Abstract: With the increasing exploitation of mineral resources, the issue of heavy metal pollution in agricultural soils surrounding mining areas has become more severe, and this trend poses significant threats to the quality of the soil, the security of the surrounding ecosystems, and the health of local human populations. Therefore, it is becoming more critical to monitor and assess heavy metal contamination in agricultural soils, especially those near mining areas. This is particularly important in complex, multi-pollutant scenarios where multiple contaminants interact and potentially have compounded adverse effects. Efficient inversion techniques are required to be developed for pollution assessment and prevention. In this study, we focus on an agricultural area near a gold mine located in Mojiang County, Yunnan Province, an area that has been severely impacted by mining activities. A total of 121 soil samples were collected from the study area, and laboratory hyperspectral data along with heavy metal content data (arsenic (As), chromium (Cr), copper (Cu), and nickel (Ni)) were obtained. The Nemerow pollution index method and the potential ecological risk index method were employed to assess and analyze heavy metal pollution in the study area. Simultaneously, the CARS-IRIV method was applied to select sensitive bands for soil heavy metals, and hyperspectral inversion models were then constructed to estimate the contents of the various heavy metals in the soil. Additionally, the spatial distribution of soil heavy metals in the study area was analyzed. The results indicated that the Nemerow pollution index method revealed severe pollution in the study area's soil, while the potential ecological risk index suggested a moderate level of ecological risk. First-order and second-order derivatives, standard normal variate, and reciprocal transformations were found to significantly enhance spectral responses. By combining the efficient band selection capability of the CARS algorithm with the variable refinement advantages of the IRIV method, sensitive bands were successfully selected, and the spectral response mechanisms of heavy metals were revealed. This method was demonstrated to be more effective in terms of the number of characteristic bands, computational runtime, and model inversion accuracy when compared to the CARS and IRIV methods used separately. Among the various inversion models, the back-propagation neural network (BPNN) was found to achieve the highest inversion accuracy for As, while the support vector machine (SVM) was identified as the optimal inversion model for Cr, Cu, and Ni. The R² values for the optimal inversion models of As, Cr, Cu, and Ni were 0.90, 0.93, 0.67, and 0.94, respectively. The root mean squared errors of external validation (RMSE) were 87.33, 142.63, 2.63, and 70.31 mg·kg⁻¹, respectively, and the relative percent differences (RPD) were 3.25, 3.84, 1.74, and 4.17, respectively. The results of spatial analysis indicated that high-value areas of heavy metals in the study area were primarily concentrated in the upper and lower parts, while low-value areas were distributed along the edges, showing an overall trend of gradual decrease from the interior to the exterior. This result reflects the spatial heterogeneity of pollution impacts caused by mining activities on the surrounding soils and reveals specific distribution patterns of soil heavy metal concentrations. This study combines traditional pollution assessment methods with hyperspectral inversion techniques to quantitatively evaluate heavy metal pollution in mountainous agricultural soils and to construct spatial distribution maps. It provides a reference for monitoring complex heavy metal pollution in agricultural soils near mining areas and offers data support for soil pollution assessment and remediation. This approach is expected to contribute to the improvement of soil pollution monitoring accuracy and is anticipated to provide valuable insights for related research and practical applications.