Research On The Rapeseed Data Processing Methods Based On Laser Point Cloud

Rapeseed is China's largest oilseed crop and main cash crop,with high economic benefits and application potential.The establishment of plant digital model based on 3D laser scanning technology,plant phenotypic measurement and 3D reconstruction are research hotspots in the direction of digital agriculture.The plant point cloud data acquired by 3D laser scanning technology not only has accurate plant spatial structure and high-precision surface information,but also contains a lot of noise in the data due to the influence of environmental factors,human factors and scanning equipment factors in the scanning process.How to characterize the obtained plant point cloud data,how to de-noising the point cloud and how to extract the phenotypic information are the key issues in the study of plant phenotype measurement,which are also the premise and basis of point cloud processing such as point cloud segmentation,3D reconstruction and 3D visualization.In this paper,3D laser scanning technology is applied to the digital research of rapeseed plants,and the feature description,point cloud denoising method and phenotypic information extraction of the obtained rapeseed point cloud data are studied.The main work and results of this paper are as follows:?1?Canola point cloud data acquisition.In this paper,a collection scheme for obtaining point cloud data of field rapeseed using ground 3D laser scanner is designed,and 3D point cloud data of field rapeseed plants are collected.?3?Construct a feature description of the point cloud.The rule vector features,curvature features,covariance matrix eigenvalues and combination features were constructed for the rapeseed point cloud data.Under different Feature descriptions,the stems,leaves and veins of rapeseed have different characteristics:The normal vector of seedling leaves has multiple vertical directions,the normal vector of flowering stems has multiple horizontal directions,Setting the range of the neighborhood radius r=1.2 Searching for the calculated curvature can well reflect the leaf vein characteristics of the seedling stage,The smaller the eigenvalue of the covariance matrix ?₂ solved by the local nearest neighbor,the closer the point is to the edge.The larger the eigenvalue ?₃ of the covariance matrix solved by the local nearest neighbor is,the higher its curvature value is;?2?Point cloud data denoising research.This paper analyzes the noise characteristics of field rapeseed crops and summarizes the noise points of field rapeseed point clouds.A point cloud denoising scheme for rapeseed crops in the field was designed:For the field rapeseed point cloud data using conditional filtering,statistical filtering and region growing filtering method to filter out different types of noise point clouds in turn,two thresholds in statistical filtering are determined experimentally,Through the experiment,the optimal range k of the neighbor point k?[6,12]is obtained,and the optimal range of the standard deviation multiple is a?[0.01,0.05].At the same time,the region growing algorithm is improved,and two conditions of combining color and height are used as classification criteria.The experimentally determined the color difference threshold P=5 of the regional growth filter and the inter-cluster color difference threshold G=5,which can well remove the field rapeseed weeds and land point clouds.?4?Study on the contour contour and leaf vein extraction of rapeseed point cloud.The algorithm of the boundary extraction algorithm for scattered points is improved.The octree is used as the spatial index.The nearest-range search is more suitable for the study of this experiment.The optimal extraction range of the blade boundary is determined to be r=1.2.The extraction method of the leaf vein and leaf boundary contour of the eigenvalue of the covariance matrix uses the Linearity L_? and Anisotropy A_? characteristics of the covariance eigenvalue combination to extract the leaf boundary and the vein,respectively,and determine the seedling.The leaf boundary extraction threshold ?_boundary=2% and the leaf vein extraction threshold at the seedling stage can get good results.