基于特征降维和深度学习的电磁信号识别方法

Electromagnetic Signal Recognition Method Based on Feature Reduction and Deep Learning

针对基于原始电磁数据通过深度学习识别方法存在计算复杂度高、特征难以物理表征的问题,提出了一种基于特征降维算法和深度学习结合的电磁信号识别方法。该方法在提取电磁信号静态物理特征的基础上,利用ReliefF算法进行特征降维并生成最优特征子集,并将特征子集数据矩阵转换为二维图像,结合不同载波频率和码元信息的电磁信号构建多类型训练样本集。进一步采用改进的残差神经网络(ResNet50-A)进行特征提取,利用识别模型完成电磁信号识别。实验结果表明:论文样本集选取18个静态物理特征所构建的特征子集的电磁信号识别率最高可以达到98.61%,明显优于其他特征子集的识别效果,验证了方法的可行性。

Aiming at the problems of high computational complexity and difficult physical representation of features in deep learning recognition method based on original electromagnetic data,an electromagnetic signal recognition method based on feature dimension reduction algorithm and deep learning is proposed.On the basis of extracting the static physical features of electromagnetic signals,this method uses the ReliefF algorithm to reduce the dimension of features and generate the optimal feature subset,and converts the feature subset data matrix into a two-dimensional image.Combined with the electromagnetic signals with different carrier frequencies and symbol information,a multi type training sample set is constructed.The improved residual neural network(ResNet50-A)is further used for feature extraction,and the recognition model is used to complete the recognition of electromagnetic signals.The experimental results show that the electromagnetic signal recognition rate of the feature subset constructed by selecting 18 static physical features from the sample set in this paper can reach 98.61%,which is significantly better than the recognition effect of other feature subsets,which verifies the feasibility of the method.