BF-YOLOv8 Based Pavement Defect Detection

Pavement defect detection refers to the various technical means to inspect, identify and evaluate road surfaces to ensure safety. Traditional target detection methods are inefficient although they are still effective in many occasions. Modern pavement defect detection is moving towards automation and intelligence, such as inspection by drones or automated vehicles, to improve the efficiency and accuracy of detection. In this paper, an improved yolov8 pavement defect detection algorithm BFyolov8 is proposed. In this method, the SAM- DETR++ fusion module is designed to enhance the model’s ability to detect small targets. The improved BAM module is added to the backbone to make the detection more efficient in paying attention to the main features of the input image. Moreover, to reduce AMP mixed-precision training, the 16-bit floating-point number (FP16) and the 32-bit floating-point number (FP32)is used to maintain accuracy as much as possible. This proposal effectively improves the accuracy and real-time performance of yolov8 in pavement defect detection applications.