Research On Typical Correlation Analysis Algorithm Of Multi - Sets Based On Fractional Order

Canonical correlation analysis(CCA) has been extended to the multiple data sets based on the idea of maximizing generalized correlation measure, named multiset canonical correlation analysis(MCCA). So the research of MCCA is the correlation between multiple data sets. The theory of pattern recognition algorithm based on the MCCA has been enriched by introducing orthogonal, kernel, supervised informations, locality preserving and so on. The idea of fractional-order embedding has been proposed to reduce the deviation of feature extraction. So we combine the idea of fractional-order embedding and theories based on MCCA, and then we research the new algorithms. Concrete work and innovation are as follows:(1) We improve the optimization model of MCCA by using orthogonal of Orthogonal multiset canonical correlation analysis(OMCCA) and Fractional-order embedding of multiset canonical correlation analysis based on fractional-order embedding(FMCC). So we propose Orthogonal multiset canonical correlation analysis based on fractional-order embedding(FOMCC) and achieve good performance in pattern recognition.(2)Because of the lack of linear calculation capacity, we improve the optimization model of FOMCC by using kernel methods of Kernel multiset canonical correlation analysis(KMCCA). So we propose Kernel multiset canonical correlation analysis based on fractional-order embedding(KFOMCC) and achieve good performance in pattern recognition.(3) We introduce fractional-order embedding and supervisory information from two different angles by using class information of samples, and then we propose two supervised learning approaches named Generalized multiset canonical correlation analysis based on fractional-order embedding(FGMCC) and Discriminative multiset canonical correlation analysis based on fractional-order embedding(FDMCC). A series of comparative experiments have been made to demonstrate the effectiveness of these two approaches.(4)We improve the optimization model of MCCA by using locality preserving and fractional-order embedding, and then we propose Locality preserving multiset canonical correlation analysis based on fractional-order embedding(FLMCC). We make many experiments to show the effectiveness of FLMCC,and then we also compare the effects of fractional factor on recognition performance.