Garbage Detection using YOLO Algorithm for
Urban Management in Bangkok
MATHUROS PANMUANG1, CHONNIKARN RODMORN2*
1Department of Educational Technology and Communications, Faculty of Technical Education,
Rajamangala University of Technology Thanyaburi,
Pathum Thani 12110,
THAILAND
2Department of Applied Statistics, Faculty of Applied Science,
King Mongkut’s University of Technology North Bangkok,
Bangkok 10800,
THAILAND
*Corresponding Author
Abstract: - Garbage problems in urban areas are becoming more serious as the population increases, resulting
in community garbage, including Bangkok, the capital of Thailand, being affected by pollution from rotten
waste. Therefore, this research aims to apply deep learning technology to detect images from CCTV cameras in
urban areas of Bangkok by using YOLO to detect images from CCTV cameras in urban areas of Bangkok,
using YOLO to detect 1,383 images of overflowing garbage bins, classified into 2 classes: garbage class and
bin class. YOLO in each version was compared, consisting of YOLOv5n, YOLOv6n, YOLOv7, and
YOLOv8n. The comparison results showed that YOLOv5n was able to classify classes with an accuracy of
94.50%, followed by YOLOv8n at 93.80%, YOLOv6n at 71.60%, and YOLOv7 at 24.60%, respectively. The
results from this research can be applied to develop a mobile or web application to notify of overflowing
garbage bins by integrating with CCTV cameras installed in communities to monitor garbage that is
overflowing or outside the bin and notify relevant agencies or the locals. This will allow for faster and more
efficient waste management.
Key-Words: - Garbage detection, Overflowing garbage bins, YOLO, Deep Learning, Machine Learning, Image
Processing.
Received: March 19, 2023. Revised: October 26, 2023. Accepted: December 27, 2023. Published: February 27, 2024.
1 Introduction
The world's population is growing and is likely to
reach 10 billion in 2050 from the current level of 7.6
billion, causing urban expansion, increasing rapidly
as the human population increases. As a result of
urban expansion, environmental problems follow,
[1]. Environmental problems have become a
problem in the world and Thailand today as people
neglect to take care of proper management of
natural resources and the environment, causing
impacts on the ecosystem and human life, [2]. Key
environmental problems are garbage or solid waste.
Currently, the amount of solid waste or waste is
increasing every year which is caused of an increase
in population, and economic and industrial
expansion, which is considered an important
problem for the community that must be managed
and solved in terms of solid waste, sewage, and
toxins that contaminate water sources, soil, and air,
[3]. Foreign studies have found that developing
countries are unable to manage solid waste
sustainably, resulting in severe environmental
degradation, [4]. At present, it is found that every
urban community in Thailand faces problems with
the density of houses and the large number of
businesses and shops, but with limited space and
natural resources remaining the same as well as an
increasing population, which results in the problem
of community waste, [5]. Bangkok, the capital of
Thailand, is the center of prosperity in every aspect,
both economically and socially. Bangkok is
therefore a city with a problem with the amount of
solid waste. The Office of the Environment's annual
work plan report for 2016 shows that each year
Bangkok has a continuous increase in the amount of
WSEAS TRANSACTIONS on COMPUTER RESEARCH
DOI: 10.37394/232018.2024.12.23
Mathuros Panmuang, Chonnikarn Rodmorn
E-ISSN: 2415-1521
236
Volume 12, 2024
solid waste and found that in 2015, Bangkok had a
high amount of solid waste of 10,167 tons/day and,
[6].
Fig. 1: Shows the average amount of solid waste from fiscal year 2013 - fiscal year 2022, [7]
The amount of waste found from a survey by the
Environment Agency from fiscal year 2013 - fiscal
year 2022 found that the amount of waste each year
is similar or may decrease in certain years, but still
considered to be a large amount (Figure 1).
