Yuriy V. Shkvarko

ℓ2−ℓ1 Structured descriptive experiment design regularization based enhancement of fractional SAR imagery

Abstract Feature-enhanced reconstruction of the reflectivity maps (remotely sensed scene images) from the low-resolution fractional SAR imagery is treated for harsh sensing scenarios with uncertainties attributed to possible imperfect sensor calibration, atmospheric turbulence and uncontrolled carrier trajectory deviations. These effects lead to the randomly perturbed signal formation operator resulting in a partial coherence of the system. A low-resolution scene image formed using the conventional matched spatial filtering method serves as a starting point for the enhancement. We commence with the descriptive experiment design regularization (DEDR) approach for solving the image enhancement inverse problem based on the l 2 -type squared error norm minimization strategy robust against the problem model uncertainties in the sense of the worst case statistical performance optimization. To exploit structural information on the desired image piecewise smoothness over the scene the l 1 structured regularization level is incorporated via aggregating the image total variation (TV) minimization approach with DEDR. In the unified DEDR-TV framework, the image gradient magnitude map sparsity and the overall image texture anisotropy properties are structured by combining the l 2 image metric with the l 1 image gradient metric in the solution space. The incorporated projections onto convex solution sets guarantee the convergence and speed up the resulting implicit iterative enhancement scheme.

Convex regularization-based hardware/software co-design for real-time enhancement of remote sensing imagery

In this paper, we address a new approach for high-resolution reconstruction and enhancement of remote sensing (RS) imagery in near-real computational time based on the aggregated hardware/software (HW/SW) co-design paradigm. The software design is aimed at the algorithmic-level decrease of the computational load of the large-scale RS image enhancement tasks via incorporating into the fixed-point iterative reconstruction/enhancement procedures the convex convergence enforcement regularization by constructing the proper projectors onto convex sets (POCS) in the solution domain. The established POCS-regularized iterative techniques are performed separately along the range and azimuth directions over the RS scene frame making an optimal use of the sparseness properties of the employed sensor system modulation format. The hardware design is oriented on employing the Xilinx Field Programmable Gate Array XC4VSX35-10ff668 and performing the image enhancement/reconstruction tasks in a computationally efficient parallel fashion that meets the near-real time imaging system requirements. Finally, we report some simulation results and discuss the implementation performance issues related to enhancement of the real-world RS imagery indicative of the significantly increased performance efficiency gained with the developed approach.

High-resolution imaging with uncertain radar measurement data: A doubly regularized compressive sensing experiment design approach

The descriptive experiment design regularization (DEDR) paradigm is aggregated with the variational analysis approach that combines the ℓ2 image metric with the ℓ1 sparse image gradient map metric structures in the solution space. The proposed ℓ2 - ℓ1 structured total variation DEDR (TV-DEDR) framework is particularly adapted for enhanced imaging with low resolution side looking airborne radar/fractional SAR sensors putting in a single optimization frame adaptive SAR image despeckling and resolution enhancement that exploits the structured desired image sparseness properties. The TV-DEDR method implemented in an implicit contractive mapping iterative fashion outperforms the competing nonparametric adaptive radar imaging techniques both in the resolution enhancement and computational complexity as verified in the simulations.

Towards super-resolution recovery of microwave sensor imagery: A unified descriptive experiment design regularization framework with projections onto nested resolution frames

We address a new approach for enhanced microwave remote sensing (RS) imaging via performing the imaging system kernel point spread function (PSF) operator refinement-based multi-scale iterative reconstructive (MSIR) image post-processing, as required for emerging feature enhanced RS missions. The high-resolution (HR) image is first reconstructed from the initial low-resolution (LR) image employing the statistically optimal minimum risk inspired descriptive experiment design regularization (DEDR) framework. Next, to approach the overall super-resolution (SR) imaging performances we incorporate into the DEDR method the additional postprocessing stage aimed at filling in the null space of the HR imaging system PSF operator via performing the corresponding projections onto the nested refined resolution greed frames. We feature the differences between the proposed DEDR-MSIR resolution refinement approach and the most competing celebrated Papoullis-Gerchberg SR method adapted for the feature enhanced RS imaging and demonstrate the advantages of the unified DEDR-MSIR approach for SR image recovery.

Multilevel descriptive experiment design regularization framework for sparsity preserving enhancement of radar imagery in harsh sensing environments

We address a new approach to a reconstructive imaging inverse problems solution as required for enhancement of low resolution real aperture radar/fractional SAR imagery in harsh sensing environments. To preserve the image and image gradient map sparsity peculiar for real-world remote sensing (RS) scenarios, we aggregate the minimum risk inspired descriptive experiment design regularization (DEDR) framework for balanced image resolution enhancement over noise suppression with two additional regularization levels: (i) the variational analysis inspired minimization of the image total variation (TV) map and (ii) the sparsity preserving regularizing projections onto convex solution sets (POCS). The new framework incorporates the TV metric structured regularization into the weighted l2 metric structured DEDR data agreement objective function and suggests the solver for the overall reconstructive imaging inverse problem employing the DEDR-TV-POCS-restructured MVDR strategy. The DEDR-TV-POCS method implemented in an implicit iterative fashion outperforms the competing nonparametric adaptive radar imaging techniques both in the resolution enhancement and computational complexity reduction as verified in the reported simulations.

