Road extraction method for heterogeneous data using sparse labels

doi:10.11947/j.AGCS.2026.20250230

Abstract

Abstract:

Labelled data are essential for road extraction from optical images; however, creating high-quality labels is labor-and time-intensive. Moreover, the transferability of network models across different regions, sensors, and imaging times is limited, restricting their broader application in spatio-temporal contexts. To address this issue, we propose a road extraction method for heterogeneous data using sparse labels that combines optical imagery with OpenStreetMap (OSM) data. Sparse road labels are generated through raster processing and coordinate alignment with OSM vector data. Then, the segment anything model (SAM) and simple linear iterative clustering (SLIC) are integrated to extract multi-level image features, thereby facilitating label dissemination through object-level processing for initial optimization. Finally, a network model was trained using both optical images and rough optimization results, and it refined the label accuracy via image-label association mapping and was further optimized with OSM data as a buffer. Experimental validation using the RoadNet and Oklahoma datasets in conjunction with the four semantic segmentation networks UNet, D-LinkNet, MANet and UNetFormer demonstrated that our proposed method outperforms existing methods in terms of both quantitative accuracy and performance, especially in challenging areas of road extraction.

Key words: OSM data, optical image, weak supervision, sparse sample, road extraction

CLC Number:

P237

Yuzhun LIN, Shuxiang WANG, Jie RUI, Fei JIN, Jianfang JIANG, Xibing ZUO, Xiao LIU, Yujie ZOU. Road extraction method for heterogeneous data using sparse labels[J]. Acta Geodaetica et Cartographica Sinica, 2026, 55(5): 881-893.

