Dual cooperative defense method for communication signal modulation intelligent identification confrontation

Through the proposed dual collaborative defense method, adversarial samples are generated using adversarial autoencoder and dynamic search strategies, and detection and defense are carried out through front and rear defense lines, the problems of poor detection effect and insufficient generalization performance of adversarial defense methods in the existing technology are solved, achieving stronger adversarial robustness and identification accuracy.

CN119691559AActive Publication Date: 2025-03-25XIDIAN UNIV
8 Cites 0 Cited by

Patent Information

Application Number
CN202411833348.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-12-13
Publication Date
2025-03-25
Estimated Expiration
2044-12-13

AI Technical Summary

Technical Problem

When facing new and unknown adversarial samples, the existing adversarial defense methods have poor detection effects, insufficient generalization performance, and single defense methods, and insufficient utilization of hidden features of the model and signal samples.

Method used

A dual collaborative defense method for intelligent identification of adversarials for communication signal modulation is proposed. The hidden features are extracted by the encoder, the adversarial autoencoder is trained to generate adversarial samples, and the dynamic search strategy is used to determine the adversarial perturbation radius, and the adversarial samples with customized tags are generated to train the adversarial sample recognition network. The adversarial sample detection network is used as the front line of defense, and the adversarial sample identification network is used as the back line of defense for secondary defense.

Benefits of technology

The dual coordinated defense of modulation identification tasks is realized, which significantly enhances the adversarial robustness and improves the accuracy of adversarial sample detection and recognition.

✦ Generated by Eureka AI based on patent content.
Patent Text Reader

Abstract

The embodiment of the invention relates to the technical field of confrontation and defense, in particular to a communication signal modulation intelligent recognition confrontation-oriented dual cooperative defense method, which comprises the following steps: respectively extracting hidden features from an original sample and a constructed confrontation sample by using an encoder, training a confrontation self-encoder based on the extracted hidden features, and obtaining a confrontation self-encoder; obtaining an adversarial sample detection network; applying a dynamic search strategy to determine an adversarial disturbance radius, generating an adversarial sample with a customized label, and training based on the adversarial sample with the customized label to obtain an adversarial sample recognition network; and taking the adversarial sample detection network as a front network, performing adversarial detection on an input sample, taking the adversarial sample identification network as a rear network, performing secondary defense on the input sample of the escape adversarial sample detection network, and completing a modulation identification task. According to the method, dual cooperative defense of the modulation identification task is realized, and the adversarial robustness is effectively enhanced.
Need to check novelty before this filing date? Find Prior Art