Rapid Non-destructive Inspection Of Aflatoxins In Rice Based On Electronic Nose And Infrared Spectroscopy

In terms of the laboratory-intensive,time-consuming or insensitive current methods for mycotoxin determination in grain,this paper used characteristic volatiles and spectra information arising in grain during storage as the pointcut,and the rice inoculated with aflatoxigenic strains was studied by electronic nose,near/mid infrared spectrum(NIR/MIR).The variation trends of characteristic volatiles from common molds in grain during different growth phases were emphatically analyzed,as well as the rapid classification models of different molds were built based on electronic nose signals.Indicative volatiles of aflatoxin(AF)were revealed.Characteristic sensors and spectral ranges associated closely with AF content were confirmed.Finally,rapid analysis models of AF contamination in agricultural products based on electronic nose,NIR and MIR spectrum were established.The main conclusions of this work were obtained as follow:1,The principal component analysis(PCA)results showed that electronic nose signal could effectively distinguish aspergillus flavus,parasitic aspergillus and penicillium in later period(5 d,13 d).The discrimination rate of linear discriminant analysis(LDA)and partial least squares discriminant analysis(PLS-DA)to aspergillus flavus,parasitic aspergillus and penicillium were 100% and 97.9%.GC-MS results indicated that 1-octen-3-ol and 3-octanone could be used as characteristic volatiles of secondary metabolites of fungi.The volatile substances of different moulds are inconsistent,and especially obvious difference at later growth stage could be observed.The results showed that the application of electronic nose and GC-MS techniques is feasible for rapid identification of common molds in grain2,The rapid analysis models of AF contamination in brown rice were established based on electronic nose signals.The PLS-DA could effectively distinguish samples contaminated with different AF contents,and the overall correct classification rate in leave-one-out cross validation was over 80%.Good correlation between electronic nose sigal and AFB1,B2,G1,G2 and total aflatoxins contents were obtained by partial least squares regression analysis(PLSR).The model of AFB1 has the highest prediction precision,as well as the correlation coefficient(r)and root-mean-square error of prediction(RMSEP)was 0.808 and 127.3 ?g/kg,respectively.The loadings analysis and GC-MS results indicated that LY2/AA,T70/2,PA/2,T30/1,P10/1,P10/2 and P40/1 presented highest contribution rates,and the changes in volatile compounds of aflatoxin-contamination samples could be mainly attributed to ketones,aldehydes,alcohols,aromatics and alkanes.The results revealed that electronic nose could be applied for classification of brown rice containmated AF with different concentrations.3,The LDA models were established based on infrared spectrum,and the correct classification rates in cross-validation for AFB1,B2,G1,G2 and total aflatoxins were all higher than 90% by NIR and MIR.Good prediction accuracy for all AF was obtained by PLSR,as well as the r and residual predictive deviation(RPD)were equal to or higher than 0.920 and 2.5,respectively,which were superior to the results of electronic nose in general.Although the slightly low performances were obtained for lower AF contents,NIR and MIR techniques will be more effective and accurate detection methods for AF contamination in agricultural products,along with improvement of instrument performance and development of algorithms.4,The preliminary analysis for AF contents in paddy samples was investigated by electronic nose,NIR and MIR.Unsatisfitied predictive performances were obtained by LDA and multivariate regression analysis.The main reasons include inhomogeneous particle size,aflatoxins distribution in samples,and so on.In order to promoting the performance of model,signifficant improvement of experiment approach and pre-treatment method of sample,strict limitation of consistency and representation of sampling,as well as feature extraction and modeling methods should be implemented for further research.