Application Of Near Infrared And Raman Spectroscopy Combined With Deep Learning In Food Analysis

The food quality safety has always been the focus of social attention,among which the raw material is the source of ensuring food quality and safety.In the supervision and management of food market,a fast and accurate method for food quality and safety analysis is urgently needed.As non-polluting and non-destructive detection methods,Near Infrared(NIR)spectroscopy and Surface-Enhanced Raman Spectroscopy(SERS)have been widely used in quantitative and qualitative detection of various substances.However,the application of the combination of spectral technology and deep learning in food quality and safety analysis needs to be developed.In this research,the quality of Lentinus edodes and the safety of fruits are evaluated and analyzed by NIR and SERS technology combined with deep learning.Lentinus edodes have been cultivated in China for thousands of years,and more than 30,000 tons of dried Lentinus edodes are exported every year.It is of great significance to strictly control the quality of Lentinus edodes.Polysaccharides are one of the most important indicators to evaluate the quality of Lentinus edodes.The traditional detection method for polysaccharide is phenol-sulfuric acid method,which is time-consuming and environmental pollution.Therefore,a rapid and non-destructive evaluation method for crude polysaccharides of Lentinus edodes was established based on NIR spectroscopy and deep learning.Firstly,the siPLS model was established in bands of 4797～3995 cm-1 and 6401～5600 cm-1.And MIR-NIR analysis verified the correlation between the selected bands and polysaccharide structure.Meanwhile the convolution and pooling operations were used in the one-dimensional convolutional neural network(1D-CNN)to capture the features of enhanced data.And the optimal super-parameters of the model were obtained by grid searching.A high-precision 1DCNN quantitative model for crude polysaccharides with R2pre of 95.50%and RMSEP of 0.1875 g/100 g was established.In addition,a quality discriminant analysis model based on polysaccharide content was established by combining principal components analysis(PCA)and multi-channel convolutional neural network-gated recurrent unit(MC-CNN-GRU)with NIR technology.Through the results of t-distributed stochastic neighbor embedding(t-SNE),it could be found that the spatial and sequential information of NIR spectra was accurately extracted by MC-CNN-GRU.Finally,the qualitative evaluation of Lentinus edodes quality was achieved with the accuracy of 95.62%.In addition,NIR spectroscopy combined with convolutional neural networks also can be used for food safety analysis.Food safety also concerns everyone’s life and health,and it is more related to the orderly operation of the market and society.Longterm consumption of food with excessive pesticide residues will cause potential threats to human health.Therefore,a rapid,sensitive and economical method for qualitative detection of mixed pesticide/fungicide residues in fruits was established based on the combination of SERS technology and convolutional neural network.Under the premise of ensuring the sensitivity of the detection method,the pretreatment process was simplified.At the same time,the detection time and cost were also greatly reduced.Compared with the results of PCA,support vector machines(SVM)and other traditional stoichiometric analysis methods,it was proved that MC-CNN-GRU had better identification effect and could realize qualitative analysis of mixed pesticide residues on fruits surface with high precision(Accuracy=99%).In summary,based on the combination of NIR and SERS technology with convolutional neural network,the research analyzed and evaluated the quality of Lentinus edodes and the mixed pesticide/fungicide residue on fruit surface,providing technical support for the application of spectral technology and deep learning in food quality and safety analysis.