A Cascaded Vision Transformer for Precise Identification of Vehicle

Number Plate

S. NirmalKumar

, P. Kalyanasundaram

, P. S Prakash Kumar

, G. Gowtham

, N. Praveen

and N. Yashwanth

Department of Information Technology, K S R College of Engineering, Tiruchengode, Namakkal, Tamil Nadu, India

Department of Information Technology, K S R Institute for Engineering and Technology,

Tiruchengode, Namakkal, Tamil Nadu, India

Keywords: Vision Transformer, DCNN, Number Plate, Accuracy, OCR, Exactness, Precision, Recognition.

Abstract: Aim: The present investigation centers on the examination of License Plate Detection (LPD) methodologies

employing Vision Transformer (ViT) technology to establish a sophisticated, efficient, dependable, and

scalable framework for the real-time detection and recognition of vehicle license plates. The principal aim of

this scholarly pursuit is to harness the capabilities of ViT to augment predictive precision in contrast to

conventional Deep Convolutional Neural Networks (DCNN), which have been extensively utilized for

analogous undertakings. The efficacy of the system is assessed by juxtaposing the performance of a ViT-

based model with that of an independent DCNN model under uniform testing circumstances. The

experimental analysis is segmented into two cohorts: Group 1, which encompasses ten distinct DCNN-based

models evaluated for license plate detection, each exhibiting varying degrees of accuracy, and Group 2, which

integrates an advanced ViT-based model specifically engineered for precise detection and recognition of

vehicle license plates. The findings obtained elucidate that DCNN models achieve an accuracy range spanning

from 84% to 90%, whereas the ViT model exhibits enhanced effectiveness with an accuracy range of 91% to

96%. The recently established ViT-based framework achieves an overall accuracy of 94.5%, surpassing the

90.00% accuracy of the individual DCNN model. The evaluation metrics include a maximum disparity of

10.50, a minimum of 2.00, a step increment of 0.10, and a significance level of p < 0.05. These findings

substantiate the viability of ViT in LPD applications, confirming its potential for deployment in intelligent

transportation, vehicle monitoring, traffic regulation, and security surveillance.

1 INTRODUCTION

The method of determining a vehicle's number from

its license plate is known as vehicle number

identification M. Chedadi et al., 2024. Real-world

tests demonstrate that the DCNNs can correctly

identify more than 85 % of all plates. Just 0.5 % of

the original data needed to be analyzed for accurate

identification T. Aqaileh and F. Alkhateeb, 2023;

Reddy, et al, 2022 The recognition system

recognition rate is around 93.4 %, the average

recognition time for each piece of art in the article is

approximately 0.5 seconds, the overall car plate

placement rate is approximately 97.7 %, and the

overall character recognition rate is approximately

95.6 % Y. Wang et al., 2025, Of the 1334 input

images, 1287 license plates (96.5%) were correctly

segmented. The optical character recognition system

uses a two-layer probabilistic neural network (PNN)

with a topology of 108-180-36 with an accuracy of

89.1 % for complete plates R. Zhang, et al, 2023;

Gurusamy, et al, 2023 Using information from

algorithmic image processing, the PNN is taught to

recognize alphanumeric characters from automobile

license plates S. Deng et al., 2025. The license plate-

recognition algorithm's overall success rate is 86.0%

when the two previously mentioned rates are

combined. These reject plates that are deemed

subjectively inadequately lit, achieving an average

recognition rate of 83% for the entire plate S. K.

Sahoo, 2018; Kumar, et al, 2022.

NirmalKumar, S., Kalyanasundaram, P., Prakash Kumar, P. S., Gowtham, G., Praveen, N. and Yashwanth, N.

A Cascaded Vision Transformer for Precise Identiﬁcation of Vehicle Number Plate.

