A high accurate and fast road crack detection algorithm based on YOLOv8

In order to solve the problem of balancing the detection accuracy and speed in the existing road crack detection algorithms, this paper proposes an improved algorithm based on YOLOv8, named RPDD. First, we use the RepViT module to replace the original backbone network structure, which enhances the feature extraction capability. Second, we employ DySample up-sampling in the neck structure, which significantly improves the processing capability for low-resolution images. Finally, we change the detection head to a novel DyHead structure and combine it with deformable convolution (DCNV4), which enhances the adaptability to various scales and complex scenes. The experimental results show that RPDD achieves significant progress compared to the baseline, including 6.8% increase in precision, 4.6% increase in recall, 7.6% increase in [email protected], and 72.9 increase in FPS, which validates its effectiveness in improving detection accuracy and speed.