【Objective】Hyperspectral technology provides a novel solution for the rapid and accurate monitoring of heavy metal content in soils. However, models developed using laboratory spectra often have limited generalizability in practical applications. Additionally, directly estimating soil heavy metal concentrations from remote sensing imagery is often hampered by factors such as weather conditions and surface environment at the time of image acquisition, which leads to reduced model accuracy and limits the ability to accurately reflect the spatial distribution of heavy metals in the study area.【Method】In this study, a tailings area in Kuanshan Town, Huize County, Yunnan Province, was selected as the research site. A total of 56 surface soil samples were collected, and both ground-based and image-based hyperspectral reflectance, as well as Pb and Zn concentrations, were obtained. First, the Direct Standardization (DS) algorithm, combined with laboratory spectra, was used to correct the GF-5 imagery. Subsequently, the Box-Cox transformation was applied to normalize the skewed distributions of Pb and Zn concentrations. Then, fractional order derivative (FOD) was performed on the corrected spectra, and the Boruta algorithm was used to identify informative spectral bands. Finally, Random Forest and XGBoost models were developed for the inversion of heavy metal concentrations.【Result】The results indicate that the DS algorithm effectively mitigated the influence of soil particle size and moisture content on image spectra. The Box-Cox transformation resolved the skewness distribution problem of Pb and Zn content. FOD effectively enhanced detailed spectral features, and the optimal feature band combinations selected by the Boruta algorithm significantly improved the inversion accuracy. Furthermore, the XGBoost demonstrated superior predictive performance in handling complex feature interactions and nonlinear regression problems. 【Conclusion】The optimal inversion model for Pb content in the tailings area was a 0.8 Order-Boruta-XGBoost model, while for Zn content it was the 1.6 Order-Boruta-XGBoost model. Both models exhibited good robustness. This study provides a reliable reference method for using hyperspectral technology to invert Pb and Zn content in mining area soils.