Attention-Assisted Dense Gated Convolutional Network: License-Plate Classification Using Deep-Learning-Based Effective Feature-Extraction Module

Abstract

Traffic-management systems depend on the identification and recognition of license plates to perform tasks such as traffic monitoring, toll collection, and regulation enforcement. However, challenges such as blurriness, dust, shadows, and various lighting conditions often hinder accurate license-plate detection. This paper addresses these challenges by using a novel hybrid deep-learning model. Initially, enhancement contrast-limited adaptive cumulative histogram equalization (ECLACHE) is utilized to improve the contrast and quality of the images. The modified Swin transformer (M-ST) is then employed to extract deep features. The extracted features are fused together using a feature-pyramid fusion module. Finally, the attention-assisted dense gated convolutional network (AADGCN) model is utilized to classify the license plates efficiently. The proposed model achieves an accuracy of 99.12%. Through comprehensive experimentation and evaluation, this paper demonstrates the effectiveness of the proposed strategy in addressing the problems of overfitting and underfitting that are common in existing detection models. These results show significant improvements in detection and recognition accuracy, demonstrating the potential of the proposed methodology to enhance the performance of intelligent traffic-management systems.