Road defect detection algorithm based on improved YOLOv10n

Roads, as a crucial component of modern transportation systems, are susceptible to defects such as cracks, potholes, and patches due to environmental erosion, vehicle loads, and construction quality. In recent years, advancements in deep learning have provided new approaches for automated road defect detection. This study proposes an optimized lightweight road defect detection algorithm based on the YOLOv10n model. First, the RDD-2022 China dataset is selected, covering five types of road defects: longitudinal cracks, transverse cracks, alligator cracks, potholes, and patches. The model is then enhanced by integrating adaptive convolution, adaptive downsampling modules, and the convolutional block attention module. Specifically, adaptive convolution dynamically adjusts the receptive field of convolutional kernels, improving adaptability to defects of varying sizes and shapes. The adaptive downsampling module mitigates the information loss problem associated with traditional pooling layers, retaining more low-level features while ensuring effective dimensionality reduction and enhancing feature representation. Additionally, the convolutional block attention module, combining channel and spatial attention, strengthens the model's focus on target areas, improving detection performance in complex backgrounds. Experimental results demonstrate that, compared to the baseline model, the proposed improvements achieve superior detection accuracy, highlighting the effectiveness of the optimized approach.