结合掩码图像建模和对比学习的道路提取方法

doi:10.16182/j.issn1004731x.joss.24-1083

系统仿真学报 ›› 2025, Vol. 37 ›› Issue (4): 922-932.doi: 10.16182/j.issn1004731x.joss.24-1083

结合掩码图像建模和对比学习的道路提取方法

伍江江, 李政宏, 沙志超, 陈浩, 彭双, 杜春, 李军

国防科技大学电子科学学院，湖南长沙 410073

收稿日期:2024-09-27 修回日期:2024-11-13 出版日期:2025-04-17 发布日期:2025-04-16
通讯作者: 李政宏
第一作者简介:伍江江(1982-)，男，副教授，博士，研究方向为卫星遥感数据存储、处理与分析。
基金资助:
国家自然科学基金(42471403)

A Method for Road Extraction Using Masked Image Modeling and Contrastive Learning

Wu Jiangjiang, Li Zhenghong, Sha Zhichao, Chen Hao, Peng Shuang, Du Chun, Li Jun

College of Electronic Science and Technology, National University of Defense Technology, Changsha 410073, China

Received:2024-09-27 Revised:2024-11-13 Online:2025-04-17 Published:2025-04-16
Contact: Li Zhenghong

摘要/Abstract

摘要：

针对遥感影像道路提取的遮挡问题，提出了结合掩码图像建模和对比学习的道路提取 方法，其模型训练过程包括掩码预训练阶段和对比训练阶段。掩码预训练阶段主要进行掩码图像重建，训练模型从部分区域被随机遮挡的图像恢复出整张图像；对比训练阶段主要对于预测错误以及置信度较低的区域进行对比学习，拉近相同类别对应的特征距离而增大不同类别特征间的距离。实验结果验证了方法的有效性和可用性。

关键词: 遥感影像, 道路提取, 掩码图像建模, 对比学习, 图像重建

Abstract:

Aiming at the occlusion problem of road extraction from remote sensing images, a road extraction method combining MIM and CL is proposed, the model training process includes a masked pre-training stage and a contrast training stage. The masked pre-training stage mainly carries out mask image reconstruction, and trains the model to recover the whole image from some areas that are randomly occluded. The comparison training stage is mainly for the prediction error and low confidence regions to learn the comparison, to narrow the distance between the features of the same category and increase the distance between the features of different categories. The experimental results verify the effectiveness and usability of this method.

Key words: remote sensing image, road extraction, masked image modeling, contrastive learning, image reconstruction

中图分类号:

TP751.1

伍江江,李政宏,沙志超等 . 结合掩码图像建模和对比学习的道路提取方法[J]. 系统仿真学报, 2025, 37(4): 922-932.

Wu Jiangjiang,Li Zhenghong,Sha Zhichao,et al . A Method for Road Extraction Using Masked Image Modeling and Contrastive Learning[J]. Journal of System Simulation, 2025, 37(4): 922-932.

