Quantitative Analysis of Total Nitrogen Content in Monoammonium Phosphate Fertilizer Using Visible-Near Infrared Spectroscopy and Least Squares Support Vector Machine

Abstract

А quantitative analysis method to determine the total nitrogen content in monoammonium phosphate (MAP) fertilizer using visible-near infrared (Vis-NIR) spectroscopy and least squares support vector machine (LS-SVM) is proposed. Sample set partitioning based on the joint x–y distance (SPXY) was used to select the calibration set. Fourteen spectral pre-processing methods were then employed to deal with the spectral data including Savitzky–Golay (SG) smoothing, fi rst derivative (D1) and second derivative (D2) with SG smoothing, multiplicative scatter correction (MSC), standard normal variate (SNV), wavelet, and combination thereof. Next, the LS-SVM model with radial basis function kernel was established with the best pre-processing method, and its performance was compared with that of partial least squares (PLS) model. The results revealed LS-SVM calibration with the discrete wavelet transform provided the best prediction for total nitrogen content in MAP fertilizer, yielding R2, root mean square error of prediction (RMSEP), and ratio of performance to deviation (RPD) values of 0.91, 0.101, and 3.34, respectively.