DCMA-Net: A dual channel multi-scale feature attention network for crack image segmentation

Cracks are a common structural damage to pavement that significantly threatens traffic safety, so regular inspections of road conditions are essential for maintaining traffic safety. Artificial intelligence (AI) technology has demonstrated significant potential in various computer vision tasks and has been widely applied to the field of crack detection in recent years. However, the crack detection remains a challenging task due to the intricacy of crack types, the presence of intensity inhomogeneities, and the difficulty in detecting edge areas within a complex background. To address these challenges, this paper proposes a novel dual channel multi-scale feature attention segmentation network (DCMA-Net) model for crack detection. Specifically, this two feature extraction channels are exploited to obtain global and local information. A multi-scale feature extraction module (MFE) is integrated into the residual structure to capture depth multi-scale information. Additionally, a hybrid attention mechanism is employed to connect the coder and decoder, allowing the network to focus on the target area from different scales and dimensions to enrich crack feature representation. To verify the superiority of the proposed method, we evaluate it on three crack datasets and compare it with state-of-the-art crack detection techniques. The experimental results demonstrate the superior segmentation performance of our proposed network compared to existing advanced methods.