Road Crack Detection Algorithm Based on Large Convolution Kernels

Road crack detection is widely used in road maintenance and bridge safety monitoring. The complex textures of road cracks lead to occlusion between targets, and environmental factors affect images, causing color shifts and blurring. To address these issues, this paper proposes the YOLOv8-LSK road crack detection algorithm. First, a Spatial Fusion Module (SMF) is designed to enhance the focus on spatial dimension information, improving the detection capability of occluded cracks. Second, a large-kernel convolution attention feature extraction module is designed to enhance the feature extraction capability of blurred and color-shifted road images. Finally, structural reparameterization (SR) is applied to the cross-scale feature fusion unit of YOLOv8-LSK, optimizing the model’s detection speed. Experimental results show that the YOLOv8-LSK algorithm achieves a detection accuracy of $\mathbf{7 2. 1 \%}$ on the RDD2022 dataset, with detection speeds of 33FPS, respectively.