Semantic segmentation of sparse 3D point cloud based on geometrical features for trellis structured apple orchard

Orchard operations, such as mechanical pruning and spraying, are heavily affected by the tree architectures. Quantified inputs are necessary information for achieving precise control of these orchard operations, for example, cutting locations for mechanical pruning, and the variable spraying rates and covering areas for precision spraying. Even with planar orchard systems, trees are growing differently as a nature of biosystem. Therefore, to achieve precision orchard operations, it is essential to obtain the tree canopy structure information. A Three-dimensional (3D) Lidar based imaging system is developed in this study to estimate the shape of the canopy. The Lidar sensor was installed on an orchard utility vehicle to drive along the tree row in an apple orchard. A total of 1138 frames of point cloud data were acquired from 69 'Tall spindle' apple trees. An algorithm was developed in MATLAB environment to segment the trellis wires, support poles, and tree trunk in a point cloud image. The results indicated that the proposed algorithm achieved overall accuracy of 88.6%, 82.1%, and 94.7% for identifying the corresponding three objects in these images. Furthermore, the canopy density and depth maps were created. The outcomes from this study could provide baseline information for precision orchard operations such as mechanical pruning and precision spraying.