DSFC-Net: A Dual-Encoder Spatial and Frequency Co-Awareness Network for Rural Road Extraction

TL;DR

Rural road extraction from high-resolution imagery is hampered by high intra-class variability, vegetation occlusion, and narrow road widths. The authors propose DSFC-Net, a dual-encoder network that fuses a CNN-based spatial stream with a Spatial-Frequency Hybrid Transformer (SFT) that decouples high- and low-frequency information via Cross-Frequency Interaction Attention. A Channel Feature Fusion Module (CFFM) adaptively reconciles the two streams to preserve local textures and global topology. Evaluations on WHU-RuR+, DeepGlobe, and Massachusetts show state-of-the-art performance and robust cross-dataset generalization, highlighting the method’s potential for reliable rural road mapping and infrastructure planning. The work advances frequency-aware segmentation for sparse, occluded rural roads and offers a practical tool for SDG monitoring and rural development planning.

Abstract

Accurate extraction of rural roads from high-resolution remote sensing imagery is essential for infrastructure planning and sustainable development. However, this task presents unique challenges in rural settings due to several factors. These include high intra-class variability and low inter-class separability from diverse surface materials, frequent vegetation occlusions that disrupt spatial continuity, and narrow road widths that exacerbate detection difficulties. Existing methods, primarily optimized for structured urban environments, often underperform in these scenarios as they overlook such distinctive characteristics. To address these challenges, we propose DSFC-Net, a dual-encoder framework that synergistically fuses spatial and frequency-domain information. Specifically, a CNN branch is employed to capture fine-grained local road boundaries and short-range continuity, while a novel Spatial-Frequency Hybrid Transformer (SFT) is introduced to robustly model global topological dependencies against vegetation occlusions. Distinct from standard attention mechanisms that suffer from frequency bias, the SFT incorporates a Cross-Frequency Interaction Attention (CFIA) module that explicitly decouples high- and low-frequency information via a Laplacian Pyramid strategy. This design enables the dynamic interaction between spatial details and frequency-aware global contexts, effectively preserving the connectivity of narrow roads. Furthermore, a Channel Feature Fusion Module (CFFM) is proposed to bridge the two branches by adaptively recalibrating channel-wise feature responses, seamlessly integrating local textures with global semantics for accurate segmentation. Comprehensive experiments on the WHU-RuR+, DeepGlobe, and Massachusetts datasets validate the superiority of DSFC-Net over state-of-the-art approaches.

Figures (7)

• Figure 1: Illustrations of the three primary challenges in extracting rural roads from remote sensing imagery. (a) High intra-class variability and low inter-class separability, where diverse road surfaces blend with surrounding bare soil or fields. (b) Occlusion by vegetation, disrupting road continuity due to trees and crops. (c) Narrow roads, resulting in detection difficulties and foreground-background imbalance.
• Figure 2: Overall architecture of the proposed DSFC-Net. The structure of DSFC-Net consists of two parallel encoders, a CNN branch using ConvNeXt-v2 convnextv2 and a Transformer branch using our proposed SFT. And the CFFM is used to fuse the output feature maps of the two encoders.
• Figure 3: The structure of proposed Cross-Frequency Interaction Attention (CFIA). Left: The example of Laplacian Pyramid decomposition strategy. Right: The detail structure of CFIA, where $\mathbf{X}_{L}$ and $\mathbf{X}_{H}$ are obtained from Laplacian Pyramid decomposition strategy.
• Figure 4: Visual comparison of the different methods on the WHU-RuR+ dataset.
• Figure 5: Visual comparison of the different methods on the DeepGlobe dataset.