The rapid development of hyperspectral imaging technology has increased the use of domestic hyperspectral images for the inversion of soil parameters in a wide range. However, the accuracy needs to be improved. Therefore, by considering the Daxigou mining area in Shaanxi Province and taking GF-5 hyperspectral satellite images and measured soil samples as data sources, we proposed an XGBoost inversion model based on genetic algorithm feature selection (GA-XGBoost). First, the preprocessed image data were transformed by continuum removal and logarithm of spectral reciprocal. Then, the Monte Carlo cross-validation method was used to remove abnormal soil samples. Finally, The XGBoost heavy metal content inversion models based on correlation coefficient and genetic algorithm feature selection were established respectively. The results show that the performance of the proposed GA-XGBoost model significantly improved compared with the XGBoost model based on correlation coefficient feature selection under the same spectral transformation. Furthermore, the GA-XGBoost model based on continuum removal transformation has the best inversion accuracy, with a root mean square error of 4.85 mg·kg^-1, goodness fit of 0.84, and relative prediction error of 2.0. The inversion results of the spatial distribution of soil Cu content in the study area using the model show that the surrounding of the mining area and both sides of the road are seriously polluted by Cu, which is consistent with the field survey results.