A resonance region method for recognition of multiple targets using the MUSIC algorithm and a time correlation technique

This paper presents, for the first time in literature, a natural resonance based electromagnetic target classification method that is useful not only for single target recognition but also for the recognition of multiple targets. In this method, the scattered wide-band electromagnetic signals are processed over an optimal late-time region by using the MUSIC algorithm to extract target features called fused MUSIC spectrum matrices (FMSM). These features are almost invariant to aspect variations and to the variations in target-to-target separation distances in the case of multiple targets. A challenging special case of multiple and identical targets is also handled in the suggested method by investigating the time correlation of a given scattered test signal as a part of the decision mechanism. A simpler ldquosingle target classifierrdquo version of this method was suggested recently in (Seemen et al., 2007) and shown to be very successful in recognizing isolated targets of various geometries and material compositions even in high-noise scenarios. Extraction of the FMSM, which is closely related to the discrete version (a matrix) of the multi-aspect fused power spectrum of a given target, is explained in detail in (Seemen et al., 2007). System poles of a given target coincide with the peak points of this FMSM map plotted in the discrete complex frequency domain. This same feature extraction procedure is used for the recognition of single targets in the present paper but crucial modifications are made in the design procedure for the characterization of multiple targets as to be explained.