面向航空影像跨域语义分割的测试时自适应方法

doi:10.13474/j.cnki.11-2246.2026.0116

摘要/Abstract

摘要： 针对遥感图像目标域训练数据难以获取,且在测试阶段无法进行完整无监督域适应训练,从而导致模型精度下降的问题,本文提出了一种用于航空遥感影像跨域语义分割的测试时自适应方法。该方法基于教师-学生网络框架,引入空间一致性约束与高置信度伪标签机制,在无需源域数据与标注的情况下,实现模型在目标域上的实时持续优化。试验结果表明,在航空遥感影像跨波段场景中,本文方法在目标域上的模型IoU由27.51%提升至42.63%,性能显著优于其他对比方法。此外,与DRDG(depth-assisted resi-DualGAN)方法结合构建多阶段自适应框架后,IoU可进一步提高5%以上。本文方法在航空遥感影像跨域语义分割中具有优越的跨域适应能力,同时在复杂跨域场景中表现出良好的扩展性、有效性与稳健性。

关键词: 航空语义分割, 无监督域适应, 测试时自适应, 迁移学习

Abstract: To address the issue of limited availability of training data in the target domain for remote sensing images and the inability to perform complete unsupervised domain adaptation during testing,which leads to decreased model accuracy,this paper proposes a test-time adaptation technique for cross-domain semantic segmentation of aerial remote sensing images.Based on a teacher-student network framework,the method introduces spatial consistency constraints and a high-confidence pseudo-labeling mechanism,enabling real-time and continuous model optimization in the target domain without requiring source domain data or annotations.Experimental results demonstrate the superior cross-domain adaptation capability of the proposed method in cross-band scenarios of aerial remote sensing imagery.On the target domain,the model's IoU increased from 27.51% to 42.63%,significantly outperforming comparative methods.Furthermore,the method exhibits strong extensibility and can be integrated with the DRDG (depth-assisted resi-DualGAN) approach to construct a multi-stage adaptive framework,which further improves IoU by over 5%.The proposed method demonstrates excellent cross-domain adaptation capability for aerial remote sensing image semantic segmentation,along with strong extensibility,effectiveness,and robustness in complex cross-domain scenarios.

Key words: aviation semantic segmentation, unsupervised domain adaptation, test-time adaptation, transfer learning

中图分类号:

P237

罗惠恒, 杨磊, 文凡, 胡金艳, 王海羽, 李卓建, 高琛. 面向航空影像跨域语义分割的测试时自适应方法[J]. 测绘通报, 2026, 0(1): 100-107.

LUO Huiheng, YANG Lei, WEN Fan, HU Jinyan, WANG Haiyu, LI Zhuojian, GAO Chen. Adaptive testing method for cross domain semantic segmentation of aerial images[J]. Bulletin of Surveying and Mapping, 2026, 0(1): 100-107.

