Abstract:

An improved kernel clustering based discriminant analysis (IKCDA) method is proposed. The data for each class is firstly partitioned into multiple clusters via the fast global kernel k-means clustering algorithm, and then the optimal projection vectors are found based on these clusters. Experimental results performing on the UCI machine learning dataset demonstrate that samples belonging to different classes are more separable by the IKCDA method. Moreover, experimental results performing on synthetic aperture radar ground stationary targets based the moving and stationary target acquisition and recognition (MSTAR) public database also indicate that the classification capabilities of the true objects classes and the rejection capabilities of the confusers classes can be bettered via the IKCDA method.