Research On Maize Seed Mildew Detection Method Based On Hyperspectral Imaging Technology

As an important food crop in China,the safety of maize is closely related to people's lives.Because of the highwater content and a large number of bacteria,fresh maize is easy to be mildewed under high temperature and high humidity conditions.Aflatoxin B1 and zearalenone are representative toxins produced during the mildew process of maize.Ingestion can lead to liver cell damage under the metabolism of peroxidase in the body.Therefore,the rapid detection and evaluation of moldy maize are necessary.This recent work describes the spectral data pre-processing method.Hyperspectral imaging technology was used to collect hyperspectral images of maize seeds.Then three preprocessing methods were used to preprocess spectral data,which are smoothing,multip le scattering correction and variable standardization respectively.Comparing the three pretreatment methods,it was concluded that variable standardization is the best pre-treatment method for spectral data to detect the mold of maize seeds.Research on the feature extraction method of maize seed hyperspectral image by comparing spectral feature extraction and texture feature extraction.There are three methods for spectral feature extraction: PCA(Principal Component Analysis),Manifold distance and SPA(Successive Projections Algorithm).The characteristic wavelength extracted by the PCA(Principal Component Analysis)is 608 nm,Manifold distance is 548 nm ? 768 mm and SPA(Successive Projections Algorithm)is 490 nm ?572m ? 735mm? 845 nm.By comparing the three feature wavelength extraction methods,it is concluded that the continuous projection method is the best spectral feature extraction method.The texture features mainly choose energy,entropy,the moment of inertia,and correlation.Through experiments,it is found that spectral features combined with texture features are the best hyperspectral image feature extraction methods.Research on the method of modeling and identifying corn seed mildew by comparing the three modeling methods of LDA(Linear Discriminant Analysis),k Nearest Neighbors(KNN)and SVM(Support Vector Machine).Through the analysis of different recognition models on the identification of corn seed mildew,it is conclud ed that SG(Image Smoothing Techniques)+ SPA(Successive Projections Algorithm)+ SVM(Support Vector Machine)and SG(Image Smoothing Techniques)+ SPA(Successive Projections Algorithm)+ LDA(Linear Discriminant Analysis)is the best recognition model.T he correct rates of identification of healthy samples and mildew samples in the validation set were 100% respectively.To sum up,this paper concludes that variable standardization combined with a continuous projection method combined with spectral feature s combined with texture features combined with SG(Image Smoothing Techniques)+ SPA(Successive Projections Algorithm)+ SVM(Support Vector Machine)and SG(Image Smoothing Techniques)+ SPA(Successive Projections Algorithm)+ LDA(Linear Discriminant Analysis)is the best modeling method.This article can be used as the basis for hyperspectral detection of corn seed mildew,which will play a role in the future hyperspectral imaging technology to detect corn seed diseases.