Classification Of Liquor Brands By Means Of Hybrid K-means-SVM Model

Liquor is a traditional product in China.To control the quality of liquor influences the development of wine industry and the health of consumers.Therefore,it is very important to discriminate the quality of liquor conveniently.In a real world application,chemometrics combined with pattern recognition technology is widely used to identify the quality of the liquor.In pattern recognition,selection of the training samples influences the performance of classification model.This thesis mainly analyzes the influence of Support Vectors(SVs)on Support Vector Machine(SVM)model.The proposed method is that the training set selected by K-means can improve the accuracy rate and generalization ability of classifier.The number of training samples influence the generalization ability of the model.In addition,it proves that the performance of SVM is only related to SVs rather than non-SVs.It can improve the performance and generalization ability of SVM by means of extracting the samples which will may be the SVs.K-means method has the advantage of low computing complexity and it can search the class centers automatically.This thesis uses this method to search the class centers and the boundary samples.It can reduce the risk of confidence that deleting the mistake classification samples avoids Overfitting.This thesis verifies the performance of the proposed method(K-SVM)on WDBC,Iris,Wine,Sonar datasets of UCI.The results show that K-SVM improves the accuracy rate compared with RS and KS samples selection methods.Firstly,detecting the Time of Flight Mass Spectrometry(TOFMS)of the different brands and batches of liquors.It can get the reliable data through the detection of the instrument in precision,repeatability and stability.The Wavelet Analysis method is applied to data noise reduction,and then selecting 350 features through PLS-VIP.And on this basis,K-SVM method is applied to discriminate the brands of liquors.Additionally,it optimizes the kernel parameters of SVM by means of Grid Search Method.This method gets a better accuracy rate than other methods in samples selection,and it caters to the requirement for liquor quality control of high accuracy.In this thesis,SPI-TOFMS is the detection method.Some data preprocessing methods are applied in the TOFMS matrix.SVM with samples selection throughK-means improves the performance in discrimination of liquor brands.The proposed method is the extension of liquor quality identification methods.And it is a feasible measure in SVM samples selection.