Optimization of a Soil Particle Content Prediction Model Based on a Combined Spectral Index and Successive Projections Algorithm Using Vis-NIR Spectroscopy

This study explores the problem of fast estimation of soil particle content by visible-near-infrared (vis-NIR) spectroscopy. Four spectral pre-processing methods were used, and it was found that the multiplicative scatter correction (MSC) preprocessing transformation can maintain the original spectral characteristics and effectively eliminate the influence of scattering, with the best effect. Compared with the spectral index (SI) established by the original spectral data, the SI established by the MSC greatly increases its correlation and sensitivity with the soil particle content. The partial least squares (PLS) and random forest (RF) models were optimized using the successive projections algorithm (SPA), and it was found that the model complexity and calculation amount were greatly reduced, and the model accuracy was improved. The reserved samples were verified, and a good prediction effect was achieved. Our results show that the combination of spectral index and successive projection algorithm can be used as an effective means for rapid prediction of soil particle content.