Road damage detection algorithm based on optimised You Only Look Once version 8

Automatic detection and identification of road damage is of vital importance for the prevention of traffic accidents. Given the shortcomings of existing target detection algorithms in processing road damage images, we propose a Yolov8 model based on dynamic snake convolution (DSConv) optimisation. By introducing an innovative fusion module C2DS, we achieve efficient extraction of elongated cracks and weak target features; at the same time, combined with the CBAM attention mechanism module, we significantly improve the accuracy and robustness of detection. In addition, we adopt F-EIOU Loss to further improve the convergence speed and positioning accuracy of the model. The experimental results on the RDD2022 dataset show that Precison, Recall, F1, and mAP are improved to 82.3%, 73.4%, 77.0%, and 77.9%, respectively, which fully validates the superior performance and accuracy of the proposed method in road damage detection. This study not only provides a reference for future lightweight improvement, but also lays a solid foundation for practical application.