Shajiang black soil faces the risk of continuous degradation, and the decrease of soil organic matter content is one of the main manifestations of soil degradation. The soil organic matter (SOM) content and corresponding spectral data from the Shajiang black soil in Shangshui County served as the primary data sources for this study. The spectrum were processed using continuous wavelet transform (CWT) and the competitive adaptive reweighting algorithm (CARS) to extract feature coefficients. Partial least squares regression (PLSR), random forest (RF), K-Nearest Neighbor (KNN), long-short memory network (LSTM) and convolutional neural network (CNN) algorithms were employed to develop estimation models for soil organic matter content. The results indicated the following: (1) Continuous wavelet processing significantly enhanced the correlation coefficients between the original spectrum and the first derivative spectrum with soil organic matter, improving by 54.11% and 5.28%, respectively; the original spectrum exhibited a greater degree of improvement than the first derivative spectrum. (2) The CARS algorithm compressed the feature coefficients of each decomposition layer except FD10 to below 6% of the total number of coefficients, substantially reducing variable dimensions and model complexity. (3) The model constructed using the third layer coefficient from the FD spectrum decomposed by CWT (CWTFD3-CARS-CNN) yielded the most effective results, with a modeling determination coefficient of 0.95. The validation determination coefficient was 0.97. which results was better than the other four machine learning methods and models constructed using spectral indices. Thus, the combination of CWT-CARS-CNN effectively enhances the modeling accuracy of soil organic matter and serves as a valuable reference for constructing accurate estimation models.
Niu et al. (Sat,) studied this question.