Apple tree canopy leaf spatial location automated extraction based on point cloud data

Abstract A fine structure inside the canopy of the apple tree determines the light distribution and is one of the important factors affecting the quality and yield of apples. Leaf spatial location (LSL) is the spatial co-ordinates of the petiole–midrib junction point. In this paper, a Trimble TX8 was used to obtain 3D point clouds of the canopy in the flowering, leaf growth stage and stable growth stages of apple tree as the research object. An LSL extraction approach using density-based spatial clustering of applications with noise (DBSCAN) and layers K-mean and median (L-KaM) methods was proposed. Firstly, the DBSCAN clustering method based on adaptive parameters is used to separate single leaves from branches. Secondly, the same point cloud is sliced into layers, and the L-KaM method is used to fit the branch center line. Finally, the Euclidean distance of each point between the single leaf and the center line is determined, and the point with the smallest Euclidean distance is the LSL point. Field experiments show that the DBSCAN-L-KaM method proposed in this study is suitable for LSL extraction during the leaf growth and stable growth stages. The maximum Euclidean distance between the manual measurements’ actual LSL value and the values automatically extracted (ELD_MaA) was less than 9 mm, and the average ELD_MaA was 1.41 mm. The average extraction rate of tall spindle training system apple trees and free spindle training system apple trees were 89.90% and 51.75%, respectively. Using the method in this paper, we obtain the whole apple tree (3-year-old) LSL value in about 2.5–3 h. The method provides theoretical basis and systemic support for the analysis of the light-intensity distribution of apple trees and the internal structure details of fruit trees.