Edge-Constrained Point Cloud Segmentation for Plants

Point clouds is the powerful and practical data representation for plant parameterization and plant growth monitoring. The most critical step in the precise extraction of plant parameters is the accurate segmentation of plant point clouds, especially in the edge regions, as this directly affects the accuracy of the parameters. This research introduces an edge extraction algorithm to obtain edge points from the point cloud and utilizes these edge points to construct an edge module to enhance the Stratified Transformer model. The proposed method can better constrain the edge and learn the edge features of the point cloud. In this research, we compare our method with various methods on Pheno4D and S3DIS datasets. Our method shows significant improvement in edge segmentation, such as stem-leaf edges and window-wall edges.