UAV-BCD: A UAV Building Change Detection Dataset

Zilu Ying, Zijun Tan, Wenba Li, Jianhong Zhou, Zhangzhao Liang, Yikui Zhai

Abstract: Remote sensing change detection (RSCD) holds significant prominence as a research topic within the realm of computer vision. However, previous RSCD datasets have been constructed based on satellite remote sensing images. Traditional satellite remote sensing images have problems such as insufficient resolution, difficult data acquisition, and complex processing processes, and there is a certain gap between data distribution and actual needs. UAVs not only have the advantages of flexibility and high-speed, but also can capture high-resolution images, which are especially suitable for high-precision RSCD in small areas. Therefore, this paper proposed a new UAV RSCD dataset — UAV Building Change Detection Dataset (UAV-BCD). The proposed dataset contains 2024 pairs of finely registered high-resolution images collected by UAVs and their corresponding pixel-level labels, which can provide a new benchmark for RSCD. We evaluate the effectiveness of UAV-BCD with the five state-of-art deep neural networks in RSCD.