Inverse synthetic aperture radar imaging exploiting dictionary learning

We consider the Inverse synthetic aperture radar (ISAR) imaging with under-sampled data in the framework of compressive sensing (CS) theory. In the existing studies of CS based ISAR imaging or sparse ISAR imaging, the scene to be imaged is simply assumed to be sparse or existing transforms, such as the wavelet transform, etc. are employed to sparsely represent certain features of the target. The imaging quality is actually limited by the sparse representation of the scene, which in the cases aforementioned may not be fully appropriate to the scene to be imaged. In this paper, we exploits the on-line and off-line dictionary learning (DL) techniques to obtain the sparse representation of the scene, respectively and then incorporate such learned dictionaries into the image reconstruction. We demonstrate the performance of the proposed DL based imaging methods using real ISAR data. The results show that the adaptive on-line dictionary learnt from the current data to be processed and the off-line dictionary learned from the previously available ISAR data are both able to better sparsely represent the targets leading to better imaging results and the off-line DL based imaging method works even better.