At present, the Environment Agency has given
serious importance to waste management and
campaigns to make people aware of their
responsibilities in sorting and disposing of waste at
designated points. However, it still faces complaints
from citizens about the amount of residual waste
that gives off a bad odor and creates an unattractive
environment and the waste disposal that does not
comply with regulations, such as failure to dispose
of waste timely or in the right place. Currently,
Bangkok has installed more than 60,000 CCTV
cameras in the area, divided into CCTV cameras to
monitor traffic conditions and CCTV cameras for
safety. Therefore, to integrate, utilizing existing
CCTV cameras for waste management will make
waste management more efficient and create greater
satisfaction for citizens.
Therefore, in this research, image processing
techniques were applied to help manage waste by
analyzing images from CCTV cameras and using
the YOLO model for processing from deep learning
to classify the images into 2 classes: overflowing
garbage bin and bin class. The results obtained from
this research will allow Bangkok to develop further
into a mobile application to notify relevant agencies
or people in the area to know whether the waste is
overflowing. It is also a process that allows the
public, government, and private sectors to
participate in sustainable waste management.
2 Related Work
The 2016 algorithm known as YOLO (You Only
Look Once) offers quick speed and excellent
accuracy. The input image is divided into a grid, and
the likelihood of the bounding box and related
objects appearing in each grid cell is predicted. It
performs bounding box estimation and object
categorization using these predictions. Because of
its excellent accuracy and quick speed, it has been
used extensively in the field of table recognition
recently, [8]. The YOLO architecture is a single-
stage object detector with three main parts: the first
part is the backbone, which is responsible for
extracting image features; the second part is the
WSEAS TRANSACTIONS on COMPUTER RESEARCH
DOI: 10.37394/232018.2024.12.23
Mathuros Panmuang, Chonnikarn Rodmorn
E-ISSN: 2415-1521
237
Volume 12, 2024
model neck, a layer between the backbone and the
head that is used to extract different feature maps of
different states within the backbone, which helps in
dealing with images with different image scales
such as the feature pyramid network; and the last
part, the model head, is responsible for detecting
objects and is used to predict images in the
bounding box and calculate confidence scores.
Currently, YOLO is so popular that it has been
developed in several versions; until now in 2023, it
has developed into YOLOv8.
There are differences in the bounding boxes
between the two dataset formats. The network of
YOLO is made up of two fully linked layers after
twenty-four convolutional layers. The following is
how the YOLO framework works. There are n × n
grids inside the image. Next, each grid is subjected
to image localization and classification. The
bounding boxes and the associated class
probabilities are predicted by YOLO. Labeled data
is needed to train the model. A one-n-dimensional
vector called is present in each grid cell. The
value is an 8-dimensional vector made up of the
following if the image is divided into a 3 × 3 grid
and three classes need to be classified, [9].
(1)
Where is a number that indicates if an object
is present in that grid cell and might be either 0 or 1.
The bounding box's coordinates are described by
(, , , and ), and the existence of a certain
class in a given grid cell is indicated by the binary
values , , and .
For research that applies YOLO to object
detection, such as employing YOLOv4 for real-time
vehicle detection with improvement for increased
efficiency, refer to, [10]. The YOLOv4-tiny model
combined with Hard Negative Sample Mining
(HNEM) uses the CSPBlock module by using the
CSPDarknet-53 as the backbone to improve the
detection rate of occluded vehicles. The experiment
results found that improving the efficiency of real-
time vehicle detection becomes more accurate
compared to the research, [11], that used YOLOv5
to categorize the parts according to their lateral and
head shapes. The creation of an image acquisition
platform with two mounted cameras and an
appropriate lighting system to produce high-quality
photos is another element of this work. The
suggested deep learning (YOLOv5) method has
demonstrated encouraging results with a mAP@0.5
of 0.996 for component classification, despite the
difficulties connected with such systems. In
contrast, the suggested image processing method
produced a maximum inaccuracy of 0.05 mm for
pitch computation and 100% accuracy for standard-
size assignments. In addition, there is an
introduction to the utilization of YOLOv6 for
transferring learning to a real-time object detection
model, [12]. Another crucial component of this
work is the suggested model's ability to recognize
every object in a scene—indoor and outdoor—and
to alert the user to close and distant things via voice
output. The Google Text-to-speech (gTTs) library is
used to get the audio response. Following 30%,
40%, and 50% pruning of the YOLOv6 baseline
model, the optimized YOLOv6 framework achieves
37.8% greater average accuracy (AP) at 1235
frames per second (FPS).
