Real-time polarization-diverse features extraction and automated target identification using neural networks

The authors have mapped the polarization-diverse features extraction problem onto the Lyapunov energy function of the Hopfield model neural network to obtain the real-time solution of the feature, i.e., the ellipticity, the tilt angle and the amplitude of the fit ellipse, derived from this parameterization. Simulated results are presented to shows its real-time capacity in the context of the polarization-diverse features extraction problem by using the Hopfield net as well as its strong target identification ability by applying the back-propagation neural network employing the resulting feature sets. The approaches considered here are expected to have a faster computational speed than those of the traditional approaches such as the LMS (least mean square) approximation method and classical pattern recognition.<>