关键词: 井冈蜜柚, 果树识别, 无人机遥感, 植被指数, 估产模型

Abstract: In order to realize the rapid and accurate extraction of tree shape information (crown width and tree height) and yield prediction of Jinggang pomelo fruit tree based on the Unmanned Aerial Vehicle (UAV) imageries. By generating Digital Orthophoto map (DOM) based on the UAV imageries, four vegetation indexes were calculated and the accuracy of crown width extraction by threshold segmentation of four vegetation index was analyzed. The sensitive vegetation index and its best classification threshold was determined to complete the extraction of vegetation area and realize the crown width extraction. Then, based on the digital elevation model (DEM) generated by UAV imageries, the tree height of fruit trees was extracted. Finally, three models of crown width, tree height and crown width + tree height were used to predict yield. The results showed that the normalized difference index (NDI) had the highest accuracy in extracting crown width. The coefficients of determination R2 between the extracted EastWest crown width and the measured value was 0.917 2, the R2 between the NorthSouth crown width and the measured value was 0.823 6, and the R2 between the mean of crown width and the measured value was 0.892 8. When extracting tree height based on DEM, it also had a good effect. The R2 between the extracted tree height and the measured value is 0.863 3, and the root mean square error (RMSE) is 0.148 m. The yield prediction results showed that the R2 of using crown width + tree height to predict the number of fruits was 0.676, and the adjusted R2 was 0.638, which had the best prediction effect.

Key words: Jinggang pomelo, fruit trees detection, UAV remote sensing, vegetation index, yield estimation model