Computationally Enhanced UAV-Based Real-Time Pothole Detection Using YOLOv7-C3ECA-DSA Algorithm

Road deterioration due to potholes has a significant impact on traffic safety and infrastructure maintenance, highlighting the need for detection systems that combine precision with real-time capabilities. Although YOLO-based algorithms have been widely adopted for their speed and efficiency in object detection, achieving a balance between high accuracy and low inference time remains a challenge, particularly in scenarios involving small objects and complex features. This study introduces YOLOv7-C3ECA-DSA, an improved YOLOv7 architecture designed to address these limitations. The model incorporates Cross-Stage Enhanced Channel Attention (C3ECA) blocks in the backbone network and Depthwise Shuffle Attention (DSA) in the detection head to enhance feature learning and boundary detection, achieving high detection accuracy and real-time inference capabilities. The experimental results demonstrate that YOLOv7-C3ECA-DSA achieves an mAP0.5 of 85.3% with an inference time of 10.9 ms, outperforming the prior methods. The proposed model also performs well under adverse vision conditions such as night and rain. However, the model performance may be limited in severe environmental conditions, such as low-contrast surfaces or heavily obscured potholes. Despite these limitations, the research significantly advances real-time pothole detection by striking an optimal balance between computational efficiency and detection accuracy. The findings underscore the effectiveness of YOLOv7-C3ECA-DSA in addressing practical challenges in real-time infrastructure monitoring, making it suitable for scalable deployment in autonomous vehicle systems and road maintenance applications.