CFRNet: Road Extraction in Remote Sensing Images Based on Cascade Fusion Network

Road extraction from remote sensing images has attracted widespread attention of researchers due to its crucial role in the fields of autopilot, urban planning, navigation, and other fields. However, the task becomes challenging as the roads in remote sensing images are easily occluded by obstacles such as shadows, buildings and trees. In this letter, a cascade fusion network for road extraction (CFRNet) in remote sensing images is proposed. Considering the lightweight characteristics of MobileNet block (MbBlock), it is used as the feature extraction module of the backbone network. To enable CFRNet to generate and fuse more features at multiscale, we design several cascade stages. Each stage includes a sub-backbone for feature extraction and a triple-level adaptive feature fusion (TAFF) module for feature fusion. This structure can more deeply and effectively fuse multiscale features with most of the parameters in the entire backbone. The experimental results demonstrate that the proposed CFRNet significantly outperforms other state-of-the-art methods on the publicly available Istanbul City Road dataset and DeepGlobe Road dataset. Specifically, it achieves an intersection over union (IoU) of 89.76%, reflecting a 5.3% improvement on the Istanbul dataset, and 67.22% with a 0.98% enhancement on the DeepGlobe Road dataset. Our code is available at https://github.com/XYQ1517/CFRNet .