| According to statistics,the decades-long decline in the number of hungry people in the world has ended.Corn is one of the main food crops,increasing the yield of corn can effectively alleviate the global food security problem.The selection and breeding of excellent crops requires the combination of crop genes and phenotypic traits,the selection of corresponding genes is according to phenotypic characteristics.Due to the problems of subjective error,low efficiency and damage of traditional measurement of plant phenotype,high-throughput phenotype extraction has become the focus of research.In this paper,a three-dimensional phenotypic extraction algorithm based on multi vision stereo of maize plants is proposed to realize automatic,high-precision and high-throughput parameter measurement.The main contributions of the paper include:(1)In view of the time-consuming 3D reconstruction,this paper uses a calibration frame to calibrate and save the pose parameters of each camera of the instantaneous imaging equipment in advance,and directly use the pose parameters of the camera in the 3D reconstruction of the motion recovery structure to reduce the time of re-calculating the pose of the camera each time.Then,some point clouds with low information content were filtered out in RGB space based on density clustering.The number of point clouds was reduced twice.After sampling and clustering filtering,the number of point clouds decreased by 99.53%,the speed increased by 98.50%,and the average absolute percentage error only increased by 4.15%.At the same time,the accuracy is guaranteed,and the running time is greatly reduced.(2)Aiming at the problem of low accuracy,this paper analyzes the principle of point cloud noise generation,filters out the noise of key parts based on the region growing algorithm merging with color information,and improves the accuracy of parameter extraction;And based on the characteristics of corn plant and prior knowledge,puts forward improvement for the skeleton extraction algorithm based on Laplace,using principal component analysis algorithm fitting stem skeleton,and using tip point distance and angle between neighbouring points and features as a filter condition,relocation leaf and leaf base frame position,skeleton optimization parameters before and after the mean absolute percentage error were down 5.89%,0.04% and 0.86%.(3)In view of the time-consuming and labor-consuming problem of manual operation,the fully automated algorithm was adopted in this paper to input 64 Angle photos taken by the instantaneous imaging equipment and directly output Excel sheets with parameters such as leaf length,leaf maximum width,leaf angle,plant height,leaf base height,minimum oriented bounding box volume,leaf area and leaf circumference.The algorithm parameters in the process do not need manual debugging,high robustness,without manual assistance.(4)Two maize inbred lines,A619 and W64 A were studied.With selecting larger error results of two maize inbred lines,compared with the manual measurement,measurement of leaf length,leaf width,the largest leaf base height of mean absolute percentage error is 6.66%,6.45% and 5.43% respectively,root mean square error were 4.64 cm,0.47 cm,3.65 cm,determination coefficient were 0.911,0.907,0.978;The growth dynamics of plant height,leaf length,leaf maximum width and leaf base height were measured and analyzed automatically.The effectiveness of the algorithm is verified. |
PDF Full Text Request