3D Phenotypic Measurement Of Crop Grain Based On Structured Light Imaging

With population growth,climate warming,and the reduction of agricultural cultivated land,China’s food security is facing severe challenges.In order to effectively regulate the relationship between grain and supply in China,finding ways to improve crop yield is one of the effective ways to solve the grain crisis.The examination of crop grains is not only an important expression of yield,but also an indispensable link in breeding and functional gene analysis.Grain type parameters such as length,width,thickness and volume are also closely related to crop yield and its own genotype.It is of great significance to study the rapid measurement of 3D phenotypic characteristic parameters of crop grains,how to realize the discrimination between varieties and shriveled grains,how to predict grain weight,and to find a plan to improve the final yield of crops.At present,the traditional grain character measurement methods mainly rely on manual measurement,which is slow and difficult to avoid the manual error caused by operation error.Most of the existing seed testing equipment rely on RGB image.The measurement characteristics are mainly grain length and grain width,which are based on2 D image characteristics,but such as grain volume,surface area and sphericity,which are based on three-dimensional characteristics,are not easy to obtain.Based on the above reasons,a new method of 3D phenotypic measurement of crop grains based on structured light imaging is proposed in this paper.Firstly,the manipulator structured light imaging scanning platform was built,and the structured light scanner was used to obtain the threedimensional point cloud data of crop grains.Then the point cloud processing algorithm for crop grains was designed,including point cloud data preprocessing,point cloud segmentation and 3D phenotypic measurement.Finally,25 3D phenotypic parameters of rice grains and 32 3D phenotypic parameters of wheat grains were extracted.Based on the orthogonal experiment,the best imaging conditions of the experimental platform were determined as the black platform,the rotation angle of the turntable was 30 °and the scanning angle was 37°.The average measurement efficiency of the experimental platform for crop grains was 12 s / grain.Comparing the length,width and thickness of crop grains measured by the algorithm with the artificial real values,the average absolute percentage errors of grain length,width and thickness measured in this study were 2.12%,2.06% and 1.95%,respectively.Based on the extracted 3D phenotypic parameters of 25 rice grains,combined with 6 different machine learning classification algorithms,the classification model of Indica and japonica rice and full and shriveled rice grains was constructed.Based on the extracted 32 3D phenotypic parameters of wheat grain,combined with five different machine learning regression algorithms,a regression model for wheat grain weight prediction was constructed.Finally,the 3D phenotypic measurement algorithm of crop grain,the full and shriveled grain recognition model and the wheat grain weight prediction model were integrated into the developed visual interactive software.In conclusion,this paper proposes a new method for 3D phenotypic measurement of crop grains,which can effectively assist the research related to crop breeding and provide a convenient measurement analysis tool for grain phenotype traits for scientific research and production applications.