Based on high-resolution remote sensing images of Chenzhou's main urban area from 2010, 2015, 2020 and 2024 years, a deep learning-based road extraction model was applied to generate binary road maps, which were subsequently vectorized. The resulting dataset includes spatial locations, lengths, widths, and other attributes of the roads. The generation of this data product involved multiple steps: first, the individual scene images were mosaicked, fused, and clipped to produce a 2m resolution remote sensing image. Then, a deep learning-based algorithm was applied to extract roads, resulting in a binary road image. For the binarized road image, it was imported into ArcGIS, and automated vectorization was performed using the ArcScan toolbar. Finally, geometric calculations were conducted on the vectorized roads to derive their lengths.