The Collection And Analysis Of Soybean Seed Phenotypic Characteristic Data For Artificial Intelligence Breeding

Soybean is one of the most important sources of oil and vegetable protein in the world.In recent decades,China's rigid demand for soybeans has grown rapidly with the increase in the proportion of protein in the mass food structure,but domestic soybean production has shown a downward trend,and its dependence on foreign countries is serious,threatening national grain and oil security.Cultivating excellent soybean varieties and increasing soybean yields have become the primary goals of soybean breeding experts.Seed testing has always played an important role in the study of crop phenotyping.By testing soybean seeds and plants,the phenotypic characteristics are obtained,and the relationship between the yield and quality of soybeans is analyzed.This is important for the design and research of breeding programs The traditional manual testing method has the disadvantages of high cost,low accuracy,and rough description.However,the existing testing method based on hyperspectral image technology is difficult to popularize due to the high price of technical equipment and inconvenience.In order to improve this situation,this thesis proposed a solution to the collection and analysis of soybean seed phenotypic characteristics for artificial intelligence breeding based on computer vision technology.This thesis focused on the collection and analysis of soybean phenotype characteristics based on computer vision technology,mainly completed the following work:(1)Designed and constructed a high-throughput,low-cost batch collection solution for soybean seed images,rapid and lossless image collection was carried out on about 100,000 soybean seeds from 200 northeast soybean hybrid lines harvested in 2018,finally more than 5,000 original soybean seed images were taken.(2)The preprocessing of the original soybean seed image is completed.First,the relevant text information contained in the collected soybean seed image was extracted through the tesseract-OCR open source engine,and a naming template was designed to generate uniform and regular soybeans seed image data set;completed image graying and image denoising,based on the Hough transform algorithm,the original soybean seed image was tested for seed real coordinates and region segmentation,and as a result a uniform and single-grain soybean seed image data set was obtained.(3)In the obtained single-grain soybean seed image data set,the color characteristics of soybean seed skin,gloss,uniformity were extracted under different color models,and the shape characteristics of soybean,like seed length,width,perimeter,area were calculated.Based on the ideas of Laplacian of Gaussian(LoG)operator edge detection and breadth-first search(BFS),this thesis firstly extracted the feature information of soybean hilum and kidney,provided.(4)Designed and implemented the soybean phenotype group data platform system,which saved the image processing results of this thesis and the artificial seeding data recorded in previous years,and provided breeding experts with simple and reliable analysis of soybean seed phenotype characteristics and data query service.The solution designed by this thesis provides a certain reference value for the research of soybean phenotype group for artificial intelligence breeding.Through computer vision technology,it realizes the rapid and non-destructive testing of soybean seeds,which solves the high cost and low efficiency of traditional artificial testing problems such as poor accuracy,and provides a richer and more accurate phenotypic parameter reference for breeding experts to carry out breeding design.