Building Detection in Earthquake Prone Urban Area with Satellite Imagery Data Using Deep Learning

Earthquakes have the potential to be extremely damaging, resulting in significant loss of life and property damage, particularly in urban areas. The ability to observe damaged buildings is critical for emergency management professionals because it allows them to quickly direct rescue teams to the right place. Currently deep learning models have achieved cutting-edge accuracy in extracting natural image shapes from large, labeled data sets. This success has included implementation in the field of remote sensing despite problems with the lack of valid label data supply, especially those that are fully segmented. The goal of this work is to create damage assessment in earthquake-prone urban areas using satellite imagery and the deep learning model, with each image pixel labeled as either a building or a non-building label. Compared to other deep learning models such as UNet and DeepLab, the results show that PSPnet is the best model in this regard, as it achieves training and testing accuracy of 94.3 percent and 89.8 percent, respectively.