基于主成分分析及最近邻算法的纺织材料识别  

Textile Material Identification Based on Principal Component Analysis and Nearest Neighbor Algorithm

作　　者：沈国康 孟想[2] 曹雅丽 潘静娴 汤辉 SHEN Guokang;MENG Xiang;CAO Yali;PAN Jingxian;TANG Hui(Zhejiang Zhongtian Textile Testing Co.,Ltd.,Zhejiang Jiaxing,314400;Shanghai University of Engineering Science,Shanghai,201620)

机构地区：[1]浙江中天纺检测有限公司,浙江嘉兴314400 [2]上海工程技术大学,上海201620

出　　处：《棉纺织技术》2018年第11期30-34,共5页Cotton Textile Technology

摘　　要：探讨了利用主成分分析与最近邻分类器对8种纯纺织物的识别分类效果。首先采集8种纯纺织物在920nm～2 528nm波长范围内的720幅高光谱图像,经过图像标定、曲线滤波、波长预选等预处理之后得到可供分析使用的高光谱原始数据。而后将720个织物小样分为训练集和测试集,结合主成分分析提取出织物小样的4个特征波长,最后利用最近邻分类器完成训练并对测试集进行分类识别。结果表明:训练集识别率达到了100%,测试集分辨率也都达到了95%,整体识别效果较好。认为:高光谱成像技术可以用于棉、涤纶、聚乙烯、羊毛、聚氯乙烯、锦纶、亚麻、蚕丝的材料识别。The identification and classification effect of eight pure fabrics by adopting principle component analysis and nearest neighbor classifier were discussed.Firstly,720 hyperspectral images of eight pure fabrics in the wave range of 920 nm^2 528 nm were collected.The original hyperspectral data were obtained by analysis study after the pre-treatment of image calibration,curve smoothing,pre-selecting wavelength and so on.Then,720 fabric samples were divided into training set and test set.Four characteristic wavelengths of fabric samples were extracted combining principle component analysis.In the end,training was finished by using nearest neighbor algorithm,the test set was identified and classified.The test results showed that the identification rate of training set was reached 100%and the distinguishing rate of test set was reached 95%.The overall identification effect was better.It is considered that hyperspectral imaging technology can be used in the identification of cotton,polyester,polyethylene,wool,polyvinyl chloride,polyamide,flax and silk.

关 键 词：纺织材料识别 高光谱成像 滤波器 主成分分析 最近邻算法 

分 类 号：TS101.8[轻工技术与工程—纺织工程]