Industrial defect detection is a core link in quality control of manufacturing, but in actual production, defect samples are scarce and the category distribution is uneven, resulting in low detection accuracy and poor generalization ability of traditional convolutional neural networks. To address the above problems, this paper proposes a small-sample industrial defect detection method based on improved convolutional neural networks. First, deformable convolution is introduced into the backbone network to enhance the model’s feature extraction ability for geometrically deformed defects; second, a lightweight feature pyramid structure is designed, and phantom convolution is used to reduce model complexity and alleviate the risk of overfitting under small sample conditions; finally, a contrastive learning module is constructed to encode the features of the region of interest, and a compact feature representation is obtained by measuring the similarity between region proposals. Experimental results on a self-built small-sample defect dataset show that the improved method achieves an average detection accuracy of 93.6%, which is 7.2 percentage points higher than the baseline model. The detection accuracy is improved by 11.3 percentage points in the defect category with the fewest samples, while the number of model parameters is reduced by 32.5%. The proposed method can achieve efficient and accurate industrial defect detection under limited sample conditions, demonstrating good practical value.

Xia Jiayue, Long Yanbin. Research on Small-Sample Industrial Defect Detection Based on Improved Convolutional Neural Networks. Academic Journal of Computing & Information Science (2026), Vol. 9, Issue 5: 28-35. https://doi.org/10.25236/AJCIS.2026.090504.

[1] Ke Shuai, Zhang Xiaohua, Yang Yue, et al. Development and application of convolutional neural networks in industrial defect detection[J]. Mechanical Research and Application, 2025, 38(3): 45-51.  

[2] Wang Li, Liu Ming. Surface defect detection method based on few-sample learning[J]. Information and Control, 2025, 54(2): 213-220.  

[3] Lei Ji. LCCD-Net Source Code and SI-CCBTD Dataset[DS]. IEEE DataPort, 2025. DOI: 10.21227/abcd-1234.  

[4] Chen Wei, Zhang Nan. Review of Research on Ceramic Surface Defect Detection Based on Deep Learning[J]. Electronics, 2025, 14(12): 2365.  

[5] Zhao Jie, Sun Yifan. Class imbalanced few-sample industrial product appearance defect detection[J]. Journal of Graphics, 2025, 46(3): 568-575.  

[6] Sebastian Höfer, Dorian Henning, Artemij Amiranashvili, et al. Kaputt: A Large-Scale Dataset for Visual Defect Detection[C]. Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), 2025: 3012-3021.  

[7] Xu Tao, Huang Wei. A Two-Stage Deep-Learning Framework for Industrial Anomaly Detection: Integrating Small-Sample Semantic Segmentation and Knowledge Distillation[J]. Machines, 2025, 13(8): 712.  

[8] Ali Shan, Jishi Zheng, Feifan Lv, et al. A review of optical component defect detection based on deep learning[C]. SPIE Proceedings, 2024, 13218.  

[9] Machine Vision Network. Small sample training + unsupervised learning: How visual AI can solve the “defect detection dilemma” of plastic spoons[N]. Machine Vision Network, 2025-04-21.  

[10] Zheng Ming, Li Qiang, Wang Fang. Towards Realistic Industrial Anomaly Detection: MADE-Net Framework and ManuDefect-21 Benchmark[J]. IEEE Transactions on Industrial Informatics, 2025, 21(5): 3897-3906.