A new feature vector using selected bispectra for signal classification with application in radar target recognition

Radially integrated bispectra (RIB), axially integrated bispectra (AIB), and circularly integrated bispectra (CIB) were used as feature vectors of signals, but many bispectra on integration paths may be redundant, and some bispectra are even baneful for signal classification. To avoid these problems, this paper proposes using selected bispectra with the maximum interclass separability as feature vectors of signals. In radar target recognition, range profiles are suitable feature vectors, but they have two main shortcomings: sensitivity to time shift and aspect dependence. Since the selected bispectra of range profiles are translation invariant and can avoid redundant and baneful bispectra as features, they are thus especially suitable for radar target recognition, which is shown by experiments.