基于改进AlexNet模型的抓毛织物质量检测方法

doi:10.13475/j.fzxb.20210602607

摘要/Abstract

摘要：

针对传统图像识别方法对抓毛织物表面特征难以提取且识别准确率低的问题,提出了一种改进AlexNet模型的抓毛织物质量检测方法,通过数据增强方法对抓毛织物数据进行扩充,构建卷积神经网络对抓毛织物的样本特征进行提取,利用SGDM、RMSProp、Adam优化算法和改变学习率相结合的实验方法,采用全新学习与迁移学习两种算法对抓毛织物图像数据集进行训练,在训练完成后,分别利用卷积神经网络的不同深度池化层提取抓毛织物样本的特征作为输入,将提取到的抓毛织物特征拟合支持向量机(SVM)分类器,最后对输入的抓毛织物图像进行分类。实验结果表明:使用卷积神经网络方法能够增加卷积层对抓毛织物表面特征的提取能力,获得具有较高分辨力的图像特征,通过数据增强和SGDM算法训练的模型,提取网络pool5层特征拟合SVM分类器,识别准确率明显提高。基于改进AlexNet模型的抓毛织物质量检测方法能够提取抓毛织物表面特征且识别率高。

关键词: 抓毛织物, 机器视觉, 卷积神经网络, 迁移学习, 数据增强, 织物质量检测

Abstract:

The traditional image recognition method is difficult to extract the surface features of the fleece fabrics leading to low recognition accuracy. This study proposed an improved AlexNet model for the quality detection method of the fleece fabrics. The convolutional neural network was used to extract the sample features of the fleece fabric, and the experimental method combining SGDM, RMSProp, Adam optimization algorithm was adopted for the study to investigate effects of changing learning rate and the use of two new learning and transfer learning algorithms in training the fleece fabric image dataset. After the completion of training, different depth pooling layers of the convolutional neural network were employed to extract the features of the fleece fabric samples. The extracted fleece fabric features were fitted to the support vector machine(SVM) classifier to analyze the input fleece fabric image. The experimental results show that the use of the convolutional neural network method can increase the ability of the convolutional layer to extract the surface features of the fleece fabric, and obtain image features with higher resolution. The model trained by the data enhancement and SGDM algorithm can extract the network pool5 layer features. With the SVM classifier, the recognition accuracy enhanced significantly. The quality detection method of fleece fabrics based on the improved AlexNet model can extract the surface features of fleece fabrics with high recognition rate.

Key words: fleece fabric, machine vision, convolutional neural network, transfer learning, data enhancement, fabric quality inspection

中图分类号:

TN911.73

金守峰, 侯一泽, 焦航, 张鹏, 李宇涛. 基于改进AlexNet模型的抓毛织物质量检测方法[J]. 纺织学报, 2022, 43(06): 133-139.

JIN Shoufeng, HOU Yize, JIAO Hang, ZHNAG Peng, LI Yutao. An improved AlexNet model for fleece fabric quality inspection[J]. Journal of Textile Research, 2022, 43(06): 133-139.

