| Fourier transform is the basis for the coherent processing in radar imaging. In some special cases, however, the utilization of super-resolution algorithms can often bring much mare benefits than those brought by Fourier transform. For example, in SAR imaging, if more details of some special areas in a large scene are needed, super-resolution algorithms offer finer resolution; in ISAR imaging of the maneuvering targets, as the Doppler spectrum is time-varying and the Time-Frequency distribution of sub-echoes are much more complex, the one-order approximated signal model is connonly used during a short data collection interval, and the resolution degradation caused by the decreased data interval can be avoided by the use of ultra-resolution algorithms. However, the excellent performances of the super-resolution methods are 慴ased on the accurate knowledge of the data model, and what抯 more, they are much mare sensitive to the errors than traditional methods. If there is a large error existing in the data model, the performance of the super-resolution algorithm will degrade extremely, and in some cases, even worse than that of the traditional methods. So, in the super-resolution radar imaging, more attention must be paid to the applicability of the data model. For instance, range curvature effect must be taken into account when the scene of the imagery is large; the Doppler drift caused by the non-uniform rotation of the object has to be considered; and the motion induced errors must be compensated further. Above all, when eigenstructure based super-resolution algorithms are employed to estimate the two dimensional(2-D) spectrum (2-D image) from the 2-D phase history data, the correlation matrix is usually to be estimated. But in radar imaging, the data is recorded only in one frame, an equivalent of one snapshot in array signal processing, so some sub-aperture smoothing average must -iv. ADSTR1~CF be applied to the full aperture data to Obtain the correlation matrix. In the estimation of the correlation matrix, there is saa~e~ contradiction in choosing the size of the sub-aperture: if the size is large, a small amount of average is performed, so the estimate accuracy of the correlation matrix is low; and if the size of the sub-aperture is small, the resoluti(x1 is liznitedby the aperture size. This problem, together with the reduction of the amount of computations, must be solved properly in super-resolution imaging. Based on the above considerations1 in this thesis, the main research work focuses on the following aspects: An efficient super-resolution algorithm for radar imaging is proposed. In eigenstructure based super-resolution algorithms1 estimates of the 2-D frequencies are obtained by searching for spectral peaks in the 2-D space. To get the correlation matrix estimate, sane kind Qf sub-aperture smoothing average must be made. To reduce the amount of c~nputation and mitigate the performance degradation in smoothing average, a computationally efficient algorithm is proposed. ?The super-resolution radar imaging of re3.ative large object is studied. In the traditional super-resolution algorithms, 2-D sinusoids are used as the signal model of scatterers, so image blurring will occur when scatterers motion through resolution cell exist in the regions relatively far from the reference point. The 2-D approximation model is established in this c...
|
PDF Full Text Request