A Survey of Recognition Model Attack Algorithms in Communication Countermeasure

Accurate reconnaissance and identification of communication radios are crucial for success in communication electronic warfare. Recently, counteracting adversarial attacks has emerged as a significant area of research. By making minor modifications to a sample, attackers can cause deep learning models to make incorrect predictions, posing significant security risks. The field of communication countermeasures is similarly threatened by adversarial samples. Attack strategies in this domain differ substantially from those in image recognition and other fields. Therefore, researching attack methods for communication countermeasures is of paramount importance. This paper introduces the concept of adversarial samples. It progresses from white-box to black-box attacks, from digital to physical attacks, and from specific vectors to general vectors, moving from simpler to more complex scenarios that closely mimic real-world conditions. Recent typical attack methods against adversarial samples are systematically reviewed. The paper analyzes and summarizes the current technical challenges in adversarial attack. Finally, it explores future development directions for sample attack strategies.