done is that there are no developed

systems for garbage collection. These past researches

have lots of drawbacks such as:

• The current method of waste sorting is

inefficient, requires a lot of workers, and may be

harmful to their health due to the pollution and

smell coming from the treatment facility.

• The computing demands on garbage

classification devices are significant because

large-scale models contain a huge number of

parameters. Due to the high level of hardware

limitation, these devices are both large and

expensive.

• In the past, workers would stand on each side of

a conveyor belt and physically sort household

garbage as it came into the waste treatment

facility. They would use their hands or

equipment to separate the different types of

waste.

In comparison to the current technology,

Convolutional Neural Network (CNN), the suggested

technique, Improved Lightweight Convolutional

Neural Network (ILCNN) increases the model's

recognition accuracy and streamlines waste

management processes. The following figure 1 shows

the system architecture and the following figure 2

shows the system flow diagram.

Intelligent garbage cans, recycling bins, and

machine vision-based intelligent garbage sorting

systems have recently made headlines, thanks to

advancements in AI technology and suggested a

network-based garbage categorization system that

utilized deep learning to automate and enhance the

accuracy of this procedure. To sort garbage into

different types, such paper, cardboard, plastic, green-

glass, etc., our system leverages an extra layer of

enhanced lightweight convolutional neural networks

(ILCNNs). A publicly accessible dataset including

images of different kinds of waste gathered from

different places has been used to train the various

state-of-the-art deep learning models. Then, we

utilized image augmentation and transfer learning

strategies to this dataset to fine-tune the ILCNN

models that had already been trained.

Figure 1: System Architecture.

ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,

COMMUNICATION, AND COMPUTING TECHNOLOGIES

Figure 2: System Flow Diagram.

This improved its recognition performance in

garbage classification tasks.

• The convolution layer makes use of the Mish

activation function to enhance the deep network's

information representation ability and

generalization performance.

• To decrease the model's parameter, count and

prevent overfitting, the classifier opts for global

average pooling rather than a complete

connection layer.

• In order to get the most out of the pre-trained

model for garbage categorization tasks, we built

it using four prevalent forms of household waste.

In the implementation phase, the original theoretical

design for the project is turned into a functioning

system. For this reason, this is the most important

step in the design of a new system that the user will

rely on and trust will work well. The implementation

step will require planning everything out, research on

the existing system and its shortcomings, devising a

plan to switch to a new system, and reflecting on the

effectiveness of that process.

3.1 Data Collection

Although the original MobileNet pre-trained model is

trained on the ImageNet dataset, it is challenging to

guarantee the effectiveness of transfer learning as the

data in ImageNet does not entirely contain the images

we need. To thus fine-tune the model, we require a

dataset including garbage photos. There isn't yet a

standard dataset for waste categorization projects.

Though it has relatively few categories, which is not

in line with the real state of residential garbage

categorization in Indian, the garbageNet dataset for

waste classification is still useful. Thus, this work

generates a data set especially used for visual garbage

sorting through network retrieval and laboratory real

scene shooting, including many scenes such single

object, multiple similar objects, complex background,

and various interference situations such lighting and

motion blur. With 4152 JPG photos overall, it is split

into four categories: kitchen garbage, recyclable

waste, hazardous waste, and other waste.

ILCNN: An Improved Lightweight Convolutional Neural Network Based Recycling Garbage Classiﬁcation Strategy with Image Processing

Technique

3.2 Data Preprocessing

Organize and preprocess the images. Among other

preprocessing methods, this can involve scaling,

normalizing, noise reduction, contrast enhancement,

and other approaches.

3.3 Data Splitting

Three subsets made out of the dataset were validation,

training, and testing. The training set is used to train

the model; the validation set is used to fine-tune the

hyper parameters; the testing set evaluates the

performance of the produced model. This

fundamental stage is needed in the development and

application of machine learning algorithms for image

classification chores. Open your preferred data

analysis tool (python or C) then import the dataset for

depicts categorization. One should verify if the

dataset follows an ordered layout.

• Perform the necessary data preparation

activities prior to data division. Examples of this

are handling waste connected to missing

images.

• Shuffling your dataset before beginning a

machine learning project helps us to avoid our

model from learning patterns that could be

connected to the sequence of the data. A dataset

could be shuffled both before and after the data

loading process.

• The model gets trained with it. Validation set is

used to adjust hyper parameters. Testing set: It

helps to evaluate the performance of the final

model.

3.4 Feature Extraction

Extract relevant information from the images and the

deep learning methods could mean learning

automatically features using convolution layers.

Machine learning methods would extract features for

traditional models using histograms of oriented

gradients. Choose sensible model architecture. For

deep learning techniques, this often involves ILCNN.

