Abstract: The tensile cracks caused by coal mining destroy the soil structure and affect the soil quality. Using hyperspectral technology to monitor the important components of the soil in the fracture zone is of great significance for accurately restoring the soil quality in the fracture zone and improving agricultural production. In this study, 90 groups of soil samples are collected in the tensile fracture area of coal mining in Zhuzhuang coal mine in Huaibei, and the spectra of soil samples are measured indoors. The reflectance values are correlated with the measured organic matter content, and the characteristic bands sensitive to organic matter are selected. Partial least squares and BP neural network are used for modeling, and the accuracy of each model is evaluated. The results show that the inversion effect of this study is ideal, and the first-order differential and partial least squares model (FD-PLSR) has the best modeling effect. The R² of FD-PLSR model are 0.876 1、0.845 9, and the RMSE of FD-PLSR model are 0.497 2、0.680 6, respectively. The research can provide some technical support for the monitoring of soil organic matter content in tension fracture zone of coal mining.

Key words: coal mining tensile cracks, soil organic matter, hyperspectral remote sensing, partial least squares, BP neural network