A review of research progress in deep learning methods for road extraction from remote sensing images
Wei Jiang
City University Malaysia/Jiangxi College of Applied Technology
Shamsul Ariffin
Sultan Idris Education University
Yanpeng Li
Jiangxi College of Applied Technology
DOI: https://doi.org/10.59429/esta.v12i3.11349
Keywords: deep learning; remote sensing imagery; road extraction; semi-supervised learning; transformer
Abstract
Driven by the combination of high-resolution remote sensing and deep learning, road extraction research has advanced along three main lines (2023-2025): semi- and weakly supervised learning, CNN-Transformer hybrid architectures, and graph/vector modeling. Semi-supervision has proven effective in robustly improving road connectivity in annotation-scarce scenarios. Hybrid architectures that fuse local texture and global context are gradually replacing single CNN/Transformer architectures in feature representation. Direct prediction of road network graphs (nodes and edges) has significantly improved topological consistency and achieved breakthroughs in large-scale inference efficiency. This paper reviews representative methods and datasets/evaluation systems, and provides a horizontal comparison based on publicly available comparison results from the past two years. It summarizes current bottlenecks and trends: Learning paradigms focused on connectivity, cross-domain generalization, and zero-shot geographic transfer, along with more economical supervisory signals, will be the primary areas of focus.
References
[1] Chen H, Li Z, Wu J, Xiong W, Du C. SemiRoadExNet: A semi-supervised network for road extraction from remote sensing imagery via adversarial learning[J]. ISPRS Journal of Photogrammetry and Remote Sensing, 2023, 198: 169–183.
[2] Zhu X, Huang X, Cao W, et al. Road Extraction from Remote Sensing Imagery with Spatial Attention Based on Swin Transformer[J]. Remote Sensing, 2024, 16(7): 1183.
[3] Mo S, Makungu D, Nhamo L. A Survey of Deep Learning Road Extraction Algorithms for High-Resolution Remote Sensing Images[J]. Sensors, 2024, 24(1): 109.
[4] Hetang C, Xue H, Le C, et al. Segment Anything Model for Road Network Graph Extraction[C]//CVPR 2024 Workshops. 2024: 2556–2566.
[5] Zhao Y, Chen Z, Zhao Z, et al. A Deeply Supervised Vertex Network for Road Network Graph Extraction in High-Resolution Images[J]. International Journal of Applied Earth Observation and Geoinformation, 2024, 133.
[6] Yu H, Zhou J, Li Q, et al. Improved Road Extraction Models through Semi-Supervised Learning[J]. ISPRS International Journal of Geo-Information, 2024, 13(10): 347.
[7] Liu W, Gao S, Zhang C, Yang B. RoadCT: A Hybrid CNN-Transformer Network for Road Extraction From Satellite Imagery[J]. IEEE Geoscience and Remote Sensing Letters, 2024, 21: 2501805.
[8] Li Z, Wang Y, Huang S, et al. RemainNet: Explore Road Extraction from Remote Sensing Images Based on Mask Image Modeling[J]. Remote Sensing, 2023, 15(17): 4215.
[9] Yang H, Zhang Y, Zhou Y, et al. A High-Resolution Remote Sensing Road Extraction Method Based on Global Spatial Learning and Fourier Frequency Domain Learning[J]. Remote Sensing, 2024, 16(20): 3896.
[10] Bai X, Guo L, Huo H, et al. Rse-Net: Road-shape Enhanced Neural Network for Road Extraction in High-Resolution Remote Sensing Image[J]. International Journal of Remote Sensing, 2023, 44(??): 1–22.
[11] Zhang J, Hu X, Wei Y, Zhang L. Road Topology Extraction From Satellite Imagery by Joint Learning of Nodes and Their Connectivity[J]. IEEE Transactions on Geoscience and Remote Sensing, 2023, 61: 5602613.
[12] Luo Z, Zhou K, Tan Y, et al. AD-RoadNet: An Auxiliary-Decoding Road Extraction Network Improving Connectivity While Preserving Multiscale Road Details[J]. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2023, 16: 8049–8062.