Botanical origin identification and adulteration quantification of honey based on Raman spectroscopy combined with convolutional neural network

Abstract

A non-destructive and reliable spectroscopic analysis method was proposed for detecting honey fraud based on Raman spectroscopy and convolutional neural network (CNN). Acacia, litchi and linden honey were adulterated with high fructose corn syrup (HFCS), rice syrup (RS), maltose syrup (MS) and commercial blended syrup (BS) at different concentrations, respectively. Spectra were collected from 60 authentic honeys and 360 adulterated honeys. Dimensionality reduction algorithms provided an overview and visualization of the spectral dataset. The strategy of this study was that the botanical origins and adulteration concentrations of unknown samples were obtained through a combination of qualitative and quantitative models, regardless of the types of honey and contaminant. The CNN classification model and chemometric algorithms achieved more than 99.76% accuracy for honey matrix identification, while the CNN quantitative models kept the coefficients of determination (R2p) and root mean square errors of prediction (RMSEP) above 0.95 and below 4.25, respectively. CNN achieved significantly better performance than chemometric algorithms and met the routine detection requirements of honey. The proposed method provides a promising alternative to combat honey fraud.