Explainable AI-Based Surface Damage Recognition for Automated Structural Health Monitoring

The safety of civil infrastructure requires intelligent and explainable systems that should operate independently to detect surface damage. The inspection process needs person- nel to perform time-consuming subjective evaluation because traditional AI systems fail to provide both generalization and interpretability. The research introduces a lightweight, explain- able Structural Health Monitoring (SHM) pipeline that uses the Segment Anything Model (SAM) for segmentation and ResNet for damage detection. The system achieves explainability through Grad-CAM visual interpretations which show the image sections that influence the model’s decision-making process. The proposed framework achieves flexibility, computational efficiency, and interpretability, enabling the trustworthy deployment of moni- toring systems in real-world settings. Experiments on a state-of- the-art benchmark demonstrate that introducing segmentation before classification significantly improves detection accuracy and reliability, particularly for complex textures or subtle defect boundaries.