Multi-aspect target classification using hidden Markov models for data fusion
Tactical targets often exhibit a monostatic response that is a function of target-sensor orientation. In SAR image formation, this aspect dependence is lost through integration over the synthetic aperture. The aspect dependent response may be recovered through directional filtering of the focused image, yielding a sequence of subaperture images. The response of a target at an unknown orientation is subsequently modeled using the feature statistics of the subaperture image sequence in conjunction with a hidden Markov model (HMM), with the states of the HMM corresponding to target-sensor intervals over which the target response is relatively invariant. The features extracted from each observation are derived from two-dimensional matching pursuits. Performance is quantified against standard parametric detection architectures for measured data.