Target recognition from limited-angle backscatter data

The problem of determining the nature of a target from limited backscatter data has two difficult features. The first is the problem of inverting scattered-field data when multiple scattering arises. There are few procedures to do this, and even fewer computationally feasible algorithms. The second problem lies in the fact that these methods expect to have scattering data taken all around a target, for all possible illumination directions. We have developed a spectral-estimation technique that can make use of prior knowledge about the target, to improve resolution when only limited data are available. It transpires that this approach is particularly effective at identifying support bounds on an actual target. These support bounds provide a better indicator of what the target is than the image derived from the same scattering data, especially when the data are too few to generate a meaningful image. From this we conclude that with a certain number of data points, and a certain degree of detail in one's prior knowledge of a target set, one can determine whether the best course of action is to "shape" the target using a dynamic prior function, or to form an image, in order to make the identification. These latter observations were carried out using real data acquired from a model of a cruise missile. Being a metallic target, there was little multiple scattering, and the inverse-scattering aspect of the identification step was not demanding.