一种优化的短波通信辐射源个体智能识别方法

An Optimized Individual Intelligent Identification Method for Shortwave Communication Emitters

针对短波频段信号环境复杂、通信辐射源个体识别中样本数量不足,造成常规个体识别方法困难、残差神经网络训练时过拟合的问题,提出了一种基于改进残差神经网络算法的短波通信辐射源个体智能识别方法。该方法将迁移学习的思想引入残差神经网络中,利用实际采集的短波通信辐射源信号,经过短时傅里叶变换(Short-Time Fourier Transform,STFT)得到的特征图谱,输入到结合迁移学习改进的残差神经网络算法中。然后,对信号采用不同方式提取信号特征、不同信噪比等条件下,进行短波通信辐射源个体识别,并与改进前残差神经网络进行对比实验。实验结果表明,对于短波通信辐射源信号,在样本数量不足、有噪声的条件下,经过STFT结合改进的残差神经网络算法相比改进前有更高的识别准确率和更好的抗噪性能。

In response to the complex signal environment in the shortwave frequency band and the insufficient number of samples in individual identification of communication radiation sources,which makes conventional individual identification methods difficult and the problem of overfitting during residual neural network training,a shortwave communication radiation source individual intelligent identification method based on an improved residual neural network algorithm is proposed.The idea of transfer learning is introduced into residual neural networks using this method.Using the actual collected shortwave communication radiation source signal,the feature map obtained throughshort time Fourier transform(STFT)is input into a residual neural network algorithm improved by combining transfer learning.Then,under the conditions of extracting signal characteristics in different ways and different signal-to-noise ratios,individual identification of shortwave communication radiation sources is carried out.Comparative experiments with the residual neural network before improvement show that,for the shortwave communication emitter signal,under the condition of insufficient sample number and noise,the STFT combined with the improved residual neural network algorithm has higher recognition accuracy and better anti-noise performance than before improvement.