Construction of Non-Destructive Measurement Technology for the Quantity of Spherical Branchlets in Caulerpa lentillifera

Inefficient evaluation methods for germplasm traits severely hinder the development, utilization, and preservation of germplasm resources. Addressing the current lack of high-throughput measurement techniques for the traits of Caulerpa lentillifera, this study established a nondestructive quantification method for the spherical branchlets count of C. lentillifera based on CT imaging and the PointNet++ model. First, X-CT systems were employed to perform 3D imaging of 196 fronds, enabling rapid and nondestructive acquisition of their three-dimensional structural data. The obtained data were converted into 3D point cloud format, and a dataset was constructed by manually annotating the point cloud data of 196 fronds using the open-source software CloudCompare. The dataset was partitioned into training, validation, and test sets at a ratio of 7∶2∶1. Through model training based on the point cloud segmentation network PointNet++, along with multiple rounds of refined data processing and hyperparameter tuning, an optimal model was successfully developed. This model achieved high-throughput and accurate quantification of spherical branchlets on fronds. Finally, the model was tested on C. lentillifera accessions with densely clustered spherical branchlets, demonstrating high goodness-of-fit, with R², RMSE, and MAPE values of 0.927 6, 4.23, and 6.37%, respectively, confirming the feasibility of this technique. This study provides an effective tool for morphological trait measurement in C. lentillifera.