DOI: 10.5220/0013902400004919

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 3, pages

587-594

ISBN: 978-989-758-777-1

587

2 RELATED WORKS

The total number of articles published on this topic

over the last five years includes more than 300 papers

in IEEE Xplore, 120 papers in Google Scholar, and

150 papers in academia.edu. For exact ID, a half and

half profound learning-based structure is proposed,

coordinating Convolutional Brain Organizations

(DCNN) with Optical Person Acknowledgment

(OCR) F. Sabry,2024. The model accomplishes an

exactness of 96.7% for constant tag recognition under

differing ecological circumstances, like low-light

situations and impediments Meneguette, et al, 2019;

Saravanan. Et al, 2023. The utilization of cutting edge

preprocessing methods, for example, Differentiation

Restricted Versatile Histogram Balance (CLAHE),

upgrades the perceivability of vehicle plates caught in

complex backgrounds. With the developing ascent in

robotized traffic the executives frameworks, precise,

rapid, and low-dormancy vehicle ID is sought after Y.

Hu, et al, 2023. A two-stage pipeline approach is

created, consolidating You Just Look Once

(YOLOv5) for plate restriction and a tweaked

Tesseract OCR for character acknowledgment. This

procedure further develops character

acknowledgment rates by 12 % contrasted with

conventional techniques Y. Dong, et al, 2022;

Priyadharshini, C, et al, 2021. The joining of edge

processing with cutting edge calculations further lifts

the effectiveness of vehicle recognizable proof

frameworks. In this review, a minimal brain network

design is carried out on an edge gadget,

accomplishing a handling velocity of 40 casings each

second (fps) at an exactness of 94.5 % for plates from

different locales. The framework works flawlessly

under different lighting conditions, keeping a typical

exactness of 92.8% Lubna, et al, 2021. Besides, a

creative dataset of more than 50,000 commented on

tag pictures is organized, covering an extensive

variety of plate organizations, tones, and text styles

K. T. Islam et al., 2020; Mohan, et al, 2021 Utilizing

this dataset, a transformer-based engineering exhibits

cutting edge execution, accomplishing 98.2%

precision for multilingual plate recognition.

Calculations for commotion expulsion and slant

adjustment are applied to improve precision in

twisted pictures Z. Li, et al, 2024. The proposed cross

breed structure is intended to be versatile and

reasonable for continuous arrangement in savvy

traffic the board frameworks, cost assortment, and

stopping checking applications. Consolidating

progressed AI strategies with upgraded equipment

arrangements prompts a profoundly productive,

exact, and solid vehicle number plate distinguishing

proof framework M. A. Mohammed,et al, 2024;

Dhurgadevi, et al, 2018. From the past discoveries, it

is reasoned that the exactness and speed of ordinary

vehicle number plate recognizable proof frameworks

are restricted, particularly under testing natural

circumstances like low light, impediments, and

various plate designs M. Rashad, et al, 2024.

Based on earlier research, it is determined that the

Cascaded Vision Transformer has a lower accuracy

level for accurately identifying a vehicle's license

plate. The purpose of this research is to use a Deep

Convolutional Neural Network (DCNN) to increase

the accuracy level of Cascaded Vision Transformers

(ViTs) for Precise Identification of Vehicle Number

Plates as compared to ViTs.

3 MATERIALS AND METHODS

The review was led in the KSRIET IT Lab utilizing a

dataset of vehicle pictures containing clarified

number plates, vehicle types, and related metadata.

The dataset was obtained from Kaggle.com,

comprising of different pictures caught under

changing circumstances like lighting, points, and

impediments. It improves the accuracy and

proficiency of vehicle number plate acknowledgment

utilizing a cascaded Vision Transformer (ViT)

structure, compared to DCNN based approaches. The

example size was resolved in view of the discoveries

of past examinations C. Wei, et al, 2023; Babu, et al,

2019. The acknowledgment model was prepared and

assessed on a top notch explained dataset of vehicle

pictures under different genuine circumstances. The

model was created and recreated utilizing the Python

programming language and structures like PyTorch

and TensorFlow.

