OBJECT IDENTIFICATION

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 10 Issue: 10 | Oct 2023 www.irjet.net p-ISSN: 2395-0072
© 2023, IRJET | Impact Factor value: 8.226 | ISO 9001:2008 Certified Journal | Page 471
OBJECT IDENTIFICATION
Prof. Heena Patil1, Anurag Shekhar2, Renuka Dingore3 , Archana sahu4
1Head of Department, Dept. of AIML Diploma, ARMIET ,Maharashtra, India
2Lecturer, Dept. of AIML Diploma, ARMIET ,Maharashtra, India
--------------------------------------------------------------------------***-------------------------------------------------------------------------
Abstract - Artificial neural networks are the best and
most popular method for classifying images and identifying
objects in images. The paper examines them as a technique
that greatly enhances the aforementioned, extremely
challenging computer calculations later section of the
publication includes a picture of the chosen object detector
we used for our introduction experiment as well as a brief
overview of its development. Also presented is a fresh way
for automatically producing brand-new domain-specific
datasets, which are essential during the training stage of
neural networks. This proposal for future study will be based
on the experiment that was completed.
Key Words: Object detection, Convolutional neural
networks, YOLO, Deep learning, Computer vision
1. INTRODUCTION
Object detection is crucial in computer systems industrial
automation, automated vehicles, and vision. Real-time
object detection is a difficult task. Deep Object detection
training is superior to classical target detection. Region
suggestion is one deep learning technique. object detection
methods that produce regions of interest network
proposals, and then categorize them. SPPnet, region-based
convolutional neural networks, fast CNN, faster-RCNN, etc.
are a few examples. Object regression detection SSD and
YOLO algorithms produce region proposals networks
while simultaneously classifying them. This paper lists the
many real-time object-detecting techniques and methods
based on "YOLO" (You Only Look Once)[1].
2. NEURAL NETWORK
Many general approaches solve problems in distinctive
ways while taking the least amount of time possible. In the
modern era, neural networks have emerged as one of these
approaches that have gained commercial traction as a
result of the significant daily advancements being made in
both hardware and software. They are now widely
employed in a variety of computer science fields, from
authentication to Arduino microcontroller interfaces to
our study on image categorization and object
recognition.Neurons are interconnected groupings of
nodes that make up neural networks. These neurons
receive multivariable linear combinations of variables
from input functions from the data, where the values are
multiplied by each function variable (i.e. weights)[2]. Later
nonlinearity is given to this linear combination, giving the
neural networks the ability to model intricate nonlinear
relationships. More layers are possible in neural networks,
where the input for one layer serves as the output for the
next. Additionally, learned datasets are used by neural
networks during the learning and detection
operations.There are several algorithms today that use
different kinds of neural networks. Their historical
development is discussed in section 2.1 after that.
2.1 History of neural network
Since 1958 [3], when Frank Rosenblatt began researching
how information from the physical environment is stored
in biological systems to be used for detection or behavioral
effects in the future, there have been easy methods for
building one of the first neural networks.Later models with
numerous sequentially non-linear layers of neurons were
created; these models date to the 1960s [4] and 1970s [5].
The gradient descent method, often known as
backpropagation [6], was initially applied to a neural
network in 1981 for supervised learning in discrete,
differentiable networks of any depth.Because neural
networks had so many different layers at the time, it was
difficult to develop them, and their advancement stalled
until the introduction of unsupervised learning [7]
techniques at the beginning of the 1990s .
There were notable advancements in this type of field
during the 1990s and 2000s of the previous century. The
agent looks into a foreign world and uses the trial-and-
error technique [8] to learn about its surroundings, getting
better with each new activity it tries. This newly created
reinforcement learning method [9] is used by the
agent.The use of neural networks in numerous fields drew
a lot of researchers in the third millennium [10], leading to
some of the greatest algorithms. Since 2009, neural
networks have excelled in several contests, particularly

