Study On ISAR High Resolution Imaging And Parameter Estimation Techniques

With the development of national economy and the transformation of modern wars, the radar imaging technique has played an irreplaceable role in both civilian and military remote sensing, mainly attributed to its distinctive capabilities of high-resolution imaging at long-range, day and night, and in all weather conditions. As one of the most important means for space, air and marine suivallence, inverse synthetic aperture radar(ISAR) imaging is paramount to non-cooperative target recognition.To satisfy the growing requirements on numerous applications, ISAR is currently developing with the trend to multiple functions, multiple dimensions and collaborative networks. The various working patterns and data acquisition modes, as well as the intricate target motions, greatly challenge the current ISAR imaging system in high-resolution image formation and target parameter extraction, etc. Under the support of several national programs, e.g. National Key Basic Research Program(973), this dissertation focuses on the problems of low resolution induced by short apertures, incoherence between sparse apertures, and cross-range scaling for(non-)maneuvering targets. The primary goal of this study is to enhance the ISAR imageries and investigate robust parameter estimation approaches for automatic target identification. Listed in the following is the outline of this dissertation:(1) High-resolution sparse image reconstruction for short-aperture ISAR data Due to the multi-mode status of the radar system or the maneuverability of the target, short apertures commonly occur in ISAR imaging. Although it is simple and efficient to obtain images in this case, the target recognition performance might be limited as a result of the the low image resolution. Therefore, a sparse reconstruction method for high-resolution image is presented based on the theory of Compressive Sensing in Chapter 2, after an overview of ISAR signal model and typical motion compensation techniques. This scheme puts an emphasis on a data-driven method for sparsity controller estimation from a statistical point of view. By deducing the regularization problem of high-resolution image retrieval, the sparsity controller is found to have an analytical form via the maximum likelihood estimation. One of the parameters, namely the noise variance, can be estimated from massive noise samples in the rough image, while the other parameter can be initialized by the pre-denoised image. It is preferable to update the sparsity controller with the resultant high-resolution image and re-solve the optimization problem. Several iterations are favorable for improving the algorithm performance. Results approve that the iteratively estimated sparsity controller can well balance the signal fidelity and sparisty during the sparse reconstruction of the high-resolution image.(2) Coherent processing and high-resolution imaging for sparse aperturesIn addition to the diversity of radar modes, external or intrinsic interference may also lead to data missing in the form of sparse apertures. After independent processing of range alignment and autofocus for each block-sparsely distributed subaperture, discrepancy in residual linear phase and complex amplitude will appear between the measured apertures. To deal with this incoherence, Chapter 3 illustrates a coherent processing method. All the envelopes of the subapertures are firstly lined up with respect to the center of mass, wherein the range cells containing prominent scatterers are extracted and regarded as all-pole models. The poles as well as the Doppler difference between the blocks are estimated by invoking a root-MUSIC method. Followed by that, the linear phase difference can be easily corrected with a referenced subaperture. Meanwhile, the coefficients of each model are solved via least-square(LS) method. Then the subapertures after shift correction can be extrapolated forward or backward to the whole aperture length, giving rise to simple calculation of the amplitude difference. Based on complete calibration, a partial Fourier Transform dictionary is constructed to recover the missing apertures by sparse signal processing scheme. Experimental results show that the image after coherent processing and missing aperture reconstruction can be distinctly improved.(3) Cross-range scaling for uniformly rotating targetsBesides high-resolution images, target recognition poses requests on the knowledge of target size. Therefore, Chapter 4 and 5 propose two different scaling methods for uniformly rotating targets. The former analyzes the impact on different scaling methods from the residual translation phase caused by motion compensation. A method is firstly devised to estimate the effective rotational velocity estimation(ERV) by cancellation of the scatterer chirp rates, which is immune to the residual phase error. Herein, some scatterer range cells are extracted and imposed with a time-varying autoregressive model. The instantaneous poles can be estimated from a short-time data, and thus sliding the time window results in the Doppler history. According to the mathematic relationship, ERV can be calculated from the Doppler rates of the scatterers locating in different range cells. After rotational phase error compensation, it is suggested to recall the autofocus method to enhance the image focus. To some extent, this method will degrade in the case of low signal-to-noise(SNR) ratio, because of its great dependency on the qualities of the extracted scatterers.For better robustness, Chapter 5 presents another ERV estimation for uniformly rotating targets based on image quality measures. The influence of the residual phase error is also considered, which is parameterized jointly with the ERV. By iteratively compensating the second-order phase error, ERV with the focused image can be obtained when the intensity-squared sharpness(ISS) of the corrected image is maximized. On the basis of that, the image can be further focused by autofocus methods. Since ISS maximization problem is a typical nonlinear least square problem, a Gauss-Newton method is exploited for high efficiency. Comparisons with several existing methods shed light on that the image-quality based method has moderate efficiency but higher accuracy and stability.(4) Cross-range scaling and high-resolution imaging for non-uniformly rotating targetsOn account of non-uniform rotation, range and azimuth is intricately coupled in phase, which engenders increasing difficulties in rotation parameter estimation. In this context, a cross-range scaling method is firstly explored by maximizing the ISS of the image via Matched Fourier Transform(MFT) in Chapter 6. The essence of this method is to replace the linear transform of FFT in RD imaging with the parametric transform of MFT. Herein, the two-dimensional coupled phase error is recursively compensated until the ISS of the MFT image converges to the maximum, yielding the optimal ERV and chirp rate of MFT. Though this image-based method performs well in low SNR situations, strong noise or coarse resolution are still disadvantageous for target recognition. To this end, the high-resolution images are reconstructed by incorporating the partial MFT within the sparse signal processing scheme.