Taking the hyperspectral data of cultivated land in typical mountainous areas of Guizhou province as the research object,a model for estimating soil organic matter(SOM)content in mountainous areas of Guizhou province was established by using spectral transformation method and machine learning.From August 2020 to March 2021,120 soil samples were collected from 13 counties and cities of Guizhou province,and the visible near-infrared spectral information of soil was detected.Five spectral data trans-formations(original spectra,first-order differential,sec-ond-order differential,first-order differential of reciprocal logarithm,continuum removal)and four types of models(partial least squares regression,support vector machine,random forest and BP neural network)were used to combine different soil organic matter content prediction models.The training samples and test samples were selected according to the ratio of 3 ∶ 1 to estimate the SOM content in mountain area.The correlation between the first-order differential data transformation and the SOM content in mountain area was high,and the highest correlation coefficient was-0.635.In the inversion model,the BP neural network model based on the first-order differential spectral transformation had the highest accuracy.The determination coefficients(R2)of the training set and the test set were 0.845 and 0.838,respectively.The root mean square error(RMSE)of the test set was 3.452.The relative a-nalysis error(RPD)reached 2.470,followed by RF,PLSR and SVM.The first-order differential method in spectral data transformation could greatly extract the SOM content information of mountain cultivated land.The BP neural network model was the optimal model for estimating the SOM content in mountain areas.The results of this study can provide theoretical reference for the monitoring of soil fertility and agricultural production in mountainous areas of Guizhou province.
维普
万方数据