基于机器视觉的碳纤维预浸料表面缺陷检测方法

doi:10.13475/j.fzxb.20190502107

摘要/Abstract

摘要：

针对碳纤维预浸料表面缺陷人工检测方法效率低、成本高、实时性差等问题,提出基于机器视觉的碳纤维预浸料表面缺陷自动检测方法。首先,在碳纤维预浸料生产线上,采用2台高分辨率线扫描相机快速连续采集图像,从中随机选择带有缺陷的图像1 000张;其次,基于大气光散射模型对图像进行去雾增强处理,以消除白色树脂的干扰;然后,改进具有19个卷积层和5个最大值池化层的YOLOv2目标检测算法,用于缺陷的检测;最后,对预处理后的图像进行网络训练提取图像特征,识别图像目标,并对训练好的网络进行实验验证。结果表明:该方法在复杂的工业环境下,具有较高的识别精度和鲁棒性,识别成功率达到94%以上,且每张图像的检测时间不超过 0.1 s,可满足工业生产中精度和实时性要求。

关键词: 机器视觉, 碳纤维预浸料, 表面缺陷检测, 图像预处理, YOLOv2算法

Abstract:

Aiming at low efficiency, high cost and poor real-time of artificial detection of surface defects of carbon fiber prepregs, an automatic detection method based on machine vision was proposed. Two high resolution line scanning cameras were used to collect images quickly and continuously in the carbon fiber production line, from which 1 000 images with defects were randomly selected. After that, the image enhancement algorithm based on the atmospheric light scattering model was used to pre-process the images to eliminate the interference of white resin. The YOLOv2 object detection network was refined with 19 convolution layers and 5 maximum pooling layers for improvement in detect detection. Finally, the pre-processed images were trained, image features were extracted, image objects were identified, and the trained network was verified. The experimental results show that the proposed method has high accuracy and robustness under complex industrial environment, the recognition success rate in this research is over 94%, and the detection time of each image is less than 0.1 s, meeting the requirements of precision and real-time in industrial production.

Key words: machine vision, carbon fiber prepreg, surface defect detection, image pre-procession, YOLOv2 algorithm

中图分类号:

TP391

路浩, 陈原. 基于机器视觉的碳纤维预浸料表面缺陷检测方法[J]. 纺织学报, 2020, 41(04): 51-57.

LU Hao, CHEN Yuan. Surface defect detection method of carbon fiber prepreg based on machine vision[J]. Journal of Textile Research, 2020, 41(04): 51-57.

