Efficient detection of building in remote sensing images using an improved YOLOv10 network

Objectives. At present, rapid detection of the location and size of building objects from remote sensing images is important for scientific research value and has practical significance for urban planning, environmental monitoring and disaster management. Methods. This paper proposes an object detection method based on improved YOLOv10 network, which incorporates Super Token Attention, RepConv and Normalized Weighted Distance to more precisely detect buildings in remote sensing images. This method improves the detection accuracy and efficiency especially for small objects. The LEVIR-CD dataset is used for model training and testing. Results. The experimental results show that the method demonstrates better accuracy on the building detection task than the traditional YOLOv10 and other methods. Conclusion. The proposed method significantly enhances the accuracy and efficiency of building detection in remote sensing images