CARNet: An effective method for SAR image interference suppression
Shunjun Wei, Hao Zhang, Xiangfeng Zeng, Zichen Zhou, Junren Shi, Xiaoling Zhang
Abstract
Synthetic aperture radar (SAR) routinely confronts the interference of radiofrequency devices in normal missions, causing ineffective imaging and seriously affecting Earth observation capability. In general, it is a great challenge to ensure interference suppression performance and image quality. To address this problem, we present an efficient method for SAR image interference suppression based on the Combined-Attention Restoration Network (CARNet). SAR image model is established, including target image, interference image, and background noise image. Specifically, we first propose a new feature extraction scheme to capture image model information over space and channels for enriching the context. Then encoder–decoder is employed to suppress interference and produce different-dimensional feature maps for target information exchange. Moreover, the image attention mechanism is introduced to calibrate the target features under the guidance of original images for essential information propagation. Besides, several attentional connections exist to prevent further loss of target details. The effectiveness of the proposed method is validated on simulated data and measured Sentinel-1 images. Compared with conventional and state-of-the-art algorithms, the results indicate that CARNet achieves better interference suppression performance and can generate high-resolution images closer to the ground truth.