Road extraction from high-resolution remote sensing images can provide vital data support for applications in urban and rural planning, traffic control, and environmental protection. However, roads in many remote sensing images are densely distributed with a very small proportion of road information against a complex background, significantly impacting the integrity and connectivity of the extracted road network structure. To address this issue, we propose a method named StripUnet for dense road extraction from remote sensing images. The designed Strip Attention Learning Module (SALM) enables the model to focus on strip-shaped roads; the designed Multi-Scale Feature Fusion Module (MSFF) is used for extracting global and contextual information from deep feature maps; the designed Strip Feature Enhancement Module (SFEM) enhances the strip features in feature maps transmitted through skip connections; and the designed Multi-Scale Snake Decoder (MSSD) utilizes dynamic snake convolution to aid the model in better reconstructing roads. The designed model is tested on the public datasets DeepGlobe and Massachusetts, achieving F1 scores of 83.75% and 80.65%, and IoUs of 73.04% and 67.96%, respectively. Compared to the latest state-of-the-art models, F1 scores improve by 1.07% and 1.11%, and IoUs increase by 1.28% and 1.07%, respectively. Experiments demonstrate that StripUnet is highly effective in dense road network extraction.

In remote sensing images, roads are usually in complex shapes and can be partially occluded by buildings, trees, and other surroundings. To extract a complete and continuous road network is still a challenging job. This paper proposes a context-aware road extraction method for remote sensing imagery based on Transformer network, in which, a foreground feature enhancement module (FFEM) is designed to further extract detailed road features such as contours from the shallowest feature map; Dual-attention module (DAM) is constructed and applied at different skip connections to make the model focus more on road features in the different level of feature maps; A Swin Transformer-based contextual information extraction module (CIEM) is built between the encoder and decoder modules to capture the global and local road contextual information so as to recover the occluded roads information as much as possible. Furthermore, a multi-scale decoder (M-Decoder) is designed to improve the feature map recovery ability of the decoder module. Experiments on the DeepGlobe road dataset are conducted to verify the efficiency of the proposed method.

Remote sensing images contain many objects that resemble road structures, making it diffcult to distinguish roads from the background. Moreover, road extraction is affected by many factors, such as lighting conditions, noise, occlusions, etc., resulting in incomplete and discontinuous road extraction. Learning discriminative road features from remote sensing images is a highly challenging task. In this paper, a novel road extraction model is proposed for remote sensing images under encoder and decoder U-Net like architecture. An axial Transformer module (ATM) is designed to learn global road features in the deepest layer with linear computational complexity regarding image size. And a multilayer attention fusion module (MLAF) is also presented to fuse multiple layers of Transformer features, obtaining more comprehensive and richer semantic information. In the skip connection, a channel attention module (CAM) is designed to weight the feature maps along the channel dimension, with the goal of improving the capability of feature representation. Extensive experiments are conducted on the DeepGlobe and Massachusetts road datasets. Compared with other methods, our proposed method in this paper realized road extraction from remote sensing images with higher accuracy and less computational cost, e.g., achieving a intersection over union (IoU) of 81.71% (1.02% improvement) and a 22.38% reduction in convergence time over the latest TransRoadNet on the Massachusetts road dataset. Ablation experiments also demonstrate the effectiveness of the designed model.

Multi-level road feature extraction from remote sensing image plays an important role in numerous applications such as autonomous driving and urban planning. However, interference from background, occlusions and road-like information making it difficult to distinguish different levels of roads. To address these issues, this study proposes a deep network named BMDCNet, which adopts LinkNet [1] as the baseline model. The Bidirectional Multi-level Road Feature Dynamic Fusion (BMDF) module is designed to replace the simple skip connections, which greatly reduce the semantic gap between the encoder and decoder. Furthermore, the Dual Context Dynamic Extraction (DCDE) module is designed to dynamically extract and integrate global and local multi-scale context information. Finally, experiments are conducted on the DeepGlobe Road Extraction Dataset and the Massachusetts Roads Dataset. The results demonstrate that compared with LinkNet, the F1 and IoU of BMDCNet increased by 1.79% and 2.68% on DeepGlobe, and by 0.35% and 0.47% on Massachusetts, respectively. Our source code is available at https://github.com/ZehuaChenLab/BMDCNet.