Yolov9-Powered Bangla License Plate Recognition: A Comparative Analysis for Optimized Performance in Localized Contexts

This study presents an advanced approach to Bangla license plate recognition using YOLOv9, the latest iteration in the YOLO (You Only Look Once) object detection series. We employ a comprehensive dataset of Bangladeshi vehicle license plates, encompassing 12 vehicle categories across four major cities. The dataset underwent extensive preprocessing, including merger, format conversion, and augmentation to optimize it for YOLOv9 architecture. We evaluate five YOLOv9 model variants (YOLOv9-T, YOLOv9-S, YOLOv9-M, YOLOv9-C, and YOLOv9-E) on this dataset, analyzing their performance in terms of accuracy, efficiency, and generalization capability. Our results demonstrate a clear correlation between model size and detection accuracy, with YOLOv9-E achieving the highest F1 score of 0.9550. We observe a trade-off between model efficiency and accuracy, with larger models offering superior detection capabilities at the cost of increased computational demands. The study provides insights into the optimal model selection for various real-world scenarios, balancing between high-throughput requirements and accuracy priorities.