3D Reconstruction And Phenotype Measurement Of Quinoa Based On Computer Vision Technology

Quinoa,which originated in South America,is rich in high-quality protein,minerals,vitamins and other nutrients,and has excellent quality of salt tolerance,low temperature,high altitude,so it has become a new healthy health food in the 21 st century.However,there is little research on quinoa in China.Crop performance traits are the result of the joint influence of plant growth environment and crop genes.How to obtain and measure crop phenotypes in a low-cost,comprehensive and nondestructive way is an important part of Phenomics research.However,the traditional measurement methods are labor-intensive and can not be non-destructive,which can not meet the needs of intelligent management of modern precision agriculture and the development of plant Phenomics.The information that 2D images can express is limited,and it can not fully express the spatial features of crops such as position and shape.But the 3D model of crop can solve this problem,which can truly reflect the growth morphology and spatial position of crops.Therefore,the accurate 3D model of crops has become a hot topic in computer vision technology and agronomy.In this context,we started to study the establishment of 3D model and measurement of phenotypic traits of quinoa.In this study,quinoa seedlings as a sample,through shooting multiple angles of 2D images,using the method based on Muti view stereo for 3D reconstruction of quinoa.Then,the method of combining statistical filtering and radius filtering is used to remove the noise points in the generated dense point cloud.Finally,the five parameters of plant height,stem diameter,leaf area,leaf length and leaf width were measured and the number formula of leaf edge contour was solved.Regression analysis is made between the results of the algorithm and the real value of manual measurement.This paper provides a reference for the study of three-dimensional reconstruction of quinoa and the extraction of quinoa characters.The main work of this study is as follows:(1)The images of each quinoa plant were acquired from four different angles: low elevation angle,direct view angle,low view angle and high view angle.The acquired image data is corrected by camera calibration.(2)The sparse point cloud and dense point cloud of quinoa were established by using structure from motion(SFM)and multi view stereo(MVS)methods.(3)The noise points caused by the accuracy of image acquisition equipment,shooting environment,such as shooting light,the material characteristics of the main body of point cloud,etc.are processed by the combination of statistical filtering and radius filtering.(4)The plant height,stem diameter,leaf area,leaf length and leaf width of quinoa point cloud were measured.After regression analysis between the result of the algorithm and the real value of manual measurement.The results showed that the phenotypic parameters,such as plant height,stem diameter,leaf area,leaf length and leaf width,measured by this method at seedling stage of quinoa had high accuracy and good consistency with those measured manually.