纺织学报 ›› 2020, Vol. 41 ›› Issue (04): 39-44.doi: 10.13475/j.fzxb.20190500406

• 纺织工程 • 上一篇    下一篇

基于改进卷积神经网络的化纤丝饼表面缺陷识别

王泽霞, 陈革, 陈振中()   

  1. 东华大学 机械工程学院, 上海 201620
  • 收稿日期:2019-05-06 修回日期:2019-12-31 出版日期:2020-04-15 发布日期:2020-04-27
  • 通讯作者: 陈振中
  • 作者简介:王泽霞(1995—),女,硕士生。主要研究方向为机器学习在化纤丝饼表面缺陷识别的应用。
  • 基金资助:
    中央高校基本科研业务费专项资金项目(18D110316);中国博士后科学基金项目(2018M630383)

Surface defect recognition of chemical fiber yarn packages based on improved convolutional neural network

WANG Zexia, CHEN Ge, CHEN Zhenzhong()   

  1. College of Mechanical Engineering, Donghua University, Shanghai 201620, China
  • Received:2019-05-06 Revised:2019-12-31 Online:2020-04-15 Published:2020-04-27
  • Contact: CHEN Zhenzhong

摘要:

针对传统人工检测化纤丝饼表面缺陷方法的不足,提出改进的卷积神经网络对正常以及3种常见缺陷丝饼进行分类识别。首先对采集的丝饼图像进行分块处理,然后利用改进的卷积神经网络进行特征提取,采用全局最大池化层代替全连接层,增强了图像对空间变换的鲁棒性,减少了模型参数,并利用softmax分类器进行分类。最后在网络学习过程中提出主动学习方法,用少量标注样本对网络进行训练,选出对提升网络性能最具价值的样本进行标注并加入到训练样本中进行训练检测。结果表明,该方法可有效实现丝饼的缺陷识别,识别准确率达到97.1%,并有效减少了网络所需的标注样本数量,节省大量的标注成本,具有一定的通用性。

关键词: 化纤丝饼, 缺陷识别, 图像分块, 卷积神经网络, 全局最大池化, 主动学习方法

Abstract:

Focusing on the disadvantages of traditional manual method for defect detection of chemical fiber yarn packages, an improved convolutional neural network was proposed to classify and recognize the normal and three common defective yarn packages. The images of collected yarn package were processed into blocks before features were extracted by using the improved convolutional neural network. The global maximum pooling layer was used instead of the full connection layer, enhancing the robustness of images to spatial translations and reducing the model parameters. Softmax classifier was employed for classification. As a result, an active learning method was proposed for network learning. Firstly, a small number of labeled samples were used to train the network, and then the most valuable samples for improving network performance were selected and labeled, which were then added to the training samples. The experimental results show that this method can effectively facilitate the defect recognition of yarn packages, achieving a recognition accuracy of 97.1%. This method effectively reduces the number of labeled samples required by the network and saves a lot of labeling costs, with a certain degree of universality.

Key words: chemical fiber yarn package, defect recognition, image blocking, convolutional neural network, global maximum pooling, active learning method

中图分类号: 

  • TP391.4

图1

算法流程图"

图2

正常以及缺陷样本"

图3

丝饼缺陷识别网络模型"

图4

典型主动学习流程"

图5

模型训练损失曲线"

图6

模型训练准确率曲线"

表1

模型测试识别准确率"

丝饼类型 样本数量 错分样本数 识别准确率/%
正常 128 2 98.4
绊丝 128 6 95.3
成型不良 128 4 96.9
油污 128 3 97.7

表2

不同连接方式的识别结果对比"

连接方式 参数内存/
MB
识别准确
率/%
单张图片检测
时间/ms
全局最大池化 21.48 97.1 82.19
全局平均池化 21.43 88.9 82.81
全连接 32.26 88.1 91.88

图7

不同挑选准则的识别准确率"

