Few-shot defect recognition in optical components with deep learning

Shao Yanhua; Xin Chen; Chu Hongyu

doi:10.11884/HPLPB202537.250066

Article Contents

Article Navigation > High Power Laser and Particle Beams > 2025 > Accepted Manuscript

Shao Yanhua, Xin Chen, Chu Hongyu. Few-shot defect recognition in optical components with deep learning[J]. High Power Laser and Particle Beams. doi: 10.11884/HPLPB202537.250066

Citation:

Shao Yanhua, Xin Chen, Chu Hongyu. Few-shot defect recognition in optical components with deep learning[J]. High Power Laser and Particle Beams. doi: 10.11884/HPLPB202537.250066

Shao Yanhua, Xin Chen, Chu Hongyu. Few-shot defect recognition in optical components with deep learning[J]. High Power Laser and Particle Beams. doi: 10.11884/HPLPB202537.250066

Citation:

Shao Yanhua, Xin Chen, Chu Hongyu. Few-shot defect recognition in optical components with deep learning[J]. High Power Laser and Particle Beams. doi: 10.11884/HPLPB202537.250066

PDF( 1459 KB)

Few-shot defect recognition in optical components with deep learning

doi: 10.11884/HPLPB202537.250066

School of Information and Control Engineering, Southwest University of Science and Technology, Mianyang 621900, China

Received Date: 2025-04-07
Accepted Date: 2025-08-22
Rev Recd Date: 2025-09-05

Available Online: 2025-09-12

Abstract

Abstract

Surface defects on optical components in high-power solid-state laser systems seriously impair the system’s operational stability and laser output performance. However, precise detection of such defects under few-shot conditions remains a critical challenge, as limited training data often restricts the generalization ability of detection models and creates an urgent need for high-performance defect detection methods adapted to this scenario. To address this issue, this study aims to design and propose an enhanced detection method dubbed ICFNetV2, which is developed based on the existing ICFNet. Its core goal is to improve the accuracy and generalization of optical component surface defect detection under few-shot scenarios. ICFNetV2 integrates data augmentation techniques with deep residual networks: its framework adopts a synergistic design of residual connection mechanisms and decoupled channel convolution operations to construct a 34-layer cascaded network—this structure mitigates gradient decay during deep network training and enhances cross-layer feature transmission efficiency. The network also incorporates spatial dropout layers and implements a data preprocessing pipeline encompassing random rotation, mirror flipping, and Gaussian noise injection, which expands the training dataset to 9 times its original size. Additionally, ablation studies were conducted to verify the efficacy of each individual network module. Experimental results demonstrate that the optimized ICFNetV2 achieves a classification accuracy of 97.4% for three typical defect types, representing a 0.7 percentage point improvement over the baseline ICFNet model. In conclusion, ICFNetV2 effectively enhances defect detection performance under few-shot conditions through architectural optimization and data augmentation. The validation from ablation studies and the observed accuracy gains confirm the effectiveness of its key modules, providing a reliable solution for surface defect detection of optical components in high-power solid-state laser systems and offering reference value for similar few-shot detection tasks.
- precision optical elements,
- deep learning,
- ResNet,
- imperfections classification,
- few-shot,
- Dropout

FullText(HTML)

References(20)

