Abstract

Analyzing phenotypic traits, diagnosing diseases, and anticipating yields are just a few of the many applications of plant organ segmentation in precision agriculture and plant phenotyping. Because plant structures are so varied and intricate, traditional methods have a hard time keeping up. By combining several data sources, such as images and point clouds, graph neural networks (GNNs) have completely altered crop organ segmentation. In this research, we present a new method for rethinking plant organ segmentation by using the powerful features of GNNs. The approach takes a look at point clouds of plant shoots and uses graph representations to capture deep structural intricacies and intricate spatial interactions. One important novelty is the use of betweenness centrality for weighting edges and vertex, which guarantees that the segmentation results are biologically significant. The model's ability to understand geometric and topological details is improved, leading to more accurate segmentation through dynamic computing and continuous updates of Forman-Ricci curvatures. This all-encompassing work opens new doors for plant phenotyping research by improving the accuracy of organ segmentation and facilitating the integration of complicated mathematical theories into biological analysis.