参考文献

[1] GAO Han,ZHAO Yang,GUO Peng,et al.Cycle and self-supervised consistency training for adapting semantic segmentation of aerial images[J].Remote Sensing,2022,14(7): 1527.
[2] 甘文祥,张远谊,李欣园.一种航空影像建筑物检测的轻量化CNN建模方法[J].地理空间信息,2023,21(6): 24-27.
[3] LI Yansheng,SHI Te,ZHANG Yongjun,et al.Learning deep semantic segmentation network under multiple weakly-supervised constraints for cross-domain remote sensing image semantic segmentation[J].ISPRS Journal of Photogrammetry and Remote Sensing,2021,175: 20-33.
[4] ZHANG Zhaoxiang,DOI K,IWASAKI A,et al.Unsupervised domain adaptation of high-resolution aerial images via correlation alignment and self training[J].IEEE Geoscience and Remote Sensing Letters,2021,18(4): 746-750.
[5] 刘海军,杨鸿海.基于非对称卷积网络的遥感影像地物检测方法研究[J].地理空间信息,2023,21(9): 73-76.
[6] 王子明.基于注意力机制的遥感影像道路提取方法[J].地理空间信息,2025,23(1): 13-16.
[7] 周晓忠,刘军廷,周海涛,等.基于自监督学习的多尺度差分高光谱异常检测[J].地理空间信息,2025,23(6): 37-42.
[8] 潘杰,刘波,邹筱瑜.基于特征异常检测与伪标签回归的无监督对抗域适应[J].电子学报,2025,53(1): 128-140.
[9] BOUDIAF M,DENTON T,VAN MERRIENBOER B,et al.In search for a generalizable method for source free domain adaptation[C]//Proceedings of the 40th International Conference on Machine Learning.Honolulu:ACM,2023: 2914-2931.
[10] NATH KUNDU J,VENKAT N,RAHUL M V,et al.Universal source-free domain adaptation[C]//Proceedings of 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR).Seattle: IEEE,2020: 4543-4552.
[11] YANG Shiqi,WANG Yaxing,VAN DE WEIJER J,et al.Generalized source-free domain adaptation[C]//Proceedings of 2021 IEEE/CVF International Conference on Computer Vision (ICCV).Montreal: IEEE,2022: 8958-8967.
[12] 陆霆洋,吕凡,周涛,等.持续测试时域自适应图像分类方法[J].中国图象图形学报,2025,30(8): 2660-2674.
[13] ZHAO Yang,GUO Peng,SUN Zihao,et al.ResiDualGAN: resize-residual DualGAN for cross-domain remote sensing images semantic segmentation[J].Remote Sensing,2023,15(5): 1428.
[14] WANG Qin,DAI Dengxin,HOYER L,et al.Domain adaptive semantic segmentation with self-supervised depth estimation[C]//Proceedings of 2021 IEEE/CVF International Conference on Computer Vision (ICCV).Montreal: IEEE,2022: 8495-8505.
[15] NIU Shuaicheng,WU Jiaxiang,ZHANG Yifan,et al.Towards stable test-time adaptation in dynamic wild world[EB/OL].[2025-10-25].https://arxiv.org/abs/2302.12400.
[16] SUN Yu,WANG Xiaolong,LIU Zhuang,et al.Test-time training with self-supervision for generalization under distribution shifts[C]//Proceedings of the 37th International Conference on Machine Learning.[S.l.]: ACM Press,2020: 9229-9248.
[17] ZHAO Yang,GUO Peng,GAO Han,et al.Depth-assisted ResiDualGAN for cross-domain aerial images semantic segmentation[J].IEEE Geoscience and Remote Sensing Letters,2023,20: 2500305.
[18] WANG Dequan,SHELHAMER E,LIU Shaoteng,et al.Tent: fully test-time adaptation by entropy minimization[EB/OL].[2025-10-25].https://arxiv.org/abs/2006.10726.
[19] YI Changan,CHEN Haotian,ZHANG Yifan,et al.From question to exploration: test-time adaptation in semantic segmentation[EB/OL].[2025-10-25].https://arxiv.org/abs/2310.05341.
[20] WANG Qin,FINK O,VAN GOOL L,et al.Continual test-time domain adaptation[C]//Proceedings of 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR).New Orleans: IEEE,2022: 7191-7201.
[22] ZOU Yang,YU Zhiding,VIJAYA KUMAR B V K,et al.Unsupervised domain adaptation for semantic segmentation via class-balanced self-training[M]//Computer Vision-ECCV 2018.Cham:Springer International Publishing,2018: 297-313.
[23] ARJOVSKY M,CHINTALA S,BOTTOU L.Wasserstein generative adversarial networks[C]//Proceedings of the 34th International Conference on Machine Learning.Sydney: ACM Press,2017: 214-223.
[24] SCHNEIDER S,RUSAK E,ECK L,et al.Improving robustness against common corruptions by covariate shift adaptation[J].Advances in Neural Information Processing Systems,2020,33: 11539-11551.
[25] HU E J,SHEN Yelong,WALLIS P,et al.LoRA: low-rank adaptation of large language models[EB/OL].[2025-10-25].https://arxiv.org/abs/2106.09685.