图/表 10

图1

图2

图3

表1

图4

表2

图5

表3

图6

图7

参考文献 21

1	Bong D B L, Lai K C, Joseph A. Automatic Road Network Recognition and Extraction for Urban Planning[J]. International Journal of Applied Science, Engineering and Technology, 2009, 5(1): 209-215.
2	Manandhar Prajowal, Prashanth Reddy Marpu, Aung Zeyar. Deep Learning Approach to Update Road Network Using VGI Data[C]//2018 International Conference on Signal Processing and Information Security (ICSPIS). Piscataway: IEEE, 2018: 1-4.
3	Sravanthi Peddinti A, Arjun Singh Chouhan, Asisa Kumar Panigrahy. Road Extraction Using Aerial Images for Future Navigation[J]. Materials Today: Proceedings, 2021, 47, Part 17: 6306-6308.
4	Kahraman I, Karas I R, Akay A E. Road Extraction Techniques from Remote Sensing Images: A Review[C]//The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Geospatial and Disaster Risk Management. Germany: Copernicus Publications 2018: 339-342.
5	Liu Pengfei, Wang Qing, Yang Gaochao, et al. Survey of Road Extraction Methods in Remote Sensing Images Based on Deep Learning[J]. PFG-Journal of Photogrammetry, Remote Sensing and Geoinformation Science, 2022, 90(2): 135-159.
6	Jaiswal A, Babu A R, Zadeh M Z, et al. A Survey on Contrastive Self-supervised Learning[J]. Technologies, 2020, 9(1): 2.
7	Xu Yingxiao, Chen Hao, Du Chun, et al. MSACon: Mining Spatial Attention-based Contextual Information for Road Extraction[J]. IEEE Transactions on Geoscience and Remote Sensing, 2021, 60: 1-17.
8	Wang Yooseung, Seo Junghoon, Jeon Taegyun. NL-LinkNet: Toward Lighter but More Accurate Road Extraction with Nonlocal Operations[J]. IEEE Geoscience and Remote Sensing Letters, 2022, 19: 1-5.
9	Mei Jie, Li Roujing, Gao Wang, et al. CoANet: Connectivity Attention Network for Road Extraction from Satellite Imagery[J]. IEEE Transactions on Image Processing, 2021, 30: 8540-8552.
10	Li Ruirui, Gao Bochuan, Xu Qizhi. Gated Auxiliary Edge Detection Task for Road Extraction with Weight-balanced Loss[J]. IEEE Geoscience and Remote Sensing Letters, 2021, 18(5): 786-790.
11	Chen Hao, Li Zhenghong, Wu Jiangjiang, et al. 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.
12	Ding Lei, Bruzzone Lorenzo. DiResNet: Direction-aware Residual Network for Road Extraction in VHR Remote Sensing Images[J]. IEEE Transactions on Geoscience and Remote Sensing, 2021, 59(12): 10243-10254.
13	Zhang Chaoning, Zhang Chenshuang, Song Junha, et al. A Survey on Masked Autoencoder for Self-supervised Learning in Vision and Beyond[EB/OL]. (2022-07-30) [2024-02-24]. .
14	Li Siyuan, Wu Di, Wu Fang, et al. Architecture-agnostic Masked Image Modeling—From ViT back to CNN[EB/OL]. (2023-06-02) [2024-02-24]. .
15	Demir I, Koperski K, Lindenbaum D, et al. DeepGlobe 2018: A Challenge to Parse the Earth Through Satellite Images[C]//2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW). Piscataway: IEEE, 2018: 172-181.
16	Mnih V. Machine Learning for Aerial Image Labeling[M]. Toronto: University of Toronto, 2013.
17	Chen L C, Zhu Yukun, Papandreou G, et al. Encoder-decoder with Atrous Separable Convolution for Semantic Image Segmentation[C]//Computer Vision-ECCV 2018. Cham: Springer International Publishing, 2018: 833-851.
18	Zhou Lichen, Zhang Chuang, Wu Ming. D-LinkNet: LinkNet with Pretrained Encoder and Dilated Convolution for High Resolution Satellite Imagery Road Extraction[C]//2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW). Piscataway: IEEE, 2018: 192-1924.
19	Chaurasia A, Culurciello E. LinkNet: Exploiting Encoder Representations for Efficient Semantic Segmentation[C]//2017 IEEE Visual Communications and Image Processing (VCIP). Piscataway: IEEE, 2017: 1-4.
20	Chen Sibao, Ji Yuxin, Tang Jin, et al. DBRANet: Road Extraction by Dual-branch Encoder and Regional Attention Decoder[J]. IEEE Geoscience and Remote Sensing Letters, 2022, 19: 1-5.
21	Yang Zhigang, Zhou Daoxiang, Yang Ying, et al. Road Extraction from Satellite Imagery by Road Context and Full-stage Feature[J]. IEEE Geoscience and Remote Sensing Letters, 2023, 20: 1-5.

