ISAR Target Feature Enhancement Based On Signal Sparse Representation

For restricted by signal bandwidth and coherence accumulated angle, radar resolution is limited. As a new radar imaging technique, inverse synthetic aperture radar (ISAR) target feature enhancement technique utilizes single radar signal or multi-radar signals obtained from different angles, frequency bands, and time to increase precision of parameters estimation of scattering model, and to get a high-resolution image. signal sparse representation is an effective data analyze method, and it can accurately represent original signal in transform domain by basis functions as less as possible. Actually, ISAR imaging is a signal representation problem, therefore signal sparse representation can be applied in ISAR target feature enhancement. Utilizing prior information of sparsity, sparest signal representation is solved on a over-complete dictionary in order to extracte scattering parameters more accurately and improve image quality greatly, which is beneficial to image analyze and target recognition.Aiming at the background of ISAR imaging, this thesis studies ISAR target feature enhancement technique based on signal sparse representation, including multi-radar signals mutual coherent processing, same angle and multi-band ISAR target feature enhancement technique, multi-angle and multi-band ISAR target feature enhancement technique and so on. Firstly, multi-radar signals mutual coherent processing is reserched. A methodbased on range poles and cross-range poles is given to estimate the referenced range errors and the angles between radars'line of sight; Because the traditional methods based on spectral estimation to estimate the amplitude-phase compensation parameters need smooth processing, and the estimated precision is not very well, multi-band signal mutual coherent processing based on sparse representation is proposed. It uses the sparsity of compensation parameters to estimate uncorrelated terms by sparse reconstruction method. This method overcomes the disadvantages of traditional methods, and improves the precision and stability of parameters estimation.Secondly, this thesis studies same angle but multi-band ISAR target feature enhancement technique in detail. Aiming to the shortcoming that the number of sacttering centers must be known predictively in spectral estimation, signal representation models of multi-band ISAR signals are constructed. After researching the limitation of Basis Pursuit that the sparest solution may not be obtained, we propose two improved methods. One is the method based on Basis Pursuit combined with AR extrapolation, and the other is Sparse Bayesian Learning method. They don't need to estimate the number of scattering centers, and the solutions obtained by the two methods is sparser than the solutuion of Basis Pursuit. For extracting polarization characteristic of target, multi-band coherent polarization Geometrical Theory of Diffraction model is constructed, and SBL is applied in full-polarization multi-band ISAR target feature enhancement to overcome the shortcoming of spectral estimation methods. The simulation experiment shows the improvement of estimated precision.Subsequently, multi-angle and multi-band ISAR target feature enhancement technique is researched detailedly. For solving the mismatch of two-dimensional poles, the extended matrix enhance and matrix pencil method is proposed to improve accuracy of parameter estimation and matching precision. Compared with modified Root-MUSIC, the performance of the method is better, but it also need to estimate the number of scattering centers. Therefore, multi-angle and multi-band ISAR signal representation model is constructed, and signal sparse reconstruction methods are used to estimate parameters of these models. Furthermore, multi-angle and multi-band ISAR target feature enhancement for edge enhanced is researched to retain sparsity and enhance edge feature of target. Aiming to the phenomenon that specular responses occur in the wide angular domain, and the precision of traditional specular angle estimated method is lower, a method based on genetic algorithm is proposed to improve estimated accuracy of model parameters, and then a method is given to estimate frequency dependent factor in the case of narrowband processing.Finally, the performances of ISAR target feature enhancement based on signal sparse representation is analysed. The effects of SNR and band interval on the estimated performance of different multi-band ISAR target feature enhancement methods are analysed based on Cramér-Rao Bound; Then we reduce ISAR target feature enhancement processings to a Compressed Sensing problem. Based on Compressed Sensing theory, the effect of sparisty and the number of sampling on multi-band ISAR target feature enhancement processing is analyzed. Furthmore, correlation measurement based on point spread function is proposed as evaluation criteria to optimize two-dimensional sampling matrix and to guide setting of multi-radar.