基于图像语义分割的冰雹致建筑外窗受损量化评估方法

In hail disasters, building exterior windows often suffer severe damage, leading to the failure of the building envelope. Traditional damage assessment methods typically rely on manual on-site inspections, which is lack of objectivity and efficiency, making it difficult to promptly and accurately quantify the damage to the building envelope caused by hail. This study proposes an intelligent quantitative assessment method for damaged building exterior windows based on image semantic segmentation. The Swin Transformer model for identifying window damage is trained using online data, and the transfer learning is applied using on-site photography data from disaster scenes to achieve intelligent quantitative identification of damaged building exterior windows. An exterior window damage index (EWDI) is introduced based on this identification. Case studies demonstrate that the proposed method can accurately identify damaged areas of building exterior windows with an accuracy exceeding 85%. The proposed EWDI effectively evaluates the extent of damage to building exterior windows, providing reliable quantitative metrics for disaster assessment and post-disaster recovery efforts.