| More than 95% of coal mining in China is underground mining,and underground mining will form an underground goaf,which often causes irreversible harm to the surrounding ecological environment.In recent years,China has actively carried out the reclamation of cultivated land in mining areas,and the research on the evaluation of the soil quality of cultivated land around mining areas has gradually increased.Soil fertility indicators such as organic matter,total nitrogen,total phosphorus,total potassium are important factors to measure the quality of cultivated land,while traditional physical and chemical properties experiments take a long time and have a certain lag.Hyperspectral remote sensing technology has certain advantages in the rapid identification of nutrient content in the cultivated land around the mining area by virtue of extremely high spatial resolution and rich spectral information.In this study,three types of arable land in different subsidence stages around Wangzhuang Coal Mine in Changzhi,Shanxi Province were taken as study area.Samples were collected in the study area,and UAVs equipped with hyperspectral cameras were used for image acquisition,and the collected samples were measured indoors.To carry out four different forms of transformation,such as reciprocal,first-order differential,second-order differential,and multivariate scattering correction.By analyzing the correlation between the spectral reflectance and the measured physical and chemical indicators,the sensitive bands with high correlation coefficients are screened out,and the four nutrient contents are established using three models of Multiple Stepwise Regression,Partial Least Squares Regression and BP neural network Forecast model,evaluate the accuracy of model prediction results,select the optimal model to bring in aerial hyperspectral images for nutrient mapping,obtain nutrient distribution within the range of cultivated land,and analyze and discuss the nutrient content of cultivated land at different stages of subsidence.Provide reference for hyperspectral remote sensing in mining area reclamation and rapid nutrient identification.The study found that the sensitive band of organic matter content is mainly concentrated in the visible light 400-600 nm and the near-infrared 800-900 nm band;the sensitive band of total nitrogen content is concentrated in the near-infrared 800-900 nm band;the sensitive band of total phosphorus content is mainly550-700 nm band;the sensitive band of total phosphorus content is mainly 700-950 nm band.The accuracy of the Partial Least Squares Regression model and the BP Neural Network model of the spectral curve information processed by the multiple scattering correction is significantly better than the multiple stepwise regression model.The organic matter inversion model reaches 0.839 and 0.884,respectively,and the total phosphorus model results R~2 respectively.0.842 and 0.864,the total potassium modeling results R~2 reached 0.863 and 0.873,respectively,which can be used to identify nutrient content.The R~2 in the total nitrogen inversion model is 0.524 and 0.619 respectively,which can only be used to estimate the nutrient content. |
