| The three-dimensional(3D)morphology of maize plants can respond to the influence of unique genotypes on the phenotype of maize plants,and the maize plant population’s threedimensional morphology can be used to further analyze the gene-phenotype correlation,which is of great research significance in the search for superior traits and effective selection and breeding.Advanced sensor technology is enabling methods for rapid and nondestructive maize population modeling and phenotypic analysis,which are maturing.Among them,laser point cloud technology has high-precision 3D scene reproduction capabilities,is non-destructive,and performs in real-time,making it an important study area in the agricultural sector.Due to the influence of planting density,laser scanning can be disturbed by occlusion between maize leaves,which seriously affects the reconstruction effect of point clouds.In addition,the roughness of real field roads and the non-rigid characteristics of maize additional difficulties for point cloud reconstruction of maize populations.In order to achieve large-scale,component-level three-dimensional point cloud reconstruction of maize populations in real field scenes,this study focuses on a self-built field mobile collection platform and a manually planted maize experimental field block as the research object,with3 D point cloud reconstruction as the research background,and conducts research on the acquisition of point cloud data for maize populations in real field environments,large-scale reconstruction,and data accuracy analysis.The primary research is described as follows:1)Design of the field mobile phenotype collection platform: To meet the requirements for acquiring maize population phenotype information,the selection of main sensors was carried out,and the working principles and parameter requirements of the sensors were analyzed.In addition,we designed a mobile phenotypic data collection platform for field use based on the planting method and growth status of maize plants.The platform can move autonomously or be remotely controlled to move between maize plant rows,and the sensing components can collect data driven by the collection mechanism.2)Single maize multi-frame point cloud registration: This article uses a batch registration method based on the Gaussian Mixture Model for 3D point cloud reconstruction of non-rigid and structurally complex maize plants.In addition,coordinate system calibration is added to the algorithm.Two sets of experiments,including reconstruction accuracy experiment and algorithm comparison experiment,were designed to validate the effectiveness and accuracy of the proposed algorithm.Both experiments showed that the Gaussian mixture model-based batch registration method used in this study has certain advantages in reconstructing sparse and less distinctive feature maize plant point clouds.3)Three-dimensional reconstruction of maize plant population point cloud: This thesis proposes a two-stage high-precision 3D point cloud reconstruction algorithm for large-scale maize plant reconstruction.The reconstruction of the plant is divided into two stages: local reconstruction and global reconstruction.The final reconstruction results show that the point cloud reconstruction of maize plants is effective and accurate,and can be used for further extraction and analysis of maize plant phenotypes.4)Maize point cloud stem and leaf segmentation and surface reconstruction: The skeleton of the reconstructed maize plant point cloud was extracted using the Laplacian operator,and a stem and leaf segmentation method based on the point cloud skeleton was proposed.The segmented leaves were reconstructed using the Alpha-shape algorithm.The reconstructed leaves better preserved the spatial morphology of the maize plant. |
PDF Full Text Request