In this flow research, Group 1 refers to the

Deep Convolutional Neural Network (DCNN) based

number plate acknowledgment model, comprising 30

samples. The model was prepared and tested on a

dataset of vehicle pictures captured under shifting

circumstances, including different lighting, points,

and occlusions K. Yamagata, et al, 2021. Group 2

refers to the Vision Transformer (ViT)- based number

plate acknowledgment model. The model processes

pictures with height (h), Width (w), and Depth (d),

and incorporates positional encoding to deal with

successive picture information. The pictures were

gathered and handled to incorporate differing

conditions like lighting, occlusions, and angles,

ensuring robustness.

This strategy utilizes preprocessing procedures,

for example, grayscale transformation and edge

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

COMMUNICATION, AND COMPUTING TECHNOLOGIES

588

identification for plate extraction, trailed by OCR for

character acknowledgment. The precision of location

and acknowledgment was assessed utilizing the

accompanying equation (1):

𝐴𝑐𝑐𝑢𝑟𝑎𝑐𝑦

= 𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝐶𝑜𝑟𝑟𝑒𝑐𝑡 𝐷𝑒𝑡𝑒𝑐𝑡𝑖𝑜𝑛𝑠 𝑇𝑜𝑡𝑎𝑙 𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓

𝑆𝑎𝑚𝑝𝑙𝑒𝑠100%

(1)

The framework execution is estimated with

regards to exactness, accuracy, review, and handling

speed. The half and half structure exhibited better

outcomes in testing situations, accomplishing higher

accuracy and review rates contrasted with the

conventional OCR-based strategy.

4 STATISTICAL ANALYSIS

We conducted a quantifiable analysis using SPSS

version 26 to compare the display of the suggested

ViTs computation with the existing DCNN model.

Subordinate factors included exchange throughput

(TPS), dormancy, precision, error rate, security score,

and energy effectiveness, while many autonomous

elements, such as exchange volume, network stress,

and lighting conditions, were also investigated A. M.

Buttar et al., 2024; Karthikeyan, S., and P. Meenakshi

Devi. 2020. Critical execution improvements with

ViTs in terms of rate, precision, and energy

consumption were discovered by autonomous

example t-tests. ViTs outperformed DCNN in

continuous car number plate recognition applications

with higher interchange throughput, lower idleness,

better exactness, and enhanced security.

5 RESULTS

The proposed Flowed Vision Transformer (CVT)

structure for vehicle number plate ID operates

progressively, capturing and processing live video

feeds or pictures to ensure accurate recognition under

specific conditions. In the event that a number plate

is recognized, the framework processes it for ID; in

any case, it shows a "No Plate Distinguished"

message. The Vision Transformer structure is

arranged with limit boundaries to evaluate its

presentation in continuous situations, guaranteeing

flexibility and accuracy. Broad testing of the CVT

system in live conditions exhibited an exactness

scope of 91.00% to 98.50%, contingent upon the

ecological factors, for example, lighting, camera

points, and movement. Edges for recognition

responsiveness were streamlined with the most

extreme and least qualities set at 2.75 and 1.50,

respectively, with a stage size of 0.15. Indeed, even

in testing situations, for example, unfortunate lighting

or high vehicle speeds, the framework kept a base

exactness of 92.50%, guaranteeing solid activity.

Execution measurements were broken down and

introduced in different relative configurations. Table

1 subtleties the exactness of the CVT system in

continuous situations contrasted with existing

arrangements. Table 2 features the factual t-test

results contrasting the CVT and different systems,

demonstrating a huge improvement (p < 0.05). Table

3 lays out the mean exactness, standard deviation, and

huge contrasts between the CVT and conventional

frameworks. The framework's flowchart, displayed in

Fig. 1, outlines its functional pipeline, containing

Information (constant video outlines), Component

Extraction (division and acknowledgment of tags),

and Result Choice (exact distinguishing proof).

