Aug 2026· Applied Sciences· 0 citations· 33 references
TL;DR
An adaptive scale-aware road extraction network, termed ASAR-Net, which jointly improves multi-scale feature representation and structural continuity and effectively improves both the semantic completeness and structural continuity of extracted road networks is proposed.
Abstract
Accurate road extraction from high-resolution remote sensing imagery plays a vital role in numerous geospatial applications, including urban planning, disaster emergency response, intelligent transportation, and map updating. However, significant variations in road width, geometry, and orientation, together with complex backgrounds such as shadows, vegetation, and occlusions, often lead to incomplete extraction and poor structural continuity. To address these challenges, this paper proposes an adaptive scale-aware road extraction network, termed ASAR-Net, which jointly improves multi-scale feature representation and structural continuity. Specifically, an Adaptive Bidirectional Enhancement Module (ABEM) is introduced in the encoder to improve the representation of roads with diverse spatial scales through adaptive scale-aware convolution and bidirectional attention. Furthermore, a Directional Fusion Module (DFM) is incorporated into the decoder to guide feature reconstruction along road orientations using dynamic snake convolution, thereby facilitating the recovery of continuous and complete road structures. Extensive experiments on two public benchmark datasets, Massachusetts Roads and DeepGlobe, demonstrate that ASAR-Net consistently outperforms several representative state-of-the-art road extraction methods in terms of mIoU and F1-score. The proposed network effectively improves both the semantic completeness and structural continuity of extracted road networks, demonstrating its robustness and effectiveness for road extraction in complex high-resolution remote sensing scenarios.
Accurate extraction of road networks from high-resolution remote sensing imagery is a fundamental task underpinning autonomous-driving navigation, urban spatial planning, and the dynamic updating of geographic information databases. Although existing road extraction methods attain outstanding pixel-level segmentation accuracy and topological integrity, most follow an accuracy-first design paradigm that relies on heavyweight backbones and increasingly complex decoders, incurring a parameter volume and storage overhead that constitute the principal bottleneck for deploying them on resource-constrained edge platforms such as unmanned aerial vehicles, mobile terminals, and onboard satellite processors. Conversely, models that pursue extreme lightweighting often fail to preserve the thin, continuous, linear structure of roads, tending to produce topological breaks in the extracted road networks. To bridge the performance gap between segmentation accuracy and model size, we propose LOA-Net, a lightweight orientation-aware road extraction network. LOA-Net introduces a Road-Aligned Deformable Convolution (RA-DCN) that adaptively aligns the sampling region with the road geometry and explicitly supervises the predicted road orientation, thereby accurately capturing road connectivity while substantially reducing the parameter count. Experiments on the CHN6-CUG and DeepGlobe benchmarks show that LOA-Net surpasses representative state-of-the-art methods on both IoU and F1, while achieving the lowest parameter count of all compared models and a computational complexity comparable to its peers, striking an excellent trade-off between segmentation performance and a mobile-friendly footprint that makes it well suited for road extraction from remote sensing imagery in resource-constrained scenarios.
Bo Huang, Yiwei Lu, Zizhuo Li et al.· Remote Sensing· 0 citations
Building extraction from high-resolution remote sensing imagery is critical for urban planning and smart city development, yet it faces challenges such as blurred boundaries, missing fine details, and severe background interference. To address these issues, this study proposes an improved model named GSU-HRNet, which integrates attention mechanisms and boundary refinement strategies on the basis of UHRNet's high-resolution parallel backbone. An enhanced Pyramid Squeeze-and-Excitation (PSE) module is embedded in the lateral feature transmission paths of each hierarchical stage, capturing multi-scale contextual information via adaptive average pooling of multiple sizes to strengthen semantic responses for buildings and suppress background noise. A Gated Bottleneck Convolution (GBC) module is further introduced in the feature fusion stage, adopting a dual-branch structure with gating mechanisms and residual connections to selectively regulate fused features, alleviate redundant feature accumulation, and improve the stability of feature representation. Experiments were conducted on the aerial imagery subset of the WHU Building Dataset (covering 450 km2 in Christchurch with 8,189 512×512 image tiles), which was split into training, validation and test sets at a ratio of 6:1:3. Ablation experiments verify the effectiveness and complementarity of PSE and GBC modules, with the combined model achieving optimal performance. Quantitative comparisons show that GSU-HRNet outperforms classical models like U-Net and PSPNet, reaching an IoU of 89.93% and an F1-score of 94.50%. Qualitative analysis demonstrates that the proposed model yields clearer building boundaries, more complete structural preservation, and reduced false detections and omissions, even in challenging scenarios with complex building structures and shadow interference. The results confirm that GS-UHRNet effectively enhances feature representation and boundary delineation accuracy, and exhibits strong generalization ability across different building extraction datasets, providing a robust solution for automated building extraction from hig-hresolution remote sensing imagery.
