Biomass accumulation during the growth and development stages of rice is one of the key factors determining the rice yield.With the continuous development of UAV remote sensing technology in recent years，quantitative remote sensing inversion of rice biomass with UAV high-definition images，multispectral and hyperspectral remote sensing data has become an important technique to quickly obtain biomass information at the critical reproductive stages of rice.The UAV hyperspectral remote sensing platform was used to obtain the hyperspectral reflectance information of rice canopy at 400 to 1 000 nm to solve the poor universality and mechanism of inversion models for rice leaf biomass.The sensitivity of parameters for PROSAIL model was analyzed，and the sensitive wavelengths were extracted with continuous projection method according to the results of analyses.On this basis，the bald eagle algorithm （BES） was used to optimize the biomass parameters of the PROSAIL model to quickly and accurately retrieve leaf biomass inversion at the critical reproductive stages of rice through combining the PROSAIL crop radiation transmission model with rice hyperspectral data.The results showed that the improved Sobol method was used to analyze the global sensitivity of rice leaf biomass，and the sensitivity range was 700-1 000 nm.Six characteristic wavelengths of rice leaf biomass，namely 750，788，898，940，962 and 999 nm，were extracted with continuous projection method for the spectra at the sensitive interval.The PROSAIL-BES numerical optimization method was constructed by combining the PROSAIL model with the BES optimization algorithm.Using the spectral reflectance of rice characteristic wavelengths as the input of model，the parameters for the PROSAIL model were corrected by PROSAIL-BES numerical optimization method.The results of leaf biomass inversion showed that R² was 0.694 and RMSE was 0.002. It is indicated that the PROSAIL-BES numerical optimization method has better accuracy of inversion compared with the inversion results of traditional machine learning models，and has better practical value and application potential in the field of rice biomass inversion.