图/表 12

图1

图2

表1

图3

图4

图5

图6

表2

图7

图8

表3

图9

参考文献 18

[1]	金守峰, 陈阳, 林强强, 等. 起绒织物表面轮廓提取及覆盖程度估计方法[J]. 棉纺织技术, 2019, 47(9): 13-17.
	JIN Shoufeng, CHEN Yang, LIN Qiangqiang, et al. Fluff fabric[J]. Cotton Textile Technology, 2019, 47(9): 13-17.
[2]	KIM S M, PPAR C K. Evaluation of fabric pilling using hybrid imaging methods[J]. Fibers and Polymers, 2006, 7(1): 71-89.
[3]	李春雷, 高广帅, 刘洲峰, 等. 应用方向梯度直方图和低秩分解的织物疵点检测算法[J]. 纺织学报, 2017, 38(3): 149-154.
	LI Chunlei, GAO Guangshuai, LIU Zhoufeng, et al. Fabric defect detection algorithm based on histogram of oriented gradient and low-rank decomposition[J]. Journal of Textile Research, 2017, 38(3): 149-154. doi: 10.1177/004051756803800207
[4]	何峰, 周亚同, 赵翔宇, 等. 纹理织物疵点窗口跳步形态学法检测[J]. 纺织学报, 2017, 38(10): 124-131.
	HE Feng, ZHOU Yatong, ZHAO Xiangyu, et al. Textured fabric defect detection based on windowed hop-step morphological algorithm[J]. Journal of Textile Research, 2017, 38(10): 124-131.
[5]	汪亚明, 崔新辉, 韩永华. 基于小波变换及Gabor滤波的起毛起球图像分割[J]. 丝绸, 2016, 53(3): 37-40.
	WANG Yaming, CUI Xinhui, HAN Yonghua. Fabric pilling image segmentation based on wavelet transform and Gabor filter[J]. Journal of Silk, 2016, 53(5): 37-40.
[6]	夏雨薇, 石美红, 贺飞跃, 等. 基于降维融合特征和集成学习的织物疵点分类[J]. 国外电子测量技术, 2019, 38(7):86-91.
	XIA Yuwei, SHI Meihong, HE Feiyue, et al. Fabric defect classification based on dimensionality reduction fusion features and ensemble learning[J]. Foreign Electronic Measurement Technology, 2019, 38(7): 86-91.
[7]	张顺, 龚怡宏, 王进军. 深度卷积神经网络的发展及其在计算机视觉领域的应用[J]. 计算机学报, 2019, 42(3): 453-482.
	ZHANG Shun, GONG Yihong, WANG Jinjun. The development of deep convolution neural network and its applications on computer vision[J]. Chinese Journal of Computers, 2019, 42(3): 453-482.
[8]	王理顺, 钟勇, 李振东, 等. 基于深度学习的织物缺陷在线检测算法[J]. 计算机应用, 2019, 39(7): 2125-2128.
	WANG Lishun, ZHONG Yong, LI Zhendong, et al. On-line fabric defect recognition algorithm based on deep learning[J]. Computer Application, 2019, 39(7): 2125-2128.
[9]	张家玮. 基于机器视觉技术的织物缺陷检测方法研究[D]. 无锡: 江南大学, 2021:1-39.
	ZHANG Jiawei. Research on fabric defect detection method based on machine vision technology[D]. Wuxi: Jiangnan University, 2021:1-39.
[10]	金守峰, 林强强, 马秋瑞, 等. 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(2): 69-76.
	JIN Shoufeng, LIN Qiangqiang, MA Qiurui, et al. Method for detecting fluff quality of fabric surface based on BP neural network[J]. Journal of Textile Research, 2020, 41(2): 69-76.
[11]	KRIZHEVSKY Alex, SUTSKEVER Ilya, HINTON Geoffrey E. ImageNet classification with deep convolutional neural networks[J]. Communications of The ACM, 2017, 60(6): 65-78. doi: 10.1145/3122803
[12]	HINTON G E, SALAKHUTDINOV R R. Reducing the dimensionality of data with neural networks[J]. Science, 2006, 313(5786): 72-85. doi: 10.1126/science.1126287
[13]	卢宏涛, 张秦川. 深度卷积神经网络在计算机视觉中的应用研究综述[J]. 数据采集与处理, 2016, 31(1): 1-17.
	ZHOU Feiyan, JIN Linpeng, DONG Jun. Review of convolutional neural network[J]. Chinese Journal of Computers, 2017, 40(6): 1229-1251.
[14]	周飞燕, 金林鹏, 董军. 卷积神经网络研究综述[J]. 计算机学报, 2017, 40(6): 1229-1251.
	ZHOU Feiyan, JIN Linpeng, DONG Jun. Review of convolutional neural network[J]. Chinese Journal of Computers, 2017, 40(6): 1229-1251.
[15]	陈思文, 刘玉江, 刘冬, 等. 基于AlexNet模型和自适应对比度增强的乳腺结节超声图像分类[J]. 计算机科学, 2019, 46(S1): 146-152.
	CHEN Siwen, LIU Yujiang, LIU Dong, et al. AlexNet model and adaptive contrast enhancement based ultrasound imaging classification[J]. Computer Science, 2019, 46(S1): 146-152.
[16]	常亮, 邓小明, 周明全, 等. 图像理解中的卷积神经网络[J]. 自动化学报, 2016, 42(9): 1300-1312.
	CHANG Liang, DENG Xiaoming, ZHOU Mingquan, et al. Convolutional neural networks in image understanding[J]. Acta Automatica Sinica, 2016, 42(9): 1300-1312.
[17]	唐浩漾, 孙梓巍, 王婧, 等. 基于VGG-19混合迁移学习模型的服饰图片识别[J]. 西安邮电大学学报, 2018, 23(6): 87-93.
	TANG Haoyang, SUN Ziwei, WANG Jing, et al. Clothing image recognition based on VGG-19 hybrid migration learning model[J]. Journal of Xi'an University of Posts and Telecommunications, 2018, 23(6): 87-93.
[18]	陈卫中, 倪宗瓒, 潘晓平, 等. 用ROC曲线确定最佳临界点和可疑值范围[J]. 现代预防医学, 2005(7): 729-731.
	CHEN Weizhong, NI Zongzan, PAN Xiaoping, et al. Receiver operating characteristic curver to determine the optimal operating point and doubtable value interval[J]. Modern Preventive Medicine, 2005(7): 729-731.

