Research On Radar Imaging Algorithm Based On Compressed Sensing

It is an important work to achieve a high-resolution imaging result in the radar system. Based on the conventional radar imaging thoery, it needs to improve the bandwidth of the transmitted signal. According to the Nyquist theorem, this requires a high sampling rate, and leads to a heavy burden of data sampling, storage, and transmission.Compressed sensing(CS) is a technique to capture a compressible signal at a rate below the Nyquist rate. It measures a small number of linear projections of a compressible signal and then reconstructs the signal from these projections using an optimization process. According to CS theory, this technique can be applied to radar imaging to release the burden of data sampling, storage, and transmission. Thus, there will be great change in applying CS to high-resolution radar imaging.In this paper, we apply CS to two kinds of typical sparse targets, the ship in ocean and the ballistic target in space, to deeply develop the technologies for radar imaging.Moreover, the applications of feature extraction and recognition for the target based on the CS-SAR imaging results are well investigated. Accordingly, this paper focuses on the following works.First, the optimization of the measurement matrix for CS is investigated. In order to acquire a more accurate reconstruction result, a lower linear coherence between columns of the measurement matrix is required. Two algorithms, based on the mutual coherence, are presented to optimize the measurement matrix. First, a completely orthogonal matrix is used to replace parts of columns of the measurement matrix to construct a semi-random matrix. Then, an iterative procedure is applied to this semirandom matrix, and it involves the design of shrinkage procedure, the reduction of rank of the Gram matrix, and the decomposition of the measurement matrix. In each iteration, the mutual coherence of the measurement matrix approaches a lower bound.After a series of iterations, the linear coherence between columns of the measurement matrix has minimum value, which is favorable in obtaining a more accurate reconstruction result.In this paper, a novel synthetic aperture radar(SAR) imaging algorithm is presented. This algorithm uses compressed sensing to reconstruct the scattering coefficients of the target scene with a low sampling rate. This algorithm is efficient, blind and based on two-dimensional CS. First, a conventional imaging algorithm, like rangeDoppler algorithm, is used to generate the complex image of the target scene from the raw signal. Then, a greedy algorithm is applied to this complex image, and it involves the selection of threshold, the estimation of the point spread function, and the removal of the complex image of the target point. Since based on two-dimensional CS, this algorithm fully utilizes the spatial sparsity. By applying the greedy algorithm to the complex image rather than the original signal and by limiting the peak search to a small set of pixels instead of the entire complex image, this algorithm improves the computational efficiency greatly. In addition, this algorithm is based on blind CS,that is, the point spread function is estimated from the signal. This means that this algorithm applies even if the radar parameters are unknown.In this paper, a novel method is presented to simulate the echoes from ballistic targets when the radar transmits chirps or other waveforms. A tool for electromagnetic calculation, like the CST Microwave Studio, is first used to simulate the frequency response of the electromagnetic scattering. The echo is then acquired from the frequency response by further processing. This technique can be used to build a library of the echoes from ballistic targets, and this library can be used to train and test the classifier of warheads and decoys.In this paper, a novel recognition algorithm based on high-resolution range profile is presented for the ballistic target. First, two features which represent the scattering characteristic are extracted from the high-resolution range profile of the ballistic target.Then, a two-dimensional feature space is constructed based on the two features. Finally, a classifier resulting from statistical property and sequence attribution is applied to this space to obtain the recognition rate and false alarm rate. Based on the simulated data, this algorithm can achieve good recognition results for the sparse targets.In addition, in real applications, the bandwidth of the transmitted signal is limited,and thus the resolution of the range profile may degrade. In this paper, CS is used to reconstruct the scattering coefficient from the echo signal, and the range profile with a higher resolution is acquired. Based on this, we can achieve the recognition result comparable with that resulting from the signal with a large bandwidth.Many sets of data are used to evaluate the performance of our algorithms. These results demonstrate that the proposed algorithms are effective for the sparse targets.In addition, by applying CS to the downsampled signal, the result which is comparable with that from fully sampled signal can be acquired for radar imaging and target recognition, and this is valuable in military and civil fields.Finally, we conclude this paper and list the potential reaserch topics in future works.