Abstract: 【Objective】Citrus yield estimation is of great significance for realizing precision production management of orchards such as fertilizer and water variable application. In order to solve the difficulty of yield estimation of large-scale orchards, the unmanned aerial vehicle (UAV) oblique photography was employed to extract the canopy characteristics of citrus trees and to predict the yield of individual fruit tree. 【Method】In this study, a small quadrotor UAV equipped with a camera was used to collect oblique photographs of the Navelina navel orange orchard using a multi-oriented mode, flying at a height of 20 meters, with the heading overlap of 75% and the side overlap of 70%. The orchard point cloud data of the orchard, namely digital surface model (DSM), was obtained based on the 3D reconstruction of the orchard images, and the non-ground points were filtered to generate digital terrain model (DTM). The canopy height model (CHM) data were obtained by subtracting DTM and low vegetation from the DSM in turn, and the canopy characteristics, such as tree height, canopy projected area and canopy volume of individual trees, were extracted. The optimal yield estimation phenological period was determined based on the correlation between the production and the canopy characteristics in the three main physiological phenological periods, i.e. the second physiological drop (SPD) period, the turning stage (TU) period and the mature period. Then, the Levenberg-Marquardt (LM) algorithm was used to establish the yield estimation models of navel orange with different combinations of canopy characteristics extracted from the optimum phenological period. Finally, the obtained yield models were evaluated, analyzed and optimized to establish the optimal yield estimation model for Navelina navel oranges, and the model was validated by the validation set. 【Result】The determination coefficients (R²) between the measured tree height, canopy projected area and canopy volume based on UAV oblique photography and manual measurements were 0.99, 0.79 and 0.88, respectively, which shown that this method could be used to obtain canopy information with higher efficiency instead of manual measurements. By analyzing the relationship between yield and canopy information at different phenological periods, the SPD was selected as the optimal yield estimation phenological period. The optimal yield estimation model of Navelina navel orange based on canopy volume in SPD was developed using LM algorithm with R²=0.7961 and RMSE=11.85. The absolute mean relative error of the sample plants in the validation set was 4.43%. 【Conclusion】The results showed that it was feasible to collect the image of Nevelina navel orange orchard in the SPD by UAV oblique photography technology, to extract the individual canopy characteristics from the orchard point cloud through three-dimensional reconstruction, to establish the yield estimation technology of Nevelina navel orange using LM algorithm.

Key words: yield estimation, Navel orange, canopy characteristics,  , oblique photography, unmanned aerial vehicle (UAV)