A network for fatigue crack segmentation and quantification using frequency-domain enhancement and linear attention mechanism

Fatigue cracks are one of the primary causes of structural failure in engineering structures. To meet the practical need for precise detection of fatigue cracks in structural strength testing, we propose a fatigue crack segmentation and quantification network-CrackDAM-Net-based on frequency domain enhancement and the Mamba linear attention mechanism. This network adopts a dual-branch parallel structure comprising a local feature encoder and a global information encoder, effectively balancing the local detailed features and global contextual information of fatigue crack images. A global information encoder combining frequency domain enhancement and Mamba linear attention is designed: the first two layers are hybrid frequency domain-attention layers, and the last two layers are pure attention layers. This encoder not only effectively extracts global features from crack images but also significantly reduces computational complexity. A multi-scale feature fusion module combining spatial attention, channel attention, and pixel attention mechanisms is proposed, which can fully mine and integrate complementary information between the local feature encoder and the global information encoder from multiple dimensions to enhance feature expression capabilities. A crack length automatic quantification method combining orthogonal projection and discrete skeleton evolution is established to achieve precise measurement of fatigue crack length in structural strength tests. Experimental results demonstrate that CrackDAM-Net achieves a good balance between accuracy and speed, with segmentation accuracy, recall rate, and F1 score reaching 83.11%, 77.56%, and 80.24%, respectively, segmentation speed reaching 37 frame/s, and fatigue crack length quantification error within ±4%, showcasing strong potential for engineering applications.