摘要
基于深度学习对信号进行特征提取和分类是当前的一个研究重点,而在工程实践中,对于非合作信号,获得样本过程极为复杂导致有效样本量少。为此提出了一种快速的原始信号样本增强方法来提升辐射源的个体识别率。首先,对原始信号样本分别进行加噪、分割重组和频移变换处理,得到增强样本集;然后,利用栈式自编码网络和卷积神经网络对增强样本集分别进行测试。测试结果表明,在小样本情况下,通过样本增强处理后,辐射源个体识别准确率有了显著的提升。
The feature extraction and classification of signals based on deep learning is a current research focus,but in engineering practice,for non-cooperative signals,the sample acquisition process is extremely complicated,which results that the effective sample is rare.Therefore,a fast original signal sample enhancement method is proposed to improve the idividual recognition rate.Firstly,the original signal samples are processed by noise addition,segmentation and recombination,and frequency-shift transformation to obtain the enhanced sample set.Then,the stack self-coding network and convolutional neural network are used to test the enhanced sample set respectively.The test results show that in the case of small samples,after sample enhancement processing,the accuracy of individual radiation source identification has been significantly improved.
作者
李刚
LI Gang(Southwest China Institute of Electronic Technology,Chengdu 610036,China)
出处
《电子质量》
2024年第6期59-64,共6页
Electronics Quality
关键词
样本增强
栈式自编码网络
个体识别
深度学习
sample enhancement
stack autoencode network
individual identification
deep learning