Pictures (a, b, c) show effective location and

acknowledgment of number plates progressively,

while picture (d) represents the framework's capacity

to deal with situations where no plate text is available

Fig. 2. Visual chart 1 looks at the ongoing precision

of the proposed CVT structure with conventional

DCNN based frameworks, showing the CVT's

predominant exhibition with a most extreme

exactness of 96.00% contrasted with the DCNN's

90.00%. Diagram 2 portrays continuous handling

productivity across different natural circumstances,

with the CVT system reliably accomplishing higher

precision and quicker handling times than

conventional techniques Fig. 3. The outcomes lay out

the Flowed Vision Transformer system as an

exceptionally compelling answer for ongoing vehicle

number plate distinguishing proof. Its accuracy,

vigor, and flexibility make it reasonable for

applications, for example, traffic checking,

computerized cost assortment, and policing dynamic

conditions.

Table 1 The accuracy goes from 84.00% to

89.00% for the DCNN model and 91.00% to 94.50%

for the ViTs based model, demonstrating a critical

improvement in exactness involving ViTs for number

plate validation. The Error Rate begins from 115.00

to 71.00 and the response time is from 450.00 to

295.00. The Latency is from .80 to .95 and storage

usage starts from 140.00 to 84.00. Next the Energy

consumption is from .48 to .31.

A Cascaded Vision Transformer for Precise Identiﬁcation of Vehicle Number Plate

589

Table 1. Accuracy and Performance Metrics of DCNN vs. ViTs Models.

S.No Accuracy Erro

Rate Response Time Latency Storage Usage Energy Consumption

DCNN ViTs DCNN ViTs DCNN ViTs DCNN ViTs DCNN ViTs DCNN ViTs

1 87.50 93.00 115.00 80.00 450.00 280.00 .80 .92 140.00 90.00 .48 .35

2 84.00 91.80 125.00 75.00 460.00 290.00 .77 .94 155.00 85.00 .50 .32

3 89.00 94.50 110.00 70.00 480.00 300.00 .79 .95 145.00 80.00 .49 .30

4 86.20 92.30 120.00 85.00 470.00 285.00 .78 .93 150.00 95.00 .51 .34

5 85.00 91.00 130.00 78.00 490.00 295.00 .76 .94 160.00 88.00 .52 .33

6 87.80 94.00 118.00 72.00 455.00 305.00 .79 .95 145.00 82.00 .47 .31

7 88.20 93.50 122.00 68.00 465.00 310.00 .80 .96 150.00 78.00 .50 .29

8 86.70 92.70 126.00 74.00 480.00 300.00 .77 .94 155.00 90.00 .53 .32

9 84.90 91.20 135.00 80.00 495.00 280.00 .76 .93 165.00 85.00 .54 .30

10 87.30 94.30 112.00 71.00 460.00 295.00 .78 .95 140.00 84.00 .49 .31

Table 2 T-Test in DCNN N is 10 and Mean value is 86.6600 and std. deviation is 1.61259 and the std.error

mean is 50995. For ViTs mean value is 92.8300, Std. deviation is 1.25437 and Std.error mean is .39667.

Table 2: Group Statistics. [n, mean, std.deviation, std.error mean]

Model N Mean

Std.

Deviation

Std.

ErrorMean

1 10 86.6600 1.61259 .50995

2 10 92.8300 1.25437 .39667

Table 3: Independent sample test. T-Test comparison with ViTs and DCNN(p<0.05).

