Nonparametric weighted MMC feature extraction method for HRRP recognition

To improve the class separability and storage efficiency for radar target high resolution range profile (HRRP) recognition task, some linear discriminant analysis (LDA) based feature extraction methods have been successfully applied. However, as an effective feature extraction method, traditional LDA encounters four main drawbacks with respect to Gaussian distribution assumption, small sample size problem, limited classes and boundary sample problem, especially, when it is applied in HRRP recognition field. In this paper, a nonparametric weighted maximum margin criterion (NWMMC) is proposed. Compared to the parametric form of other LDA based methods, NWMMC can find the discriminant direction without assuming the data follow the Gaussian distribution. Meanwhile it defines a new criterion to measure the class separability which can solve the small sample size problem and it provides weight to samples near the boundary which tends to increase the class separability. The HRRPs of experiment were obtained from the scattering center model of four different targets. Simulations were presented to evaluate the recognition performance with this method.