[1]杨立然,张小龙,赵振兵,等.基于原型引导与双重对比学习的无源域自适应[J].智能系统学报,2026,21(5):1130-1141.[doi:10.11992/tis.202508033]
　YANG Liran,ZHANG Xiaolong,ZHAO Zhenbing,et al.Prototype guidance and dual-level contrastive learning for source-free domain adaptation[J].CAAI transactions on intelligent systems,2026,21(5):1130-1141.[doi:10.11992/tis.202508033]
点击复制

基于原型引导与双重对比学习的无源域自适应

《智能系统学报》[ISSN 1673-4785/CN 23-1538/TP] 卷: 21 期数: 2026年第5期 页码: 1130-1141 栏目: 学术论文—机器学习 出版日期: 2026-09-05
Title:
Prototype guidance and dual-level contrastive learning for source-free domain adaptation
作者:
杨立然1,2,3, 张小龙2, 赵振兵1,4, 翟永杰1,2, 苏攀2,5
1. 华北电力大学 燕赵电力实验室, 河北 保定 071003;
2. 华北电力大学 控制与计算机工程学院, 河北 保定 071003;
3. 华北电力大学 复杂能源系统智能计算教育部工程研究中心, 河北 保定 071003;
4. 华北电力大学 电气与电子工程学院, 河北 保定 071003;
5. 华北电力大学 河北省能源电力知识计算重点实验室, 河北 保定 071003
Author(s):
YANG Liran1,2,3, ZHANG Xiaolong2, ZHAO Zhenbing1,4, ZHAI Yongjie1,2, SU Pan2,5
1. Yanzhao Electric Power Laboratory, North China Electric Power University, Baoding 071003, China;
2. School of Control and Computer Engineering, North China Electric Power University, Baoding 071003, China;
3. Engineering Research Center of Intelligent Computing for Complex Energy Systems of Ministry of Education, North China Electric Power University, Baoding 071003, China;
4. School of Electrical and Electronic Engineering, North China Electric Power University, Baoding 071003, China;
5. Hebei Key Laboratory of Knowledge Computing for Energy and Power, North China Electric Power University, Baoding 071003, China
关键词:
无源域自适应; 原型引导; 对比学习; 伪标签; 特征融合; 原型正交; 自训练; 特征对齐
Keywords:
source-free domain adaptation; prototype guidance; contrastive learning; pseudo-label; feature fusion; prototype orthogonality; self-training; feature alignment
分类号:
TP391
DOI:
10.11992/tis.202508033
摘要:
无源域自适应(source-free domain adaptation, SFDA)旨在源域数据不可访问的情况下，实现预训练模型从源域到目标域的有效迁移。现有SFDA方法大多采用自训练学习策略，但其易受噪声伪标签的影响，模型难以准确捕捉数据关键特征，导致性能下降。为此，本文提出一种基于原型引导与双重对比学习的无源域自适应方法(prototype guidance and dual-level contrastive learning, PDCL)。PDCL基于源模型提取源域各类特征，构建目标域初始类别级原型表示。通过特征融合策略对齐源域各类别特征与目标域样本特征，生成目标域伪标签。引入原型正交损失，通过最大化类间原型的正交性以增强判别性。PDCL采用双重对比学习策略：在决策空间进行实例级对比，拉近近邻样本并推远异类样本；在特征空间进行原型级对比，增强特征空间的结构一致性。在Office-31、Office-Home及DomainNet-126基准数据集上的实验结果表明，PDCL在SFDA中相较现有先进方法实现了性能提升，验证了其有效性。
Abstract:
Source-free domain adaptation (SFDA) aims to enable effective knowledge transfer of pre-trained models from the source domain to the target domain when the source samples are inaccessible. Most SFDA methods adopt self-training learning, but they are susceptible to noisy pseudo-labels, making it difficult for models to accurately capture intrinsic features and leading to performance degradation. To address this issue, we propose a novel SFDA method named prototype guidance and dual-level contrastive learning (PDCL). Specifically, PDCL first extracts features from the source domain using the pre-trained source model to construct initial class-level prototype representations for the target domain. Subsequently, it aligns the features of source and target samples via a feature fusion strategy, thereby generating pseudo-labels for the target domain. Secondly, it introduces a prototype orthogonality loss to enhance discriminability by maximizing the orthogonality between class prototypes. Finally, PDCL employs a dual-level contrastive learning strategy, where instance-level contrast is conducted in the decision space to pull near neighbors closer and push outliers away, while prototype-level contrast is conducted in the feature space to enhance structural consistency. Experimental results on Office-31, Office-Home, and DomainNet-126 datasets show that PDCL outperforms state-of-the-art methods, which validates its effectiveness.
参考文献/References:
[1] DONG Shi, WANG Ping, ABBAS K. A survey on deep learning and its applications[J]. Computer science review, 2021, 40: 100379
[2] LIU Xiaofeng, YOO C, XING Fangxu, et al. Deep unsupervised domain adaptation: a review of recent advances and perspectives[J]. APSIPA transactions on signal and information processing, 2022, 11(1): 1-51
