Rapid Non-Destructive Detection Of Aflatoxin In Wheat Based On Spectroscopic And Imaging Techniques

Wheat is one of the three major staple foods in China,and its quality and safety are related to national stability and national health.Fungal toxins are one of the important threats to food security,and the global annual loss of wheat due to fungal infections amounts to 62 million tons.The development of spectroscopy and imaging technology provides new means for rapid and nondestructive detection of mycotoxin infections in wheat.In this paper,a rapid and nondestructive detection method of Aspergillus flavus and its toxins in wheat was investigated by multispectral imaging and near-infrared spectroscopy combined with various machine learning algorithms,and the main work is as follows:(1)Rapid nondestructive detection of Aspergillus flavus in wheat was accomplished based on multispectral imaging technology.Based on the multispectral imaging technique,we obtained the spectral images of wheat with different levels of A.flavus infection,obtained the color,morphological and spectral characteristics of wheat by ROI(Return of investment)region extraction,analyzed the characteristics of wheat with different levels of A.flavus infection by comparing three different pre-processing methods of t-SNE,GA and PCA,and obtained the best prediction of A.flavus infection by combining BPNN,KNN,SVM and PLS methods.model.The test results showed that t-SNE-BPNN predicted the best results,and the correlation coefficient(R)and root mean square error(RMSE)of the prediction set reached the level of 0.942 and 1.688 CFU/kg,respectively;the prediction accuracy of wheat infection level reached 95.8 %.(2)The rapid nondestructive detection of aflatoxin B1(AFB1)in wheat was accomplished based on spectroscopic techniques.The characteristics of wheat with different AFB1 concentrations were analyzed by comparing three different pretreatment methods,t-SNE,GA and PCA,with multispectral imaging,NIR detection and other analytical methods,and the best prediction models for different AFB1 concentrations in wheat were obtained by combining BPNN,KNN,SVM and PLS methods.The test results showed that the prediction accuracy of different AFB1 contamination levels in wheat by PCA-BPNN modeling reached 95 %.(3)Design of a software system for rapid detection of wheat fungi and toxins.This experiment develops a software system for rapid detection of wheat fungi and toxins,combining various spectral pre-processing and feature extraction algorithms with four machine learning algorithms to achieve rapid analysis of the obtained spectral data.The rapid and non-destructive detection of Aspergillus flavus fungus and its toxin in wheat by MSI and NIR techniques provides a new and effective way to monitor the production processing and storage of wheat.