Feature-level fusion recognition based on complex-valued independent component analysis
Yanjie Wang · Optics and Precision Engineering · 2009
The basic principle and algorithm of the complex valued Independent Component Analysis(ICA) are introduced,and a new target recognition algorithm based on the complex valued ICA is proposed to be applied to the multi-sensor fusion target recognition. Two images from different cameras are composed into a complex value training group matrix,then the Fast ICA(FICA) algorithm is performed on the matrix to get independent components (ICs).After extracting the features of training set and testing set based on ICs,the linear discriminant analysis is adopted to train target features to find out a reasonable classification threshold.Finally,the targets in testing set are classfied and recognized.Experiments show that the recognition rate obtained by the proposed algorithm is 92.1%,which is more excellent than those of the traditional ICA in 78.1% and PCA in 76.2%.