Hyperspectral retrieval model of soil organic matter content based on particle swarm optimization-support vector machines

To monitor the soil organic matter in the reclamation area of coal mines, the relationship between soil organic matter content and soil spectra in the reclamation area of coal mines was studied, and a quantitative retrieval model was established and validated in order to implement the organic matter content detection in this paper. After the preprocessing of the original spectral, the correlation of the organic matter content and reflectance spectra was analyzed, and 450 nm, 500 nm, 650 nm, 770 nm, 1460 nm and 2140 nm wavelength were extracted as feature bands. Using the multiple linear regression (MLR), partial least squares regression (PLSR) and particle swarm optimization support vector machine regression (PSO-SVM) methods, the hyperspectral quantitative retrieval models for soil organic matter content were built. The results show the coefficient of determination (R²) of MLR, PLSR and PSO-SVM were 0.79, 0.83 and 0.85 respectively, and the root mean square error of prediction (RMSEP) were 5.26, 4.93 and 4.76 respectively. The results demonstrate that the stability and predictive ability of PSO-SVM model are better than those of the MLR and PLSR model.