Abstract: To achieve accurate inversion of soil organic carbon (SOC) content in tobacco-growing soil under limited sensitive bands, this study utilized 106 soil samples from three counties in Qujing City as research subjects. An inversion model was used to predict SOC content in tobacco-growing soil based on the combination of sensitive bands and their wavelet features. A model for inverting SOC content based on continuous wavelet analysis was established using a hyperspectral imaging device. Results indicated that the regression model established with the hyperspectral data optimized by Savitzky-Golay filtering showed an average increase of 44% in the coefficient of determination compared to that before the optimization. Using the correlation coefficient method, it was found that 733 nm was the optimal spectral band for SOC in tobacco planting, with a coefficient of determination (R²) value of 0.75. By applying continuous wavelet transform to the optimized hyperspectral reflectance data, it was observed that the inversion accuracy increased with wavelet features at scales from 1 to 10 under the 733 nm band, stabilizing after scale 6, where the R² value exceeded 0.8. Four machine learning models were developed using the optimized 733 nm spectral data combined with its wavelet features extracted at scales 6 to 10 as input variables. Among these, the Random Forest model produced the best prediction results, with R² values of 0.88, root mean square errors of 1.65 g/kg, relative prediction deviations of 3, and running time of 0.21 s, respectively. Combining sensitive bands with their own high-scale wavelet features enables rapid, non-destructive, and accurate inversion of SOC content in tobacco-growing soil. These results provide a reference for the development of algorithms for SOC sensors.