Use of NIR spectroscopy and LS-SVM model for the discrimination of varieties of soil

Abstract

Near-infrared spectroscopy (NIR) combined with chemomatrics method was employed in this paper. The objectives were to investigate the feasibility of using VIS/SWNIR spectroscopy to discriminate soil of different kinds, and to validate the performance of selected sensitive ebands. Spectrums in the NIR regions (4003.563-12496.67 cm-1) were collected from 380 samples and the data was expressed as absorbance, the logarithm of the reciprocal of reflectance(log 1/R).240 samples were randomly collected as modeling, and the others were used to check the model's performance. Principal components analysis (PCA) tested the clustering of these four kinds of soil, which made a qualitative analysis for the discrimination, but for the sake of speedup the calculating time, the mathematics analysis of support vector machine classification and 10-folds cross-validation were used to model, and Based on SVM, the recognition ratio of 98% (0.2 as threshold value) was obtained. Compared with the PLS result of 79% (0.2 as threshold value), SVM classification is apparently more effective. At the same time, the sensitive bands of soil varieties were caculated,in which we found the 801-972 nm can predict well with the result of 90% from LS-SVM.The prediction results of 99% indicated that the NIR can mainly represent the characteristics of soil of different kind based on SVM model. © 2009 Springer Science+Business Media, LLC.