Removal of Micro-Doppler Effect in ISAR Imaging Based on Data-Driven Deep Network

The micro-Doppler (m-D) effect induced by micro-motion parts of the target degrades the quality of inverse synthetic aperture radar (ISAR) imaging results. To solve this issue, this article proposes a data-driven deep network framework to remove the m-D effect in ISAR imaging, so as to obtain better imaging results. First, a new training data generation method is proposed, which is suitable for both point targets and block targets. Through this method, a plenty of m-D ISAR images and label images without m-D effect can be obtained, which can be used as training data. Then, we improve the UNet as our training network, namely, IUNet. After that, training data are used to train the IUNet. After training, the network learns the mapping relationship between training data, and the network model is saved. Finally, the well-trained network can remove the m-D effects in test ISAR images. The experiments of simulated data and measured data show that the proposed method is superior in terms of eliminating m-D effect than other methods and has good performance.