Shi He, Shiye Zhang, Xiujuan Liang et al.· GEOINFORMATICS· 0 citations
Abstract. Road network is an important infrastructure of urban spatial structure and traffic system. Its accurate acquisition is of great significance for urban traffic analysis, automatic driving map construction and disaster emergency response. With the wide acquisition of high-resolution remote sensing images, automatic extraction of road masks from remote sensing images has become an important research direction in the field of remote sensing image understanding. However, the existing deep learning methods still face the problems of obvious modal differences and insufficient modeling of road structure continuity in remote sensing scenes. To solve the above problems, this paper proposes a remote sensing image road segmentation model LR-SAM based on SAM (Segment Anything Model). In this model, the LoRA (Low Rank Adaptation) fine-tuning strategy is introduced to achieve efficient parameter updating, and the MS multi-scale feature interaction module is designed in the coding phase to enhance the expression ability of the linear structure and fine-grained information of the road. At the same time, the original prompt encoder is removed and a lightweight AD decoder is constructed to achieve multi-scale feature fusion. In the reasoning stage, TTA (Test Time Augmentation) strategy is introduced to improve the stability and segmentation accuracy of the model. Experimental results based on CHN6-CUG and SATMTB datasets show that the proposed method achieves 97.20% and 85.06% mIoU and 96.67% and 84.94% F1-score, respectively, which is significantly better than the mainstream road segmentation methods, and verifies the effectiveness of the proposed improvement points.
Bingquan Yao, Haolin Liu, Wenjuan Mao et al.· The International Archives o...· 0 citations
To address the challenges of vegetation interference, background confusion, and road fragmentation caused by the narrow and elongated structures of forest roads in complex remote sensing imagery, this study proposed an improved U-Net-based model, namely HAA-UNet, for automatic forest road extraction. The proposed model integrates a VGG16 encoder, an Atrous Spatial Pyramid Pooling (ASPP) module, and a Hybrid Dilated Convolution (HDCC) module to enhance feature representation and spatial detail reconstruction of road targets under complex forest environments. Experimental results demonstrated that HAA-UNet achieved superior segmentation performance on the forest road dataset of Xichang City, Sichuan Province, with Precision, Recall, F1-score, and mIoU values of 87.24%, 87.83%, 87.53%, and 79.85%, respectively, outperforming all comparison models. These results indicate that the proposed method effectively improves road continuity and boundary delineation in complex forest scenes. Furthermore, the extracted road information was integrated into the Forest Fire Risk Index (FFRI) assessment framework, demonstrating that accurate road data can improve the spatial characterization of fire risk and provide reliable data support for forest fire risk assessment and forest resource management.
Road-network extraction from very high-resolution (VHR) remote-sensing imagery remains a challenging task owing to the structural sparsity, topological complexity, and severe occlusions of road networks. Conventional graph-based approaches preserve topological consistency yet incur considerable computational overhead, whereas prevailing convolutional neural network (CNN) and Transformer architectures struggle to reconcile long-range contextual modeling with computational efficiency. To address these limitations, this study proposes RFM-UNet, a hybrid frequency and state–space network designed for road-network segmentation. Specifically, the encoder integrates Mamba blocks with an Anisotropic Directional Attention (ADA) module to jointly capture local geometric cues and global dependencies at linear computational complexity. In addition, a Multi-Scale Adaptive Fusion Module (MAFM) is introduced to dynamically recalibrate multi-stage features, thereby suppressing cross-scale interference and preserving the connectivity of narrow roads. To enhance robustness against shadow-induced occlusions, a Dual-Spectrum Aggregation Module (DualSpec) decouples the phase and amplitude spectra in the frequency domain and fuses them with spatial features, effectively mitigating spurious responses and background noise characterized by similar textures. Quantitative and qualitative experiments on three public datasets demonstrate that RFM-UNet consistently outperforms current state-of-the-art methods.
Pu Song, Peng Yu, Xiaojing Zhong et al.· Remote Sensing· 0 citations
Timely and reliable mapping of landslide-affected areas from high-spatial-resolution optical imagery is essential for disaster investigation and post-event assessment. However, this task remains challenging because landslides usually exhibit large-scale variations, irregular boundaries, and strong spectral–textural similarities with surrounding bare-surface objects, which often cause missed detections, false positives, incomplete delineation, and inaccurate boundary localization. To address these problems, this paper presents a Scale-View Interactive Attention Network, named SIA-Net, for RGB-based landslide segmentation. First, a Multi-Scale Attention Module (MSAM) is constructed to encourage information exchange among features with different spatial resolutions. By doing so, the network can better represent both small scattered landslide patches and large continuous landslide bodies. Second, a Multi-View Attention Module (MVAM) is introduced to aggregate contextual cues from multiple receptive field views. This design strengthens the model’s ability to distinguish landslides from visually confusing objects, including bare soil, roads, riverbanks, and terrain shadows. In addition, a Convolutional Block Attention Module (CBAM) is incorporated during feature reconstruction to enhance landslide-related channel and spatial responses, thereby improving segmentation completeness and boundary localization. Experiments on the CAS Landslide Dataset (CLD) and GVLM Dataset show that SIA-Net provides more accurate landslide masks than the compared segmentation networks under the adopted benchmark settings. These results indicate that integrating scale-level interaction, view-level contextual modeling, and attention-guided decoding can effectively improve landslide extraction in complex optical remote sensing scenes.
