Method of Kiwifruit Orchard Leaf Density Based on Point Cloud Semantic Segmentation
This study proposed a LiDAR-based kiwifruit orchard leaf density measurement method to provide more precise guidance for kiwifruit orchard canopy spraying.Firstly,ten kiwifruit canopies with different densities were scanned using lidar to create a point cloud dataset.RandLA-Net neural network model and six-fold cross-validation method was used to semantically segment leaves,branches and T-frame point clouds of kiwifruit canopy;Then the surface area of the canopy point cloud containing only leaf information was calculated,compared with the measured canopy leaf attributes(leaf area,the number of leaf)after artificial defoliation by regression analysis.The relationship between the surface area of the leaf point cloud and the attributes of real canopy leaf was obtained.Finally,the 3 m×3 m kiwifruit canopy area was divided in-to 400 small grid areas of 225 cm2 to generate the canopy leaf density map.The results show that the RandLA-Net net-work model can effectively segment the branches,leaves and T-frame point cloud of the kiwifruit canopy.The average OA of the model reaches over 92％,and the average mIoU is 81.4％.A high correlation(R=0.78)was obtained in the re-gression analysis of the real leaf area and the leaf point cloud surface area.Relationship is y=1.491 x+8315.For each ki-wifruit canopy point cloud data,the canopy leaf density map was generated by predicting the real leaf area and calculating the canopy volume of the grid area through the Alpha-shape algorithm.All indicators of the developed kiwifruit canopy leaf density measurement method based on point cloud semantic segmentation meet the expected requirements,and it can provide more accurate guidance for orchard canopy spraying.
万方数据
维普
