主成分分析和SIMCA的甘蓝与杂草光谱识别方法研究  

Research on Spectra Recognition Method for Cabbages and Weeds Based on PCA and SIMCA

作　　者：祖琴[1,2,3] 邓巍[1,2] 王秀[1,2] 赵春江[1,2] 

机构地区：[1]北京农业智能装备技术研究中心,北京100097 [2]国家农业智能装备工程技术研究中心,北京100097 [3]贵州大学电气工程学院,贵州贵阳550025

出　　处：《光谱学与光谱分析》2013年第10期2745-2750,共6页Spectroscopy and Spectral Analysis

基　　金：国家"十二五"科技支撑计划项目(2011BAD20B07);国家(863计划)项目(2012AA101904);国家(948计划)项目(2011-G32)资助

摘　　要：为了提高杂草识别的准确性和快速性,利用光谱反射率差异区分作物与杂草。首先利用SG卷积求导法与多元散射校正法的不同组合对原始光谱数据进行预处理,然后利用主成分分析法(PCA)对各类植物进行聚类分析,并根据主成分分析结果中各个最佳主成分对应的载荷图,提取对各类植物识别敏感的特征波长,最后以特征波长为输入变量,用簇类的独立软模式(SIMCA)分类法对各类植物进行分类识别。甘蓝与杂草的分类结果表明,在1阶3次51点SG卷积求导法加上多元散射校正法(MSC)的最佳预处理基础上,根据主成分分析中前3个主成分,提取到23个特征波长,以它们为输入变量,利用SIMCA方法进行分类时,建模集和预测集的识别率分别达到98.6%和100%。In order to improve the accuracy and efficiency of weed identification, the difference of spectral reflectance was em- ployed to distinguish between crops and weeds. Firstly, the different combinations of Savitzky-Golay （SG） convolutional deriva- tion and muhiplicative scattering correction （MSC） method were applied to preprocess the raw spectral data. Then the clustering analysis of various types of plants was completed by using principal component analysis （PCA） method, and the feature wave- lengths which were sensitive for classifying various types of plants were extracted according to the corresponding loading plots of the optimal principal components in PCA results. Finally, setting the feature wavelengths as the input variables, the soft inde- pendent modeling of class analogy （SIMCA） classification method was used to identify the various types of plants. The experi- mental results of classifying cabbages and weeds showed that on the basis of the optimal pretreatment by a synthetic application of MSC and SG convotutional derivation with SG＇s parameters set as lrd order derivation, 3th degree polynomial and 51 smoot- hing points, 23 feature wavelengths were extracted in accordance with the top three principal components in PCA results. When SIMCA method was used for classification while the previously selected 23 feature wavelengths were set as the input variables, the classification rates of the modeling set and the prediction set were respectively up to 98. 6% and 100%.

关 键 词：主成分分析 特征波长 杂草识别 多元散射校正 聚类 

分 类 号：O657.3[理学—分析化学] S45[理学—化学]