2026-10-14T11:46:50Z https://www.scidb.cn/oai

10.57760/sciencedb.30821 2026-03-30T11:39:39Z

2026-03-30 CHN10-CUG High-Resolution China Road Dataset doi:10.57760/sciencedb.30821 en CHN10-CUG is a large-scale benchmark dataset specifically designed for road extraction tasks from high-resolution remote sensing imagery, constructed by the team at China University of Geosciences (Wuhan). The dataset covers ten representative Chinese cities with diverse geographical environments: Beijing, Shanghai, Guangzhou, Shenzhen, Wuhan, Hangzhou, Qingdao, Dalian, Hong Kong, and Macau. It comprises 8,020 RGB images at 0.5-meter resolution, each with dimensions of 512×512 pixels. All images have been meticulously annotated by remote sensing experts, providing three types of labels: road surface, centerline, and graph structure. These annotations encompass 20 road subcategories, including highways, main roads, secondary roads, residential roads, bicycle paths, and pedestrian walkways, fully reflecting the complexity and diversity of China's urban road networks. The construction of CHN10-CUG aims to address the prominent limitations of existing public datasets, such as limited regional coverage, single road types, and simplistic annotation formats. Compared with existing datasets like DeepGlobe, CHN10-CUG demonstrates significant advantages in regional diversity (from high-density core urban areas to mountainous coastal cities), road category richness (20 subcategories), and annotation completeness (surface + centerline + graph structure). It authentically represents road features under complex backgrounds, including occlusions, shadows, spectral variations, and topological structures, providing a more challenging and representative data foundation for road extraction research. Road extraction; Benchmark dataset; High-resolution remote sensing; Deep learning; Semantic segmentation Zhu Qiqi Zhang Tiansheng PUBLIC https://creativecommons.org/licenses/by-nc-nd/4.0/ dataset Science Data Bank