Abstract: The photosynthetic capacity of a crop plays a decisive role in its yield. The quantum yield (QY) of soybean flag leaf is also very important to assess photosynthetic efficiency. High-throughput QY inversion can rapidly, non-destructively, and efficiently monitor the physicochemical changes in the soybean flag leaf during photosynthesis using a UAV multispectral. The objective of this study was to investigate the correlation between the vegetation indices and QY, and then to invert the QY values using the highly correlated vegetation indices. The inversion models were also constructed for high accuracy with the multiple versus single vegetation indices. Eight vegetation indices were calculated, including the Normalized Difference Vegetation Index (NDVI), green NDVI (GNDVI), Enhanced Vegetation Index (EVI), Leaf Chlorophyll Index (LCI), Soil Adjusted Vegetation Index (SAVI), green SAVI (GSAVI), Optimized SAVI (OSAVI), and Normalized Difference Red Edge (NDRE). The high throughput of the spectral collection was used in the five bands of the soybean canopy. Pearson correlation coefficients were also utilized to determine the correlations between the single vegetation indices and QY values of the soybean flag leaf. Six vegetation indices with high correlations were then selected as NDVI, GNVDI, LCI, SAVI, OSAVI, and NDRE. The single-index inversion models of the six highly correlated vegetation indices were constructed using five models, including the Support Vector Regression (SVR), Partial Least Squares Regression (PLSR), Random Forest (RF), Adaptive Boosting (AdaBoost), and Gradient Boosted Decision Tree (GBDT). The simulation was then evaluated using three evaluation indexes, including the coefficient of determination (R2), root-mean-square deviation (RMSE), and relative percent difference (RPD). Five models were evaluated to select the vegetation indices with the better inversion for the QY of the soybean flag leaf. The single vegetation index modelling showed that the NDVI, GNDVI, LCI, and NDRE performed better inversion for the QY of the soybean flag leaf. A comprehensive analysis was made to verify the evaluation indexes of each modelling. The SVR, AdaBoost, and GBDT modelling were more suitable for this case, compared with the PLSR and RF. The integrated learning-based AdaBoost improved the accuracy and robustness of the model, particularly with the validation set R2 of 0.982, RMSE of 0.089, and RPD of 7.29, indicating the standard of a Class A model, compared with the traditional inversion SVR. Among them, the NDVI presented the strongest correlation with the QY values of soybean flag leaf, especially with a Pearson correlation coefficient of 0.956. The R2 values were all greater than 0.7 for the fitting between the NDVI, GNDVI, NDRE, LCI, and QY values, indicating being suitable for the QY inversion model. The R2 values in the test set were 0.959, 0.954, 0.962, 0.982, and 0.967, respectively, using the traditional inversion SVR, PLSR, integrated learning-based random forest, AdaBoost, and GBDT algorithms. Therefore, the learning algorithm based on integrated learning can be expected to further improve the accuracy and robustness of the inversion model. Compared with the inversion models with the single and multiple vegetation indices, the combination of multiple vegetation indices can be used to improve the prediction accuracy of the inversion model. More importantly, the R2 and RPD of the AdaBoost model were improved by 0.149 and 4.645, respectively, while the RMSE was reduced by 0.306. The multiple vegetation indices and the AdaBoost can be used to construct the much more effective multispectral inversion model for the photosynthesis QY of the soybean flag leaf. The finding can also provide an advanced method to assess the high-throughput photosynthetic efficiency using remote sensing.