Post-seismic structural damage segmentation using YOLOv8-seg model

A customized YOLOv8-seg architecture is hired in this study to automatically detect and segment post-earthquake damage, such as cracks, spalling, reinforcement exposure, crushing, buckling, and structural failure, that appears on bridge piers tested using slow and fast cyclic tests, shaking table tests, and real-time hybrid simulations. Using a hybrid loss function, the YOLOv8-seg model processes 32×32 and 256×256-pixel image patches, extracted from 124 large RGB images, for cracks and other seismic damage categories, respectively. Training is conducted on the image patches and their corresponding labeled annotations, distinguishing between seismic damage and background (non-damage) pixels. The model is trained with a batch size of 16, utilizing the Adamax optimizer, an exponential learning rate scheduler, and weight decay techniques to improve training stability and performance. The results demonstrate that the generated mask patches closely resemble the actual damage patterns.