Damage assessment of bridges using segmentation and shape analysis from aerial imagery

Rapid and quantitative assessment of bridge damage immediately after a disaster is critical for effective emergency response and early recovery. In this study, a method was developed to apply the YOLOv8 segmentation model to aerial imagery for detecting and classifying bridges based on the presence or absence of damage and to estimate bridge length and width from the detected regions to evaluate changes in shape dimensions. The results confirmed that for undamaged bridges, both bridge length and width could be estimated with high accuracy. In contrast, for damaged bridges, debris accumulation and inundation often resulted in only partial detection of the bridge, leading to a tendency for dimensional underestimation, particularly in the bridge length direction. Furthermore, it was confirmed that damage caused changes in the relative scaling relationship between bridge length and width, disrupting the original balance between these dimensions. These findings indicate that the observed dimensional shrinkage is not merely an estimation error but a quantitative characteristic of damage itself. This suggests that analysing changes in shape dimensions is a potentially effective method for detecting the presence of bridge damage.