Meanwhile, YOLOv7 is employed to introduce a
novel technique for detecting dragon fruit. This
technique goes beyond merely locating the fruit; it
also identifies the endpoints at its head and root,
[13]. YOLOv7 outperforms specific previous
models in this regard. Furthermore, the study
applied YOLOv8 to detect waste visually,
leveraging the efficiency of YOLOv8, the latest
object detection model in the YOLO series,
specifically for automated waste sorting, [14].
YOLOv8 is used for automated waste sorting to
improve safety and efficiency in waste treatment
procedures. The outcomes show that YOLOv8 is an
effective tool for enhancing waste management
procedures since it outperforms state-of-the-art
waste detection and classification algorithms.
However, research on the use of YOLO for
identifying different things was also conducted, and
it was discovered that deep learning techniques are
used for garbage detection. The study also creates a
garbage image categorization system based on deep
learning, [15]. Its primary goals are to compare deep
learning neural network models, identify the best
classifier, create online apps, and implement neural
networks. According to the results,
inceptionResnetv2 has an 89% detection accuracy
and a 0.8 loss value. It performs better in terms of
detection when compared to the two migration
models mentioned above.
Based on a review of related research, the
researcher applies YOLO for real-time CCTV image
detection to sort out overflowing garbage bins in
Bangkok by comparing the YOLO architecture in
each version including YOLOv5n, YOLOv6n,
YOLOv7, and YOLOv8n to get the model with the
best accuracy.
WSEAS TRANSACTIONS on COMPUTER RESEARCH
DOI: 10.37394/232018.2024.12.23
Mathuros Panmuang, Chonnikarn Rodmorn
E-ISSN: 2415-1521
238
Volume 12, 2024
3 Methodology
3.1 Data Collection and Preparation of
Datasets
The image datasets from the internet and video
images of Bangkok's public garbage dumps were
used for testing in this study. Then, the images were
enhanced using three image augmentation
techniques: horizontal image inversion, brightness
adjustment, and 90-degree rotation. This produced
1,383 images, which were then labeled into two
classes using the Roboflow tool: bin images and
garbage images. Later, the data set was divided into
3 parts: training with 1,105 images, validation with
139 images, and testing with 139 images, or
equivalent to 80:10:10. An example of the data set is
shown in Figure 2.
Fig. 2: Example of data set
3.2 Model Training
The images were resized to the same size, 640×640
pixels, in 3 channels. Then, the images were trained
with the validation dataset to measure the
performance of the model and set the threshold
equal to 0.25, the IoU (Intersection over union)
value equal to 0.50, trained for 100 epochs. In each
model, the researcher selected the model with the
smallest size to reduce the training time as when
testing the model, it was found that the smallest size
is sufficient for classification. Therefore, the models
used include YOLOv5n, YOLOv6n, YOLOv7, and
YOLOv8n.
YOLOv5n: The most recent network model in
the YOLOv5 series is the YOLOv5n. On the one
hand, the YOLOv5n network model features fast
reasoning speed and good detection accuracy. In
contrast, the YOLOv5n network model's weight file
is very small—roughly 75% less than YOLOv5s—
meaning that YOLOv5n is well suited for
deployment to embedded devices for real-time
detection because automatic driving lane curvature
target identification relies heavily on the precision,
real-timeliness, and lightweight nature of the model,
[16].
YOLOv6n: YOLOv6 through the development
of a parameterizable network structure based on
RepVGG-EfficientRep and Rep-PAN, as well as
improvements to the backbone network, neck,
detection head, and training method. Therefore,
YOLOv6n reduces more parameters and processing,
but the average recognition accuracy is significantly
lower as a result. Regarding models operating on
devices with high-performance computing, [17].
YOLOv7: Additionally, YOLOv7 makes use of
the idea of deep supervision. To guide the weight of
the external network, it provides an additional
auxiliary head structure in the middle network layer.
