Research On The Detection Method Of Soybean Seed Vitality By Near Infrared Spectroscop

Soybeans are closely related to people’s daily lives in agricultural production,and the vitality of soybean seeds directly affects the yield of soybeans.The current seed market is mixed,and disputes arising from seed quality issues are not uncommon.Detecting the vitality of seeds through traditional methods is time-consuming and labor-intensive,with strong subjectivity and poor scientificity.Therefore,studying a non-destructive and accurate method for detecting soybean seed vitality has enormous potential benefits.This study applies near-infrared spectroscopy technology to identify the vitality of soybean seeds based on measured near-infrared spectroscopy data.The main content of this study is as follows:(1)Basic data collection.Taking soybean seed Wudou 188 harvested in Daqing City,Heilongjiang Province in 2020 as the research object,in order to obtain soybean seeds with different vitality levels,high-temperature and high humidity artificial aging experiments were conducted on them.Six aging levels were set at 1-day intervals,and the non aged seeds were set as the control group.30 sets of data were collected for each aging level,and a total of 210spectral data were collected in the experiment.Conduct germination experiments on soybean seeds after collecting spectral data to obtain their corresponding germination rate data.(2)Spectral data processing.Apply Monte Carlo cross validation(MCCV)to eliminate abnormal samples from the collected soybean seed spectral data.After removing one abnormal sample,the partial least squares modeling method is applied to compare the performance of five different preprocessing methods and three feature extraction algorithms to determine the optimal processing method.(3)Model establishment and result evaluation.Based on near-infrared spectral data and corresponding soybean seed aging level data,a multi-level prediction model for soybean seed vitality is established by combining BP neural network model(BPNN),support vector machine(SVM),and K-nearest neighbor(KNN)methods.Based on near-infrared spectroscopy data and soybean seed germination rate,a quantitative measurement and prediction model for soybean seed germination rate was established by combining BPNN,SVM,and KNN methods.And compare the parameters of the three types of models to obtain the optimal model.The accuracy of the multi level KNN prediction model for soybean seed vitality ultimately established in this study is 97.14%,F1score is 0.98.The relative analysis error(RPD)of the quantitative SVM prediction model for soybean seed germination rate is 4.14,the root mean square error(RMSEC)of the training set is 0.0216,the determination coefficient(_cR~2)of the training set is 0.9648,the root mean square error(RMSEP)of the testing set is 0.0247,and the determination coefficient(R_p~2)of the testing set is 0.9362.The model obtained in this study has good performance,providing a method and theoretical basis for rapid detection of soybean seed vitality,and can be used for reference in other crop vitality research work.