Post-earthquake detection of surface spalling and cracks in masonry buildings based on computer vision

Rapid post-earthquake detection of visible damage to buildings is essential for damage assessment and ensuring safety. The inefficiency of traditional methods, such as manual visual inspection and measurement, necessitates automatic solutions, especially for masonry buildings, which are widely used and more vulnerable in economically underdeveloped areas. In this study, an automatic damage detection method combining machine learning and image processing techniques was proposed to identify and quantify cracks and spalling on the plastered surfaces of masonry walls. An image classification model based on deep convolutional neural networks (CNNs) was developed, which was combined with the sliding window technique and Otsu threshold segmentation method to generate binary images of the damage regions from captured images. This method streamlines data preparation by avoiding pixel-level annotation, significantly reducing manual labeling effort compared to direct training of segmentation models. Subsequently, based on the differences in geometry and size between the crack region and spalling region, morphological operations were applied to separate them. Finally, the maximum inscribed circle algorithm and the pixel counting method were used to measure the maximum crack width and spalling area respectively. The measurement results indicate that the proposed method can effectively identify the crack and spalling regions on the plastered surfaces of masonry buildings from photography images, as well as achieve pixel-level segmentation of the damage and accurately quantify crack widths and spalling areas. The proposed method provides a reliable solution for large-scale damage detection in the early stages following the earthquake.