Research On Liquor Alcohol Detection Platform Based On Raman Spectroscopy

In this paper,through the Raman spectroscopy technique,spectral data information was collected from 150 liquor samples using a Raman spectrometer with a wavelength range of 350-1 000nm.Among them,white wine samples were selected from XX brand alcohol with 43%,53%,and 56%alcohol concentration.Based on the classification accuracy rate,I have analyzed and compared of various detection models,the SPA-BP neural network(Successive Projection Algorithm-Back Propagation Neural Network)model was selected as the alcoholic concentration detection model.Combined with the Graphical User Interface(GUI)tool of MATLAB,the detection platform for identifying the alcohol content of liquor is designed to realize the intelligent detection of the alcoholicity of liquor in a fast and non-destructive way.The main research results of this paper are as follows:(1)Raman spectral data information collection for three different alcoholic liquor samples,using Wavelet Transform(WT),Standardized Normal Variate(SNV),Multiple Scattering Correction(MSC),SG Savitzky Golay Smoothing(SGs)method to preprocess the original Raman spectrum.Using all the sample data as the training set through the establishment of Partial Least Squares-Discriminant Analysis(PLS-DA)model,the SNV pre-processing method is the best way.The correlation index R2 is 0.9364,the root means square error RMSEc is 0.1343.(2)Using the sample set partitioning based on joint xy distance(SPXY)method,the alcohol content is 43%liquor sample(category number is 1),53%liquor sample(category number is 2),56%liquor sample(category number is 3),divided into 3:1 training set and prediction set.Among them,114 samples of the training set are used for the establishment of the classification model,and 36 samples of the prediction set are used for the verification of the classification model.Principal Component Analysis(PCA)and Continuous Projection Algorithm(SPA)are used to reduce the dimensionality of the preprocessed data.The number of characteristic wavelengths extracted by the SPA algorithm is 10,and the PCA algorithm reduces the data dimension is 8.(3)Establishing the K Nearest Neighbor(KNN)classification model for full spectrum,PCA and SPA.The results show that SPA-KNN can represent the full spectrum data to establish the classification model.The training sets' and prediction sets' correct rate are respectively 83.33%and 80.56%;BP neural network classification model of PCA and SPA was established.The results show that SPA-BP neural network is the optimal classification model,and the correct rate of training sets and prediction sets are 93.86%and 94.44%,respectively.By comparing the correct rates of SPA-KNN and SPA-BP neural network classification models,it is concluded that the SPA-BP neural network model is better than the SPA-KNN model.Therefore,the SPA-BP neural network model is applied to the design of liquor alcohol detection platform.(4)Using MATLAB GUI to design a white alcoholic alcohol online detection platform,including raw data,data preprocessing,data dimensionality reduction,detection model establishment new sample prediction,five modules,through the operation of interface related functions,to achieve liquor alcohol on-line detection.The platform has been verified the effectiveness of the detection platform through experiments.