Research And Application Of LS-SVM Algorithm Based On Multidimensional And Multispectral Fingerprints

At present,food quality and safety remain difficult in China.With improving food safety regulations,strengthening supervision and formulating food standards,it is necesssary to improvment the accuracy and efficiency of food safety testing methods.In order to supplement food safety standards and testing methods,promote the development of food safety standards and testing systems,this paper presents the LS-SVM optimization method based on multidimensional and multispectral fingerprints.Firstly,to resolve the problem of inaccurate test results on complex sample pretreatment,utilizing correlation analysis,PCA methods for data cleaning,fusion and feature selection to build multidimensional multispectral feature matrices.The experimental results show that a good feature matrix is beneficial to data mining,which can construct a clear hyperplane and reduce the complexity of the modeling process.Secondly,this thesis mainly deals with the modeling and identification of liquor.For the same group of liquor data,the characteristic matrix constructed by the same processing method is combined with partial least squares regression algorithm and LSSVM algorithm to identify liquor category.In this experiment,the validity of multidimensional and multispectral is fully reflected in the model of partial least squares regression,and the superiority of LS-SVM algorithm is also seen.In order to further verify the validity of the multidimensional and multispectral fingerprinting combined with the LS-SVM algorithm,the modeling and identification experiments of vinegar were carried out.In this experiment,the effectiveness of multidimensional and multispectral fingerprints was verified again.Finally,the artificial fish swarm intelligence algorithm is used to optimize the LSSVM model of multidimensional and multispectral fingerprint.Using the fast convergence characteristic of artificial fish swarm intelligence algorithm,the optimal LS-SVM penalty parameter C and kernel function parameters are quickly found.The experiment shows that the least squares support vector machine method can be used to analyze and evaluate multi-dimensional and multi-spectral chemical fingerprint.It may analyze the food quality on complex components,with high accuracy and stability.The method complements the fingerprint discriminant theory.