Diagonal Clustering-based discriminant analysis for synthetic aperture radar Automatic Target Recognition

A novel subspace analysis method called diagonal clustering- based discriminant analysis (DiaCDA) is proposed. Unlike principal component analysis (PCA), fisher linear discriminant (FLD) and clustering-based discriminant analysis (CDA), DiaCDA directly seeks the optimal projection vectors from 2-dimensional diagonal image matrices not from 1-dimensional original image vectors. Moreover, the advantage of DiaCDA over 2-dimensional PCA (2DPCA), 2-dimensional FLD (2DFLD) and 2- dimensional CDA (2DCDA) is that DiaCDA seeks the projection vectors by interlacing both row and column information of the original images, while 2DPCA, 2DFLD and 2DCDA seek the projection vectors by using only row information of the original images. Experimental results conducted on the moving and stationary target acquisition and recognition (MSTAR) public database show that DiaCDA has obtained higher recognition rates or at least the same ones as 2DPCA, 2DFLD, and 2DCDA. In addition, the recognition performances can be further improved and feature dimensions can be reduced sharply by combining DiaCDA with 2DCDA.