N.S. Akinshin, A.V. Boldin, R.P. Bistrov, A.V. Khomyakov
The decision rule for recognition of two classes of targets is determined when using statistics of Poincaré sphere zero polarizations as a criterion. To reduce the influence of underlying surface reflections on recognition reliability, signal characteristics adaptive adjustment of the classifier is used. Optimal adaptive procedure of classification is reduced to comparison of the likelihood ratio logarithm with variable threshold. Exact values of decision statistics are replaced by their assessments obtained at the stage of training.
Analytical relationship for probability of correct recognition of two classes of targets is obtained.
Assessment of decision statistics distribution parameters at the stage of training results in 14-15 dB gain in comparison with the unknown parameters case.