浮空器主缆绳表面的小样本学习缺陷检测研究

陈亮希; 张燕龙; 陈兴玉; 田富君; 郭磊; 周金文; 查珊珊; 黄健; 孙丙宇

doi:10.3969/j.issn.1671-7872.2022.03.011

浮空器主缆绳表面的小样本学习缺陷检测研究

1. 中国电子科技集团公司第三十八研究所, 安徽合肥 230088;
2. 安徽大学计算机科学与技术学院, 安徽合肥 230601;
3. 中国科学院合肥智能机械研究所, 安徽合肥 230031

基金项目:

国家自然科学基金项目(61773360)

详细信息

作者简介:
陈亮希(1992—)，男，湖南常德人，工程师，主要研究方向为数字孪生、机器视觉。

中图分类号: TP391.41
计量
- 文章访问数: 68
- HTML全文浏览量: 0
- PDF下载量: 4
出版历程
- 收稿日期: 2021-04-29
- 网络出版日期: 2022-09-28

Research on Surface Defect Detection of Aerostat Main Cable Based on Few-shot Metric Learning

1. The 38th Institute of China Electronics Technology Group Corporation, Hefei 230088, China;
2. School of Computer Science and Technology, Anhui University, Hefei 230601, China;
3. Institute of Intelligence Machines, Chinese Academy of Sciences, Hefei 230031, China

摘要

摘要: 针对浮空器主缆绳表面缺陷样本获取困难的特点,提出一种小样本度量学习方法来检测主缆绳的表面缺陷。小样本学习由特征编码和度量两模块组成,特征编码模块采用预训练的卷积神经网络在通用图像集上提取样本特征,度量模块用来度量与未知类别样本最相似的样本从而完成缺陷种类的分类;由于辅助数据和检测数据差异较大,将微调策略引入小样本学习方法。实验结果表明:本文方法的缺陷检测准确率达93.17%,相比传统机器学习和深度学习方法,准确率大幅提升;微调策略可进一步提升本文方法缆绳表面缺陷检测性能,准确率由93.17%提升至93.85%;在钢材缺陷分类数据集NEU上本文方法也可获得91.22%的准确率。
- 缺陷检测 /
- 小样本学习 /
- 度量学习 /
- 浮空器
Abstract: In view of the difficulty in obtaining the surface defect samples of the main cable of the aerostat, a few-shot learning method was proposed to detect the surface defects of the main cable. Few-shot metric learning consists of feature encoder module and metric module. The feature encoder module was used to extract features from a common used image dataset by a pre-training convolutional neural network (CNN), and the metric module was used to measure the samples that were the most similar to the unknown class samples, so as to complete the classification of defect. The experimental results show that the defect detection accuracy of this method is 93.17%, which is greatly improved compared with traditional machine learning and deep learning methods. The fine tuning strategy can further improve the detection performance of cable surface defects by this method, and the accuracy can be improved from 93.17% to 93.85%. On the steel defect classification dataset NEU, the accuracy of this method can also be 91.22%.
- defect detection /
- few-shot learning /
- metric learning /
- aerostat

HTML全文

参考文献(17)

[1]	SIMONYAN K, ZISSERMAN A. Very deep convolutional networks for large-scale image recognition[J/OL]. Computer Science,[2015-04-10] (2021-04-03). https://arxiv.org/abs/1409.1556?lipi=urn%253Ali%253Apage%253Ad_flag ship3_pulse_read%253BWbgfrO62TNyM82Dut7H2Ew%253D%253D.
[2]	HE K M, ZHANG X Y, REN S Q, et al. Deep residual learning for image recognition[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Las Vegas:CVPR, 2016:770-778.
[3]	SNELL J, SWERSKY K, ZEMEL R. Prototypical networks for few-shot learning[C]//201731st Conference on Neural Information Processing Systems (NIPS 2017).Long Beach, CA, USA:NIPS, 2017:4080-4090.
[4]	米春风,卢琨,汪文艳,等.基于机器视觉的热轧带钢表面缺陷检测研究进展[J].安徽工业大学学报(自然科学版),2022, 39(2):180-188.
[5]	LI J, DENG X, KAN Z, et al. Review of machine vision research in surface defect detection of agricultural products[J]. Journal of Agricultural Mechanization Research, 2013, 7:221-224.
[6]	ZHAO J, TIAN H. The applications of potato external quality detection using machine vision[J]. Journal of Graphics, 2017, 38(3):382-387.
[7]	HUANG S H, PAN Y C. Automated visual inspection in the semiconductor industry:a survey[J]. Computers in Industry, 2014, 66:1-10.
[8]	彭方进.一种高鲁棒性的钢轨表面缺陷检测算法[J].中国机械工程,2019, 30(3):266-270.
[9]	龚中良,杨张鹏,梁力,等.基于机器视觉的柑橘表面缺陷检测[J].江苏农业科学,2019, 47(7):236-239.
[10]	SONG Y, WANG H, LI Z, et al. Vision based transmission line broken strand detection and robot behaviour planning[J]. Jiqiren/robot, 2015, 37(2):204-211, 223.
[11]	孙实超,肖潇,李栋,等.利用参数关联性的输电线检测方法[J].广西大学学报(自然科学版),2015, 40(6):1461-1468.
[12]	王万国,张晶晶,韩军,等.基于无人机图像的输电线断股与异物缺陷检测方法[J].计算机应用,2015, 35(8):2404-2408.
[13]	黄新波,章小玲,张烨,等.基于径向基概率神经网络的输电导线缺陷状态识别[J].电力系统自动化,44(3):201-210.
[14]	SAIZ F A, SERRANO I, BARANDIARÁN I. et al. A robust and fast deep learning-based method for defect classification in steel surfaces[C]//2018 International Conference on Intelligent Systems (IS).Portuguese Islane Madeira:IS, 2018:455-460.
[15]	LIEN P C, ZHAO Q. Product surface defect detection based on deep learning[C]//201816th IEEE Int Conf on Dependable, Autonom and Secure Comp. Athens, Greece:IEEE, 2018:250-255.
[16]	VINYALS O, BLUNDELL C, LILLICRAP T, et al. Matching networks for one shot learning[C]//NIPS'16:Proceedings of the 30th International Conference on Neural Information Processing Systems. Barcelona, Spain:NIPS, 2016:3637-3645.
[17]	SONG K, YAN Y. A noise robust method based on completed local binary patterns for hot-rolled steel strip surface defects[J]. Applied Surface Science, 2013, 285(Part B):858-864.