those involving pattern recognition.When convolutional
neural networks were created for the image classification
task on the ImageNet challenge by Alex Krizhevsky et al. in
2012, the pattern recognition was significantly improved
[11]. He and his team triumphed in the contest and
produced a cutting-edge image classification technique
that is still in use today.
2.2 Datasets
There are many different datasets available today for
machine learning, but we'll focus on picture datasets
because they're crucial for tasks like object detection and
image categorization.Since their meaning is only
understood when they contain a vast amount of data,
creating image datasets takes a fair bit of effort. Objects
are labeled and precisely located using bounding boxes to
produce the picture datasets required for object detection
and image classification. These fully automated solutions
for tagging and locating things are no longer available.The
community is interested in developing a mechanism that
automates the production of these datasets since we want
to focus our research on domain-specific environments.In
our two trials, we experimentally tested the object
detector for images within the YOLO architecture and the
convolutional neural network, both of which we plan to be
used in its construction (section 3.2). We want to compile
photographs from the internet of objects belonging to the
same classes in a range of shapes and hues against a
transparent or solid-colored background[12]. To collect
more images for the training phase, we might later extract
certain objects from the images and programmatically
change their brightness, light settings, shadows, etc. As
seen in the following graphic, our goal is to place those
objects onto backgrounds that were generated at random,
with random placements and overlapping (Fig.1).
Fig -1: Background created at random, using elements that
are randomly placed and overlapped.
3. THE EXPERIMENTS AND DETECTOR YOLO
The object detector known as YOLO was developed by
Redmon, J., et al. [13]. We employed the YOLO authors'
state-of-the-art Darknet neural network in our research
since they claim [14] that it is the fastest and most
accurate image object detector currently available.
3.1 Detector
Every cell in the grid predicts B bounding boxes and their
confidence, and each image is partitioned by YOLO into a
grid of size S x S. The precision and dependability of the
bounding box used to locate and categorize an object
depends on its confidence level. The confidence of an
object is described as follows:
P(𝑂𝑏𝑗𝑒𝑐𝑡) ∗ 𝐼𝑂𝑈𝑝𝑟𝑒𝑑 𝑡𝑟𝑢𝑡ℎ (1)
3.2 Experiments
On a dataset that had already been trained using COCO, we
ran two tests using the detector YOLO [15]. In the first, we
demonstrated how the detector operates using the image
below (Fig. 2), and in the second, we put the detector to
the test on a set of 500 photos to experimentally verify its
performance.
various resolutions for image detectors In this study, we
compared object detection and image categorization using
the same image processed on the Intel Core i7-7700K
processor (Table 1) and GeForce GTX 1070 graphics card
(Table 2).Fig.2. Used Image for this experiment .
Comparing the two tables reveals that the processor
processes data classify images, and detects objects
substantially more slowly than the graphic card.
Additionally, when the resolution increases, more things
are detected, which is a result of improved image clarity.
Fig -2: Used Image for this experiment.

Table-1: Processor Testing for Object Detection.
Resolution Objects detected Time in ms
368x274 9 1649.214
766x577 10 1653.339
1009x757 12 2
2026x1522 14 1699.288
4042x3034 15 1530.113
Table-2: Testing for Objects Detection on the Graphic Card
Resolution Objects detected Time in ms
368x274 9 184.574
766x577 10 222.665
1009x757 12 6
2026x1522 14 178.224
4042x3034 15 164.799
Applying the detector to the 500-image sequence. We
expanded our initial experiment to include the detection of
items on the 500-image series. Additionally, based on the
findings of our last study, Different resolutions were no
longer used because they had no effect[16]. We also
reported the average time and quantity of detected items
for the entire set of photos. Similar to the prior
experiment, we employed processing from the processor
and graphics card. The following tables display the
findings (Table 3 and Table 4)
Table-3: Testing of Object Detection on 500 Images on the
Processor.
Property Objects
detected
FLOPS Time in
ms
quickest
detection
18 64.867 2288.822
The last rapid
detection
18 64.867 1301.273
average period 14.23 64.867 1363.503
the most things 42 64.867 1605.41
the smallest
items
3 64.867 1872.51
Table-4: Testing Object Detection in 500 Images on the
Graphic Card
Property Objects
detected
FLOPS Time in
ms
quickest
detection
18 64.867 210.742
The last rapid
detection
18 64.867 157.003
average
period
14.23 64.867 122.299
the most
things
42 64.867 169.685
the smallest
items
3 64.867 188.625
The processing time for the series of 500 photos
ranges from 1301.273 to 2288.822 milliseconds, with an
average detection time of 1686.303 milliseconds for the
processor and an average detection time of 169.670
milliseconds for the graphic card[17].
According to the results of this experiment, the quantity of
items discovered has no bearing on the speed of detection
(both the quickly and slowly processed photos include the
same quantity of objects), and both the quickly and slowly
processed images take nearly the same length of time.
4. FUTURE WORK
In the future, we plan to process a sizable number of
photos using the YOLO detector to train an automated
dataset-generating system. We will direct the processes
toward a graphic card based on our findings. The card we
were using managed 5 FPS.
Additionally, fully new methodologies for the automated
production of domain-specific datasets will be designed as
part of future research. We anticipate that the method will
play a significant role in speeding up the process of
building new datasets, particularly during the labeling
stage where each object on an image needs to be
accurately placed within its bounding box. This method
should fully eliminate human involvement in the labeling
process, which is now done by hand. The technique would
also be used for real-time detections and a variety of other
jobs, for example teaching pupils how to study certain
objects in the same way that youngsters do from their very
first days of existence or identifying specific species of a
particular kind.The last point I want to make is how

