Abstract:

Since the classical kernel matching pursuit (KMP) learner can only address the binary classification problems, a KMP based on Hadamard correct codes is proposed, borowing the idea of errorcorrecting output codes (ECOC). This algorithm can be extended to multiclass classification through Hadamard correct codes. Moreover, the correcting ability of the correct codes can enhance the generation capability of the classifier. The illustrative experiments implemented on UCI datasets and three kinds of aero target high resolution range profile (HRRP) data set demonstrate that the proposed algorithm outperforms two other coding methods in both classification performance and robust ness.