Research On Nondestructive Detection Of Soybean Seed Vigor And Viability Based On Hyperspectral Image Technology

Seed vigor and viability are the main evaluation indicators of soybean seed quality.Quick and accurate detection of soybean seed vigor and viability before sowing is beneficial to know the germination of seeds in advance,so as to screen out nonliving soybean for increasing soybean yield.Traditional seed vigor detection methods include enzyme activity measurement,soaking solution conductivity measurement,etc.,and the seed viability detection methods include tetrazole measurement,red ink dyeing method,etc.The above methods are time-consuming and easy to damage seeds.Besides,the existing non-destructive detection methods mainly contain image processing technology,near-infrared spectroscopy technology and infrared thermal imaging technology,which have the problems of incomplete information,low detection accuracy and large error.However,the hyperspectral image technology has the advantages of complete spatial information,wide detection area,image and spectrum integration,solving the problems of incomplete information,strong randomness and low detection accuracy of other detection technologies.Therefore,the hyperspectral image technology was used to detect the vigor and viability of soybean seeds in this paper.The main research contents and results are as follows:(1)The collection of experimental data and the analysis of the detection mechanism were completed.Firstly,samples of soybean seeds with different vigor and viability gradients were prepared by the artificial accelerated aging experiment,and the vigor and viability of soybean seeds were determined by the germination experiment.Secondly,the visible-near infrared hyperspectral imaging system was used to obtain the hyperspectral images of soybean seeds.Thirdly,the entire sample area was taken as the region of interest(ROI)and the average spectrum in the ROI was used as the original spectral data of the sample.Finally,the correlation between the changes in the internal cell structure,organic matter content and spectral characteristics of soybean seeds at different levels of vigor and viability was analyzed,thereby proving that it is feasible to detect the vigor and viability of soybean seeds using hyperspectral image technology.(2)The accurate identification of soybean seed vigor based on hyperspectral image technology was realized.Multiplicative scatter correction(MSC)was implied to preprocess the spectral data,reducing the effect of light scattering.The stacked auto-encoder(SAE)was used to reduce the dimension of spectral data and59-dimensional feature spectral data was obtained.The support vector machine(SVM)models of seed vigor identification were established based on the full spectrum and characteristic spectrum data extracted by SAE respectively.Among them,the performance of the SAE-SVM model was superior to the model based on the full spectrum data.Bird swarm algorithm(BSA),whale optimization algorithm(WOA)and grasshopper optimization algorithm(GOA)were used to optimize the parameters C and g of the SVM model.The results showed that SAE-BSA-SVM had the best effect,and the classification accuracy of the correction and validation set reached100%and 98.67%respectively.Therefore,it is feasible to use hyperspectral image technology combined with SAE-BSA-SVM model to accurately identify the vigor of soybean seeds.(3)The accurate detection of soybean seed viability based on hyperspectral image technology was realized.MSC was used to preprocess the spectral data to remove the influence of baseline drift.Iteratively variable subset optimization(IVSO),iteratively retaining informative variables(IRIV)and IVSO-IRIV algorithms were used to extract characteristic wavelengths respectively.Based on the full spectrum and IVSO,IRIV and IRIV-IVSO characteristic spectral data,support vector regression(SVR)models were established respectively.By contrast,IRIV-IVSO-SVR model had the best prediction effect.BSA,WOA and GOA algorithms were respectively used to optimize the SVR model based on IRIV-IVSO characteristic spectral data.The results showed that WOA-SVR model had the best effect,and R_P~2 and RMSEP were 0.951and 0.047.Therefore,the combination of hyperspectral image technology and IRIV-IVSO-WOA-SVR model can achieve accurate quantitative detection of soybean seed viability.In this paper,soybean seed vigor and viability were accurately detected by hyperspectral image technology combined with chemometric modeling methods.This study provides a reference for other seed vigor and viability detection,and provides a theoretical basis for the development of seed vigor and viability detection instruments.