Forecasting soil parameters based on NIR and SVM

By using SVM (Support Vector Machine) algorithm, the spectra of real-time soil samples were processed. Therefore the soil TN (Soil Total Nitrogen) and SOM (Soil Organic Matter) regression models were acquired and analysis on model accuracy which changed with the parameters of SVM was conducted. First, 150 soil samples were collected from the experimental field of China Agricultural University. Then the spectrum of each original soil sample was detected with the NIR (Near Infrared Reflectance) spectrograph. And the contents of TN and SOM in each sample were measured by using laboratory analysis methods. Finally, the regression models for TN and SOM were established based on SVM, and the evaluation about the extent was studied, which parameters of SVM could produce an effect on the predicting models. Research showed that it was befitting for the TN and SOM to established prediction models based on the whole spectra by using SVM. For soil TN, the calibration R2 reached to 0.9224 and validation R2 was 0.3667 based on the denoised spectra. For SOM, the calibration R2 reached to 0. 9179 and validation R2 was 0. 4152 based on the original NIR spectra. And the predicting accuracies for both TN model and SOM model were acceptable with low root-mean-square deviation of regression and validation.