Image Super-Resolution via Block Extraction and Sparse Representation

Super-Resolution (SR) has many applications in several issues of the image processing by obtaining High-Resolution (HR) images from Low-Resolution (LR) images. In this paper, a SR technique that can increase the resolution in images of different nature is proposed. Our approach in obtaining SR image, first uses Lanczos interpolation of initial LR image, then edge features are extracted via convolution of an image with two different filters; following, the most informative features are performed employing principal component analysis (PCA). In next step, preprocessed image presented in blocks is used, where for an each block its sparse representation is performed using LR dictionary and another HR dictionary. In final step, the SR blocks are reconstructed resulting in improved SR image. Experimental results demonstrate the effectiveness of our method in comparison to state-of-the art techniques in terms of objective criteria PSNR, MAE and SSIM values as well as in subjective visual performance. Additionally, the proposed technique significantly reduces computational time in SR reconstruction.

Solving Enhanced Radar Imaging Inverse Problems: From Descriptive Regularization to Feature Structured Superresolution Sensing

Resolution enhancement (RE) and superresolution (SR) enhancement of the remote sensing imagery provided by conventional low-resolution (LR) coherent real aperture radar or fractional SAR sensors operating in real-world scenarios with the model data statistics unknown to the observer belong to a class of nonlinear uncertain inverse problems. The classical Bayesian statistical inference and modern compressed sensing-based approaches are not properly adapted to such inverse problems as the latter require robust spatially selective image despeckling adaptively balanced over RE with preservation of salient radar/SAR image features. In this paper, we propose to treat the RE/SR radar imaging inverse problems in a multistage descriptive experiment design regularization (DEDR) setting that logically unifies composite RE optimization tasks into a four-level structured DEDR framework. First, the conventional despeckled LR image formed via performing the matched spatial filtering of the recorded trajectory signals is considered as initial LR data for further feature enhanced processing. Second, the minimum risk inspired robust adaptive beamforming method is DEDR restructured and unified with the convergence guaranteed and sparsity promoting composite projections onto convex sets aimed at feature enhanced recovery of the high-resolution despeckled image. The third level suggests transition to the nested refined SR scales with preservation of the consistency space. Finally, the SR recovery of the fine image features is performed consequently in each nested refined image frame via discrete wavelet domain postprocessing-based sparsity promoting denoising with consistency preservation. The new multistage SR-DEDR technique conjugates excellent despeckling and SR enhancement performances corroborated via reported computer simulations.

Maximum entropy neural networks for feature enhanced imaging with collaborative microwave multi-sensor data fusion

We present a collaborative neural network (NN) computing-oriented approach for feature enhanced reconstruction of microwave remote sensing (RS) imagery via sensor data fusion. Two reconstruction/fusion frameworks are proposed and featured. Both unify the maximum entropy and descriptive experiment design regularization (DEDR) paradigms but employ different NN-based fusion (NNF) strategies. The first one addressed as RS-NNF(1) aggregates the adaptively weighted DEDR-structured individual sensor image recovery objective functions, while the second one addressed as RS-NNF(2) performs cooperative multi-sensor statistical recovery performances enhancement-oriented fusion. The simulations corroborate superiority of both proposed technics over the conventional non-collaborative RS image fusion with RS-NNF(2) on top.

Multi-polarimetric SAR adapted virtual beamforming-based techniques for feature enhanced tomography of forested areas

Modern sum of Kronecker Products decomposition technique for multi-polarimetric multi-baseline SAR tomography of forested scenes results in rank-deficient data covariance matrices (CV) restricting the usage of the conventional Capon beamforming-based focusing technique due to the ill-posedness of the involved CV inversions. To overcome these limitations, in this paper we address and compare three modified robust adaptive virtual beamforming-based SAR-adapted tomographic imaging approaches addressed as: robust Capon beamforming, norm constrained Capon beamforming and double constrained robust Capon beamforming (DCRCB) techniques. Those manifest considerable feature enhanced tomographic imaging capabilities with DCRCB on top that we corroborate via processing real-world airborne TomoSAR data.

SAR tomographic imaging technique based on fusion of the Prony-inspired parametric and MVDR-inspired non-parametric DOA spatial spectral estimators

In this paper, the synthetic aperture radar tomography (SARTom) vertical distribution estimation problem is treated within the direction of arrival (DOA) estimation framework. Super-resolution parametric DOA estimation methods improve the vertical resolution and mitigate the effect of sidelobes. Nevertheless, these techniques have the main drawback related to the assumption that the scene is composed by a finite number of point-type backscattering sources. On the other hand, the minimum variance distortionless response (MVDR) inspired non-parametric DOA estimation methods are better suited to cope with scenarios characterized by the presence of distributed scatterers. In this work, we propose to decompose the SARTom vertical distribution estimation problem into two paradigms: parametric DOA estimation for point-type scatterers and non-parametric recovery of the spatial spectrum pattern (SSP) of the spatially distributed scattering components. The principal innovative contribution of this study relates to the proposition for fusion of the Prony-inspired parametric and MVDRinspired non-parametric DOA estimation paradigms through the use of the spectral positional invariance property of the point-type targets, which holds with the extended Prony model.