Mtech_final_year_project

The early identification and assessment of structural cracks in buildings is critical for preventing progressive damage and ensuring public safety. Traditional manual inspection methods are time-consuming, subjective, and prone to human error, particularly in large scale infrastructure monitoring. This project presents an Explainable Deep Learning-based Crack Detection and Severity Assessment System designed for real-world building environments. Unlike existing approaches that are primarily trained on concrete-only datasets and focus solely on binary crack classification, the proposed system addresses multi-surface building conditions, including plastered walls, painted surfaces, brick masonry, and concrete structures. A hybrid deep learning pipeline combining a segmentation model (U-Net / YOLO-Seg / SegFormer) and morphological post-processing is employed to accurately localize cracks and measure geometric attributes such as length and width, enabling severity classification into hairline, moderate, and severe categories. To enhance trust and interpretability, Explainable AI (XAI) techniques such as Grad-CAM and integrated gradient visualization are integrated to highlight model decision regions and differentiate true cracks from noise, shadows, stains, and texture variations. The system is trained on a combination of publicly available crack datasets and custom real-world image data collected from building environments, enabling robust performance across diverse lighting, texture, and surface conditions. Experimental evaluations demonstrate strong performance in crack segmentation accuracy, severity estimation, and generalization across unseen building surfaces. The proposed solution provides a reliable and transparent tool for civil engineers, facility managers, and inspection agencies, paving the way toward automated structural health monitoring and safer infrastructure management.