Study On Detection Methods Of Aflatoxin B₁ In Maize Based On Raman Spectroscopy Technique

Maize is an important global food crop that occupies a pivotal position in human society.Aflatoxin B₁（AFB₁）is one of the mycotoxins widely present in maize,which seriously threatens human life and health.Therefore,the implementation of rapid detection of AFB₁ in maize will not only protect the rights of consumers,but also guide farmers and processors to take measures to reduce its contamination.Raman spectroscopy has the merits of being non-destructive,highly efficient and non-polluting,and has been extensively implemented for the detection of harmful substances in agricultural products.In view of this,this study proposes to carry out a qualitative identification of the degree of AFB₁ contamination in maize based on Raman spectroscopy and a quantitative method for the determination of AFB₁ content in maize.The main work is as follows:（1）Study on the identification of the degree of AFB₁ contamination in maize based on Raman spectroscopy technique.A portable Raman spectrometer was used to perform Raman spectral characterization of moldy maize samples.And the raw Raman spectra obtained were properly preprocessed.Principal component analysis（PCA）was used to reconstruct the features of the preprocessed Raman spectral data,and extreme learning machine（ELM）,k nearest neighbor（KNN）and support vector machines（SVM）recognition models based on the reconstructed principal components（PCs）were developed respectively.At the same time,a rational one-dimensional convolutional neural network（1DCNN）structure was designed to implement the self-learning and recognition model correction of Raman spectra after pre-processing.The results show that both the best 1DCNN model and the best SVM model achieve 100%prediction accuracy for independent samples.The results show that the qualitative identification of the level of AFB₁contamination in maize can be achieved with high accuracy using Raman spectroscopy combined with appropriate chemometric methods.Furthermore,the results can provide a methodological reference for rapid monitoring of the level of mycotoxin contamination in cereals.（2）Study on detection method of AFB₁ content in maize based on Raman spectroscopy technique.The competitive adaptive reweighted sampling（CARS）and variable combination population analysis（VCPA）algorithms were used to optimize the preprocessed Raman spectral features respectively.The partial least squares（PLS）and SVM detection models were constructed respectively based on optimized features.Meanwhile,according to the preprocessed Raman spectrum features,a reasonable 1DCNN structure was designed to realize the feature self-learning and quantitative model correction of Raman spectra.The research results show that compared with the best PLS and SVM models,the 1DCNN model obtained the best detection performance.The determination coefficient of prediction（R_P²）and the root mean square error of prediction（RMSEP）are 0.9540 and 4.3823μg?kg^-1,respectively.The results of the investigation show that it is feasible to use Raman spectroscopy combined with appropriate chemometric methods to achieve quantitative analysis of maize AFB₁,and at the same time broaden the application range of deep learning in the field of spectroscopic chemometric analysis.In this study,deep learning was introduced into the chemometric analysis of Raman spectroscopy,and the qualitative identification of maize AFB₁ pollution degree and the quantitative detection of AFB₁ content were successfully realized.The research results provide a method reference for the successful application of deep learning in the field of spectral chemometrics.