Canonical Correlation Analysis Algorithm And Its Application On Tea Origin Recognition

The color,smell,taste of the tea are associated with its origin,and they affect the price of tea.Therefore,recognizing the origin of tea has an important practical significance.In this paper,canonical correlation analysis algorithm and its expansion algorithm are used to analyze the data sets by reducing dimension and fusion.Through the results,a cost-effective method for recognizing origin has been investigated.The main work of this paper is as follows:First of all,the source and background of the problem are introduced in this paper.And taking Wuyi Yan tea as the research object.The data sets are used to recognize the origin of tea.There are 7 sets of data sets that are introduced,such as: amino acid content,catechins content,electronic tongue data,element content,isotope content,electronic nose data and near infrared data.Among them,electronic tongue data,electronic nose data and near infrared data are the main research object.Because these data can easily get and cost less.Then by using SVM and naive Bayes classification method to recognize the origin of Wuyi yan tea samples,it can be concluded that element content and isotope content can make more accurate result.In order to find an effective feature fusion method,a variety of canonical correlation analysis algorithm have been used.The sample data were tested by multiple correlation canonical correlation analysis,kernel canonical correlation analysis,and discriminant canonical correlation analysis.Then the fusion results were tested by SVM and naive Bayes classification.The best method of feature fusion is discriminant canonical correlation analysis.Besides,electronic tongue data,electronic nose data and amino acid content are the best combination since its accuracy is up to95.2%.This results can be applied in practical problems.As the dimension of near infrared data is too high,it is difficult to directly use.Sparse canonical correlation analysis algorithm is used.And alternating direction method of multipliers are used to solve sparse canonical correlation analysis.The dimension of near-infrared data has been reduced from 4148 to 40.Near infrared data and other data sets are used for feature fusion.It can be concluded that using amino acid content,isotope content and near infrared data can get the accuracy of 95.2%.Thus using less data gets better recognition results.