Metrics

Viewed

Full text

Abstract

Cited

Shared

Discussed

名称	说明	参数总数
Data(图像输入)	227×227×3图像	-
Conv1(卷积)	核数:96 11×11×3卷积步幅[4 4]填充[0 0 0 0]	34 944
Conv2(卷积)	核数:2组1 285×5×48卷积步幅[1 1]填充[2 2 2 2]	307 456
Conv3(卷积)	核数:3 843×3×256卷积步幅[1 1]填充[1 1 1 1]	88 5120
Conv4(卷积)	核数:2组1 923×3×192卷积步幅[1 1]填充[1 1 1 1]	663 936
Conv5(卷积)	核数:2组1 283×3×192卷积步幅[1 1]填充[1 1 1 1]	442 624
FCL6(全连接)	4 098全连接层	3 775 2832
FCL7(全连接)	4 098全连接层	1 6781 312
FCL8(全连接)	1 000全连接层	4 097 000
Output(分类输出)	-	-

训练方式	训练优化算法	学习率	过拟合率	训练平均耗时/min	测试集准确率/%
训练方式	训练优化算法	学习率	过拟合率	训练平均耗时/min	抓毛织物A	抓毛织物B	抓毛织物C	抓毛织物D	抓毛织物E	抓毛织物F
全新学习	SGDM	0.000 1 0.001	1.013 1.019	31 27	88.35 75.21	82.11 70.25	89.35 76.31	88.52 75.35	88.37 77.97	86.24 79.24
	RMSProp	0.000 1 0.001	1.028 1.032	36 33	83.58 79.35	83.64 75.24	88.57 82.15	82.35 70.37	86.17 75.64	87.68 80.79
	Adam	0.000 1 0.001	1.123 1.106	37 35	85.25 73.03	76.35 77.27	84.38 80.65	81.25 80.02	83.68 80.42	84.13 82.47
迁移学习	SGDM	0.000 1 0.001	1.010 1.015	21 18	99.85 98.75	98.28 98.35	99.34 97.35	99.98 98.27	99.98 99.87	99.43 98.94
	RMSProp	0.000 1 0.001	1.016 1.020	27 23	95.36 94.49	92.41 91.68	91.27 94.35	91.26 88.34	93.06 92.01	92.15 91.56
	Adam	0.000 1 0.001	1.113 1.119	28 26	92.35 90.35	90.35 91.25	91.25 74.21	89.08 92.25	91.02 87.82	94.25 89.54

学习方式	数据增强	学习率	pool5层的输出拟合SVM分类器抓毛织物表面测试准确率/%						平均识别率/%
学习方式	数据增强	学习率	抓毛织物A	抓毛织物B	抓毛织物C	抓毛织物D	抓毛织物E	抓毛织物F	平均识别率/%
全新学习	是	0.000 1	98.12	99.02	98.96	98.14	97.08	98.89	98.37
迁移学习	是	0.000 1	99.37	99.08	99.13	99.09	99.75	99.54	99.33