Abstract:

Support vector data description (SVDD) is usually used to distinguish two classes that the target class can obtain sufficient samples and the nontarget class involves various kinds of objects. However, while used in the field of radar target recognition, the classification ability of SVDD rapidly weakens with an increase in noise energy. In order to deal with such problem, the reason why noise results in the weakening of SVDD is particularly described, and a method of radar target classification based on adaptive SVDD is proposed. The proposed method constructs an adaptive model between signal to noise ratio (SNR) value and optimal hypersphere radius using a receiver operating characteristic curve, which can adaptively choose the decision thresholds of different SNR values in target classification. The experiment results demonstrate that the adaptive SVDD algorithm greatly improves the classification performance of targets in low SNR condition compared with the classical SVDD algorithm.