These mono-modality object recognition approaches
achieve performance as well as real-time inference.
However, these object detection models only use
one stream. Consequently, the advantages of each
stream—such as precise edges and suitable
illumination in infrared images and the object's
color and detail information in red-and-green
images—cannot be utilized by these models.
Enhancing object detection performance across all
streams requires appropriate feature exploitation,
[18].
YOLOv8n: The four components of YOLOv8n
are the input, backbone, neck, and head. The C2f
and CBS modules, which extracted features from
the input image, made up the majority of the
Backbone. The C2f module allowed for the
gathering of rich gradient information while being
WSEAS TRANSACTIONS on COMPUTER RESEARCH
DOI: 10.37394/232018.2024.12.23
Mathuros Panmuang, Chonnikarn Rodmorn
E-ISSN: 2415-1521
239
Volume 12, 2024
lightweight. The neck implemented feature pyramid
networks and path aggregation network topologies,
which improved the network's capacity for feature
fusion by merging features of various scales. Three
distinct scale detection branches were present in the
head, and non-maximum suppression (NMS) was
used to determine the ideal detection box, [19].
3.3 Model Performance Evaluation
The performance of the model is measured by
determining the precision, recall, and mean average
precision (mAP), which are values obtained from
the testing method to determine the predictive value
of the accuracy of the data, [20], as detailed below:
1) Precision is a measure of the accuracy of the
prediction. If the value is large, it means that the
evaluation method has high accuracy as shown in
Equation 2.
(2)
2) Recall is a measure of the accuracy of the
model by considering each class separately as
shown in Equation 3.
(3)
3) Mean Average Precision (mAP) is the average
of the Precision and Recall of objects in the image
as shown in Equation 4.
TP (True positive) is the positive correct
prediction value and TN (True negative) is the
negative correct prediction value, while FP (False
positive) is the positive false prediction value and
FN (False negative) is the negative false prediction
value.
4 Result and Discussion
To compare each version of YOLO, the resources of
the computer are determined on the same basis:
CPU: Intel(R) Core(TM) i7-9700
CPU@3.00GHz
GPU: NVIDIA® GeForce® GTX 1660 Ti
RAM: 16 GB
Storage: 220 GB SSD.
Operating System: Windows 11 Home Single
Language
We also used Google Colab with Tesla T4 GPU
to train the neural networks, and 100 epochs were
trained for every model, and hyperparameters were
configured to default. However, for YOLOv7, after
training with 100 epochs, it was found that the
performance value was quite low (unable to
recognize the class of the test data set). Therefore,
we increased the number of epochs from 100 to 200,
and the results of increasing the number of trains
showed that the efficiency increased. The
comparison results are shown as follows:
Table 1. Model performance comparison results
Model
Precision
Recall
mAP:0.5
mAP:0.95
YOLOv5n
0.938
0.890
0.945
0.633
YOLOv6n
0.717
0.617
0.716
0.370
YOLOv7
0.340
0.319
0.246
0.082
YOLOv7
(200 epochs)
0.893
0.758
0.868
0.466
YOLOv8n
0.941
0.876
0.938
0.707
From Table 1, the comparison of model
performance results showed that YOLOv5n had the
highest mAP:0.5 value; that is, the model was able
to classify classes with an accuracy of 94.50%,
which when compared to YOLOv8n, the value was
not much different. If we consider using the test data
set to test the model to predict at a confidence level
of 0.50, it is found that YOLOv5n and YOLOv8n
can predict more accurately than YOLOv6n and
YOLOv7, as shown in Figure 3.
(A) YOLOv5n
(4)
WSEAS TRANSACTIONS on COMPUTER RESEARCH
DOI: 10.37394/232018.2024.12.23
Mathuros Panmuang, Chonnikarn Rodmorn
E-ISSN: 2415-1521
240
Volume 12, 2024
(B) YOLOv6n
(C) YOLOv7
(D) YOLOv8n
Fig. 3: Results obtained from experiments on the
test dataset.
