• ISSN 0258-2724
  • CN 51-1277/U
  • EI Compendex
  • Scopus 收录
  • 全国中文核心期刊
  • 中国科技论文统计源期刊
  • 中国科学引文数据库来源期刊

基于时频Grad-CAM的调制识别网络可解释分析

梁先明 倪帆 陈文洁 张家树

梁先明, 倪帆, 陈文洁, 张家树. 基于时频Grad-CAM的调制识别网络可解释分析[J]. 西南交通大学学报, 2024, 59(5): 1215-1224. doi: 10.3969/j.issn.0258-2724.20210791
引用本文: 梁先明, 倪帆, 陈文洁, 张家树. 基于时频Grad-CAM的调制识别网络可解释分析[J]. 西南交通大学学报, 2024, 59(5): 1215-1224. doi: 10.3969/j.issn.0258-2724.20210791
LIANG Xianming, NI Fan, CHEN Wenjie, ZHANG Jiashu. Interpretability of Modulation Recognition Network Based on Time-Frequency Gradient-Weighted Class Activation Mapping[J]. Journal of Southwest Jiaotong University, 2024, 59(5): 1215-1224. doi: 10.3969/j.issn.0258-2724.20210791
Citation: LIANG Xianming, NI Fan, CHEN Wenjie, ZHANG Jiashu. Interpretability of Modulation Recognition Network Based on Time-Frequency Gradient-Weighted Class Activation Mapping[J]. Journal of Southwest Jiaotong University, 2024, 59(5): 1215-1224. doi: 10.3969/j.issn.0258-2724.20210791

基于时频Grad-CAM的调制识别网络可解释分析

doi: 10.3969/j.issn.0258-2724.20210791
基金项目: 国家自然科学基金项目(62071396);四川省自然科学基金项目(2022NSFSC0531)
详细信息
    作者简介:

    梁先明(1976—),男,高级工程师,研究方向为通信信号处理与分析、智能信号分析,E-mail:liangxm8873@126.com

    通讯作者:

    张家树(1965—),男,教授,博士,研究方向为神经网络与机器智能、信号与信息智能处理、信息安全技术,E-mail:jszhang@swjtu.edu.cn

  • 中图分类号: TP301.6

Interpretability of Modulation Recognition Network Based on Time-Frequency Gradient-Weighted Class Activation Mapping

  • 摘要:

    针对时频深度学习调制识别方法存在可解释性差的问题,提出一种基于时频梯度加权类激活映射(Grad-CAM)的调制识别网络可解释框架. 该框架通过时频Grad-CAM可视化深度模型中隐含层的关键特征,从视觉上解释网络隐含层提取的时频深度特征对于正确与错误识别中的作用,揭示低信噪比环境下网络性能下降的内在机理,并通过量化和排序网络中每层不同卷积核的贡献值来判断网络的冗余程度. 仿真实验结果验证了基于时频Grad-CAM的调制识别网络可解释性框架的有效性;可解释分析结果表明,在低信噪比环境下,网络特征提取区域有大量噪声存在,且本文所测试的调制识别网络冗余程度较为严重.

     

  • 图 1  可解释调制识别网络设计

    Figure 1.  Design of interpretable modulation recognition network

    图 2  12种调制信号的二维时频特征图

    Figure 2.  2D time-frequency characteristic diagrams of 12 modulation signals

    图 3  SE-ResNet50结构

    Figure 3.  SE-ResNet50 structure

    图 4  二维时频特征提取的调制识别整体准确率

    Figure 4.  Overall accuracy of modulation recognition based on 2D time-frequency feature extraction

    图 5  0 dB下SE-ResNet识别结果的混淆矩阵

    Figure 5.  Confusion matrix of recognition results with SE-ResNet at 0 dB

    图 6  −10 dB下SE-ResNet识别结果的混淆矩阵

    Figure 6.  Confusion matrix of recognition results with SE-Res Net at −10 dB

    图 7  类激活图在不同时频特征上的映射结果

    Figure 7.  Mapping results of class activation graphs on different time-frequency features

    图 8  4条BPSK信号时频图在SE-ResNet下的梯度类激活映射

    Figure 8.  Gradient-weighted class activation mapping of time-frequency graphs from four BPSK modulated signals with SE-ResNet

    图 9  每类信号时频特征对于不同预测类别的梯度类激活映射

    Figure 9.  Gradient class activation mapping of time-frequency characteristics from each class of signals for different prediction classes

    图 10  8PSK调制信号的时频特征在不同信噪比下的SE-ResNet50隐藏层最后一层的类激活图

    Figure 10.  Class activation graphs of time-frequency characteristics of 8PSK modulated signals on the last hidden layers with SE-ResNet50 under different SNR ratios

