SAR Imaging with Undersampled Data Via Matrix Completion
Dong Yang,Guisheng Liao,Shengqi Zhu,Xi Yang,Xuepan Zhang
DOI: https://doi.org/10.1109/lgrs.2014.2300170
IF: 5.343
2014-01-01
IEEE Geoscience and Remote Sensing Letters
Abstract:High-resolution synthetic aperture radar (SAR) imagery of a wide area of surveillance is a difficult large-data problem. In the past few years, researchers have applied compressive sensing (CS) to SAR, as it exploits redundancy in signals. To further extend the sparse problem from the vector to the matrix, a new theory called matrix completion (MC) has attracted much attention, which can complete a matrix from a small set of corrupted entries based on the assumption that the matrix is essentially of low rank. Inspired by this technique, a novel SAR imaging algorithm is proposed in this letter to deal with the undersampled data. After representing the data of a range cell as a matrix, the phase is compensated to keep the matrix holding the property of low rank. Subsequently, MC can be utilized to recover the full-aperture data in the new constructed matrix. Since the data are completely unsampled in the corresponding azimuth cells, the proposed method has effectively conquered the restriction of previous applications that each received channel must have a small number of samples. The final results in both simulation and real-data experiments show that the targets can be well focused even in the scenario of discarding a large percentage of the received pulses. Moreover, when compared with CS, the method is not required to design the complicated measurement matrix.