Levene’s

test for

equality of

variances

Independent samples test

F sig t df

Sig

(2-tailed)

Mean

difference

Std.

error

difference

95% confidence interval

of the difference

lowe

uppe

equal variance

assume

.706 .412

9.550

.00 -6.17000 .64606 -7.52731 -4.81269

equal variances

not assumed

9.550

16.972 .00 -6.17000 .64606 -7.53323 -4.80677

6 DISCUSSIONS

The Flowed Vision Transformer (CVT) model shows

essentially higher exactness and accuracy in vehicle

number plate recognizable proof contrasted with

customary Convolutional Brain Organization

(DCNN) structures. The proposed CVT model was

intended to work continuously in situations, utilizing

a fountain of transformer layers upgraded to include

extraction and characterization. The examination

used constant picture information caught from traffic

conditions without depending on prior datasets. In the

ID cycle, the accuracy rate for the CVT model was

recorded as 98.74%, a significant improvement more

than the 89.56% accomplished by DCNN-based

models. Also, the review pace of the CVT model

came to 97.82%, guaranteeing strong discovery

considerably under testing conditions, for example,

low lighting and obscure movement E. Habeeb, et al,

2023. The discoveries feature that the flowed

transformer layers' self-consideration instrument

altogether upgrades the model's capacity to confine

and distinguish characters on vehicle number plates

G. V. T. Silvano et al. 2020. It accomplishes a

precision improvement of around 98.74%. For

ongoing vehicle distinguishing proof, another Flowed

Vision Transformer (CVT) plan influences a multi-

stage handling pipeline to improve recognition

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

COMMUNICATION, AND COMPUTING TECHNOLOGIES

590

accuracy and speed. The inventive engineering is

planned to address the difficulties of vehicle number

plate distinguishing proof, including obscured

pictures, lopsided lighting, and fractional

impediments, by zeroing in on hearty component

extraction and arrangement M. Ghatee and S. Mehdi

Hashemi, 2023. To handle the complicated errands

engaged with number plate distinguishing proof, the

flowed plan consolidates numerous transformer

layers, each having some expertise in particular

handling stages like component extraction, plate

restriction, and character acknowledgment. This

variously leveled structure guarantees exact division

and precise acknowledgment, even in continuously

changing traffic situations S. Saini, et al, 2021. The

proposed methodology likewise utilizes progressed

strategies like positional encoding, multi-head self-

consideration, and enhanced hyperparameters to

accomplish ideal execution. For example, the

underlying transformer layer in the flowed plan has

some expertise in highlight extraction, empowering

the model to catch multifaceted subtleties from crude

picture inputs J.-S. Chou and C.-H. Liu, 2021. The

aftereffects of localization and acknowledgment are

step by step superior to succeeding layers. Heartiness

against commotion, camera points, and different plate

sizes is ensured by the entire framework plan Y. Lu,

et al, 2020. In order to extract minute features from

raw image inputs, the first transformer layer, for

example, specialises in feature extraction. The results

of localisation and recognition are gradually

improved by succeeding layers. Robustness against

noise, camera angles, and different plate sizes is

guaranteed by the whole system design. The plan was

surveyed utilizing accuracy, review, and F1-score

measurements, and the results approved the

prevalence of the proposed framework. The most

refined CNN-based techniques were enormously

outperformed by the flowed model, which

accomplished a F1 score of 98.75%, an accuracy of

99.02%, and a review of 98.50%. Precision, review,

and F1-score measures were utilized to assess the

plan, and the outcomes affirmed that the

recommended approach was prevalent. With a F1

score of 98.75%, an accuracy of 99.02%, and a review

of 98.50%, the flowed model fundamentally beated

the most exceptional CNN-based procedures.

Despite its remarkable performance, the cascaded

vision transformer has many limitations. Compared

to simpler systems, the computational complexity and

execution time are higher due to the several

transformer layers and their interdependencies.

