Soil moisture, dielectric constant, and compaction degree are key physical parameters of soil, playing a great significance role in digital soil mapping (DSM), precision agriculture, and ecological research. A typical red clay farmland area in Kunming was selected as the study object. High-resolution multispectral remote sensing images were synchronously acquired. Field measurements of surface soil volumetric water content, dielectric constant, and compaction degree were conducted. Vegetation indices were calculated from the image data. Pearson correlation analysis and the variable importance in projection (VIP) algorithm were used to extract key spectral features. Multiple linear regression (MLR) and random forest (RF) models were constructed for quantitative inversion of soil parameters and for mapping. The results show that the VIP algorithm can effectively extract spectral features that significantly contribute to parameter prediction. The predictive performance of the RF model is found to be significantly better than that of the MLR model, with an R² value greater than 0.94. The spatial differentiation characteristics of moisture, dielectric constant, and compaction degree are clearly revealed by the spatial distribution maps. A high consistency is observed between the spatial distributions of moisture and dielectric constant, reflecting their physical correlation. More complex spatial heterogeneity is exhibited by compaction degree. The inversion model based on UAV multispectral imagery is shown to enable high-precision simultaneous inversion of red soil physical parameters, which is of great value for improving the information management of soil resources and supporting sustainable agriculture.