The large-scale and high-intensity mining of coal leads to the formation of cracks in the ground, which destroy the structure of the soil and affect the soil quality. To quickly estimate the content of soil organic matter (SOM) in the coal stretching fissure area, soil samples were collected from the fissure area at Zhuzhuang coal mine, Huaibei City, China. The spectrum and SOM content of the soil samples were then determined. Inverse log reflectance (LR), first order differential reflectance (FD), and continuum removal(CR) were performed on the original spectrum. Then, the difference index, ratio index, and normalized difference index of any two band combinations were calculated. The Pearson correlation coefficient (PCC) was combined with the maximum relevance minimum redundancy (mRMR) algorithm to extract one-dimensional spectral bands and two-dimensional spectral indices, respectively. In the end, the two algorithms, PLSR and eXtreme Gradient Boosting (XGBoost), were used to construct a SOM content estimation model for the mining fissure area. The accuracy of the model was then tested and evaluated. The results show that: (1) The high intensity mining of coal leads to the formation of cracks in the ground, which destroys the structure of the soil and affects the soil quality. It also accelerates the loss of SOM and fine soil particles in the soil, and the coefficient of variation of SOM in the study area reaches 61.32%; (2)Regardless of one-dimensional spectral band or spectral index, the model based on FD spectrum has the highest accuracy; (2) The correlation between the two-dimensional spectral index and SOM content is significantly better than that of the one-dimensional spectral band, and the prediction model based on the spectral index has higher prediction accuracy; (3) Compared with the one-dimensional spectral band, the difference index (DI), ratio index (RI), and normalized difference index (NDI) have a stronger correlation with SOM, and the FD-DI index has the highest correlation, with a correlation coefficient of 0.88; (4) The accuracy of XGBoost based model is better than PLSR model, among which FD-NDI-XGBoost model has the highest accuracy, and its R², RMSE and RPD are 0.83, 0.49 mg·kg^-1 and 2.44, respectively. The experimental results can provide a technical reference for the hyperspectral estimation of SOM content in the coal mining tensile fracture area.