Completion of three-dimensional point cloud geometry of cotton leaves
Cotton is one of the main economic pillars in Xinjiang.It is of great significance to obtain phenotypic data of cotton plants automatically by using 3D reconstruction method for real-time monitoring of cotton growth.Due to the limitation of occlusion,light and resolution,the obtained point clouds of cotton plants are incomplete.Therefore,PF-Net model based on deep learning is used to com-plete the geometry of incomplete cotton leaf point cloud.Firstly,multi-view cotton images were collected to reconstruct the 3D point cloud,and then the cotton leaf point cloud dataset was constructed by clustering,farthest point sampling,normalization and data en-hancement,and the data was put into the PF-Net model to complete the geometric form completion of the cotton leaf point cloud.The experimental results show that the chamfering distance values of the point cloud model after completion of three different degrees of missing data(25%,50%and 75%)are 11.8×10-3,9.9×10-3 and 10.8×10-3,respectively,and the cotton leaf point cloud can keep the original geometry after completion.Therefore,the provision of a comprehensive 3D model of cotton plant leaves through the utiliza-tion of point cloud completion algorithms holds immense significance in furnishing complete geometric information for the automated measurement of cotton plant phenotypes.
PF-Net modelcotton leaf point cloudspoint cloud completion3D reconstruction
国家科技期刊平台
NSTL
万方数据
