Maximum-Likelihood Estimation for Hybrid Specular/Diffuse Models of Radar Imaging and Target Recognition

We derive maximum-likelihood estimators for the glint coefficient of specular radar targets and for the scattering coefficient of diffuse radar targets. The estimator for diffuse targets is a thresholded version of the squared magnitude of the received data. The estimator for specular targets is more complicated, involving the solution of a nonlinear equation involving ratios of modified Bessel functions. We also consider a hybrid model consisting of both specular and diffuse components. Several iterative algorithms for maximum-likelihood estimation, including an expectation-maximization algorithm, are proposed. CramerRao bounds on estimator performance are derived. Diffuse target models have been previously explored for designing both radar image formation algorithms and model-based automatic target recognition algorithms. We explore reformulating both applications in terms of the purely specular and hybrid specular/diffuse models. For radar imaging, the hybrid model is found to be overparameterized; the maximum-likelihood estimation procedure for this hybrid model always chooses the diffuse component to be zero. For automatic target recognition, we give an example involving high-resolution range profile data.