Design And Implementation Of Soybean Phenotype Acquisition Platform "wisdom Of Soybean Online"

In recent years,the development of computer technology has promoted the rise of the "Internet+" industry.At the same time,digital agriculture has also made some progress.The online phenotype extraction technology with convolutional neural network as the core is expected to promote agricultural research The breeding work is of great significance for increasing soybean yield and promoting the development of smart agriculture.Based on the realization of the construction of web development platform and the convenience of information management,this paper aims to obtain the phenotypes of soybean maturity and growth periods,takes images of soybean plants as the carrier,and uses machine vision and deep learning algorithms as the core to design And developed a webbased automatic acquisition platform for soybean phenotype.In the breeding process,the recording and tracking of soybean phenotypic data is an indispensable link,but phenotyping is usually a manual task,which is time-consuming,laborintensive and costly.In addition,the observer has the subjective initiative,so the visual assessment of many plant phenotypes is very difficult and error-prone.It is very urgent to realize the automatic extraction of phenotypes.In line with the original intention of applying information management and image analysis technology to the field of precise soybean phenotypes,based on the idea of industrial engineering optimization and improvement,this research will establish a "Smart Bean Online" phenotyping platform to realize the automatic acquisition of soybean phenotypes,and at the same time Better improve and realize the management of soybean information.This article has done the following three parts of work around the design and development of the "Zhidou Online" platform:(1)Realize the information management function based on the database.With the development of plant phenotype research,the gradual accumulation of acquired information has led to a corresponding increase in the difficulty of information management.In addition to data,there is a large amount of image information,which needs to be scientifically stored and managed.To this end,a soybean image information database is designed to support importing pictures and provide users with classified query and batch download,to realize the orderly management of complicated pictures,and at the same time as the basis for online phenotype acquisition,providing a carrier for subsequent analysis.(2)Realize the function of automatically acquiring the phenotype of soybeans during the growth period.There are many phenomena in the reproductive growth process of soybeans that affect the yield.Among them,the shedding of flower and pod is considered to be a major limiting factor affecting the increase in yield.For this reason,the phenotype acquisition part of the growth period mainly realizes the identification and counting of soybean flower and pod.Realizes the online application of soybean flower recognition and counting based on Yolo＿v3 and soybean pod recognition and counting based on Faster R-CNN,which can accurately recognize and count the number of flowers and pods in soybean growth period pictures in various complex scenes.After testing the picture recognition The accuracy is above 0.94.(3)Realize the function of automatically acquiring the phenotype of soybeans in the mature period.The phenotypic data of the maturity period includes soybean stems,pods,and beans.The phenotype of the soybean stems is the most direct manifestation of the strength of soybeans.The related phenotypes of the beans and pods are also key information related to the quality of the yield.To this end,this paper proposes a deep convolutional neural network based on instance segmentation,which can automatically segment images of soybean stems,pods and grains and extract phenotypic parameters.The m AP obtained after 60,000 iterations is above 95,and a strong correlation coefficient is obtained by comparing with manual measurement data,which provides experimental data support instead of manual measurement.The database in the platform can provide a carrier for online phenotype acquisition,which is the basis of intelligent species testing;online phenotype acquisition can provide great convenience for phenotypic data acquisition,and automated online phenotype acquisition is the "smart" platform The most direct embodiment.The system provides a platform for joint research and analysis,provides new ideas for in-depth research on online phenotype acquisition,and also provides data and theoretical support for cultivating excellent soybean varieties.