[3] 李庆勇, 何军, 张春晓. 基于分类差异与信息熵对抗的无监督域适应算法[J]. 智能系统学报, 2021, 16(6): 999-1006 LI Qingyong, HE Jun, ZHANG Chunxiao. Unsupervised domain adaptation algorithm based on classification discrepancy and information entropy[J]. CAAI transactions on intelligent systems, 2021, 16(6): 999-1006
[4] LI Jingjing, YU Zhiqi, DU Zhekai, et al. A comprehensive survey on source-free domain adaptation[J]. IEEE transactions on pattern analysis and machine intelligence, 2024, 46(8): 5743-5762
[5] LI Rui, JIAO Qianfen, CAO Wenming, et al. Model adaptation: unsupervised domain adaptation without source data[C]//2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Seattle: IEEE, 2020: 9638-9647.
[6] LIANG Jian, HU Dapeng, FENG Jiashi. Do we really need to access the source data? source hypothesis transfer for unsupervised domain adaptation[C]// Proceedings of International Conference on Machine Learning. New York: PMLR, 2020: 6028-6039.
[7] LONG Mingsheng, CAO Zhangjie, WANG Jianmin, et al. Conditional adversarial domain adaptation[C]//Proceedings of the 32nd International Conference on Neural Information Processing Systems. Montréal: ACM, 2018: 1647-1657.
[8] LONG Mingsheng, CAO Yue, CAO Zhangjie, et al. Transferable representation learning with deep adaptation networks[J]. IEEE transactions on pattern analysis and machine intelligence, 2019, 41(12): 3071-3085
[9] 曹艺, 郭茂祖, 吴伟宁. 融合扩增技术的无监督域适应方法[J]. 软件学报, 2025, 36(7): 3253-3270 CAO Yi, GUO Maozu, WU Weining. Unsupervised domain adaptation method with augmentation technology[J]. Journal of software, 2025, 36(7): 3253-3270
[10] HOU Yunzhong, ZHENG Liang. Visualizing adapted knowledge in domain transfer[C]//2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Nashville: IEEE, 2021: 13819-13828.
[11] DING Yuhe, SHENG Lijun, LIANG Jian, et al. ProxyMix: proxy-based mixup training with label refinery for source-free domain adaptation[J]. Neural networks, 2023, 167: 92-103
[12] 王帆, 韩忠义, 尹义龙. 伪标签不确定性估计的源域无关鲁棒域自适应[J]. 软件学报, 2022, 33(4): 1183-1199 WANG Fan, HAN Zhongyi, YIN Yilong. Source free robust domain adaptation based on pseudo label uncertainty estimation[J]. Journal of software, 2022, 33(4): 1183-1199
[13] ZHOU Lihua, LI Nianxin, YE Mao, et al. Source-free domain adaptation with class prototype discovery[J]. Pattern recognition, 2024, 145: 109974
[14] JUI S, VAN DE WEIJER J, WANG Kai, et al. Attracting and dispersing: a simple approach for source-free domain adaptation[C]//Advances in Neural Information Processing Systems 35. New Orleans: NeurIPS, 2022: 5802-5815.
[15] WANG Rui, WU Zuxuan, WENG Zejia, et al. Cross-domain contrastive learning for unsupervised domain adaptation[J]. IEEE transactions on multimedia, 2023, 25: 1665-1673
[16] CHHABRA S, VENKATESWARA H, LI Baoxin. Generative alignment of posterior probabilities for source-free domain adaptation[C]//2023 IEEE/CVF Winter Conference on Applications of Computer Vision. Waikoloa: IEEE, 2023: 4114-4123.
[17] 赵磊, 张慧铭, 邢卫, 等. 基于自相似与对比学习的图像跨域转换算法[J]. 计算机研究与发展, 2023, 60(4): 930-946 ZHAO Lei, ZHANG Huiming, XING Wei, et al. Image cross-domain translation algorithm based on self-similarity and contrastive learning[J]. Journal of computer research and development, 2023, 60(4): 930-946
[18] QIU Zhen, ZHANG Yifan, LIN Hongbin, et al. Source-free domain adaptation via avatar prototype generation and adaptation[C]//Proceedings of the Thirtieth International Joint Conference on Artificial Intelligence. Montreal: IJCAI, 2021: 2921-2927.
[19] HUANG Jiaxing, GUAN Dayan, XIAO Aoran, et al. Model adaptation: historical contrastive learning for unsupervised domain adaptation without source data[C]//Proceedings of the 35th International Conference on Neural Information Processing Systems. Vancouver: Curran Associates Inc. , 2021: 3635-3649.
[20] ZHAO Xuejun, STANISLAWSKI R, GARDONI P, et al. Adaptive contrastive learning with label consistency for source data free unsupervised domain adaptation[J]. Sensors, 2022, 22(11): 4238