Based on the experimental results, we will focus
our presentation on YOLOv5n and YOLOv8n as
they are the models with the highest accuracy
values. It can be seen that the confusion matrix
results have similar classification values, with
YOLOv5n being able to predict 90% of garbage
images with 96% of bin images, while YOLOv8n
can predict 96% of garbage images with accuracy
and predict the accuracy of 91% of garbage images
and 84% of bin images, Table 2 and Figure 4.
Table 2. Confusion matrix results of YOLOv5n and
YOLOv8n
True Label
YOLOv5n
Predicted
Garbage
Bin
Background
Garbage
0.90
0.00
0.73
Bin
0.00
0.96
0.27
Background
0.10
0.04
0.00
YOLOv8n
Garbage
0.91
0.00
0.72
Bin
0.00
0.84
0.28
Background
0.09
0.06
0.00
Fig. 4: mAP rate for comparison of YOLOv5n and
YOLOv8n
We can apply the results from this experiment to
develop a mobile or web application to notify of
overflowing garbage bins as the resulting model can
accurately classify or predict garbage and bins. The
establishment of this system should be synchronized
with CCTV cameras deployed in different
communities throughout Bangkok, aiming to
observe overflowing garbage bins and promptly
alert relevant agencies or individuals in the vicinity.
This integration facilitates quicker and more
efficient responses, [21]. Although using a different
version of YOLO, the goal is to focus on garbage
detection in urban communities. This research
applied a refined YOLOv3 network model utilized
for garbage identification and classification. Based
on the dataset gathered for this purpose, the network
has been optimized. The findings indicate that the
suggested strategy might make a significant
contribution to smart cities’ ability to manage
garbage more effectively. Additionally, the study
strives to enhance the precision of automatic
garbage recycling, accelerate computational
processes, and reduce the model size, all while
ensuring practical applicability in real-world
garbage recycling scenarios, [22]. This research
presents a unique YOLO-based neural network
model using Variational Autoencoder (VAE). A
decoder, a convolutional predictor, and a
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Y O L O V 5 N Y O L O V 8 N Y O L O V 5 N Y O L O V 8 N Y O L O V 5 N Y O L O V 8 N
GARBAGE BIN B A C K G R O U N D
PREDICTED
Garbage Bin Background
Predicted
YOLOv5n YOLOv8n
Garbage
YOLOv5n YOLOv8n
Bin
YOLOv5n YOLOv8n
Background
WSEAS TRANSACTIONS on COMPUTER RESEARCH
DOI: 10.37394/232018.2024.12.23
Mathuros Panmuang, Chonnikarn Rodmorn
E-ISSN: 2415-1521
241
Volume 12, 2024
convolutional feature extractor make up the model.
After training, this model outperforms current
models like YOLOv1 and Fast R-CNN, with a
correct rate of 69.70% and 32.1 million parameters,
processing at a pace of 60 frames per second (FPS).
Moreover, the study affirms that using YOLO in
garbage detection proves to be an effective means of
controlling garbage pollution, [23]. This research
primarily enhances the spatial pyramid pooling with
average pooling, mish activation function,
concatenated densely connected neural network, and
hyperparameter optimization, this research provided
an optimized YoLOv4-tiny model to detect floating
junk. With a size of 16.4 MB and better results of
74.89% mean average precision, the suggested
model is the best compromise among the other
models. In terms of model size, detection time, and
memory space, the suggested model performs well
and may be integrated into low-cost devices.
5 Conclusion and Future Work
This research applied YOLO to detect 1,383 images
of overflowing garbage bins, classified into 2
classes: garbage class and bin class. Currently, as
YOLO has been developed in several versions, we
compare each version, including YOLOv5n,
YOLOv6n, YOLOv7, and YOLOv8n, using the
smallest hyperparameter size as it has features to
classify images. The comparison results showed that
YOLOv5n had the highest mAP:0.5 value, and the
model was able to classify with an accuracy of
94.50%, which when compared with YOLOv8n, the
values were not much different. Using the test
dataset to test the model to predict at a confidence
level of 0.50, it is found that YOLOv5n and
YOLOv8n can predict more accurately than
YOLOv6n and YOLOv7. Considering the confusion
matrix results, YOLOv5n can predict the accuracy
of 90% of garbage images and 96% of bin images
while YOLOv8n can predict 91% of garbage images
and 84% of bin images.
