人工智能教学研究:以生成模型下的射频指纹数据增强为例
DOI:
https://doi.org/10.65436/hssj.v1i6.40Keywords:
去噪扩散概率模型(DDPM);数据增强;射频指纹;生成对抗网络;变分自编码器Abstract
摘 要:AI生成内容(AIGC)技术的迅猛发展为突破数据瓶颈提供了新范式。其中,去噪扩散概率模型(DDPM)因其通过渐进式加噪与去噪的马尔可夫过程学习数据分布,具备训练稳定、生成质量高的优势,在复杂数据建模中展现出巨大潜力。本文将DDPM应用于射频指纹识别(RFF)领域,针对小样本场景下深度模型因数据稀缺导致的过拟合与性能下降问题,提出基于DDPM的数据增强框架,旨在合成高保真射频指纹信号以扩充训练集。与生成对抗网络(GAN)、变分自编码器(VAE)等主流生成模型相比,DDPM能更精准地捕捉和重建无线设备硬件损伤的细微特征。实验结果表明:在每类仅10个原始样本的条件下,DDPM生成的信号在关键评价指标上均显著优于WGAN、VAEGAN和DCGAN——均方根误差(RMSE)低至0.197(较次优降低64.4%),峰值信噪比(PSNR)达14.003(提升160%),频谱相干性(FSCS)达0.983(提升58.3%),Fréchet距离(FD)仅39.4(降低93.6%)。IQ图、频谱图和星座图的可视化对比进一步证实了DDPM生成信号在时域、频域的高保真度。本研究表明,基于DDPM的AIGC方法能有效生成高鉴别性射频指纹信号,为小样本RFF识别提供了高效的数据增强解决方案。
基金项目:安徽省教育厅自然科学研究项目(重大)(2024AH040217);省级质量工程-教学研究项目(2023jyxm1009;2023xjzlts117;2023sdxx145);基于人工智能下的家禽康养与疾病预警技术研究(S202413620055);基于自监督学习与对抗增强的小样本信号的智能感知研究(2025SK019);基于深度学习的生猪识别与康养研究(S202513620083);安徽省智慧农业技术与装备重点实验室开放基金(AEC2026016;AEC2025009)
References
[1] Dhariwal P, Nichol A. Diffusion models beat GANs on image synthesis[C]//Advances in Neural Information Processing Systems. Red Hook: Curran Associates, 2021, 34:8780-8794.
[2] Liu S, et al. Rolling bearing fault diagnosis using variational autoencoding generative adversarial networks with deep regret analysis[J]. Measurement, 2021, 168:108371.
[3] Li Y, Zou W, Jiang L. Fault diagnosis of rotating machinery based on combination of Wasserstein generative adversarial networks and long short term memory fully convolutional network[J]. Measurement, 2022, 191:110826.
[4] Zhou K, Diehl E, Tang J. Deep convolutional generative adversarial network with semi-supervised learning enabled physics elucidation for extended gear fault diagnosis under data limitations[J]. Mechanical Systems and Signal Processing, 2023, 185:109772.
[5] Fan C, et al. A novel lightweight DDPM-based data augmentation method for rotating machinery fault diagnosis with small sample[J]. Mechanical Systems and Signal Processing, 2025, 232:112741.
[6] Lin K, et al. RF Distillation Diffusion Model: An Efficient RFF Data Augmentation Method[C]//ICASSP 2025 IEEE International Conference on Acoustics, Speech and Signal Processing. Piscataway: IEEE, 2025.
[7] Chi G, et al. RF-diffusion: Radio signal generation via time-frequency diffusion[C]//Proceedings of the 30th Annual International Conference on Mobile Computing and Networking. New York: ACM, 2024.
Downloads
Published
How to Cite
Issue
Section
License
Copyright (c) 2026 李广博, 李慢, 简蕊, 刘雯秀, 关曼玉

This work is licensed under a Creative Commons Attribution 4.0 International License.