Sparse Targets Of ISAR High-resolution Imaging

For the traditional inverse synthetic aperture radar(ISAR) imaging method, the azimuth resolution is limited by coherent processing interval, high-resolution ISAR image has important significance in the target classification and recognition. Due to the non-cooperation and strong maneuverability imaging target, relying on increased ISAR coherent integration time to improve the image azimuth resolution and imaging algorithms for radar systems is unrealistic. This article is based on the strong scattering point in the ISAR image is only a very small part, that is, the ISAR image is sparse prior knowledge, analyzed using compressed sensing, cosparse and Gaussian mixture model in wavelet domain, in short coherent processing interval the realization of super-resolution ISAR imaging feasibility, and were given based on the theory of super-resolution ISAR imaging of these methods. Through the measured ISAR data to validate the experimental results of these algorithms show our algorithm can get high quality super-resolution ISAR images.In this article, we propose a multilayer compressed sensing imaging structure based on the traditional compressed sensing imaging. We divide ISAR imaging process into several stages, so that the azimuth resolution ISAR image gradually increase. At each stage we regard the ISAR imaging as a sparse coding process, the scattering coefficient is coded coefficients. Encoding and decoding matrix to form a matrix of Multilayer Perceptron, using the error back propagation algorithm(BP) decoding matrix and optimized encoding matrix, making reconstruction of ISAR echo data error is minimized. After the encoding process to join the excitation function, which greatly simplifies the difficulty of controlling the sparsity sparse selection. Decoding matrix as a sparse dictionary, based on compressed sensing theory, the use smoothing 0? algorithm(SL0), recover the ISAR image, and then position the inverse Fourier transform, the equivalent echo data, as the next stage inputs. Azimuth resolution at each stage were higher than the previous stage, the final output of the super-resolution ISAR images.Then we present a ISAR super-resolution imaging method based on cosparse. Unliketraditional imaging methods based on compressed sensing is an inverse problem, we construct a forward imaging model solving. We regulate and parsing operations Operators combine learning phase error, learning to get analytical operator can be mapped to a high dimensional Doppler echo space. Furthermore, in order to strengthen the robustness of the algorithm to noise, we joined the regularization term, with enhanced Lagrange algorithm to approximate the signal de- noising. To take advantage of the sparsity of ISAR images, we were inspired by orthogonal matching pursuit algorithm, and made some improvements to adapt our forward model, using a modified orthogonal matching pursuit algorithm recovers the target of strong scattering points, get focus is good, less noise, high-quality super-resolution ISAR images.Finally, this article presents a wavelet domain based on Gaussian mixture model ISAR super-resolution imaging methods. The traditional ISAR imaging based on compressed sensing is the use of constraints to achieve de-noising, when the low SNR conditions, compressed sensing put some noise point as a strong scattering target recovered. In super-resolution imaging of sparse basis, we recovered from the scattering coefficient Doppler domain to wavelet domain, the wavelet low frequency part is smooth part ISAR image using wavelet low- frequency coefficients established Gaussian mixture model with expectation maximization algorithm(EM) to estimate the parameters, the successful separation of the target and the noise points, background point, then reverse adjustment resolve operator, operator makes parsing operation has been optimized, and then use the theory of sparse wavelet coefficients to recover the ISAR image after wavelet ISAR inverse transform of the image, so that we can still get a clear ISAR at very low SN R conditions super-resolution images.