Atmospheric turbulence-degraded images in typical practical application scenarios are always disturbed by severe additive noise. Severe additive noise corrupts the prior assumptions of most baseline deconvolution methods. Existing methods either ignore the additive noise term during optimization or perform denoising and deblurring completely independently. However, their performances are not high because they do not conform to the prior that multiple degradation factors are tightly coupled. This paper proposes a Noise Suppression-based Restoration Network (NSRN) for turbulence-degraded images, in which the noise suppression module is designed to learn low-rank subspaces from turbulence-degraded images, the attention-based asymmetric U-NET module is designed for blurred-image deconvolution, and the Fine Deep Back-Projection (FDBP) module is used for multi-level feature fusion to reconstruct a sharp image. Furthermore, an improved curriculum learning strategy is proposed, which trains the network gradually to achieve superior performance through a local-to-global, easy-to-difficult learning method. Based on NSRN, we achieve state-of-the-art performance with PSNR of 30.1 dB and SSIM of 0.9 on the simulated dataset and better visual results on the real images.

Blind Restoration of Atmospheric Turbulence-Degraded Images Based on Curriculum Learning

Atmospheric turbulence-degraded images in typical practical application scenarios are always disturbed by severe additive noise. Severe additive noise corrupts the prior assumptions of most baseline deconvolution methods. Existing methods either ignore the additive noise term during optimization or perform denoising and deblurring completely independently. However, their performances are not high because they do not conform to the prior that multiple degradation factors are tightly coupled. This paper proposes a Noise Suppression-based Restoration Network (NSRN) for turbulence-degraded images, in which the noise suppression module is designed to learn low-rank subspaces from turbulence-degraded images, the attention-based asymmetric U-NET module is designed for blurred-image deconvolution, and the Fine Deep Back-Projection (FDBP) module is used for multi-level feature fusion to reconstruct a sharp image. Furthermore, an improved curriculum learning strategy is proposed, which trains the network gradually to achieve superior performance through a local-to-global, easy-to-difficult learning method. Based on NSRN, we achieve state-of-the-art performance with PSNR of 30.1 dB and SSIM of 0.9 on the simulated dataset and better visual results on the real images. 

/pdf/blind-restoration-of-atmospheric-turbulence-degraded-images-3ia7zl32.pdf

Satellite based synthetic aperture radars (SAR) are widely used for surface remote sensing. In last years the creation of space-based multi-position SAR (MPSAR) has become more relevant. A review of the existing multi-position spaceborn SAR systems has been performed. An implementation of a satellite-based MPSAR system has been suggested. The results of mathematical modeling of the satellite-based MPSAR has been presented.

Issues of developing multi-position satellite based SAR

The problem of the optimal signal processing algorithm synthesizing in synthetic aperture radars equipped with a mode of power lines selection is solved by the maximum likelihood method. Polarization and Doppler differences are selected as the distinguishing features of power lines from the earth's surface. The useful signal at the output of the coherent receiver is specified in an analytical form with a complex envelope and high-frequency harmonic oscillations. The background radiation of the underlying surface is presented in stochastic form and is completely determined by the matrix of correlation functions. The internal noise of a multichannel receiver is described by a delta-correlated random Gaussian process with the same power spectral density in different observation channels. Due to the fact that the background radiation of the earth surface is correlated in time and at different polarizations the likelihood functional is recorded with the inverse matrix of inverse correlation functions. The obtained optimal algorithm consists in suppression of passive interference in decor-leveling filters, optimal coherent accumulation of trajectory signals for synthesizing the on-board antenna aperture and their optimal filtering in filters matched with the reference signal. Simultaneous use of Doppler and polarization differences between wanted signals and passive interference allows effectively allocate power lines against the background of reflections from the ground in multichannel synthetic aperture radars. The proposed sequence of operations on the received oscillations can be a functional addition to the operating modes of aerospace-based cognitive radars adapting in accordance with the covariance scattering matrix of the underlying surface known in advance or measured during remote sensing. The block diagram of a multichannel onboard polarization radar with synthesized antenna aperture is designed on the basis of the obtained results. The development of such a radar is relevant for a collision warning system for helicopter pilots especially in bad weather conditions and at night.

Structure Optimization of the Multi-Channel On-Board Radar with Antenna Aperture Synthesis and Algorithm for Power Line Selection on the Background of the Earth Surface

Usage of a multiposition system can improve technical characteristics of radar. Arrangement of multiple transmitting positions can significantly improve the angular resolution of radar. This paper presents method of multiposition radar aperture synthesis that operating in the range of distances, in the near field. The described method involves sequential scanning of space both in angle and in distance. This paper presents the results of mathematical modeling of multiposition radar transmitter using this method.

Methods of multiposition radar aperture synthesis

In a case of significant aircraft trajectory instability it’s difficult to define the synthesis parameters to obtain the best available resolution. A radar image synthesized with incorrect parameters looks blurry. There are algorithms which allow to define the synthesis parameters to improve image quality. Most of the algorithms are used in digital photography and microscopy. Radar images have the features that distinguish them from a photo thus the additional research of the robustness and the focus measure quality have to be performed.Different methods comparison is presented in the paper. The methods are tested on the radar images containing bright point targets, low contrast targets, etc. The most robust method which is suitable for the most of settings is defined by comparison.

Defining of the Robust Criteria for Radar Image Focus Measure

During the development of synthetic aperture radar (SAR) for operational monitoring of earth surface state, the need for several special antennas has arisen. The chosen metric of earth surface state is radar vegetation index (RVI). An ability to measure RVI requires the antennas to use multiple polarizations with a possibility to switch between them quickly. This article covers the process of antenna development from modeling to measurement and field testing, as well as ways to solve problems arising in the process.

Multiple Bands Dual Polarization Antenna System for Ground Surface State Estimation

The paper proposes an architecture of lossy multiplier for FPGA implementation. The proposed multiplier requires less logic resources but produces some error. This multiplier is being studied as a part of digital processing blocks: FIR-filters and FFT-processor. The results are verified using mathematical modeling tools and FPGA synthesis tools.

Ilya A. Kuzmin

Papers

Defining of the Robust Criteria for Radar Image Focus Measure

Multiple Bands Dual Polarization Antenna System for Ground Surface State Estimation

Issues of developing multi-position satellite based SAR

Methods of multiposition radar aperture synthesis

FPGA Implementation of Digital Processing Blocks Using Optimal Resolution Multipliers