To prevent counterfeit liquor on the market, the efficient and quick method is significant, the instrument detection methods combined with pattern recognition methods to identify the quality of liquor are increasingly widespread. Feature selection and recognition algorithms to create and optimize are the critical mission of pattern recognition. The essential characteristics of feature selection is the optimization problem, the experimental design is an important tool for solving optimization problems. Support vector machine (SVM) is an excellent machine learning methods. In this paper, it researched the classification of the liquor brand based on the experimental design and SVM method.Firstly, the paper used the orthogonal experiment design method (OED) to select feature from the three data sets among UCI benchmark database, experimental results verified the application of this method is feasible and effective for feature selection, and then the experiment design method is applied to select liquor mass feature, and the paper was comparative analysis of other feature selection methods. The paper used the single photon ionization time of flight mass spectrometry (SPI-TOFMS) with headspace sampling way to collect liquor spectrum. It generated six clusters 430 spectrum features from the raw spectra by appropriate mathematical conversion. By Fisher ratio the features were reduced to 200 feature dimensions, then it used the orthogonal design method and SVM classifier to feature selection, experiment obtained 68 optimized characteristics, the results showed the OED method could effectively remove redundant feature of liquor data sets.This paper analyzed the impact of key parameters for SVM, leading to the importance of parameter optimization. It teased three optimization algorithms at present stage and optimized SVM parameter for the selected liquor feature data with the OED method, analyzing the effects of different parameter optimization methods for SVM classifier performance. When using Grid search method, the search time is too long, this paper adopted an improved strategy, i.e., roughly searched with a larger step size in a larger parameters scope, then fine searched in the vicinity area of the results has been optimized. The experiment demonstrated the effectiveness of this method.In this paper, it used SPI-TOFMS detection methods, and combined the OED and SVM method to establish a rapid and accurate method to identify categories of liquor brands. The study of this paper was not only a useful complement to existing methods for identification of quality wine, but also new applications of this method in the field of wine classification. |