Improving SAR Imaging by Superpixel-Based Compressed Sensing and Backprojection Processing

Radars mounted on unmanned aerial vehicles (UAVs) enable high flexibility in data acquisition since arbitrary and especially nonlinear flight trajectories can be achieved. Compared to linear flight trajectories, nonlinear trajectories achieve better image quality and allow the scene and targets to be observed from different angles, thereby capturing angle-dependent features. However, the inversion of synthetic aperture radar (SAR) data for nonlinear apertures and sample acquisition is computationally taxing and artifacts may arise in the image due to the aperture’s structure. To suppress aperture artifacts and clutter in the SAR image, a backprojection imaging algorithm combined with compressed sensing (CS) methods is proposed. In this work, an efficient way to calculate the backprojection in a CS framework is described that does not require interpolation steps. Since SAR images are neither strictly sparse nor noise-dominated, an adapted CS algorithm is proposed that accounts for clutter and extended targets that are spread across multiple image pixels. In addition, an image segmentation approach is presented that enables faster processing while being robust to extended targets located at segment edges. With the proposed approach, SAR images of arbitrary flight paths can be processed efficiently with fewer acquisition points while achieving better clutter suppression at the same image quality compared to equally dense measurement sets.