[1] 王松林, 相恒学, 徐锦龙, 等. 通用合成纤维功能化基础问题与发展趋势[J]. 纺织学报, 2018,39(3):167-174.
WANG Songlin, XIANG Hengxue, XU Jinlong, et al. Basic issues and development trends on general synthetic fibers with high functionalization[J]. Journal of Textile Research, 2018,39(3):167-174.
[2] 高哲, 蒋高明, 马丕波, 等. 碳纤维多轴向经编复合材料的应用与发展[J]. 纺织学报, 2013,34(12):144-151.
GAO Zhe, JIANG Gaoming, MA Pibo, et al. Application and development of carbon fiber multi-axial warp-knitted fabric reinforced composites[J]. Journal of Textile Research, 2013,34(12):144-151.
[3] 张志宇. 浅谈由设备原因引起的DTY锭位缺陷[J]. 黑龙江纺织, 2010(1):15-16.
ZHANG Zhiyu. Talking about the DTY spindle position defect caused by equipment[J]. Heilongjiang Textile, 2010(1):15-16.
[4] 陆奕辰, 王蕾, 唐千惠, 等. 应用图像处理的纱线黑板毛羽量检测与评价[J]. 纺织学报, 2018,39(8):144-149.
LU Yichen, WANG Lei, TANG Qianhui, et al. Detection and evaluation on yarn hairiness of blackboard with image processing[J]. Journal of Textile Research, 2018,39(8):144-149.
[5] 杨曼, 李仁忠, 刘阳阳, 等. 基于改进迭代匹配滤波的织物疵点检测[J]. 西安工程大学学报, 2017,31(3):383-389.
YANG Man, LI Renzhong, LIU Yangyang, et al. Fabric defect detection based on improved iterative match filter algorithm[J]. Journal of Xi'an Polytechnic University, 2017,31(3):383-389.
[6] 李文羽. 基于机器视觉和图像处理的色织物疵点自动检测研究[D]. 上海: 东华大学, 2014: 60-76.
LI Wenyu. Research on automatic detection for yarn-dyed fabric defect based on machine vision and image processing[D]. Shanghai: Donghua University, 2014: 60-76.
[7] 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(1):147-152.
JING Junfeng, GUO Gen. Yarn packages hairiness detection based on machine vision[J]. Journal of Textile Research, 2019,40(1):147-152.
[8] KRIZHEVSKY A, SUTSKEVER I, HINTON G E. Imagenet classification with deep convolutional neural networks[C]//Annual conference on neural information processing systems. Cambridge: MIT Press, 2012: 1097-1105.
[9] WANG Shuihua, LV Yiding, SUI Yuxiu, et al. Alcoholism detection by data augmentation and convolutional neural network with stochastic pooling[J]. Journal of Medical Systems, 2017,42(1):845-856.
[10] WANG Meng, HUA Xiansheng. Active learning in multimedia annotation and retrieval: a survey[J]. ACM Transaction on Intelligent Systems and Technology, 2011,2(2):1-10.
[11] 杨文柱, 田潇潇, 王思乐, 等. 主动学习算法研究进展[J]. 河北大学学报(自然科学版), 2017,37(2):216-224.
YANG Wenzhu, TIAN Xiaoxiao, WANG Sile, et al. Recent advances in active learning algorithms[J]. Journal of Hebei University (Nature Science Edition), 2017,37(2):216-224.
[12] 徐艳. 基于主动学习的图像标注方法研究[D]. 锦州: 辽宁工业大学, 2014: 37-39.
XU Yan. Reserch on image annotation based active learning[D]. Jinzhou: Liaoning University of Technology, 2014: 37-39.
[1] 王奕文, 罗戎蕾, 康宇哲. 基于卷积神经网络的汉服关键尺寸自动测量[J]. 纺织学报, 2020, 41(12): 124-129.
[2] 邵金鑫, 张宝昌, 曹继鹏. 基于图像处理与深度学习方法的棉纤维梳理过程纤维检测识别技术[J]. 纺织学报, 2020, 41(07): 40-46.
[3] 王晓华, 姚炜铭, 王文杰, 张蕾, 李鹏飞. 基于改进YOLO深度卷积神经网络的缝纫手势检测[J]. 纺织学报, 2020, 41(04): 142-148.
[4] 贾小军, 叶利华, 邓洪涛, 刘子豪, 陆锋杰. 基于卷积神经网络的蓝印花布纹样基元分类 [J]. 纺织学报, 2020, 41(01): 110-117.
[5] 孙洁, 丁笑君, 杜磊, 李秦曼, 邹奉元. 基于卷积神经网络的织物图像特征提取与检索研究进展[J]. 纺织学报, 2019, 40(12): 146-151.
[6] 刘正东, 刘以涵, 王首人. 西装识别的深度学习方法[J]. 纺织学报, 2019, 40(04): 158-164.
[7] 吴欢, 丁笑君, 李秦曼, 杜磊, 邹奉元. 采用卷积神经网络 CaffeNet 模型的女裤廓形分类[J]. 纺织学报, 2019, 40(04): 117-121.
[8] 汪珊娜 张华熊 康锋. 基于卷积神经网络的领带花型情感分类[J]. 纺织学报, 2018, 39(08): 117-123.
[9] 王雯雯 高畅 刘基宏. 应用卷积神经网络的细纱断纱锭位识别[J]. 纺织学报, 2018, 39(06): 136-141.
[10] 景军锋 范晓婷 李鹏飞 洪良. 应用深度卷积神经网络的色织物缺陷检测[J]. 纺织学报, 2017, 38(02): 68-74.
[11] 万振凯;王占刚. 基于神经网络的复合材料缺陷超声波检测研究[J]. 纺织学报, 2010, 31(2): 54-59.
Viewed
Full text


Abstract

Cited

  Shared   
  Discussed   
No Suggested Reading articles found!