SLPA: Single-Line Pixel Attack on Specific Emitter Identification Using Time-Frequency Spectrogram

W. Q. Li

doi:10.1109/TVT.2024.3405547

SLPA: Single-Line Pixel Attack on Specific Emitter Identification Using Time-Frequency Spectrogram

2024-10-01

0

PRE

AI

DOI：10.1109/TVT.2024.3405547

Original

Original request for help

Save

Abstract

En 中文

Specific emitter identification (SEI) can identify and characterize individual radio frequency emitters, such as radar systems or communication devices. The application of deep learning (DL) techniques in the domain of SEI has gained significant attention due to its potential for enhancing the accuracy and efficiency of SEI. Recent studies show that DL-based SEI is susceptible to adversarial attacks. For SEI based on the time-frequency spectrogram, we propose a single-line pixel attack (SLPA) method. The attacker adds the same perturbation to the time-frequency spectrogram samples line by line and queries the model recognition results, selecting the RF signal corresponding to the line with the highest number of successful attacks as the optimal perturbation signal. This is a practical black-box attack that does not require synchronization and knowledge of model parameters. Under the same perturbation power, our proposed method can achieve an attack success rate of 35.05%, which is much higher than the 20.10% of the white-box attack fast gradient sign method (FGSM) that requires gradient information of the target model.

Keywords:

Time-frequency analysis

Perturbation methods

Spectrogram

Glass box

Closed box

Training

Synchronization

Black-box attack

specific emitter identification (SEI)

time-frequency spectrogram

universal adversarial perturbation

SLPA: Single-Line Pixel Attack on Specific Emitter Identification Using Time-Frequency Spectrogram

Abstract

Journal

Organization