References

[1]	陈彦舟. 激光惯性约束聚变发展态势研究[C]//中国核科学技术进展报告(第八卷)中国核学会2023年学术年会论文集第7册核科技情报研究同位素. 2023: 35-41 Chen Yanzhou. Research on development trends of laser inertial confinement fusion[C]//Proceedings of the Progress Report on China Nuclear Science & Technology (Vol. 8). 2023: 35-41
[2]	史伟. 高精度洁净度检测方法研究[D]. 成都: 四川大学, 2000: 9-15 Shi Wei. Research on hign precision purification testing methods[D]. Chengdu: Sichuan University, 2000: 9-15
[3]	楚红雨. 基于机器视觉的高功率激光装置光学元件表面缺陷检测技术研究[D]. 重庆: 重庆大学, 2011.
[4]	Stowers I F. Optical cleanliness specifications and cleanliness verification[J]. Proceedings of SPIE, 1999, 3782(1): 525-530.
[5]	陈达, 章轩, 赵圣斌, 等. 基于深度学习的激光熔覆层表面缺陷识别研究[J]. 光学学报, 2025, 45: 0915001 doi: 10.3788/AOS250462 Chen Da, Zhang Xuan, Zhao Shengbin, et al. Surface defect recognition of laser cladding layer based on deep learning[J]. Acta Optica Sinica, 2025, 45: 0915001 doi: 10.3788/AOS250462
[6]	刘斌, 王孝坤, 程强, 等. 复杂曲面光学元件高精度面形检测技术概述[J]. 南通大学学报(自然科学版), 2024, 23(1): 1-27 Liu Bin, Wang Xiaokun, Cheng Qiang, et al. Overview of high precision surface measurement for optical element with complex curved surface[J]. Journal of Nantong University (Natural Science Edition), 2024, 23(1): 1-27
[7]	邵延华, 冯玉沛, 张晓强, 等. 基于深度学习的光学元件表面疵病识别[J]. 强激光与粒子束, 2022, 34: 112002 doi: 10.11884/HPLPB202234.220023 Shao Yanhua, Feng Yupei, Zhang Xiaoqiang, et al. Using deep learning for surface defects identification of optical components[J]. High Power Laser and Particle Beams, 2022, 34: 112002 doi: 10.11884/HPLPB202234.220023
[8]	冯虎, 宋克臣, 崔文琦, 等. 基于元学习的带钢表面缺陷小样本语义分割[J]. 东北大学学报(自然科学版), 2024, 45(3): 354-360 Feng Hu, Song Kechen, Cui Wenqi, et al. Few-shot semantic segmentation of strip steel surface defects based on meta-learning[J]. Journal of Northeastern University (Natural Science), 2024, 45(3): 354-360
[9]	范勇, 陈念年, 高玲玲, 等. 大口径精密表面疵病的数字化检测系统[J]. 强激光与粒子束, 2009, 21(7): 1032-1036 Fan Yong, Chen Niannian, Gao Lingling, et al. Digital detection system of surface defects for large aperture optical elements[J]. High Power Laser and Particle Beams, 2009, 21(7): 1032-1036
[10]	Peng Hanyang, Yu Yue, Yu Shiqi. Re-thinking the effectiveness of batch normalization and beyond[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2024, 46(1): 465-478. doi: 10.1109/TPAMI.2023.3319005
[11]	Oh J, Kim H, Baik S, et al. Batch normalization tells you which filter is important[C]//Proceedings of the 2022 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV). 2022: 3351-3360.
[12]	Dubey S R, Singh S K, Chaudhuri B B. Activation functions in deep learning: a comprehensive survey and benchmark[J]. Neurocomputing, 2021, 503: 92-108.
[13]	Srivastava N, Hinton G, Krizhevsky A, et al. Dropout: a simple way to prevent neural networks from overfitting[J]. Journal of Machine Learning Research, 2014, 15(1): 1929-1958.
[14]	He Kaiming, Zhang Xiangyu, Ren Shaoqing, et al. Deep residual learning for image recognition[C]//Proceedings of the 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). 2016: 770-778.
[15]	朱小龙, 王翊, 谢志江, 等. 精密光学元件表面洁净度成像检测系统[J]. 西南交通大学学报, 2009, 44(6): 958-962 Zhu Xiaolong, Wang Yi, Xie Zhijiang, et al. Surface cleanliness level detection by imaging method for precision optical elements[J]. Journal of Southwest Jiaotong University, 2009, 44(6): 958-962
[16]	赵春溢, 郭洪涛, 郭涛, 等. 一种风机叶片图像采集及缺陷检测系统[J]. 红外技术, 2020, 42(12): 1203-1210 Zhao Chunyi, Guo Hongtao, Guo Tao, et al. Defect detection system based on UAV images for wind turbine blades[J]. Infrared Technology, 2020, 42(12): 1203-1210
[17]	黄俊, 张娜娜, 章惠. 基于优化LeNet-5的近红外图像中的静默活体人脸检测[J]. 红外技术, 2021, 43(9): 845-851 Huang Jun, Zhang Na’na, Zhang Hui. Silent live face detection in near-infrared images based on optimized LeNet-5[J]. Infrared Technology, 2021, 43(9): 845-851
[18]	Lecun Y, Bottou L, Bengio Y, et al. Gradient-based learning applied to document recognition[J]. Proceedings of the IEEE, 1998, 86(11): 2278-2324. doi: 10.1109/5.726791
[19]	Krizhevsky A, Sutskever I, Hinton G E. ImageNet classification with deep convolutional neural networks[C]//Proceedings of the 26th International Conference on Neural Information Processing Systems. 2012: 1097-1105.
[20]	Simonyan K, Zisserman A. Very deep convolutional networks for large-scale image recognition[DB/OL]. arXiv preprint arXiv: 1409.1556, 2014.