摘要
利用地物光谱仪Fieldspec FR2500采集玉米、谷子、狗尾草、牛筋草、马唐、葎草、圆叶牵牛、裂叶牵牛、马齿苋和小藜等10种植物的叶片光谱反射率,将这10种植物按照作物、单子叶杂草和双子叶杂草分成三类,在Unscrambler软件中对光谱数据做主成分分析,观察三类样本的聚类情况;在此基础上,根据波长变量对于主成分的载荷值,提取对三类植物识别敏感的特征波长;在SAS软件中,以分类变量和特征波长变量作为输入变量,利用DISCRIM过程进行非参数判别分析。结果表明,利用552、647、730、765、962、1093、1325、1410、1660 nm等9个特征波长组合可有效区分作物中单/双子叶杂草。其中,针对单子叶杂草的建模集和预测集的识别正确率分别达到90%和85%,双子叶杂草的建模集和预测集的识别正确率均达到100%,整体样本的建模集和预测集的识别正确率分别达到96.1%和95.3%。
Leaf reflectance spectra of ten plant species, including maize, foxtail millet, setafia, goosegrass, crab- grass, humulus, ipomoea, pharbitis, purslane and pigweed, were collected using the ASD spectrometer. Those plants were classified into three categories: crops, monocotyledonous weeds and dicotyledonous weeds. The princi- pal component analysis (PCA) was used to process the spectral data in the unscramble to observe the clustering patterns of the three categories. Subsequently, the characteristic wavelengths available, sensitive to distinguishingthe three categories, were extracted according to the loading weights of the wavelength variables for each principal component (PC). Finally, the nonparametric discriminant analysis was conducted by the DISCRIM procedure u- sing SAS software, in which two input variables were required including the categorical variable and the character- istic wavelength variable. The results show that the monocotyledonous and dicotyledonous weeds can be differentia- ted effectively from crops utilizing the combination of nine characteristic wavelengths : 552, 647,730, 765,962, 1093, 1325, 1410, 1660 nm. The recognition accuracies of the modeling and prediction are 99% and 85% for the monocotyledonous weeds, while both of them reached 100% for the dicotyledonous weeds, and they reached 96.1% and 95.3% in the whole sample.
出处
《红外与激光工程》
EI
CSCD
北大核心
2013年第S01期208-213,共6页
Infrared and Laser Engineering
基金
中国博士后科学基金特别资助项目(2013T60189)
公益性行业(农业)科研专项(20130331)
安徽省高等学校省级自然科学研究(KJ2013A026)
安徽省自然科学基金(1308085QC58)
安徽大学博士科研启动经费
北京市农林科学院博士后基金
关键词
玉米
单
双子叶杂草
主成分分析
判别分析
特征波长
corn
monocotyledonous/dicotyledonous weed
principal component analysis
discriminant analysis
characteristic wavelength