Hyperspectral Quantitative Inversion of Soil Organic Matter Based on IPLS-UVE Band Optimization

Affected by the local natural environment, sandy soil has the characteristics of low content and poor stability. Reasonably increasing the utilization rate of this kind of soil is one of the effective ways to improve crop planting. Compared with traditional measurement methods, hyperspectral quantitative inversion modeling can quickly and accurately predict soil organic matter (SOM)content. After preprocessing the hyperspectral information by Savitzky-Golay and first-order differential processing (SG-1D), this method uses interval partial least squares (IPLS) to model the whole waveband sequentially, and select the characteristic band between 600-800nm for the first time, then combined with the uninformed variable elimination method (UVE) to better optimize the feature vector. The results verify that the IPLS-UVE model has a great improvement in the prediction accuracy of the inversion model under the band selection. The number of bands is reduced from the original 204 to 30, and the prediction set determination coefficient R
²
=0.917, and the prediction of root mean square error RMSE=1.748 g kg
⁻¹
, and the predicted relative analysis error RPD=3.474. The results illustrate that this method can optimize the selection of soil characteristic bands and thus effectively improve its SOM prediction.