Hyperspectral Imaging Identifications For Cereal Fungi And Detection Methods For Moldy Maize Kernels

Fungi are widely distributed in nature,with various species and complex composition.Cereal is a good substrate for fungal growth because of its rich nutrients such as sugar,protein and starch.Due to the effect of fungi,the moldy cereal has caused serious economic loss,and the toxic metabolites produced by fungi in cereals have threatened the life and health of human and animal.Detection of moldy cereals is important to ensure food safety and downstream industry safety.The traditional detection methods cannot meet the requirements of on-line and batch detection.Hyperspectral imaging(HSI)technology,which combines the advantages of traditional imaging and spectral technology,can simultaneously detect the internal and external quality of agricultural and livestock products.It is a research hotspot in the field of nondestructive testing of agricultural products.Based on the purpose of detection of moldy maize,five species of fungi(Aspergillus niger,Aspergillus flavus,Aspergillus parasiticus,Aspergillus glaucus,and Penicillium)were selected for investigation.The HSI images of the five kinds of fungal colonies growing on rose bengal medium(RBM)and maize meal ager medium(MAM)were recorded daily after inoculation up to 6 days.Combined with multi-variable data analysis methods,the growth and development characteristics of single colony of each fungus under different media conditions and the difference of fungal colonies were analyzed.The growth process of each fungus on the medium and different types of maize substrate was tracked.The characteristics of fungi growth stage,fungi species and the time node of moldy early detection were identified.Finally,a hyperspectral imaging discriminant model for naturally moldy maize kernels was established and the detection of aflatoxin content in maize kernel was preliminarily explored.Software based on this discriminant model for identification of moldy maize and toxin content was developed,the early detection of moldy maize kernels was achieved.This paper evaluated the growth characteristics of fungi.The growth curve fitted by the pixel number of fungal single colony at different time points can show the growth stage of the corresponding fungi.The average reflectance spectrum of fungal colonies can show the growth status of the corresponding fungi.After principal component analysis(PCA),four growth zones within the colonies could be visualized and the fungal growth time of 6 days could be divided into 4 stages.In the fungal colony species discrimination,only Aspergillus niger could be discriminated.A multi-variable analysis method was used to establish the fungal species identification model and the discriminant models of growth time and species of fungi inoculated on different substrates.After comparison of different modeling methods,Successive Projections Algorithm(SPA)-Support Vector Machine(SVM)were used to build the fungal species identification model.The classification result of single fungal colony species was over 93.96%.The fungal growth time identification results showed that the earliest detection time for the fungi was 1 day after inoculation.The fungal species discrimination accuracy was over 98.00%,but the discriminant for the mixed fungal species in medium was ineffective.The discriminant model of natural moldy maize kernels was established using SVM based on the characteristic wavelengths extracted from SPA,the identification accuracies were over 99.50%.The classification model of AFBi content levels was established using SVM,with an accuracy of 82.50%.Further,the AFB1 content predictive model was constructed and the correlation coefficient reached 0.70.Applying the predictive model on the hyperspectral image make the growth process of fungi,the fungal species discrimination,the distribution of fungi on the kernels and the location of moldy maize kernels are more intuitive.The results indicated that,HSI technology combined with multi-variable data analysis could be used to trace the growth process of fungi,identify fungal species and detect moldy maize kernels.The research can provide new ideas and methods for non-destructive optical characteristics detection of other cereal fungi.The selected characteristic wavelengths and the discriminated models could not only simplify the model but also lay the foundation of the development of maize rapid selecting equipment.