Train the given model on the provided training set.

Tune your chosen hyper settings based off of

performance on the validation set. Training involves

adjusting weights of the models so that the loss

function is minimized. Assess the model using the

test set. Common performance measures are

accuracy and other related metrics. Every model can

be optimized further to increase the performance of

the model on the test set. This could mean tuning

hyper parameters, collecting more data, using data

augmentation, regularizing the model to prevent

overfitting, etc. Assess model prediction and errors

to understand the behavior of model. This phase may

highlight the errors of the model and facilitate future

development of the model.

3.5 ILCNN Architecture

After preprocessing the image waste classification,

we selected an ILCNN architecture that performs well

for waste image classification. For classification, we

selected EfficientNetV2B1, and included weights for

this model pretrained from an EfficientNetV2B1-like

dataset. These types of models can be found in deep

learning utilities such as TensorFlow or Pytorch. We

removed the previous classification head from

EfficientNetV2B1 pretrained model and added our

own classification head with the same number of

garbage classes in our dataset. To extract hierarchical

and discriminative features from the garbage images

we utilized ILCNN layers. To tune the model weights

during training we need to use an optimizer such, as

Adam or SGD, and an appropriate loss function, such

as categorical cross-entropy, for multi-class

classification. Finally, to deliver the final waste

classifications, we transferred the extracted features

into the ILCNN architectural scheme, again by

overlaying fully connected layers.

3.6 Training and Testing

After the model is developed, it has to be trained

using a large number of images that have the

necessary objects tagged. Keep in mind that in order

for the EfficientNetV2B1 model to learn to

differentiate between the different classes

appropriately, the data must be balanced. Once the

data is prepared, it has to be put into the

EfficientNetV2B1 model. Dataset size determines

whether this is best done in batches or in one

continuous run. Next, a suitable optimizer, such as

Adam or SGD, has to be used to train the model. In

order for the model to learn to identify different

objects in the images, its weights are modified

continually during the training phase. After then, the

testing set may be used to assess the model's

correctness. There are a number of ways to evaluate

the model's efficacy, including recall, accuracy, and

F1 score. One way to measure the model's

performance is to see how many photos it accurately

labels. It is also possible to test the model on new data

to determine how well it generalizes. The model's

performance with unobserved data will be shown by

this.

ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,

COMMUNICATION, AND COMPUTING TECHNOLOGIES

4 RESULTS AND DISCUSSION

The goal of this study was to evaluate the efficacy of

the proposed scheme by cross-validating it with the

traditional deep learning model, Convolutional

Neural Network (CNN), and classifying waste as

organic or recyclable using a deep learning-based

model called ILCNN. The research demonstrated that

ILCNN was capable of distinguishing between

different types of garbage, thereby facilitating the

development of more efficient recycling methods.

This study suggested an automated classification

method that utilizes ILCNN for efficient image

recognition. Using an optimization method with a

higher level of classification accuracy, the model that

has been presented emphasizes the role of ILCNN in

automating recycling tasks.

(a)

(b)

Figure 3: (A) Dataset View and (B) Data Visualization.

The objective of this investigation was to

categorize photographs of waste products into seven

categories: cardboard, glass, metal, organic, paper,

plastic, and refuse. Transfer learning methods were

employed to enhance the model's performance,

thereby emphasizing ILCNN's adaptability in

managing a variety of refuse categories. In a research

article, an ILCNN model was presented that achieved

ILCNN: An Improved Lightweight Convolutional Neural Network Based Recycling Garbage Classiﬁcation Strategy with Image Processing

Technique

a classification accuracy of 98.63% for a variety of

refuse categories. ILCNN's proficiency in automating

garbage categorization duties is illustrated by this

remarkable precision. The results emphasize the

effectiveness of ILCNN in accurately classifying

waste products, and these studies collectively

demonstrate the extent to which ILCNN automates

garbage classification, thereby promoting more

sustainable environmental practices and effective

recycling methods. A clear and concise representation

of the dataset view and the data visualization

perspective is provided in the following figure, which

is referred to as Figure 3.

The training accuracy and validation accuracy

assessments of the suggested scheme known as

ILCNN are depicted in the figure that can be seen

below, which holds the designation Figure 4.

Figure 4: Training and Validation Accuracy.

The training loss ratio and validation loss ratio of

the proposed scheme known as ILCNN are depicted

in the figure that can be seen below, which is referred

to as Figure 5.

Figure 5: Training and Validation Loss.

The web page for administrator authentication,

the website for picture uploading, and the page for

result prediction are all depicted in the following

images: Figure 6, Figure 7, and Figure 8. These

representations demonstrate the output of the

suggested method.

Figure 6: Administrator Authentication.

ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,

COMMUNICATION, AND COMPUTING TECHNOLOGIES

Figure 7: Image Uploading Port.

Figure 8: Result Prediction.

The prediction accuracy of the proposed scheme,

which is referred to as ILCNN, is depicted in the

following figure, which is referred to as Figure 9. In

order to evaluate the prediction accuracy of the

suggested scheme, it is cross-validated with the

traditional deep learning model known as CNN. A

descriptive representation of the same may be found

in the table that follows, which is referred to as Table-

ILCNN: An Improved Lightweight Convolutional Neural Network Based Recycling Garbage Classiﬁcation Strategy with Image Processing

Technique

Table 1: Analysis of Prediction Accuracy Between Ilcnn and Cnn.

Iterations

CNN (%)

ILCNN (%)

93.64

98.71

92.15

98.63

93.76

97.49

100

92.55

97.82

125

93.09

98.19

150

92.61

98.74

175

92.61

98.05

200

92.51

98.26

225

92.47

97.93

250

91.69

98.19

Figure 9: Prediction Accuracy Evaluation.

5 CONCLUSIONS AND FUTURE

SCOPE

The garbage detection and classification model is

built on top of an enhanced attention mechanism,

activation function, and classification layer

mechanism on the ILCNN. A recognition accuracy of

96.55% is achieved by the enhanced model on the

self-constructed garbage dataset. Based on the

findings, the suggested system may automate the

categorization work, which in turn reduces operating

expenses and human error, greatly improving the

efficiency of waste management procedures. By

substituting global average pooling for the fully

connected layer, the suggested garbage recognition

model ILCNN demonstrates how to decrease the

model's parameter count while simultaneously

improving the recognition result. To increase the

model's identification performance, the Mish

activation function allows for more precise feature

extraction from the target and better use of the

retrieved visual data.

Going forward, we intend to train the model using

a wider variety of garbage image data in order to

better prepare it for real-world application settings.

Eventually, garbage detection and classification with

little power consumption and great accuracy, serving

as a benchmark for researchers and engineers in the

upcoming years. The system's practical use in waste

management might be further enhanced by enabling

real-time waste categorization in varied situations

through its implementation on edge devices.

REFERENCES

Al Mahmud Al Mamun, et al., "Garbage classification

using convolutional neural networks (CNNs)", Material

Science & Engineering International Journal, 2023.

He Bai, et al., "Design of Garbage Classification System

Based on Artificial Intelligence Technology",

International Conference on Aviation Safety and

Information Technology, 2021.

Kirit Rathod, et al., "Garbage Classification based on Dense

Network (GCDN) using Transfer Learning and

Modified Hyper Parameter", International Journal of

Intelligent Systems and Applications in Engineering,

2024.

Kishan PS, et al., "Garbage Classification and Detection for

Urban Management", IJCTT Journal, 2021.

Qingqiang Chen, et al., "Garbage Classification Detection

Based on Improved YOLOV4", Journal of Computer

and Communications, 2020.

Shanshan Meng, et al., "A Study of Garbage Classification

with Convolutional Neural Networks", Indo – Taiwan

2nd International Conference on Computing, Analytics

and Networks, 2020.

Shoufeng Jin, et al., "Garbage detection and classification

using a new deep learning-based machine vision system

as a tool for sustainable waste recycling", Waste

Management, 2023.

Sri Kruthika M, et al., "Garbage Classification: A Deep

Learning Perspective", International Research Journal

on Advanced Engineering Hub, 2024.

Volkan Kaya, "Classification of waste materials with a

smart garbage system for sustainable development: a

novel model", Frontiers in Environmental Science,

2023.

Wei Liu, et al., "Image Recognition for Garbage

Classification Based on Transfer Learning and Model

Fusion", Mathematical Problems in Engineering, 2022.

Wenbo Liu, et al., "A Garbage Intelligent Classification and

Recycling System Based on Deep Learning", Procedia

Computer Science, 2024.

Yang Shen, et al., "Domestic Garbage Classification and

Incentive-Based Policies in China: An Empirical

Analysis", Water, 2023.

Yu Song, et al., "DEEPBIN: Deep Learning Based Garbage

Classification for Households Using Sustainable

ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,

COMMUNICATION, AND COMPUTING TECHNOLOGIES

Natural Technologies", Journal of Grid Computing,

2023.

Yujin Chen, et al., "Classification and recycling of

recyclable garbage based on deep learning", Journal of

Cleaner Production, 2023.

Zhichao Chen, et al., "Garbage classification system based

on improved ShuffleNet v2", Resources, Conservation

and Recycling, 2022.

ILCNN: An Improved Lightweight Convolutional Neural Network Based Recycling Garbage Classiﬁcation Strategy with Image Processing

Technique