TAU-Net: Trident Aggregation U-Net for Fatigue Crack Segmentation

The accurate detection of early cracks during aircraft structural fatigue tests serves as a pivotal factor in studying the damage tolerance behavior of aircraft structures, thereby safeguarding the operational safety of aircraft. Aiming at the loss of crack contour information and excessive background interference in the fatigue crack segmentation task, an aircraft structure crack segmentation method based on TAU-Net is proposed. Firstly, the trident aggregation structure is designed to improve the segmentation quality of crack boundaries and tips. Secondly, by embedding the ECA (Efficient Channel Attention) module into the contracting path, the background interference, which is highly similar to crack, to the network is suppressed. The experimental results on the self-collected dataset shows that the precision, recall and mIoU of the proposed TAU-Net is better than other methods, which verifies the feasibility of the network.