WHU-RRoad: A high-resolution remote sensing benchmark dataset for rural road extraction

Rapid and accurate extraction of road infrastructure from high-resolution remote sensing satellite imagery is important for traffic planning, construction, and management. In recent years, road extraction methods have developed rapidly, benefiting from the application of data-driven deep learning based models and the various urban road datasets. However, there are still application bottlenecks when directly transferring the current research from urban to rural areas. Specifically, most road datasets are designed for urban areas, and only a small number of rural scenes are included, without complex rural scenes. Due to the huge style differences between urban and rural roads in different geographical areas, it is difficult to apply the current datasets to rural road extraction. In this article, a large-scale high-resolution remote sensing road dataset, termed WHU-RRoad, is introduced for rural road extraction, which contains 27770 pairs of 1024 × 1024 satellite images with resolution of 0.3m and corresponding road annotations, covering a 2620.71 km2 rural area in central China. In addition, a comprehensive analysis of the performance of the current state-of-the-art deep learning based road extraction methods on the WHU-RRoad dataset is provided, where the experimental results illustrate that the proposed WHU-RRoad dataset is a challenging dataset for large-scale rural road extraction. At the same time, the WHU-RRoad dataset can meet the application requirements of rural road construction and has great application potential.