A Method for Sesame ( Sesamum indicum L.) Organ Segmentation and Phenotypic Parameter Extraction Based on CAVF-PointNet+.

Abstract

Efficient and non-destructive extraction of organ-level phenotypic parameters of sesame ( Sesamum indicum L.) plants is a key bottleneck in current sesame phenotyping research. To address this issue, this study proposes a method for organ segmentation and phenotypic parameter extraction based on CAVF-PointNet++ and geometric clustering. First, this method constructs a high-precision 3D point cloud using multi-view RGB image sequences. Based on the PointNet++ model, a CAVF-PointNet++ model is designed to perform feature learning on point cloud data and realize the automatic segmentation of stems, petioles, and leaves. Meanwhile, different leaves are segmented using curvature-density clustering technology. Based on the results of segmentation, this study extracted a total of six organ-level phenotypic parameters, including plant height, stem diameter, leaf length, leaf width, leaf angle, and leaf area. The experimental results show that in the segmentation tasks of stems, petioles, and leaves, the overall accuracy of CAVF-PointNet++ reaches 96.93%, and the mean intersection over union is 82.56%, which are 1.72% and 3.64% higher than those of PointNet++, demonstrating excellent segmentation performance. Compared with the results of manual segmentation of different leaves, the proposed clustering method achieves high levels in terms of precision, recall, and F1-score, and the segmentation results are highly consistent. In terms of phenotypic parameter measurement, the coefficients of determination between manual measurement values and algorithmic measurement values are 0.984, 0.926, 0.962, 0.942, 0.914, and 0.984 in sequence, with root-mean-square errors of 5.9 cm, 1.24 mm, 1.9 cm, 1.2 cm, 3.5°, and 6.22 cm 2 , respectively. The measurement results of the proposed method show a strong correlation with the actual values, providing strong technical support for sesame phenotyping research and precision agriculture. It is expected to provide reference and support for the automated 3D phenotypic analysis of other crops in the future.

Plant phenotyping relevance

3D画像・点群分割と幾何クラスタリングにより、ゴマの器官分割および6種類の表現型形質抽出法を開発し、手動測定との精度検証も行っているため、方法が中心的である。

Code and data availability

The paper describes sesame point cloud data, CAVF-PointNet++ code/models, and phenotypic measurements, but no public deposit or URL is provided; the Data Availability Statement says data are available only upon reasonable request to the corresponding author.

No evidence-backed public reproduction asset is currently recorded.