Classification and Identification of Frequency-Hopping Signals Based on Jacobi Salient Map for Adversarial Sample...

Area of Science:

• Electrical Engineering
• Computer Science
• Cybersecurity

Background:

• Modern electronic countermeasures increasingly focus on frequency-hopping (FH) communication.
• Deep neural networks (DNNs) are used to classify intercepted FH signals, enabling targeted interference.
• This poses a significant challenge to maintaining robust FH communication performance.

Purpose of the Study:

• To develop an advanced adversarial sample generation method for FH communication systems.
• To counter the threat of DNN-based signal classification and targeted interference.
• To enhance the efficiency and stealthiness of adversarial attacks in electronic warfare.

Main Methods:

• A novel batch feature point targetless adversarial sample generation method based on the Jacobi saliency map (BPNT-JSMA) is proposed.
• The method builds upon traditional JSMA by generating feature saliency maps.
• It perturbs the top 8% of salient feature points in batches and increases perturbation limits to avoid extreme single-point values.

Main Results:

• Experimental results demonstrate the effectiveness of BPNT-JSMA in a white-box environment.
• The proposed method maintains a high attack success rate compared to traditional JSMA.
• BPNT-JSMA significantly enhances attack efficiency and improves the stealthiness of adversarial samples.

Conclusions:

• BPNT-JSMA offers a superior approach for generating adversarial samples against FH communication systems.
• The method effectively addresses the challenges posed by DNN-based signal identification.
• This research contributes to advancing electronic countermeasures and securing FH communications.