Using hyperspectral sensor to recognize monocotyledonous/dicotyledonous weeds and crops
Dongyan Zhang,Qin Zu,Wei Deng,Xiu Wang
DOI: https://doi.org/10.3969/j.issn.1007-2276.2013.z1.040
2013-01-01
Abstract:Leaf reflectance spectra of ten plant species, including maize, foxtail millet, setaria, goosegrass, crabgrass, 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 principal 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 distinguishing the 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 characteristic wavelength variable. The results show that the monocotyledonous and dicotyledonous weeds can be differentiated 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.