Nonetheless, Bangkok or related organizations
can utilize the model to expand, for instance, by
developing a smartphone application that links to
CCTV cameras and grants public officials and
citizens access to the system. When the system
notices that garbage bins are overflowing, it will
promptly alert the local populace or relevant
agencies. This is to allow people to take part and
learn about the consequences of pollution from
overflowing garbage bins. Additionally, it facilitates
effective pollution reduction and timely waste
management.
Therefore, we intend to create an alert system
that will subsequently sound upon CCTV cameras
detecting overfilled bins. Notifications will be sent
to the appropriate authorities and the neighboring
residents, along with an image and location of the
overflowing garbage bins.
Acknowledgement:
This research was funded by the Faculty of Applied
Science, King Mongkut’s University of Technology
North Bangkok. Contract no. 652115.
References:
[1] A. Singh, Managing the uncertainty problems
of municipal solid waste disposal, Journal of
Environmental Management, Vol.240, 2019,
pp. 259-265.
[2] P. Nathapakti, N. Kulnides, and N.
Supatanyaporn, Environmental problem, laws,
consideration, and environmental case
affecting the establishment of the
environmental court in Phuket province, Suan
Sunandha Academic & Research Review,
Vol.13, No.2, 2020, pp. 1-12.
[3] W. Suppasedsak, Municipal solid waste
problem and its solution with waste to energy
power plant project, The National Defence
College of Thailand Journal, Vol.61, No.3,
2019, pp. 107-119.
[4] R. M. Sebastian, D. Kumar, and B. J. Alappat,
A technique to quantify incinerability of
municipal solid waste, Resources,
Conservation & Recycling, Vol.140, 2019, pp.
286-296.
[5] S. Srimeuang and P. Yolchayawong,
Networking of participation in solving
community waste problems in Nong Khaem
district, Bangkok, Journal of Graduate Schol,
Pitchayatat, Vol.15, No.1, 2020, pp. 203-211.
[6] P. Phodjanawichaikuland and J. Mekparyup,
The study of the waste amount in Bangkok
with cluster analysis, Burapha Science
Journal, Vol.22, No.1, 2019, pp. 308-318.
[7] Environment Department Bangkok
Metropolitan Administration, Solid waste
statistics, the fiscal year 2022, [Online].
https://webportal.bangkok.go.th/public/user_fi
les_editor/231/WasteData/2565/12.September
2022.pdf (Accessed Date: April 5, 2023).
[8] H. Choi and J. Jeong, Realtime detection of
table objects and text areas for OCR
preprocessing, WSEAS Transactions on
Information Science and Applications, Vol.20,
WSEAS TRANSACTIONS on COMPUTER RESEARCH
DOI: 10.37394/232018.2024.12.23
Mathuros Panmuang, Chonnikarn Rodmorn
E-ISSN: 2415-1521
242
Volume 12, 2024
2023, pp. 197-205,
https://doi.org/10.37394/23209.2023.20.23.
[9] K. Pathak, A. Arya, P. Hatti, V. Handragal,
and K. Lee, A study of different disease
detection and classification techniques using
deep Learning for the cannabis plant,
International Journal of Computing and
Digital Systems, Vol.10, No.1, 2021, pp. 54-
62.
[10] S. Kumar, M. Jailia, and S. Varshney,
Improved YOLOv4 approach: A real time
occluded vehicle detection, International
Journal of Computing and Digital Systems,
Vol.12, No.1, 2022, pp. 489-497.
[11] F. Mushtaq, K. Ramesh, S. Deshmukh, T.
Ray, C. Parimi, P. Tandon, and P.K. Jha,
Nuts&bolts: YOLO-v5 and image processing
based component identification system,
Engineering Applications of Artificial
Intelligence, Vol.118, 2023.
