首页 > , Vol. , Issue () : -
Natural disasters occur frequently in Yunnan, China, causing huge losses of life and property. For this reason, this paper uses the object detection technology of deep learning based on remote sensing images to quickly search for the location of damaged buildings caused by natural disasters, and then provide assistance for disaster relief. However, there are still some challenges in the field of damaged building detection, such as the less data of earthquake-damaged buildings and the weak features of the target to be detected. Therefore, this paper constructs a UAV remote sensing images-based large-scale high-resolution earthquake-damaged buildings database (UEDB). It mainly collected 4,598 remote sensing images in the disaster area of Yangbi Yi Autonomous County, Dali Bai Autonomous Prefecture, Yunnan Province. And this dataset contains 76,012 building instances, and each instance is labeled in three formats: object location box label, object segmentation label, and object boundary label. Aiming at the current problems, this paper proposes an earthquake-damaged buildings real-time detection model (EDBRDM), which includes an object feature alignment module (OFAM), a feature difference calculation module (FDCM) and an object boundary constraint based-position box detection module (OBCPB). The proposed model achieves an accuracy of 86% on the test dataset of UEDB, and it has also been verified in actual scenes at different locations and achieved well results.