基于Mask R-CNN深度学习的羊绒羊毛纤维识别技术

doi:10.19398/j.att.202104002

现代纺织技术 ›› 2022, Vol. 30 ›› Issue (2): 36-40.DOI: 10.19398/j.att.202104002

基于Mask R-CNN深度学习的羊绒羊毛纤维识别技术

从明芳¹, 李子印², 卢鸯¹(), 韩高锋¹, 谢凌佳¹, 王启真²

1.浙江省轻工业品质量检验研究院,杭州 310018
2.中国计量大学光学与电子科技学院,杭州 310018

收稿日期:2021-04-02 出版日期:2022-03-10 网络出版日期:2021-08-04
通讯作者: 卢鸯,E-mail: 69270913@qq.com
作者简介:从明芳(1986-),女,浙江杭州人,工程师,硕士,主要从事纺织品检测方面的研究。
基金资助:
国家市场监督管理总局科技计划项目(2019MK031)

Recognition technology of cashmere and wool fibers based on mask R-CNN deep learning

CONG Mingfang¹, LI Ziyin², LU Yang¹(), HAN Gaofeng¹, XIE Lingjia¹, WANG Qizhen²

1. Zhejiang Light Industrial Products Inspection and Research Institute, Hangzhou 310018, China
2. College of Optical and Electronic Technology, China Jiliang University, Hangzhou 310018, China

Received:2021-04-02 Published:2022-03-10 Online:2021-08-04

摘要/Abstract

摘要：

为提高羊绒羊毛纤维定量的自动化程度,引入基于掩模区域卷积神经网络(Mask R-CNN)深度学习技术,对通过光学显微镜采集的图片进行图片处理、算法模型优化,以及学习和训练,建立起山羊绒和绵羊毛的自动识别模型。采用测试集对所建立的模型进行了验证测试,结果表明,对山羊绒和绵羊毛纤维的自动识别正确率达到95%以上,证实了所建立的识别技术的可行性。

关键词: 山羊绒, 绵羊毛, 图像处理, 深度学习, Mask R-CNN

Abstract:

To enhance the level of automation for wool fiber quantification, the mask R-CNN deep learning technology was introduced for the processing of the pictures collected by the optical microscope, optimization of algorithm model, learning and training. An automatic recognition model of cashmere and sheep wool was established. Through the verification test using a test set, it was found that the accuracy of automatic recognition of cashmere and sheep wool fibers reached more than 95%, confirming the feasibility of the recognition technology developed in this paper.

Key words: cashmere, sheep wool, image processing, deep learning, mask R-CNN

中图分类号:

TS102.3

从明芳, 李子印, 卢鸯, 韩高锋, 谢凌佳, 王启真. 基于Mask R-CNN深度学习的羊绒羊毛纤维识别技术[J]. 现代纺织技术, 2022, 30(2): 36-40.

CONG Mingfang, LI Ziyin, LU Yang, HAN Gaofeng, XIE Lingjia, WANG Qizhen. Recognition technology of cashmere and wool fibers based on mask R-CNN deep learning[J]. Advanced Textile Technology, 2022, 30(2): 36-40.

图/表 8

图1 显微图片中纤维的分离图示

Fig.1 Diagram of fiber separation in micrograph

图2 绵羊毛与山羊绒显微图像特征

Fig.2 Microscopic image characteristics of wool and cashmere

图3 Mask R-CNN模型结构示意

Fig.3 Structure diagram of Mask R-CNN model

表1 各数据集纤维数量

Tab.1 Fiber quantity in each data set

数据集	图像张数/张	纤维根数/根
训练集	4198	15952
验证集	1199	4976
测试集	600	2274

表2 模型超参数设定

Tab.2 Model hyperparameter setting

超参数	值
DATALOADER	4
ANCHOR_ANGLES	[-90,0,90]
ANCHOR_ RATIOS	[0.33,1.0,3.0]
MAX_ITER	60000
STEPS	(36000,50000)
RESNETS DEPTH	50
BASE_LR	0.00012
SOLVER	Adam
Adam_P1	0.9
Adam_P2	0.999
Adam_T	10^-8
NUM_CLASSES	4

表3 基于Mask R-CNN模型的纤维测试集识别率

Tab.3 Recognition rate of fiber test set based on Mask R-CNN model

纤维	测试纤维数/根	与人工识别一致的根数/根	结果一致性/%
羊毛	955	925	96.859
羊绒	1015	970	95.567
其他	304	302	99.342
总计	2274	2161	96.614