图/表 12

图1

图2

图3

图4

表1

图5

图6

图7

表2

表3

图8

表4

参考文献 18

[1]	高奇. 新形势下我国碳纤维产业发展探究[J]. 合成纤维工业, 2019,42(5):1-7.
	GAO Qi. Research on the development of China's carbon fiber industry in the new situation[J]. China Synthetic Fiber Industry, 2019,42(5):1-7.
[2]	顾佳杰. 碳纤维产业研究发展报告[J]. 上海化工, 2019,44(3):32-36.
	GU Jiajie. Carbon fiber industry development report[J]. Shanghai Chemical, 2019,44(3):32-36.
[3]	杨玉娥, 张文习. 碳纤维复合材料的无损检测综述[J]. 济南大学学报(自然科学版), 2015,29(6):471-476.
	YANG Yu'e, ZHANG Wenxi. Summary of nondestructive testing of carbon fiber composites[J]. Journal of University of Jinan (Science and Technology), 2015,29(6):471-476.
[4]	倪金辉. 基于机器视觉的预浸纱缺陷检测系统的研究[D]. 南京: 南京航空航天大学, 2015: 1-5.
	NI Jinhui. Defect detection system for prepreg based on machine vision[D]. Nanjing: Nanjing University of Aeronautics and Astronautics, 2015: 1-5.
[5]	李文羽, 程隆棣. 基于机器视觉和图像处理的织物疵点检测研究新进展[J]. 纺织学报, 2014,35(3):158-164.
	LI Wenyu, CHENG Longdi. New progress in fabric defect detection based on machine vision and image processing[J]. Journal of Textile Reaearch, 2014,35(3):158-164.
[6]	张浩. 基于深度学习的表面缺陷检测方法研究[D]. 苏州: 苏州大学, 2018: 1-3.
	ZHANG Hao. Study on surface defect detection method based on deep learning[D]. Suzhou: Soochow University, 2018: 1-3.
[7]	REDMON Joseph, DIVVALA Santosh, GIRSHICK Ross, et al. You look only once: unified, real-time objection detection [C]//2016 IEEE conference on computer vision and pattern recognition(CVPR). Las Vegas: IEEE, 2016: 779-788.
[8]	REDMON Joseph, FARHADI Ali. YOLO9000: better, faster, stronger [C]//2017 IEEE conference on computer vision and pattern recognition(CVPR). Venice: IEEE, 2017: 1-9.
[9]	REN Shaoqing, HE Kaiming, GIRSHICK Ross. Faster R-CNN: towards real-time object detection with region proposal networks[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017,39(6):1137-1149.
[10]	牟新刚, 蔡逸超, 周晓, 等. 基于机器视觉的筒子纱缺陷在线检测系统[J]. 纺织学报, 2018,39(1):139-145.
	MOU Xingang, CAI Yichao, ZHOU Xiao, et al. Online yarn cone defects detection system based on machine vision[J]. Journal of Textile Reasearch, 2018,39(1):139-145.
[11]	HE Kaiming, SUN Jian, TANG Xiaoou. Single image haze removal using dark channel prior [C]//2009 IEEE conference on computer vision and pattern recogni-tion(CVPR). Miami: IEEE, 2009: 1956-1963.
[12]	WANG Xuelong, GAO Ying, DONG Junyu. Surface defects detection of paper dish based on Mask R-CNN [C]//Third international workshop on pattern recognition(IWPR). Jinan: Proceeding of Spie, 2018: 1-6.
[13]	LIN Min, CHEN Qiang, YAN Shuichen. Network in network [C]//International conference on learning representations. Banff: ICLR, 2014: 1-10.
[14]	IOFFE Sergey, SZEGEDY Christian. Batch normalization: accelerating deep network training by reducing internal covariate shift [C]//International conference on macine learning(ICML). Lille: IMLS, 2015: 448-456.
[15]	CHA Youngjin, CHOI Wooram. Deep learning based crack damage detection using convolutional neural networks[J]. Computer-Aided Civil and Infrastucture Engineering, 2017,32:361-378.
[16]	LENG Jianwei, GUO Kai. Research on monocular visual gesture positioning based on YOLOv2[C]//Proceeding of the 37th chinese control conference. Shanghai: Shanghai Systems Science Press, 2018: 9101-9106.
[17]	ZHANG Hongwei, ZHANG Lingjie, LI Pengfei, et al. Yarn-dyed fabric defect detection with YOLOv2 based on deep convolution neural networks [C]//Data driven control and learning systems conference. Enshi: IEEE, 2018: 170-174.
[18]	何晓昀, 韦平, 张林. 基于深度学习的籽棉中异性纤维检测方法[J]. 纺织学报, 2018,39(6):131-135.
	HE Xiaoyun, WEI Ping, ZHANG Lin. Detection method of foreign fibers in seed cotton based on deep-learning[J]. Journal of Textile Reasearch, 2018,39(6):131-135.