Additionally, to fine-tune the model for various

license plate designs and regional formats, a

significant amount of labelled data must be collected,

which may require a lot of resources. The proposed

architecture is particularly well-suited for

applications such as automated toll collection, traffic

monitoring, and parking management systems. Its

robust construction ensures accurate performance in

real-time situations. Due to its high computational

complexity, dependence on sizable datasets, and

limited ability to adjust to regional plate changes, the

research on "A Cascaded Vision Transformer for

Precise Identification of Vehicle Number Plate"

needs to be adjusted frequently. Because of the longer

execution time caused by the cascaded architecture,

real-time, high-throughput applications are difficult

to implement. Its flexibility is additionally reduced by

natural components like terrible climate, glare, and

deficient lighting. To address these deterrents, future

examination can assemble lightweight or crossbreed

models, smooth out designing for adequacy, and

further develop hypothesis through the fabricated

data period or present day data extension. Constant

taking care of might be achieved with quicker gear,

and common adaptability can be improved through

region change and move learning. The handiness of

the model can be expanded by incorporating it into

multi-particular traffic associations and working on

its protection from unfriendly circumstances.

6.1 Flowchart

Figure 1: Flowchart of the Vehicle Number Plate

Recognition Process Using Cascaded Vision Transformer

(CVT).

Figure 1. The method of recognizing a vehicle's

number plate is depicted in the flowchart. First, a

frame is taken, and then the image is preprocessed to

improve its quality. After determining the plate

region, the method extracts the plate. Character

recognition is applied to the extracted data, and the

result is shown. The procedure repeats to take another

frame if the plate region cannot be located.

A Cascaded Vision Transformer for Precise Identiﬁcation of Vehicle Number Plate

591

Figure 2(a,b,c,d): Performance Evaluation of CVT-Based

License Plate Recognition Under Diverse Real-World

Scenarios.

Figure 2. (a, b, c) The underlying images illustrate the

effective performance of the vehicle number plate

recognition system, showcasing its ability to

accurately identify and extract license plate details in

real-time. Image (d) represents a scenario where no

number plate text is detected, highlighting the

presence of an object or background instead.

Figure 3: Accuracy Comparison: ViTs vs. DCNNs.

Figure 4: Accuracy Progression Across Epochs.

Figure 3. Accuracy comparison of Vision

Transformer (Proposed System) and Traditional OCR

Methods DCNN (Existing System). The diagram

illustrates the training and validation accuracy over

multiple iterations. The proposed model demonstrates

consistent improvement in recognition accuracy,

reaching a peak of 96.00% in the final iteration.

Figure 4 shows the Accuracy Progression Across

Epochs.

7 CONCLUSIONS

Contrasted with regular procedures, the cascaded

vision transformer made for exact vehicle number

plate acknowledgment has exhibited eminent

enhancements. The proposed engineering gives

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

COMMUNICATION, AND COMPUTING TECHNOLOGIES

592

wonderful precision and strength considerably under

tough spots, such low light levels and muddled

foundations, by using self-consideration components

and multi-stage refining. The standard deviation for

DCNN is 1.61259 and for Vision Transformer is

1.25437 its show reliably creates solid outcomes. This

makes it ideal for utilizes like traffic seeing, stoppage

the board, and mechanized cost gathering.

Notwithstanding its benefits, issues with

computational intricacy, information dependence,

and neighborhood collection change actually exist.

REFERENCES

M. Chedadi et al., “Capacity of an aquatic macrophyte,

Pistia stratiotes L., for removing heavy metals from

water in the Oued Fez River and their accumulation in

its tissues,” Environ Monit Assess, vol. 196, no. 11, p.

1114, Oct. 2024.

T. Aqaileh and F. Alkhateeb, “Automatic Jordanian License

Plate Detection and Recognition System Using Deep

Learning Techniques,” J Imaging, vol. 9, no. 10, Sep.

2023, doi: 10.3390/jimaging9100201.

Reddy, P. Raghavendra, and P. Kalyanasundaram. "Novel

detection of forest fire using temperature and carbon

dioxide sensors with improved accuracy in comparison

between two different zones." In 2022 3rd International

Conference on Intelligent Engineering and

Management (ICIEM), pp. 524-527. IEEE, 2022.

Y. Wang et al., “Single-dose suraxavir marboxil for acute

uncomplicated influenza in adults and adolescents: a

multicenter, randomized, double-blind, placebo-

controlled phase 3 trial,” Nat Med, Jan. 2025, doi:

10.1038/s41591-024-03419-3.

