Identification Of Geographical Origin,Freshness And Adulteration Of Huajiao By Near Infrared Spectroscopy

Huajiao is one of the important economic crops in China and occupies an important position in the modern agriculture.In this paper,near infrared(NIR)spectroscopy was applied in non-destructive and fast identification of geographical origin,freshness and adulteration of Huajiao aiming to explore the rapid detection methods of Huajiao quality.NIR analytical technique has a lot of advantages such as low cost,high efficiency,no pretreatment and environment-friendly.NIR identification models of geographical origin,freshness and adulteration of Huajiao could be used to predict the Huajiao quality,which are very good for establishing quality traceability system of Huajiao and protecting famous brand.It is of great significance to provide powerful tools for market regulation,protect the interests of producers and consumers.The main research contents and results are as follows:(1)Research on NIR qualitative analysis of Huajiao geographical origins was carried out and identification models based on convolutional neural networks(CNN),discriminant partial least squares(DPLS),support vector machine(SVM)and radical basic function neural network(RBF-NN)were built and compared.Spectra data of samples from 8 regions were obtained by a near infrared spectrometer and all samples were divided into two parts,calibration set and validation set.The influence using different activation functions were discussed on the CNN models.The effects of various pretreatments on the DPLS models were compared.The effects of Grid,genetic algorithm(GA)and particle swarm optimization(PSO)searching radial basis function parameters of SVM models and the effects of different wavelet pretreatment on RBF-NN models were compared.Wavelet denoising as pretreatment method,ReLU as activation function,the correct identification rate(CIR)of calibration set of CNN model was 98.17%,and the CIR of validation set was 95.12%.MSC as pretreatment method,the CIR of validation set of DPLS model was 97.56%.Using GA and PSO methods to search best radial basis function parameters could increase the CIR of SVM models up to 100%.The CIR of validation set of RBF-NN model was 95.12%when without pretreatment.The results showed that it is feasible to identify and trace the origin of Huajiao by NIR technology.(2)Research on NIR qualitative analysis of Huajiao freshness was carried out and identification models based on sparse representation based classification(SRC),DPLS,SVM and RBF-NN were built and compared.Red Huajiao samples and green Huajiao samples with different storage periods were collected and spectra data were obtained.The feasibility of using SRC in the identification of Huajiao freshness was discussed.The effects of various pretreatments on the DPLS models were compared.The effects of Grid,GA and PSO searching radial basis function parameters of SVM models and the effects of different wavelet pretreatment on RBF-NN models were compared.Result indicated that the SRC method could build Huajiao freshness identification model with fast testing speed and high accuracy.The CIR of validation set of red Huajiao and green Huajiao were 97.5%and 100%,respectively.Mean centering as pretreatment method,the CIR of validation set of DPLS model was 100%.Using grid,GA and PSO methods to search best radial basis function parameters of SVM models could be identified totally.The CIR of validation set of red Huajiao RBF-NN model and green Huajiao RBF-NN model were 90%and 86.84%separately when without pretreatment.(3)Research on the qualitative analysis and quantitative analysis of adulterated Huajiao powder was carried out.Wheat bran,rice bran,corn flour and rosin powder were mixed with red Huajiao powder and green Huajiao powder separately and NIR spectra of pure and adulterated Huajiao powder were acquired.Kinds of qualitative models with different spectra pretreatment were established using DPLS and SVM analysis.Results showed that,when selecting feature bands the CIR of the validation set of DPLS model and SVM model could be improved up to 100%and 99.13%.SIMPLS method was adopted to establish the adulteration quantitative analysis models of Huajiao powder.R~2 of Huajiao powder,wheat bran,rice bran,corn flour and losin powder of the validation set were 0.971,0.969,0.948,0.967 and 0.995 respectively.The predictive value had a good correlation with the actual adulterated concentration.Even1 wt/wt.%adulterated concentration could be detected accurately.It showed that NIR technology is a good method to identify adulterated Huajiao powder qualitatively and quantitatively.