SAR图像对抗攻击的进展与展望

合成孔径雷达（synthetic aperture radar，SAR）在民用和军用领域得到广泛应用。对SAR电子干扰一直是军事侦察领域对抗博弈的重点。不同于人工易识别的干扰技术，基于人工难察觉的微扰样本的对抗攻击，近年来在光学图像处理等计算机视觉领域得到了广泛研究。目前，对抗样本生成技术也已逐步应用于SAR图像对抗攻击，给SAR信息安全带来了新挑战。为此，对SAR图像对抗攻击技术方法的研究进展进行总结，主要包括图像对抗攻击的基本模型和方式，SAR图像对抗攻击原理与方法。针对典型深度学习目标检测算法，开展了对抗攻击下SAR图像目标检测性能分析，验证了对抗攻击的效果。最后，从算法脆弱机制与算法加固、融合机理的对抗攻击方法、对新体制成像雷达对抗攻击、对抗攻击识别与防御等4个方面对SAR图像对抗攻击领域的研究进行了展望。

SAR image adversarial attack:developments and perspectives

Synthetic aperture radar (SAR) has been widely used in both civil and military fields. Electronic jamming of SAR has always been the focus of confrontation game in military reconnaissance field. Different from the artificial easy-to-identify jamming technology, the adversarial attack based on artificial hard-to-notice perturbation samples has been widely researched in the field of computer vision such as optical image processing in recent years. At present, the adversarial sample generation technique has also been gradually applied to SAR image adversarial attacks, which brings new challenges to SAR information security. This paper briefly summarized the research progress of SAR image adversarial attack techniques and methods, mainly including the basic models and solutions of image adversarial attack, SAR image adversarial attack principles and methods. Then, the performance analysis of SAR image target detection under adversarial attack was carried out for typical deep learning target detection algorithms. The efficiency of the adversarial attack was verified. Finally,prospects for SAR image adversarial attack were given from four aspects: the algorithm vulnerability mechanism and algorithm reinforcement, the electromagnetic scattering mechanism driven adversarial attack technique, the adversarial attack for advanced imaging radar configurations, and the identification and defence of adversarial attacks.