| 摘要: |
| 为了解决电磁频谱异常检测精度不高的问题，在深度卷积神经对抗网络（Deep Convolution Generative Adversarial Network，DCGAN）的基础上加入了编码器（Encoder）用来重构频谱数据。编码器首先将真实频谱数据编码为低维特征表示，生成器通过学习编码后的低维特征生成重构频谱数据，判别器负责将重构频谱数据与真实频谱数据进行区分，并通过对抗性训练逐渐提高模型重构频谱数据的能力，最后计算重构频谱数据与真实频谱数据的均方误差，判别异常。实验结果表明，该模型能够在多个频段下实现有效的电磁频谱异常检测，在TV频段下，干信比为-5 dB时，相比于现有电磁频谱异常检测方法，所提方法的平均检测性能提升了18%以上。 |
| 关键词: 电磁频谱异常检测 深度卷积对抗网络(DCGAN) 频谱重构 |
| DOI：10.20079/j.issn.1001-893x.231107003 |
|
| 基金项目:国家科学自然基金资助项目（62371463） |
|
| Spectrum Anomaly Detection Based on DCGAN |
| JI Haipeng,ZHANG Jiang,QIAO Xiaoqiang,ZHANG Tao |
| (1.School of Electronic and Information Engineering,Nanjing University of Information Science and Technology,Nanjing 210044,China;2.The 63rd Research Institute of National University of Defense Technology,Nanjing 210007,China) |
| Abstract: |
| In order to solve the problem of low accuracy in detecting electromagnetic spectrum anomalies,an encoder is added to the deep convolutional generative adversarial network(DCGAN) to reconstruct spectrum data.Firstly,the encoder encodes the real spectrum data into low dimensional feature representations,and the generator generates reconstructed spectrum data by learning the encoded low dimensional features.The discriminator is responsible for distinguishing the reconstructed spectrum data from the real spectrum data,and gradually improving the model’s ability to reconstruct spectrum data through adversarial training.Finally,the mean square error between the reconstructed spectrum data and the real spectrum data is calculated to identify anomalies.The experimental results show that the model can achieve effective electromagnetic spectrum anomaly detection in multiple frequency bands.In the TV frequency band,when the interference-to-signal ratio is -5 dB,the detection performance of this method is improved by more than 18% compared with that of existing electromagnetic spectrum anomaly detection methods. |
| Key words: spectrum anomaly detection deep convolutional generative adversarial network(DCGAN) spectrum reconstruction |
