Identification of Baijiu based on the Raman spectroscopy and back-propagation neural network optimized using genetic algorithm

The traditional quality testing methods of Baijiu have several challenges including long duration of testing that increases the workload of the laboratory. In this experiment, Raman spectroscopy was used to efficiently control the quality of Baijiu. The spectroscopy were primarily preprocessed using the wavelet domain denoising technique and the principal component analysis (PCA). The rapid identification and classification of 496 samples of Baijiu with various alcohol concentrations (AC), flavor grade (FG), and production year (PY) were achieved using a neural network model constructed based on Raman spectroscopy and genetic algorithm-back propagation (GA-BP). The results indicated that in comparison to the BP neural network, the best GA-BP network model achieved 95.68%, 91.37%, and 93.96% correct recognition rates for the three indicators namely, AC, FG, and PY respectively. Additionally, the average correct recognition rate of the model was enhanced by 3.62%, 9.4%, and 8.45% for AC, FG, and PY respectively. In this study, for the first time, AC, FG, and PY of Baijiu were combined to construct a model that enabled the scientific identification of Baijiu samples in unique circumstances and it also provided a new reference method for the identification, detection, and classification of Baijiu.