A pyramid auxiliary supervised U-Net model for road crack detection with dual-attention mechanism

The application of road crack detection technology plays a pivotal role in the domain of transportation infrastructure management. However, the diversity of crack morphologies within images and the complexity of background noise still pose significant challenges to automated detection technologies. This necessitates that deep learning models possess more precise feature extraction capabilities and resistance to noise interference. In this paper, we propose a pyramid auxiliary supervised U-Net model with Dual-Attention mechanism. Pyramid auxiliary supervision module is integrated into the U-Net model, alleviating information loss at the encoder end due to pooling operations, thereby enhancing its global perception capability. Besides, within dual-attention module, our model learns crucial segmentation features both at the pixel and channel levels. These enable our model to avoid noise interference and achieve a higher level of precision in crack pixel segmentation. To substantiate the superiority and generalizability of our model, we conducted a comprehensive performance evaluation using public datasets. The experimental results indicate that our model surpasses current great methods. Additionally, we performed ablation studies to confirm the efficacy of the proposed modules.