An automated skeleton extraction method for 3D point-cloud phenotyping of Schima Superba seedlings

Aiming to address the issues of low efficiency and large errors in the manual measurement process of phenotypic parameters in Schima Superba seedlings, an automated non-destructive method for acquiring phenotypic parameters based on three-dimensional point clouds is proposed, which includes the main steps of alignment, skeleton extraction, and automatic phenotypic calculation. Aiming to overcome the technical challenges of stem and leaf separation in Schima Superba, a density-weighted voxel centroid method is proposed to extract skeleton points, combined with minimum spanning tree (MST) and principal component analysis (PCA) techniques to accurately identify the stem skeleton point cloud, effectively addressing the problem of stem-leaf separation. The separation process encounters difficulties at the stem-leaf junction, resulting in suboptimal separation accuracy. An improved K-means++ algorithm is proposed to initially estimate the number of adhering leaves based on coarse segmentation, followed by fine segmentation to achieve higher precision in leaf segmentation, effectively improving the accuracy and efficiency of the segmentation process. Following the completion of stem and leaf segmentation, a fully automated phenotypic characterization method based on the segmented point cloud is proposed for the first time. The method automatically outputs relevant phenotypic parameters, including plant height, stem length, stem diameter, and leaf area. The predicted correlation coefficients for the experimental phenotypes were 0.994, 0.992, 0.938, and 0.873, meeting the requirements for on-site measurement of phenotypic parameters in Schima Superba and providing strong technical support for plantation management and cultivar improvement.