在多品种、小批量智能制造生产线中,工业异常样本稀缺且产品迭代速度快,传统有监督深度学习模型面临训练数据难以获取、换线部署周期长的问题. 针对该问题,本文提出一种基于视觉-语言大模型的解耦残差提示异常检测模型.该模型在语言端采用文本初始化残差提示学习,将专家语义描述转化为可计算的决策先验,再经深度连续提示微调,解耦全局与局部异常语义.视觉端设计V-V自注意力机制,用于增强局部特征内聚性,提升微小异常的定位能力. 在MVTec AD和VisA基准数据集上,零样本条件下该方法的图像级受试者工作特征曲线下面积(AUROC)分别为91.4%和83.8%,像素级AUROC分别为91.7%和96.0%,优于现有主流方法.
[1] 苏虎,张家斌,张博豪,等.基于视觉感知的表面缺陷检测综述[J].计算机集成制造系统, 2023, 29(1): 169-191. DOI: 10.13196/j.cims.2023.01.015.
[2] Li Z, Yan Y H, Wang X H, et al. A survey of deep learning for industrial visual anomaly detection[J]. Artif Intell Rev, 2025, 58(9): 279. DOI:10.1007/s10462-025-11287-7.
[3] Mao Y, Chen Z Y, Liu Y, et al. A survey on industrial image anomaly detection: methods, benchmarks and rethinks[J]. Measurement, 2025, 256: 118377. DOI:10.1016/j.measurement.2025.118377.
[4] Shukla V, Shukla A, Surya Prakash S K, et al. A systematic survey: role of deep learning-based image anomaly detection in industrial inspection contexts[J]. Front Robot AI, 2025, 12: 1554196. DOI:10.3389/frobt.2025.1554196.
[5] Saeed U, Jan S U, Ahmad J, et al. Generative adversarial networks-enabled anomaly detection systems: a survey[J]. Expert Syst Appl, 2026, 296: 128978. DOI:10.1016/j.eswa.2025.128978.
[6] Defard T, Setkov A, Loesch A, et al. PaDiM: a patch distribution modeling framework for anomaly detection and localization[C] //Pattern Recognition. ICPR International Workshops and Challenges. ACM, 2021: 475-489. DOI:10.1007/978-3-030-68799-1_35.
[7] Roth K, Pemula L, Zepeda J, et al. Towards total recall in industrial anomaly detection[C] //2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition(CVPR). June 18-24, 2022. New Orleans, LA, USA. IEEE, 2022: 14298-14308. DOI:10.1109/cvpr52688.2022.01392.
[8] Song S. Contrastive learning-based one-class classification for intelligent manufacturing system[J]. Machines, 2025, 13(12): 1109. DOI:10.3390/machines13121109.
[9] Paolini D, Dini P, Soldaini E, et al. One-class anomaly detection for industrial applications: a comparative survey and experimental study[J]. Computers, 2025, 14(7): 281. DOI:10.3390/computers14070281.
[10] Liznerski P, Ruff L, Vandermeulen R A, et al. Explainable deep one-class classification[PP/OL]. V3. arXiv(2021-03-18)[2026-03-25]. https://doi.org/10.48550/arXiv.2007.01760.
[11] Radford A, Kim J W, Hallacy C, et al. Learning transferable visual models from natural language supervision[C] //Proceedings of the 38th International Conference on Machine Learning. 2021, 139: 8748-8763.
[12] Li X F, Zhang Z Z, Tan X, et al. PromptAD: learning prompts with only normal samples for few-shot anomaly detection[C] //2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition(CVPR). June 16-22, 2024. Seattle, WA, USA. IEEE, 2024: 16848-16858. DOI:10.1109/cvpr52733.2024.01594.
[13] Chen X H, Zhang J N, Tian G Z, et al. CLIP-AD: a language-guided staged dual-path model for zero-shot anomaly detection[PP/OL]. V2. arXiv(2024-03-02)[2026-03-25]. https://doi.org/10.48550/arXiv.2311.00453.
[14] Zhou K Y, Yang J K, Loy C C, et al. Learning to prompt for vision-language models[J]. Int J Comput Vis, 2022, 130(9): 2337-2348. DOI:10.1007/s11263-022-01653-1.
[15] Zhou Q H, Pang G S, Tian Y, et al. AnomalyCLIP: object-agnostic prompt learning for zero-shot anomaly detection[C] //The Twelfth International Conference on Learning Representations. 2023.
[16] Cao Y K, Zhang J N, Frittoli L, et al. AdaCLIP: adapting CLIP with hybrid learnable prompts for zero-shot anomaly detection[C] //Computer Vision-ECCV 2024. ACM, 2024: 55-72. DOI:10.1007/978-3-031-72761-0_4.
[17] Dosovitskiy A, Beyer L, Kolesnikov A, et al. An image is worth 16x16 words: transformers for image recognition at scale[C] //Proceedings of the 9th International Conference on Learning Representations(ICLR). 2021.
[18] Liu Z, Lin Y T, Cao Y, et al. Swin transformer: hierarchical vision transformer using shifted windows[C] //2021 IEEE/CVF International Conference on Computer Vision(ICCV). October 10-17, 2021, Montreal, QC, Canada. IEEE, 2021: 9992-10002. DOI:10.1109/ICCV48922.2021.00986.
[19] Wang W H, Xie E Z, Li X, et al. Pyramid vision transformer: a versatile backbone for dense prediction without convolutions[C] //2021 IEEE/CVF International Conference on Computer Vision(ICCV). October 10-17, 2021, Montreal, QC, Canada. IEEE, 2021: 548-558. DOI:10.1109/ICCV48922.2021.00061.
[20] Batzner K, Heckler L, König R. EfficientAD: accurate visual anomaly detection at millisecond-level latencies[C] //2024 IEEE/CVF Winter Conference on Applications of Computer Vision(WACV). January 3-8, 2024. Waikoloa, HI, USA. IEEE, 2024: 127-137. DOI:10.1109/wacv57701.2024.00020.