相关文章 15

[1]	朱世根, 杨宏贤, 白云峰, 丁浩, 朱巧莲. 长条状细薄带钩零件变形自动检测系统[J]. 纺织学报, 2020, 41(10): 158-163.
[2]	张建新, 李琦. 基于机器视觉的筒子纱密度在线检测系统[J]. 纺织学报, 2020, 41(06): 141-146.
[3]	王文胜, 李天剑, 冉宇辰, 卢影, 黄民. 筒子纱纱笼纱杆的定位检测方法[J]. 纺织学报, 2020, 41(03): 160-167.
[4]	金守峰, 林强强, 马秋瑞, 张浩. 基于BP 神经网络的织物表面绒毛质量的检测方法[J]. 纺织学报, 2020, 41(02): 69-76.
[5]	景军锋, 张君扬, 张缓缓, 苏泽斌. 梯度空间下的丝饼表面缺陷检测[J]. 纺织学报, 2020, 41(02): 44-51.
[6]	孙卫红, 阮棉奖, 邵铁锋, 梁曼. 基于机器视觉的生丝抱合性能检测方法[J]. 纺织学报, 2019, 40(08): 164-168.
[7]	景军锋, 张星星. 基于机器视觉的玻璃纤维管纱毛羽检测[J]. 纺织学报, 2019, 40(05): 157-162.
[8]	徐洋, 朱治潮, 盛晓伟, 余智祺, 孙以泽. 基于机器视觉的鞋面特征点自动识别改进方法[J]. 纺织学报, 2019, 40(03): 168-174.
[9]	景军锋, 郭根. 基于机器视觉的丝饼毛羽检测[J]. 纺织学报, 2019, 40(01): 147-152.
[10]	牟新刚蔡逸超周晓陈国良. 基于机器视觉的筒子纱缺陷在线检测系统[J]. 纺织学报, 2018, 39(01): 139-145.
[11]	杨松林马帅丁朝鹏范红丽薛欢欢. 应用机器视觉的织物表面绒毛率测试系统[J]. 纺织学报, 2017, 38(06): 118-123.
[12]	蒋双歌王蕾王静刘建立高卫东. 织物折皱回复的各向异性表征[J]. 纺织学报, 2016, 37(06): 42-47.
[13]	李文羽程隆棣. 基于机器视觉和图像处理的织物疵点检测研究新进展[J]. 纺织学报, 2014, 35(3): 158-0.
[14]	王蕾, 刘建立, 高卫东. 基于机器视觉的织物洗涤后尺寸稳定性评价[J]. 纺织学报, 2012, 33(7): 43-47.
[15]	霍铖宇华强卢达范龙保李新徐刚. 基于虚拟仪器的针织机械用高速电磁阀测试系统[J]. 纺织学报, 2011, 32(10): 134-0.

Metrics

Viewed

Full text

Abstract

Cited

Shared

Discussed

类型	滤波器个数	卷积核/步长	输出特征尺寸
卷积层1	32	3×3	224像素×224像素
最大池化层1	1	2×2/2	112像素×112像素
卷积层2	64	3×3	112像素×112像素
最大池化层2	1	2×2/2	56像素×56像素
卷积层3	128	3×3	56像素×56像素
卷积层4	64	1×1	56像素×56像素
卷积层5	128	3×3	56像素×56像素
最大池化层3	1	2×2/2	28像素×28像素
卷积层6	256	3×3	28像素×28像素
卷积层7	128	1×1	28像素×28像素
卷积层8	256	3×3	28像素×28像素
最大池化层4	1	2×2/2	14像素×14像素
卷积层9	512	3×3	14像素×14像素
卷积层10	256	1×1	14像素×14像素
卷积层11	512	3×3	14像素×14像素
卷积层12	256	1×1	14像素×14像素
卷积层13	512	3×3	14像素×14像素
最大池化层5	1	2×2/2	7像素×7像素
卷积层14	1 024	3×3	7像素×7像素
卷积层15	512	1×1	7像素×7像素
卷积层16	1 024	3×3	7像素×7像素
卷积层17	512	1×1	7像素×7像素
卷积层18	1 024	3×3	7像素×7像素
卷积层19	1 000	1×1	7像素×7像素
平均池化		全局	1 000

YOLOv2模型	数据集	mAP	训练时间/h
YOLO-VOC	训练数据集	76.8	5
Tiny-YOLO	训练数据集	75.4	3

图像分辨率/像素	数据集	mAP	处理速度/(帧·s^-1)
288×288	训练数据集	69.6	82
416×416	训练数据集	73.5	55
544×544	训练数据集	74.2	40

缺陷类型	检出率	误检率
裂缝	98.0	0.0
毛团	96.0	1.0
孔洞	97.0	1.0