Add to citation manager EndNote|Reference Manager|ProCite|BibTeX|RefWorks

URL: http://xb.chinasmp.com/EN/10.11947/j.AGCS.2026.20250230

http://xb.chinasmp.com/EN/Y2026/V55/I5/881

Figures/Tables 8

Fig. 1

Fig. 2

Tab. 1

Tab. 2

Fig. 3

Fig. 4

Fig. 5

Fig. 6

References 35

[1]	ACHANTA R, SHAJI A, SMITH K, et al. SLIC superpixels compared to state-of-the-art superpixel methods[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2012, 34(11): 2274-2282.
[2]	LENINISHA S, VANI K. Water flow based geometric active deformable model for road network[J]. ISPRS Journal of Photogrammetry and Remote Sensing, 2015, 102: 140-147.
[3]	RONNEBERGER O, FISCHER P, BROX T. U-Net: convolutional networks for biomedical image segmentation[C]//Proceedings of 2015 Medical Image Computing and Computer-Assisted Intervention. Berlin: Springer, 2015: 234-241.
[4]	BADRINARAYANAN V, KENDALL A, CIPOLLA R. SegNet: a deep convolutional encoder-decoder architecture for image segmentation[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 39(12): 2481-2495.
[5]	CHENG Mingming, GUO Menghao, HOU Qibin, et al. SegNeXt: rethinking convolutional attention design for semantic segmentation[C]//Proceedings of 2022 Advances in Neural Information Processing Systems. New Orleans: Neural Information Processing Systems Foundation, Inc., 2022: 1140-1156.
[6]	ZHOU Lichen, ZHANG Chuang, WU Ming. D-LinkNet: LinkNet with pretrained encoder and dilated convolution for high resolution satellite imagery road extraction[C]//Proceedings of 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops. Salt Lake City: IEEE, 2018: 192-1924.
[7]	CHEN Xin, YU Anzhu, SUN Qun, et al. Updating road maps at city scale with remote sensed images and existing vector maps[J]. IEEE Transactions on Geoscience and Remote Sensing, 2024, 62: 5616521.
[8]	SUN Tao, DI Zonglin, CHE Pengyu, et al. Leveraging crowdsourced GPS data for road extraction from aerial imagery[C]//Proceedings of 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Long Beach: IEEE, 2020: 7501-7510.
[9]	BATRA A, SINGH S, PANG Guan, et al. Improved road connectivity by joint learning of orientation and segmentation[C]//Proceedings of 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Long Beach: IEEE, 2020: 10377-10385.
[10]	LI Xingang, WANG Yuebin, ZHANG Liqiang, et al. Topology-enhanced urban road extraction via a geographic feature-enhanced network[J]. IEEE Transactions on Geoscience and Remote Sensing, 2020, 58(12): 8819-8830.
[11]	XU Hongzhang, HE Hongjie, ZHANG Ying, et al. A comparative study of loss functions for road segmentation in remotely sensed road datasets[J]. International Journal of Applied Earth Observation and Geoinformation, 2023, 116: 103159.
[12]	LIU Yunyu, YUAN Jinpeng. ERSNet: lightweight attention-guided network for remote sensing scene image classification[J]. Journal of Geodesy and Geoinformation Science, 2025, 8(1): 30-46.
[13]	CHEN Zhaozheng, SUN Qianru. Extracting class activation maps from non-discriminative features as well[C]//Proceedings of 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Vancouver: IEEE, 2023: 3135-3144.
[14]	OH Y, KIM B, HAM B. Background-aware pooling and noise-aware loss for weakly-supervised semantic segmentation[C]//Proceedings of 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Nashville: IEEE, 2021: 6909-6918.
[15]	XU Jingshan, ZHOU Chuanwei, CUI Zhen, et al. Scribble-supervised semantic segmentation inference[C]//Proceedings of 2021 IEEE/CVF International Conference on Computer Vision. Montreal: IEEE, 2022: 15334-15343.
[16]	XU Lian, OUYANG Wanli, BENNAMOUN M, et al. Multi-class Token Transformer for weakly supervised semantic segmentation[C]//Proceedings of 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New Orleans: IEEE, 2022: 4300-4309.
[17]	CHEN Tao, YAO Yazhou, ZHANG Lei, et al. Saliency guided inter-and intra-class relation constraints for weakly supervised semantic segmentation[EB/OL]. [2025-06-01]. https://arxiv.org/abs/2206.09554.
[18]	ZHANG Fei, GU Chaochen, ZHANG Chenyue, et al. Complementary patch for weakly supervised semantic segmentation[C]//Proceedings of 2021 IEEE/CVF International Conference on Computer Vision. Montreal: IEEE, 2022: 7222-7231.
[19]	DAI Jifeng, HE Kaiming, SUN Jian. BoxSup: exploiting bounding boxes to supervise convolutional networks for semantic segmentation[C]//Proceedings of 2015 IEEE International Conference on Computer Vision. Santiago: IEEE, 2016: 1635-1643.
[20]	KULHARIA V, CHANDRA S, AGRAWAL A, et al. Box2Seg: attention weighted loss and discriminative feature learning for weakly supervised segmentation[C]//Proceedings of 2020 European Conference on Computer Vision. Berlin: Springer_Verlag, 2020: 290-308.
[21]	LIN Di, DAI Jifeng, JIA Jiaya, et al. ScribbleSup: scribble-supervised convolutional networks for semantic segmentation[C]//Proceedings of 2016 IEEE Conference on Computer Vision and Pattern Recognition. Las Vegas: IEEE, 2016: 3159-3167.
[22]	LIANG Zhiyuan, WANG Tiancai, ZHANG Xiangyu, et al. Tree energy loss: towards sparsely annotated semantic segmentation[C]//Proceedings of 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New Orleans: IEEE, 2022: 16886-16895.
[23]	KIRILLOV A, MINTUN E, RAVI N, et al. Segment anything[C]//Proceedings of 2023 IEEE/CVF International Conference on Computer Vision. Paris: IEEE, 2024: 3992-4003.
[24]	DING Lei, ZHU Kun, PENG Daifeng, et al. Adapting segment anything model for change detection in VHR remote sensing images[J]. IEEE Transactions on Geoscience and Remote Sensing, 2024, 62: 5611711.
[25]	LIU Yahui, YAO Jian, LU Xiaohu, et al. RoadNet: learning to comprehensively analyze road networks in complex urban scenes from high-resolution remotely sensed images[J]. IEEE Transactions on Geoscience and Remote Sensing, 2019, 57(4): 2043-2056.
[26]	MURPHY K P. Machine learning: a probabilistic perspective[M]. Cambridge: MIT Press, 2012.
[27]	MILLETARI F, NAVAB N, AHMADI S A. V-Net: fully convolutional neural networks for volumetric medical image segmentation[C]//Proceedings of 2016 International Conference on 3D Vision. Stanford: IEEE, 2016: 565-571.
[28]	LIN Yuzhun, JIN Fei, WANG Dandi, et al. Dual-task network for road extraction from high-resolution remote sensing images[J]. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2023, 16: 66-78.
[29]	KINGMA D, BA J. Adam: a method for stochastic optimization[C]//Proceedings of 2015 International Conference on Learning Representations. San Diego: ICLR, 2015.
[30]	WEI Yao, JI Shunping. Scribble-based weakly supervised deep learning for road surface extraction from remote sensing images[J]. IEEE Transactions on Geoscience and Remote Sensing, 2022, 60: 5602312.
[31]	WU Songbing, DU Chun, CHEN Hao, et al. Road extraction from very high resolution images using weakly labeled OpenStreetMap centerline[J]. ISPRS International Journal of Geo-Information, 2019, 8(11): 478.
[32]	LI Rui, ZHENG Shunyi, ZHANG Ce, et al. Multiattention network for semantic segmentation of fine-resolution remote sensing images[J]. IEEE Transactions on Geoscience and Remote Sensing, 2022, 60: 5607713.
[33]	WANG Libo, LI Rui, ZHANG Ce, et al. UNetFormer: a UNet-like transformer for efficient semantic segmentation of remote sensing urban scene imagery[J]. ISPRS Journal of Photogrammetry and Remote Sensing, 2022, 190: 196-214.
[34]	DEMIR I, KOPERSKI K, LINDENBAUM D, et al. DeepGlobe 2018: a challenge to parse the Earth through satellite images[C]//Proceedings of 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops. Salt Lake City: IEEE, 2018: 172-181.
[35]	ZHU Qiqi, ZHANG Yanan, WANG Lizeng, et al. A global context-aware and batch-independent network for road extraction from VHR satellite imagery[J]. ISPRS Journal of Photogrammetry and Remote Sensing, 2021, 175: 353-365.