图4 识别结果输出图示

Fig.4 Graphical representation of automatic recognition result

表4 测试集自动识别结果

Tab.4 The result of test set automatically identified

指标	羊毛	羊绒	羊毛/羊绒
总根数/根	140	128	195
识别正确根数/根	134	123	187
正确率/%	95.7	96.1	95.9

参考文献 10

[1]	顾昊, 王满, 闫畅, 等. 羊绒纤维鉴别方法综述[J]. 中国纤检, 2017(1):69-71.
	GU Hao, WANG Man, YAN Chang, et al. Review on the detection methods of cashmere fiber[J]. China Fiber Inspection, 2017(1):69-71.
[2]	马海, 吴桂芳, 李志东, 等. 山羊绒纤维鉴别技术的现状与发展趋势[J]. 毛纺科技, 2017, 45(3):5-9.
	MA Hai, WU Guifang, LI Zhidong, et al. Current situation and development trend of cashmere identification technology[J]. Wool Textile Journal, 2017, 45(3):5-9.
[3]	刘心如, 张利平, 王建福, 等. 可见-近红外漫反射光谱技术对羊毛和羊绒的鉴别研究[J]. 光谱学与光谱分析, 2013, 33(8):2092-2095.
	LIU Xinru, ZHANG Liping, WANG Jianfu, et al. Use of visible and near infrared reflectance spectroscopy to identify the cashmere and wool[J]. Spectroscopy and Spectral Analysis, 2013, 33(8) :2092-2095.
[4]	陈国培, 赖心田, 唐复润, 等. 羊毛及羊绒制品实时荧光PCR鉴定方法研究[J]. 纺织导报, 2013(2):88-91.
	CHEN Guopei, LAI Xintian, TANG Furun, et al. Real-time PCR identification of wool and cashmere products[J]. China Textile Leader, 2013(2):88-91.
[5]	董晓东, 杨素英, 丁楠, 等. DNA分析技术在动物纤维鉴别领域的研究进展[J]. 中国纤检, 2015(17):70-72.
	DONG Xiaodong, YANG Suying, DING Nan, et al. The research advances of DNA technology in identiifcation for animal fibers[J]. China Fiber Inspection, 2015(17):70-72.
[6]	季益萍, 王瑞, 杨锁廷, 等. 计算机图像识别技术检测羊绒羊毛混纺比[J]. 毛纺科技, 2008, 36(10):58-60.
	JI Yiping, WANG Rui, YANG Suoting, et al. Identifying blend ratio of wool and cashmere by computer image recognition technology[J]. Wool Textile Journal, 2008, 36(10):58-60.
[7]	邢文宇, 邓娜, 辛斌杰, 等. 基于多特征融合图像分析技术的羊毛与羊绒鉴别[J]. 纺织学报, 2019, 40(3):146-152.
	XING Wenyu, DENG Na, XIN Binjie, et al. Identification of wool and cashmere based on multi-feature fusion image analysis technology[J]. Journal of Textile Research, 2019, 40(3):146-152.
[8]	王军庆, 马佩佩, 李保荣, 等. 基于图像处理的羊毛/山羊绒鉴别方法研究进展[J]. 纺织科技进展, 2020(9):8-11.
	WANG Junqing, MA Peipei, LI Baorong, et al. Research progress of wool/cashmere identification method based on image processing[J]. Progress in Textile Science & Technology, 2020(9):8-11.
[9]	刘爽, 许忠保, 李春桥, 等. 基于数字图像处理的羊绒与羊毛识别研究[J]. 棉纺织技术, 2019, 47(10):30-34.
	LIU Shuang, XU Zhongbao, LI Chunqiao, et al. Study on the recognition of cashmere and wool based on digital image processing[J]. Cotton Textile Technology, 2019, 47(10) : 30-34.
[10]	盛冠忠, 李龙. 关于山羊绒鉴别有关问题的探讨[J]. 毛纺科技, 2007, 35(12):52-55.
	SHENG Guanzhong, LI Long. Study on distinguishing problem of cashmere fiber[J]. Wool Textile Journal, 2007, 35(12):52-55.

基于Mask R-CNN深度学习的羊绒羊毛纤维识别技术

Recognition technology of cashmere and wool fibers based on mask R-CNN deep learning

RichHTML

PDF

可视化

摘要/Abstract

引用本文

使用本文

图/表 8

参考文献 10

相关文章 0

编辑推荐

Metrics