Reconstruction of remote sensing imagery via fused multimode neural network computing

Abstract

We address a new multimode system/method fusion oriented neural network (NN) computing approach to enhancement of conventional low resolution remote sensing (RS) radar and/or fractional synthetic aperture radar imagery. First, the squared error norm objective function minimization-based descriptive experiment design regularization (DEDR) framework is adapted to the Hopfield-type neural network computing-based feature enhancing image reconstruction from the low resolution initial RS imagery. Second, the DEDR framework is aggregated with the variational analysis inspired total variation (TV) minimization modality aimed at anisotropic feature-enhanced image recovery with locally selective information fusion adaptively balanced over speckle and noise suppression. The DEDR and the TV enhancement modalities are fused into the TV-structured maximum entropy neural network (MENN) technique. The developed DEDR-TV-structured MENN-implemented RS image enhancement method outperforms the recently proposed competing approaches both in the achievable resolution enhancement over noise suppression and the convergence rates that is corroborated via the reported simulations.