网络模型	标签优化方法	P↑	R↑	F₁值↑	IoU↑	Com↑	Eor↓
UNet	ScRoadExtract	77.96	91.55	84.22	72.74	85.37	8.24
	WeaklyOSM	81.13	89.92	85.30	74.37	84.91	5.61
	本文方法	86.57	90.10	88.30	79.05	82.99	4.57
D-LinkNet	ScRoadExtract	78.85	92.75	85.24	74.28	86.87	7.73
	WeaklyOSM	81.32	90.66	85.74	75.04	85.66	5.99
	本文方法	96.31	80.74	87.84	78.32	81.28	1.94
MANet	ScRoadExtract	80.98	92.09	86.18	75.72	86.14	6.16
	WeaklyOSM	82.44	90.52	86.29	75.89	85.16	4.36
	本文方法	97.46	80.08	87.91	78.43	81.31	1.13
UNetFormer	ScRoadExtract	80.13	92.59	85.91	75.30	87.08	6.15
	WeaklyOSM	82.02	91.02	86.28	75.88	85.59	4.31
	本文方法	96.88	81.67	88.62	79.57	81.15	2.40

网络模型	标签优化方法	P↑	R↑	F₁值↑	IoU↑	Com↑	Eor↓
UNet	ScRoadExtract	25.59	72.96	37.89	23.37	70.40	42.42
	WeaklyOSM	24.67	66.98	36.06	22.00	65.60	46.51
	本文方法	64.72	58.83	61.64	44.55	54.57	29.03
D-LinkNet	ScRoadExtract	26.66	80.12	40.00	25.00	77.62	40.32
	WeaklyOSM	24.08	73.88	36.32	22.19	72.13	42.51
	本文方法	62.76	61.23	61.98	44.91	54.32	19.68
MANet	ScRoadExtract	28.21	89.15	42.86	27.27	83.38	37.22
	WeaklyOSM	24.25	90.92	38.29	23.68	85.86	42.89
	本文方法	59.65	72.26	65.35	48.53	65.38	21.47
UNetFormer	ScRoadExtract	28.09	83.94	42.09	26.65	78.90	35.98
	WeaklyOSM	25.81	84.98	39.59	24.68	79.55	38.72
	本文方法	60.73	69.70	64.90	48.04	62.36	23.32