It is essential to estimate above-ground biomass (AGB) quickly and accurately, and AGB is an important indicator of crop growth evaluation and yield prediction. Due to the saturation of AGB in multiple growth periods estimated by traditional vegetation indexes (VIs). Therefore, the study attempts to use VIs combined high-frequency information extracted by image wavelet decomposition (IWD) based on discrete wavelet transform (DWT) technology and wavelet coefficients extracted by continuous wavelet transform (CWT) technology, explore the estimation capabilities of VIs, VIs + IWD and VIs + CWT for AGB. Firstly, the hyperspectral and digital images of the unmanned aerial vehicle (UAV) and measured AGB were acquired during the potato budding stage, tuber formation stage, tuber growth stage, and starch accumulation stage. Secondly, three high-frequency information were extractedby using digital images through IWD technology, wavelet coefficients were extracted by using hyperspectral reflectance through CWT technology and six hyperspectral vegetation indexes were constructed. Then, the correlation between vegetation index, high-frequency information and wavelet coefficients and AGB was analyzed, and the top 10 bands with high absolute values of correlation coefficients at different scales were selected. Finally, the partial least square regression (PLSR) was used to construct and compare AGB estimation models with VIs, VIs + IWD and VIs + CWT. The results showed that: (1) 6 vegetation indexes, 3 high-frequency information and 10 wavelet coefficients selected in each growth period were significantly correlated with AGB, and the correlation decreased after increased in the whole growth period, in which the wavelet coefficients was the highest, the nextwas high frequency information, and the vegetation index was the lowest. (2) The three estimation models of each growth period were compared and analyzed, the estimation effect of VIs + CWT was the best, and that of VIs was the worst, indicating that the model based on wavelet analysis has wide applicability and strong stability. (3) The AGB estimation models constructed by PLSR method with three variables in each growth period reached the highest accuracy in the tuber growth period (VIs: modeling R-2 = 0.70, RMSE = 98.88 kg . hm(-2), NRMSE = 11.63%; VIs + IWD: modeling R-2 = 0.78, RMSE = 86.45 kg . hm(-2), NRMSE = 10.17%; VIs + CWT modeling R-2 = 0.85, RMSE = 74.25 kg . hm(-2), NRMSE = 9.27%). The PLSR method through VIs combined with IWD and CWT technology were used to improve the accuracy of AGB estimation, which provide a reliable reference for agricultural guidance and management.
