<p>In order to reduce the influence of bare soil background on the original hyperspectral reflectance when covered by low-density vegetation and the influence of multiple scattering effect on the original hyperspectral reflectance when covered by high-density vegetation. This study took soybean leaf area index (LAI) and the hyperspectral data with soil background reflectance as the research objects, and performed first-order and second-order differential transformations on the hyperspectral data before and after the soil-vegetation separation (3SV) algorithm. Five different forms of spectral indexes were constructed using the correlation matrix method. On this basis, the first four spectral indexes with the highest correlation with soybean LAI were combined as input variables to construct three soybean LAI estimation models: Back propagation neural network (BP), support vector machine (SVM), and random forest (RF). The research results show that after correcting the soil background contribution, the spectral data is closer to the characteristics of pure vegetation contribution reflectance. the spectral index containing the difference form has a correlation of more than 0.650 in each order, and has the best correlation. The correlation of spectral indices after correcting the soil background contribution reflectance is significantly improved, which is 1%-15% higher than the correlation of the original spectral data. The optimal soybean LAI estimation model is an RF model based on the 3SV algorithm combined with Normalized Difference Vegetation Index (NDVI), Triangular Vegetation Index (TVI), Soil-Adjusted Vegetation Index (SAVI), and Chlorophyll Index (CI) under original, first-order, and second-order differential processing, with a total of 12 (combination 6) input variables. The R² correlation of the training set and the validation set is greater than 0.700, reaching an extremely significant level. Compared with the existing soybean LAI estimation methods, the soybean hyperspectral LAI estimation based on soil-vegetation separation algorithm and derivative spectrum corrects the influence of soil background contribution reflectance and multiple scattering effects compared with existing estimation methods such as optical vegetation index method and active remote sensing method, and has higher estimation accuracy. The research findings can provide a theoretical basis for improving the accuracy of soybean hyperspectral LAI estimation and offer practical guidance for sustainable practices in soybean production.</p>