[21] XIA Haifeng, ZHAO Handong, DING Zhengming. Adaptive adversarial network for source-free domain adaptation[C]//2021 IEEE/CVF International Conference on Computer Vision. Montreal: IEEE, 2022: 8990-8999.
[22] M?LLER R, KORNBLITH S, HINTON G. When does label smoothing help? [EB/OL]. (2019-06-06)[2025-08-28]. https://arxiv.org/abs/1906.02629.
[23] ZHANG Yixin, WANG Zilei, HE Weinan. Class relationship embedded learning for source-free unsupervised domain adaptation[C]//2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Vancouver: IEEE, 2023: 7619-7629.
[24] SAENKO K, KULIS B, FRITZ M, et al. Adapting visual category models to new domains[C]//Computer Vision – ECCV 2010. Berlin: Springer, 2010: 213-226.
[25] VENKATESWARA H, EUSEBIO J, CHAKRABORTY S, et al. Deep hashing network for unsupervised domain adaptation[C]//2017 IEEE Conference on Computer Vision and Pattern Recognition. Honolulu: IEEE, 2017: 5385-5394.
[26] PENG Xingchao, BAI Qinxun, XIA Xide, et al. Moment matching for multi-source domain adaptation[C]//2019 IEEE/CVF International Conference on Computer Vision. Seoul: IEEE, 2019: 1406-1415.
[27] LITRICO M, DEL BUE A, MORERIO P. Guiding pseudo-labels with uncertainty estimation for source-free unsupervised domain adaptation[C]//2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Vancouver: IEEE, 2023: 7640-7650.
[28] ZHANG Yuchen, LIU Tianle, LONG Mingsheng, et al. Bridging theory and algorithm for domain adaptation[C]//Proceedings of the 36th International Conference on Machine Learning. Long Beach: PMLR, 2019: 7404-7413.
[29] JIN Ying, WANG Ximei, LONG Mingsheng, et al. Minimum class confusion for versatile domain adaptation[C]//Computer Vision–ECCV 2020. Cham: Springer, 2020: 464-480.
[30] KARIM N, MITHUN N C, RAJVANSHI A, et al. C-SFDA: a curriculum learning aided self-training framework for efficient source free domain adaptation[C]//2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Vancouver: IEEE, 2023: 24120-24131.
[31] XING Bowei, YING Xianghua, WANG Ruibin. Rectifying self-training with neighborhood consistency and proximity for source-free domain adaptation[J]. Neurocomputing, 2024, 606: 128425
[32] CHEN Dian, WANG Dequan, DARRELL T, et al. Contrastive test-time adaptation[C]//2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New Orleans: IEEE, 2022: 295-305.
[33] PENG Lin, HE Yuhang, WANG Shaokun, et al. Global self-sustaining and local inheritance for source-free unsupervised domain adaptation[J]. Pattern recognition, 2024, 155: 110679
[34] HE Kaiming, ZHANG Xiangyu, REN Shaoqing, et al. Deep residual learning for image recognition[C]//2016 IEEE Conference on Computer Vision and Pattern Recognition. Las Vegas: IEEE, 2016: 770-778.
[35] DENG Jia, DONG Wei, SOCHER R, et al. ImageNet: a large-scale hierarchical image database[C]//2009 IEEE Conference on Computer Vision and Pattern Recognition. Miami: IEEE, 2009: 248-255.
[36] BOTTOU L. Stochastic gradient descent tricks[M]//Neural Networks: Tricks of the Trade. Berlin: Springer, 2012: 421-436.
相似文献/References:
[1]杨宇宇,杨霄,潘在宇,等.基于原型引导与自适应特征融合的域适应语义分割[J].智能系统学报,2025,20(1):150.[doi:10.11992/tis.202403010]
　YANG Yuyu,YANG Xiao,PAN Zaiyu,et al.Domain adaptive semantic segmentation based on prototype-guided and adaptive feature fusion[J].CAAI transactions on intelligent systems,2025,20():150.[doi:10.11992/tis.202403010]

备注/Memo

收稿日期:2025-8-28。
基金项目:国家自然科学基金联合基金项目重点支持项目(U21A20486)；河北省自然科学基金项目(F2024502002)；中央高校基本科研业务费专项资金项目(2024MS129).
作者简介:杨立然，讲师，博士，主要研究方向为计算机视觉和迁移学习，主持河北省自然科学基金项目1项、中央高校基本科研业务费专项资金项目2项。发表学术论文10余篇。E-mail：yangliran@ncepu.edu.cn。;张小龙，硕士研究生，主要研究方向为迁移学习和无源域自适应。E-mail：zhangxiaolong@ncepu.edu.cn。;苏攀，副教授，主要研究方向为知识工程、模糊系统、电力视觉。主持国家自然科学基金项目、宁波市“科技创新2025”重大专项等项目。发表学术论文50余篇。E-mail：supan@ncepu.edu.cn。
通讯作者:苏攀. E-mail：supan@ncepu.edu.cn

更新日期/Last Update: 2026-09-05
Copyright © 《 智能系统学报》 编辑部
地址：（150001）黑龙江省哈尔滨市南岗区南通大街145-1号楼 电话：0451- 82534001、82518134　邮箱：tis@vip.sina.com