基于机器嗅觉的柑橘品种无损检测与识别  

Non-destructive Testing and Identification of Citrus Varieties Based on Machine Olfaction

作　　者：彭珂 骆德汉[1] 夏必亮 

机构地区：[1]广东工业大学信息工程学院,广东广州510006

出　　处：《江西农业大学学报》2017年第5期1017-1024,共8页Acta Agriculturae Universitatis Jiangxiensis

基　　金：国家自然科学基金项目(61571140);广东省科技计划项目(2016B030303011);广东省教育厅仪器重点培育项目(15ZK0130);广州市科技计划项目(201607010247)~~

摘　　要：为了探索一种不依赖人的感官评定而识别柑橘品种的新方法,提出了一种基于机器嗅觉的柑橘品种分类识别方法。使用PEN3电子鼻对外形相似但品种不同的柑橘水果进行无损气味采集,通过局部切空间排列算法(local tangent space alignment,LTSA)对高维水果气味数据进行降维处理,再运用线性判别分析方法(linear discriminant analysis,LDA)对降维后的数据进行判别分析,最终达到对不同品种柑橘水果分类识别的目的。实验结果表明,采用LTSA+LDA的方法能够对柑橘品种进行有效识别,对皇帝柑、脐橙和砂糖桔的识别率分别是91.1%、93.3%和91.1%,总体识别率为91.8%,该方法在柑橘品种识别方面具有良好的应用前景。In order to explore a new method for identifying citrus varieties without relying on human sensory assessment,a method for classification and identification of citrus varieties based on machine olfaction was proposed.PEN3 electronic nose was used to collect the odor of citrus fruits with similar shape but different varieties.The method of local tangent space alignment（ LTSA） was used to reduce the dimension of high-dimensional fruit odor data,then the linear discriminant analysis（ LDA） was used to analyze the data after dimension reduction.Finally,different varieties of citrus fruits were classified and identified. The results show that the method of LTSA＋ LDA can effectively identify citrus varieties,the recognition rate of Citrus reticulata Blanco cv. Hanggan,Citrus sinensis L.Osbeck and Citrus reticulata Blanco cv.Shiyue Ju are 91.1%,93.3% and 91.1%,respectively,and the overall recognition rate is 91.8%.This method has a good application prospect in citrus variety identification.

关 键 词：机器嗅觉 柑橘 品种识别 无损检测 LTSA+LDA 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术] S666.2[自动化与计算机技术—计算机科学与技术]