Research On Point Cloud Data Processing Methodology For Phenotypic Parameters Measurement Of Rapeseed Plant At Maturity Stage

Today,rapeseed is grown for the production of animal feeds,edible vegetable oils,and biodiesel.It was the third-leading source of vegetable oil in the world in 2000,after soybean and palm oil.It is the world’s second-leading source of protein meal.However,the planting area of rapeseed plants is decreasing continuously in Hubei Province and even in the whole China,because of the large-scale import of other oil crops,reduction of labor force,low level of planting mechanization,and low planting efficiency.Therefore,how to increase output and realize mechanized production to the maximum extent is the main goal of breeding and cultivation for rapeseed plant.As the mostly concerned problem in researches of crop output,crop phenotype is one of the most important problems that interest the researchers.At present,although tremendous research fruits on automatic measurements of crop phenotype parameters have been achieved,only few researches on rapeseed have been reported.Meanwhile,several problems need to be solved,including: ○1 There are few researches on phenotype parameters of rapeseed plant type based on modern optics technology.○2 Fully automatic measurement of related phenotype parameters is difficult.○3 Most of common automatic measurement methods of phenotype parameters based on digital images are achieved in the two-dimensional space,which is not applicable for rapeseed plants(e.g.quantitative statistics),because severe occlusions exist among branches.Therefore,automatic measurement technology of phenotype parameters of mature rapeseed plants has become an issue that requires urgent solution.As the emergence of 3D laser scanning technology,it has been applied in more and more fields.It can acquire more accurate 3D information of objects.Hence,point cloud data of mature rapeseed plants were collected and processed for automatic measurements of key phenotype parameters.The main research contents and fruits are as follows:(1)Acquisition of point cloud data.A 3D laser scanner was used to collect the 3D point cloud data of mature rapeseed plants.(2)Preprocessing of point cloud data.since the point cloud data were disordered,and redundant information were kept in the original datasets,necessary preprocessing procedures were implemented,including point data organization and storage,point cloud simplification,and point cloud filtering.These provided good data foundation for the following procedures.(3)Segmentation of crop organs.Firstly,a segmentation methodology based on normal vector characteristics was utilized for the rapeseed point clouds.The segmentation accuracy under the optimal threshold could reach as high as 95.16% for rapeseed branches.However,unsatisfactory performance was received when the segmentation algorithm was implemented for whole rapeseed plant.Only 89.63% pods could be segmented for the whole rapeseed plant.Besides,it also required excessive labor efforts in the assessment of the segmentation results.To solve the problem,the segmentation algorithm based on characteristic statistics was researched,which not only reduced labor efforts,but also achieved satisfying segmentation effect.The segmentation accuracy of point cloud data of rapeseed branches reached 100% and the segmentation accuracy of rapeseed organs reached 97.04% for the whole rapeseed plant.Meanwhile,the plant skeleton could be extracted successfully.(4)Measurement of phenotype parameters: statistics of rapeseed quantity,rapeseed volume and plant height were carried out based on a series of processing of point cloud data.○1 Statistics of rapeseed quantity.The point cloud segmentation algorithm based on normal vector characteristics separated 59 pods under the optimal conditions,while the point cloud segmentation algorithm based on characteristic statistics separated 62 pod successively.The pod quantity for the whole rapeseed plant is 135.The point cloud segmentation algorithm based on normal vector characteristics separated 121 pods under the optimal conditions,while the point cloud segmentation algorithm based on characteristic statistics separated 131 pods successively.○2 Measurement of rapeseed volume.Firstly,triangular meshing of point cloud data of rapeseed was accomplished by the triangularization algorithm.Secondly,some holes were repaired.Finally,pod volume was measured automatically on the complete triangularization model.○3 Measurement of plant height.3D coordinates of the two ends for the main stem were acquired directly by the program and plant height could be calculated.The accuracy was 99.20% compared to the reference value.(5)The prototype system of 3D point cloud data processing of rapeseed plant was developed,which supported both.pcd and.xyz data formats.Specifically,it could realize transformation of different data formats,automatic read of 3D coordinates,downsampling and denoising of point cloud data,automatic segmentation of rapeseed organs,automatic extraction of plant stems,storage of processing results,visualization,and statistics on running efficiency of the algorithm,etc.Finally,a series of experiments have proved that the proposed processing methodologies on 3D point cloud data are applicable,which can realize automatic segmentation of organs,real-time statistics on rapeseed quantity,automatic extraction of plant stems,and automatic measurement of key phenotype parameters.This provides a feasible way to high throughout and genetic breeding of rapeseed crops.