A Weather-Adaptive Convolutional Neural Network Framework for Better License Plate Detection.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Intelligent Transport Systems

Background:

• Automatic License Plate Recognition (ALPR) systems are crucial for transportation management and security.
• Environmental factors like rain, fog, and pollution degrade ALPR performance.
• Existing ALPR systems lack robustness in adverse weather conditions.

Purpose of the Study:

• To develop a weather-adaptive Convolutional Neural Network (CNN) framework for enhanced ALPR.
• To improve the robustness and accuracy of license plate detection in adverse weather.
• To leverage the YOLOv10 model for superior performance in challenging environments.

Main Methods:

• Implemented a weather-adaptive CNN framework utilizing the YOLOv10 object detection model.
• Employed weather-specific data augmentation techniques to train the model.
• Evaluated performance using precision, recall, F1, mAP50, and mAP50-95 metrics.

Main Results:

• Demonstrated significant improvements in ALPR detection performance under adverse weather conditions.
• The weather-adaptive framework showed enhanced robustness across diverse environmental scenarios.
• Achieved higher precision, recall, and mAP scores compared to baseline models.

Conclusions:

• The proposed weather-adaptive CNN framework offers a promising solution for resilient ALPR systems.
• This approach effectively addresses the challenges posed by adverse weather in ALPR applications.
• The study highlights the potential for deploying robust ALPR in complex environmental settings.