方法	P	R	F₁	IoU
DeepLab v3+	0.793 5	0.738 2	0.764 9	0.619 2
D-LinkNet	0.745 3	0.773 2	0.759 0	0.611 6
NL-LinkNet	0.739 4	0.779 9	0.759 1	0.611 8
RoadExNet	0.738 6	0.783 8	0.760 5	0.613 6
DBRANet	0.747 6	0.780 7	0.763 8	0.617 8
RCFSNet	0.765 7	0.772 4	0.769 0	0.624 7
本文	0.768 4	0.803 0	0.790 8	0.654 0

方法	P	R	F₁	IoU
DeepLab v3+	0.839 1	0.711 0	0.769 7	0.625 7
D-LinkNet	0.791 4	0.768 1	0.779 6	0.638 7
NL-LinkNet	0.804 5	0.758 1	0.780 6	0.640 2
RoadExNet	0.803 8	0.759 8	0.781 2	0.640 9
DBRANet	0.788 3	0.779 5	0.783 9	0.644 6
RCFSNet	0.792 5	0.775 9	0.784 1	0.644 9
本文	0.786 5	0.785 9	0.786 2	0.647 7

数据集	方法	P	R	F₁	IoU
DeepGlobe Road Extraction	Without MIM&CL	0.761 2	0.783 9	0.772 4	0.629 2
	Without MIM	0.758 4	0.789 4	0.773 6	0.630 8
	Without CL	0.766 1	0.782 4	0.774 2	0.631 6
	Ours	0.768 4	0.803 0	0.790 8	0.654 0
Massachusetts Road	Without MIM&CL	0.798 1	0.755 2	0.776 1	0.634 1
	Without MIM	0.798 6	0.765 8	0.781 9	0.641 9
	Without CL	0.803 1	0.761 8	0.781 9	0.641 9
	Ours	0.786 5	0.785 9	0.786 2	0.647 7

[1]	孙红, 张玉香, 凌岳览. 基于损失提取反馈注意网络的图像超分辨率重建研究[J]. 系统仿真学报, 2023, 35(2): 308-317.
[2]	王鑫, 徐明君, 肖坚, 徐立中. 基于融合视觉词袋的高分遥感水体提取算法[J]. 系统仿真学报, 2022, 34(5): 1033-1043.
[3]	张立峰, 王会忍. 基于卷积神经网络及有限元仿真的电容层析成像图像重建[J]. 系统仿真学报, 2022, 34(4): 712-718.
[4]	仝卫国, 曾世超, 张立峰, 侯哲, 郭佳跃. 基于深度残差神经网络的电阻层析成像及流型辨识方法[J]. 系统仿真学报, 2022, 34(09): 2028-2036.
[5]	仝卫国, 庞雪纯, 朱赓宏. 基于卷积神经网络的气液两相流流型识别方法[J]. 系统仿真学报, 2021, 33(4): 883-891.
[6]	陈宇, 夏宗基, 周雨佳. 基于修正稀疏拟牛顿的电容层析成像重建算法[J]. 系统仿真学报, 2019, 31(5): 819-827.
[7]	马敏, 郭琪, 闫超奇. 基于广义正则化的ECT图像重建算法[J]. 系统仿真学报, 2017, 29(8): 1851-1858.
[8]	夏惊涛, 王群书, 马继明, 黑东炜, 盛亮, 魏福利, 宋顾周. 多层球壳构件FBP重建差异化滤波设计与仿真[J]. 系统仿真学报, 2016, 28(12): 3061-3066.
[9]	刘彬, 陈向宁, 郭连朋. 基于无人机遥感影像的三维战场构建技术研究[J]. 系统仿真学报, 2015, 27(6): 1268-1273.

结合掩码图像建模和对比学习的道路提取方法

A Method for Road Extraction Using Masked Image Modeling and Contrastive Learning

RichHTML

PDF (PC)

可视化

摘要/Abstract

引用本文

使用本文

图/表 10

参考文献 21

相关文章 9

编辑推荐

Metrics

本文评价