In order to quickly and accurately obtain aboveground biomass (AGB), potato was taken as research object, and the hyperspectral images of unmanned aerial vehicle (UAV) and measured aboveground biomass were obtained in budding period, tuber formation period, tuber growth period, starch accumulation period and mature period. Firstly, the canopy reflectance data of potato at each growth stage were extracted from hyperspectral image. Secondly, the 0~2 order differential (the interval was 0.2) of canopy spectral reflectance were calculated by fractional differential method. The correlation between canopy spectral data and aboveground biomass was analyzed, and the first 9 differential bands with high correlation were selected. Finally, the potato AGB estimation model of the whole, different varieties, densities and fertilization based on fractional differential spectrum was constructed and compared by using multiple linear regression (MLR), random forest (RF) and artificial neural network (ANN). The results showed that the order of the maximum absolute value of correlation coefficient in〖JP2〗 each growth stage was different, the maximum value in budding stage was 0.8 order differential (470nm), the maximum value in tuber formation stage was 1.8 order differential (710nm), the maximum value in tuber growth stage and starch accumulation stage was 1.6 order differential (718nm, 722nm and 766nm), and the maximum value in mature stage was 10 order differential (622nm). The correlation between hyperspectral fractional differential and AGB was higher than that of integer differential, and fractional differential can improve the estimation accuracy of potato AGB. Comparison and analysis of potato AGB estimation models at different growth periods, different varieties, densities, and fertilization were carried out. AGB estimation by three methods with 9 differential bands as independent variables all performed best in the tuber growth period. The model obtained by MLR under each condition had the highest accuracy and the strongest stability, followed by the RF model, and the ANN model had the worst performance. The accuracy of AGB model constructed by three methods in different growth stages were as follows: tuber growth period, tuber formation stage, starch accumulation period, budding stage and mature stage.