陈仁祥,范想佳,杨黎霞,张翔,王磊,张怀超

• 1:重庆交通大学交通工程应用机器人重庆市工程实验室
• 2:重庆科技大学经济与金融学院

摘要(Abstract)：

针对钢轨表面缺陷检测在跨域场景下精度显著下降的问题，提出基于隐式风格离散对抗学习(ISDA)的“结构-风格”双分离协同优化框架。ISDA通过特征解耦机制，将输入显式映射到域不变的结构子空间与域相关的风格子空间，实现跨域变化因素的正交隔离。构建多模态隐式风格增强机制，融合特征统计插值、归一化噪声与频域扰动，从多粒度对风格表示进行随机建模，以提升源域风格多样性与目标域流形覆盖能力。同时，引入离散对抗机制，对结构特征进行向量量化，在离散空间中生成结构保持但语义对立的对抗样本，从而提升模型的判别能力与结构鲁棒性。将扰动风格向量与离散结构特征融合解码生成图像，并与原始样本共同用于分割训练，形成“风格扰动-结构对抗-分割优化”的闭环协同学习策略。推理阶段仅依托结构解耦与特征融合模块即可实现高效分割，无需额外域自适应。实验结果表明，ISDA在5个钢轨缺陷数据集的跨域检测任务中，相较于最优全监督模型，在未见域上平均交并比提升11.06%,并在与主流域泛化方法的对比中表现出显著性能优势。

关键词(KeyWords)： 钢轨表面缺陷;特征解耦;隐式风格增强;离散对抗;跨域检测

Abstract:

Keywords:

基金项目(Foundation): 国家自然科学基金(52475548);; 重庆市自然科学基金创新发展联合基金(CSTB2025NSCQ-LZX0113);; 道路与铁道工程安全保障省部共建教育部重点实验室开放课题(STDTKF202204);; 重庆市研究生联合培养基地项目(JDLHPYJD2024006)

作者(Author): 陈仁祥,范想佳,杨黎霞,张翔,王磊,张怀超

参考文献(References)：

• [1] Nguyen S D,Tran T S,Tran V P,et al.Deep learning-based crack detection:a survey [J].International Journal of Pavement Research and Technology,2023,16(4):943-967.
• [2] 陈仁祥，潘升，杨黎霞，等.基于注意力引导多尺度降噪卷积神经网络的钢轨表面缺陷图像降噪 [J].铁道学报，2024,46(5):123-131.Chen Renxiang,Pan Sheng,Yang Lixia,et al.Noise reduction of rail surface defect images based on attention-guided poly-scale denoising convolutional neural networks [J].Journal of the China Railway Society,2024,46(5):123-131.
• [3] 赵宏伟，郑嘉俊，赵鑫欣，等.基于双模态深度学习的钢轨表面缺陷检测方法 [J].计算机工程与应用，2023,59(7):285-293.Zhao Hongwei,Zheng Jiajun,Zhao Xinxin,et al.Rail surface defect method based on bimodal-modal deep learning [J].Computer Engineering and Applications,2023,59(7):285-293.
• [4] Wang J P,Xu G L,Yan F J,et al.Defect transformer:an efficient hybrid transformer architecture for surface defect detection [J].Measurement,2023,211:112614.
• [5] Pan S J,Yang Q.A survey on transfer learning [J].IEEE Transactions on Knowledge and Data Engineering,2010,22(10):1345-1359.
• [6] Blanchard G,Lee G,Scott C.Generalizing from several related classification tasks to a new unlabeled sample [C]//Proceedings of the 25th International Conference on Neural Information Processing Systems.New York:ACM,2011:2178-2186.
• [7] Ma S,Song K C,Niu M H,et al.Cross-scale fusion and domain adversarial network for generalizable rail surface defect segmentation on unseen datasets [J].Journal of Intelligent Manufacturing,2024,35(1):367-386.
• [8] 闵永智，李嘉峰，王果.基于改进DCGAN的钢轨表面缺陷图像扩充方法 [J].铁道学报，2023,45(12):123-130.Min Yongzhi,Li Jiafeng,Wang Guo.Image expansion method for rail surface defects based on improved DCGAN [J].Journal of the China Railway Society,2023,45(12):123-130.
• [9] Ballas A,Diou C.CycleMix:mixing source domains for domain generalization in style-dependent data [C]//Proceedings of the 13th Hellenic Conference on Artificial Intelligence.New York:ACM,2024:1-4.
• [10] Zhang Z X,Zhang L M.Unsupervised pixel-level detection of rail surface defects using multistep domain adaptation [J].IEEE Transactions on Systems,Man,and Cybernetics:Systems,2023,53(9):5784-5795.
• [11] Zhou K Y,Yang Y X,Qiao Y,et al.MixStyle neural networks for domain generalization and adaptation [J].International Journal of Computer Vision,2024,132(3):822-836.
• [12] Ma S,Song K C,Niu M H,et al.Feature-based domain disentanglement and randomization:a generalized framework for rail surface defect segmentation in unseen scenarios [J].Advanced Engineering Informatics,2024,59:102274.
• [13] Nam H,Lee H,Park J,et al.Reducing domain gap by reducing style bias [PP/OL].arXiv(2019-10-25)[2026-01-21].https://arxiv.org/abs/1910.11645.
• [14] Mao X F,Chen Y F,Duan R J,et al.Enhance the visual representation via discrete adversarial training [C]//Advances in Neural Information Processing Systems 35.New Orleans:NeurIPS,2022:7520-7533.
• [15] Van Den Oord A,Vinyals O,Kavukcuoglu K.Neural discrete representation learning [PP/OL].arXiv (2018-05-30)[2026-01-21].https://arxiv.org/pdf/1711.00937.
• [16] Yu H M,Li Q Y,Tan Y Q,et al.A coarse-to-fine model for rail surface defect detection [J].IEEE Transactions on Instrumentation and Measurement,2019,68(3):656-666.
• [17] Niu M H,Wang Y Y,Song K C,et al.An adaptive pyramid graph and variation residual-based anomaly detection network for rail surface defects [J].IEEE Transactions on Instrumentation and Measurement,2021,70:5020013.
• [18] Song G R,Song K C,Yan Y H.Saliency detection for strip steel surface defects using multiple constraints and improved texture features [J].Optics and Lasers in Engineering,2020,128:106000.
• [19] Niu H W,Xie L H,Lin J H,et al.Exploring semantic consistency and style diversity for domain generalized semantic segmentation [C]//Proceedings of the Thirty-Ninth AAAI Conference on Artificial Intelligence and Thirty-Seventh Conference on Innovative Applications of Artificial Intelligence and Fifteenth Symposium on Educational Advances in Artificial Intelligence.New York:ACM,2025:6245-6253.
• [20] Gao Y F,Xia W,Hu D D,et al.DeSAM:decoupled segment anything model for Generalizable medical image segmentation[C]//Medical Image Computing and Computer Assisted Intervention-MICCAI 2024.Cham:Springer Nature Switzerland,2024:509-519.
• [21] Choi S,Jung S,Yun H,et al.RobustNet:improving domain generalization in urban-scene segmentation via instance selective whitening [C]//2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition.Piscataway:IEEE,2021:11575-11585.