[12] C. Gupta, N.S. Gill, P. Gulia, and J.
Chatterjee, A novel finetuned YOLOv6
transfer learning model for real-time object
detection, Journal of Real-Time Image
Processing, Vol.20, 2023, pp. 1-19.
[13] J. Zhou, Y. Zhang, and J. Wang, A dragon
fruit picking detection method based on
YOLOv7 and PSP-Ellipse, Intelligent Sensing
and Machine Vision in Precision Agriculture,
Vol.23, No.8, 2023.
[14] R. Bawankule, V. Gaikwad, I. Kulkarni, S.
Kulkarni, A. Jadhav, and N. Ranjan, Visual
detection of waste using YOLOv8,
Proceedings of the International Conference
on Sustainable Computing and Smart Systems
(ICSCSS 2023), IEEE Xplore Part Number:
CFP23DJ3-ART.
[15] F. Li, and L. Wang, Application of deep
learning based on garbage image
classification, WSEAS Transactions on
Computers, Vol.21, 2022, pp. 277-282,
https://doi.org/10.37394/23205.2022.21.34.
[16] Y. Liu, Y. Wang, Y. Li, Q. Li, and J. Wang,
SETR-YOLOv5n: A lightweight low-light
lane curvature detection method based on
fractional-order fusion model. IEEE Access,
Vol.10, 2022, pp. 93003-93016.
[17] C. Zheng, Stack-YOLO: A friendly-hardware
real-time object detection algorithm, IEEE
Access, Vol.11, 2023, pp. 62522-62534.
[18] L. Yang, Investigation of You Only Look
Once Networks for Vision-based Small
Object Detection, International Journal of
Advanced Computer Science and Applications
(IJACSA), Vol.14, No.4, 2023, pp. 69-82.
[19] Z. Wang, Z. Hua, Y. Wen, S. Zhang, X. Xu,
and H. Song, E-YOLO: Recognition of
oestrus cow based on improved YOLOv8n
model, Expert Systems with Applications,
Vol.238, Part E, 15, 2024, pp. 1-17.
[20] A. Malta, M. Mendes, and T. Farinha,
Augmented reality maintenance assistant
using YOLOv5, Applied Sciences, Vol.11
No.11, 2021, pp. 475.
[21] B. D. Carolis, F. Ladogana, and N.
Macchiarulo, YOLO TrashNet: garbage
detection in video streams, 2020 IEEE
Conference on Evolving and Adaptive
Intelligent Systems (EAIS), Bari, Italy.
[22] A. Ye, B. Pang, Y. Jin, and J. Cui, A YOLO-
based neural network with VAE for intelligent
garbage detection and classification, ACAI
'20: Proceedings of the 2020 3rd
International Conference on Algorithms,
Computing and Artificial Intelligence,
December 2020 Article No.: 73, pp. 1-7.
[23] N. A. Zailan, A. S. M. Khairuddin, K.
Hasikin, M. H. Junos, and U. Khairuddin, An
automatic garbage detection using optimized
YOLO model, Signal, Image and Video
Processing, 2023.
Contribution of Individual Authors to the
Creation of a Scientific Article (Ghostwriting
Policy)
The authors equally contributed to the present
research, at all stages from the formulation of the
problem to the final findings and solution.
Sources of Funding for Research Presented in a
Scientific Article or Scientific Article Itself
This research was funded by the Faculty of Applied
Science, King Mongkut’s University of Technology
North Bangkok. Contract no. 652115.
Conflict of Interest
The authors have no conflicts of interest to declare
that are relevant to the content of this article.
Creative Commons Attribution License 4.0
(Attribution 4.0 International, CC BY 4.0)
This article is published under the terms of the
Creative Commons Attribution License 4.0
https://creativecommons.org/licenses/by/4.0/deed.en
_US
WSEAS TRANSACTIONS on COMPUTER RESEARCH
DOI: 10.37394/232018.2024.12.23
Mathuros Panmuang, Chonnikarn Rodmorn
E-ISSN: 2415-1521
243
Volume 12, 2024