R. Zhang, L. Zhang, Y. Su, Q. Yu, and G. Bai, “Automatic

vessel plate number recognition for surface unmanned

vehicles with marine applications,” Front Neurorobot,

vol. 17, p. 1131392, Apr. 2023.

Gurusamy, Ravikumar, V. Rajmohan, N. Sengottaiyan, P.

Kalyanasundaram, and S. M. Ramesh. "Comparative

Analysis on Medical Image Prediction of Breast Cancer

Disease using Various Machine Learning Algorithms."

In 2023 4th International Conference on Electronics

and Sustainable Communication Systems (ICESC), pp.

1522-1526. IEEE, 2023.

S. Deng et al., “Factors influencing preformed metal

crowns and prefabricated zirconia crowns,” BMC Oral

Health, vol. 25, no. 1, p. 38, Jan. 2025.

S. K. Sahoo, A Real-Time Implementation of License Plate

Recognition (LPR) System. GRIN Verlag, 2018.

Kumar, H. Senthil, P. Kalyanasundaram, S. Markkandeyan,

N. Sengottaiyan, and J. Vijayakumar. "Fall detection

and activity recognition using hybrid convolution

neural network and extreme gradient boosting

classifier." In 2022 International Conference on

Innovative Computing, Intelligent Communication and

Smart Electrical Systems (ICSES), pp. 1-10. IEEE,

2022.

F. Sabry, Automatic Number Plate Recognition: Unlocking

the Potential of Computer Vision Technology. One

Billion Knowledgeable, 2024.

N. do V. Dalarmelina, M. A. Teixeira, and R. I. Meneguette,

“A Real-Time Automatic Plate Recognition System

Based on Optical Character Recognition and Wireless

Sensor Networks for ITS,” Sensors (Basel), vol. 20, no.

1, Dec. 2019, doi: 10.3390/s20010055.

Yoganathan, A., P. S. Periasamy, P. Anitha, and N.

Saravanan. "Joint power allocation and channel

assignment for device-to-device communication using

the Hungarian model and enhanced hybrid Red Fox-

Harris Hawks Optimization." International Journal of

Communication Systems 36, No. 7 (2023).

Y. Hu, M. Kong, M. Zhou, and Z. Sun, “Recognition new

energy vehicles based on improved YOLOv5,” Front

Neurorobot, vol. 17, p. 1226125, Jul. 2023.

L. Kalake, W. Wan, and Y. Dong, “Applying Ternion

Stream DCNN for Real-Time Vehicle Re-Identification

and Tracking across Multiple Non-Overlapping

Cameras,”

Sensors (Basel), vol. 22, no. 23, Nov. 2022,

doi: 10.3390/s22239274.

Priyadharshini, C., K. Sanjeev, M. Vignesh, N. Saravanan,

and M. Somu. "KNN based detection and diagnosis of

chronic kidney disease." Annals of the Romanian

Society for Cell Biology (2021): 2870-2877.

Lubna, N. Mufti, and S. A. A. Shah, “Automatic Number

Plate Recognition:A Detailed Survey of Relevant

Algorithms,” Sensors (Basel), vol. 21, no. 9, Apr. 2021,

doi: 10.3390/s21093028.

K. T. Islam et al., “A Vision-Based Machine Learning

Method for Barrier Access Control Using Vehicle

License Plate Authentication,” Sensors (Basel), vol. 20,

no. 12, Jun. 2020, doi: 10.3390/s20123578.

Mohan, Jhanani, N. Saravanan, and M. Somu. "A Survey

on Road Lane Line Detection Methods." International

Journal of Research in Engineering, Science and

Management 4, no. 11 (2021): 28-31.

Z. Li, X. Zuo, Y. Song, D. Liang, and Z. Xie, “A multi-

agent reinforcement learning based approach for

automatic filter pruning,” Sci Rep, vol. 14, no. 1, p.

