Abstract: Land cover classification of remote sensing imagery is a crucial tool for obtaining surface information. It is essential for accurately understanding the characteristics and changes of the Earth's surface. Given the limitations of existing land cover classification datasets, such as single data sources and insufficient spectral information, this study constructs a multi-source remote sensing land cover classification dataset with rich spectral information and high spatial resolution. The dataset was collected using unmanned aerial vehicles (UAVs) equipped with a multispectral camera and a dual-light thermal infrared camera over the National Defense Park in Wuhan. Data processing and annotation were conducted using ArcGIS and Labelme software. This dataset includes visible light, thermal infrared, and multispectral images with five single bands, covering six typical land cover types: water bodies, vegetation, bare land, buildings, brick roads, and asphalt roads. In total, it contains a total of 612 image tiles and corresponding labels. This dataset provides data support for research fields such as multi-source data fusion and analysis, remote sensing land cover extraction, and recognition.
Keywords: multi-source data; land cover classification; multispectral images