    图 11  SE-ResNet、ResNet50和VGGNet对于同种调制信号时频特征的类激活图

    Figure 11.  Class activation graphs of SE-ResNet, ResNet50 and VGGNet for the same modulated signal

    图 12  4种深度网络最后一层的卷积核贡献值排序

    Figure 12.  Ranking of kernel contribution values on the last convolution layer of four deep networks

    图 13  VGG16最后一层卷积层每个卷积核对应的贡献值

    Figure 13.  Contribution values of each kernel on the last convolution layer of VGG16

  • [1] WEI W, MENDEL J M. Maximum-likelihood classification for digital amplitude-phase modulations[J]. IEEE Transactions on Communications, 2000, 48(2): 189-193. doi: 10.1109/26.823550
    [2] POLYDOROS A, KIM K. On the detection and classification of quadrature digital modulations in broad-band noise[J]. IEEE Transactions on Communications, 1990, 38(8): 1199-1211. doi: 10.1109/26.58753
    [3] HUAN C Y, POLYDOROS A. Likelihood methods for MPSK modulation classification[J]. IEEE Transactions on Communications, 1995, 43(2/3/4): 1493-1504. doi: 10.1109/26.380199
    [4] NANDI A K, AZZOUZ E E. Algorithms for automatic modulation recognition of communication signals[J]. IEEE Transactions on Communications, 1998, 46(4): 431-436. doi: 10.1109/26.664294
    [5] ZAERIN M, SEYFE B. Multiuser modulation classification based on cumulants in additive white Gaussian noise channel[J]. IET Signal Processing, 2012, 6(9): 815-823. doi: 10.1049/iet-spr.2011.0357
    [6] 李红光,郭英,眭萍,等. 基于时频特征的卷积神经网络跳频调制识别[J]. 浙江大学学报(工学版),2020,54(10): 1945-1954.

    LI Hongguang, GUO Ying, SUI Ping, et al. Frequency hopping modulation recognition of convolutional neural network based on time-frequency characteristics[J]. Journal of Zhejiang University (Engineering Science), 2020, 54(10): 1945-1954.
    [7] ZHANG H J, LE RUYET D, TERRÉ M. Spectral correlation of multicarrier modulated signals and its application for signal detection[J]. EURASIP Journal on Advances in Signal Processing, 2010, 2010: 794246.1-794246.14.
    [8] SCHREYOGG C, KITTEL K, KRESSEL U, et al. Robust classification of modulation types using spectral features applied to HMM[C]//1997 IEEE Military Communications Conference (MILCOM 97). Monterey: IEEE, 1997: 1377-1381.
    [9] MOBASSERI B G. Constellation shape as a robust signature for digital modulation recognition[C]//IEEE Military Communications Conference (MILCOM 99). Atlantic City: IEEE, 1999: 442-446.
    [10] 李楠. 双谱特征和深度学习在信号调制识别中的应用[J]. 弹箭与制导学报,2019,39(5): 81-84,88.

    LI Nan. Application of bi-spectral feature and depth learning in signal modulation recognition[J]. Journal of Projectiles, Rockets, Missiles and Guidance, 2019, 39(5): 81-84,88.
    [11] O’SHEA T, HOYDIS J. An introduction to deep learning for the physical layer[J]. IEEE Transactions on Cognitive Communications and Networking, 2017, 3(4): 563-575. doi: 10.1109/TCCN.2017.2758370
    [12] HU S S, PEI Y Y, LIANG P P, et al. Robust modulation classification under uncertain noise condition using recurrent neural network[C]//2018 IEEE Global Communications Conference (GLOBECOM). Abu Dhabi: IEEE, 2018: 8647582.1-8647582.7.
    [13] 吴亚聪. 基于深度注意胶囊的智能信号识别方法研究[D]. 西安: 西安电子科技大学,2019.
    [14] ZHOU B L, KHOSLA A, LAPEDRIZA A, et al. Learning deep features for discriminative localization[C]//2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Las Vegas: IEEE, 2016: 2921-2929.
    [15] SELVARAJU R R, COGSWELL M, DAS A, et al. Grad-CAM: visual explanations from deep networks via gradient-based localization[C]//2017 IEEE International Conference on Computer Vision (ICCV). Venice: IEEE, 2017: 618-626.
    [16] CHEN J Y, MIAO S H, ZHENG H B, et al. Feature explainable deep classification for signal modulation recognition[C]//The 46th Annual Conference of the IEEE Industrial Electronics Society (IECON 2020). Singapore: IEEE, 2020: 3543-3548.
  • 加载中
图(13)
计量
  • 文章访问数:  298
  • HTML全文浏览量:  97
  • PDF下载量:  42
  • 被引次数: 0
出版历程
  • 收稿日期:  2021-09-10
  • 修回日期:  2022-05-16
  • 网络出版日期:  2024-06-14
  • 刊出日期:  2022-06-09

目录

    /

    返回文章
    返回