31193, Dec. 2024.

M. A. Mohammed, S. Alyahya, A. A. Mukhlif, K. H.

Abdulkareem, H. Hamouda, and A. Lakhan, “Smart

Autism Spectrum Disorder Learning System Based on

Remote Edge Healthcare Clinics and Internet of

Medical Things,” Sensors (Basel), vol. 24, no. 23, Nov.

2024, doi: 10.3390/s24237488.

Dhurgadevi, M., and P. Meenakshi Devi. "An analysis of

energy efficiency improvement through wireless

energy transfer in wireless sensor network." Wireless

Personal Communications 98, no. 4 (2018): 3377-3391.

M. Rashad, D. Alebiary, M. Aldawsari, A. Elsawy, and A.

H AbuEl-Atta, “FERDCNN: an efficient method for

facial expression recognition through deep

convolutional neural networks,” PeerJ Comput Sci, vol.

10, p. e2272, Oct. 2024.

A Cascaded Vision Transformer for Precise Identiﬁcation of Vehicle Number Plate

593

C. Wei, Z. Tan, Q. Qing, R. Zeng, and G. Wen, “Fast

Helmet and License Plate Detection Based on

Lightweight YOLOv5,” Sensors (Basel), vol. 23, no. 9,

Apr. 2023, doi: 10.3390/s23094335.

Babu, AV Santhosh, P. Meenakshi Devi, B. Sharmila, and

D. Suganya. "Performance analysis on cluster-based

intrusion detection techniques for energy efficient and

secured data communication in MANET."

International Journal of Information Systems and

Change Management 11, no. 1 (2019): 56-69.

K. Yamagata, J. Kwon, T. Kawashima, W. Shimoda, and

M. Sakamoto, “Computer Vision System for

Expressing Texture Using Sound-Symbolic Words,”

Front Psychol, vol. 12, p. 654779, Oct. 2021.

A. M. Buttar et al., “Enhanced neurological anomaly

detection in MRI images using deep convolutional

neural networks,” Front Med (Lausanne), vol. 11, p.

1504545, Dec. 2024.

Karthikeyan, S., and P. Meenakshi Devi. "An attempt to

enhance the time of reply for web service composition

with QoS." International Journal of Enterprise

Network Management 11, no. 4 (2020): 289-303.

E. Habeeb, T. Papadopoulos, A. R. Lewin, and D. Knowles,

“Assessment of Anticoagulant Initiation in Patients

with New-Onset Atrial Fibrillation During Emergency

Department Visit-Point-by-Point Response,” Clin Appl

Thromb Hemost, vol. 29, p. 10760296231172493, Jan-

Dec 2023.

G. V. T. Silvano et al., “Artificial Mercosur license plates

dataset,” Data Brief, vol. 33, p. 106554, Dec. 2020.

M. Ghatee and S. Mehdi Hashemi, Artificial Intelligence

and Smart Vehicles: First International Conference,

ICAISV 2023, Tehran, Iran, May 24-25, 2023, Revised

Selected Papers. Springer Nature, 2023.

S. Saini, K. Lata, and A. Sharma, Advances in Image and

Data Processing Using VLSI Design: Smart Vision

Systems. IOP Publishing Limited, 2021.

J.-S. Chou and C.-H. Liu, “Automated Sensing System for

Real-Time Recognition of Trucks in River Dredging

Areas Using Computer Vision and Convolutional Deep

Learning,” Sensors (Basel), vol. 21, no. 2, Jan. 2021,

doi: 10.3390/s21020555.

Y. Lu, N. Vincent, P. C. Yuen, W.-S. Zheng, F. Cheriet, and

C. Y. Suen, Pattern Recognition and Artificial

Intelligence: International Conference, ICPRAI 2020,

Zhongshan, China, October 19–23, 2020, Proceedings.

Springer Nature, 2020.

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

COMMUNICATION, AND COMPUTING TECHNOLOGIES

594