Enhancing crack detection and severity assessment in historical Tabiya basins using U-Net and adaptive thresholding.

The conservation of the historical Tabiya water basins remains paramount, with consideration being their cultural and architectural importance, though structural degeneration like surface cracking poses a formidable challenge to conservation work. Since the traditional methods of inspection are often subjective, tedious, and prone to error, these limitations are tackled in this study by means of presenting an automated system for surface crack detection and segmentation based on artificial intelligence and computer vision techniques. High-resolution images were captured on-site using a Canon EOS 1100D camera and analyzed within a comparative deep learning framework using four models, namely U-Net with MobileNetV2, ResNet-50, InceptionV3, and EfficientNetB7 backbones. The proposed system performs crack detection and segmentation, as well as quantitative measurements, including crack length, width, and severity assessment through skeletonization, a crack length estimation algorithm, and a crack width extraction method. Experimental results indicated that the MobileNetV2-based model outperformed all other tested architectures, with an accuracy of 98.7%, a recall of 98.2%, a precision of 99.1%, and an F1-score of 98.6%. Furthermore, the developed framework has also been deployed as a web application that allows users to upload or drag and drop images and select from four available models for automated analysis. This integrated system represents a strong, precise, and user-friendly tool for the digital preservation and structural monitoring of heritage water infrastructure.