challenging it is to build these datasets because the
majority of object labeling and positioning in the images is
done manually. Our approach would enable researchers in
many different domains to acquire notably improved
findings because their datasets are typically fairly tiny and
could have an impact on the result’s accuracy, as is
indicated in these studies [18, 19].We might train neural
networks in specific contexts using our approach for
automatically generating domain-specific datasets, which
would substantially help in identifying not just the class of
an object but also its types and subclasses. A detected
flower, for example, would be more properly classified as a
forget-me-not, and a detected tree, as a baobab[20].
5. CONCLUSION
In this study, we suggest employing the YOLO network
model to detect objects. We put the image that has
deteriorated during the degenerative model training
process. Investigations reveal that by Using degraded
photos as training data, a network can learn more features
and become more adaptable to complicated scene types.
The outcomes demonstrate that the model enhances the
object detection's average precision. Better generalization
and more resilience are characteristics of the model that
was developed using the degraded training sets.
REFERENCES
[1] Diwan, Tausif, G. Anirudh, and Jitendra V.
Tembhurne. “Object detection using YOLO:
challenges, architectural successors, datasets and
applications.” Multimedia Tools and Applications
(2022): 1-33.
[2] Ahmad, Tanvir, et al. “Object detection through
modified YOLO neural network.” Scientific
Programming 2020 (2020).
[3] Lee, Yong-Hwan, and Youngseop Kim. “Comparison
of CNN and YOLO for Object Detection.” Journal of
the semiconductor & display technology 19.1
(2020): 85-92.
[4] Liu, Chengji, et al. “Object detection based on YOLO
network.” 2018 IEEE 4th Information Technology
and Mechatronics Engineering Conference (ITOEC).
IEEE, 2018.
[5] Long, Xiang, et al. “PP-YOLO: An effective and
efficient implementation of object detector.” arXiv
preprint arXiv:2007.12099 (2020).
[6] Huang, Rachel, Jonathan Pedoeem, and Cuixian
Chen. “YOLO-LITE: a real-time object detection
algorithm optimized for non-GPU computers.” 2018
IEEE International Conference on Big Data (Big
Data). IEEE, 2018.
[7] Liu, Wenyu, et al. “Image-adaptive YOLO for object
detection in adverse weather conditions.”
Proceedings of the AAAI Conference on Artificial
Intelligence. Vol. 36. No. 2. 2022.
[8] Huang, Zhanchao, et al. “DC-SPP-YOLO: Dense
connection and spatial pyramid pooling based YOLO
for object detection.” Information Sciences 522
(2020): 241-258.
[9] Nguyen, Duy Thanh, et al. “A high-throughput and
power-efficient FPGA implementation of YOLO CNN
for object detection.” IEEE Transactions on Very
Large Scale Integration (VLSI) Systems 27.8 (2019):
1861-1873.
[10] Krišto, Mate, Marina Ivasic-Kos, and Miran Pobar.
“Thermal object detection in difficult weather
conditions using YOLO.” IEEE access 8 (2020):
125459-125476.
[11] Wong, Alexander, et al. “Yolo nano: a highly compact
you only look once convolutional neural network for
object detection.” 2019 Fifth Workshop on Energy
Efficient Machine Learning and Cognitive
Computing-NeurIPS Edition (EMC2-NIPS). IEEE,
2019.
[12] Lee, Yong-Hwan, and Youngseop Kim. “Comparison
of CNN and YOLO for Object Detection.” Journal of
the semiconductor & display technology 19.1
(2020): 85-92.
[13] Yin, Yunhua, Huifang Li, and Wei Fu. “Faster-YOLO:
An accurate and faster object detection method.”
Digital Signal Processing 102 (2020): 102756.
[14] Lee, Jeonghun, and Kwang-il Hwang. “YOLO with
adaptive frame control for real-time object detection
applications.” Multimedia Tools and Applications
81.25 (2022): 36375-36396.
[15] Lu, Yonghui, Langwen Zhang, and Wei Xie. “YOLO-
compact: an efficient YOLO network for single
category real-time object detection.” 2020 Chinese
control and decision conference (CCDC). IEEE, 2020.

IEEE, 2018.
IEEE, 2018.
[18] Du, Juan. “Understanding of object detection based
on CNN family and YOLO.” Journal of Physics:
Conference Series. Vol. 1004. No. 1. IOP Publishing,
2018.
[19] Zhang, Shuo, et al. “Tiny YOLO optimization oriented
bus passenger object detection.” Chinese Journal of
Electronics 29.1 (2020): 132-138.
[20] Bathija, Akansha, and Grishma Sharma. “Visual
object detection and tracking using Yolo and sort.”
International Journal of Engineering Research
Technology 8.11 (2019).

OBJECT IDENTIFICATION

More Related Content

Similar to OBJECT IDENTIFICATION (20)

More from IRJET Journal (20)

Recently uploaded (20)

OBJECT IDENTIFICATION