基于ResNet-Peca的信号分选方法

Signal Sorting Method Based on ResNet-Peca

针对传统的信号分选方法在信噪比过低时,调制信号识别效果不佳的问题,提出一种融合通道注意力机制(ECA)和位置注意力机制(PAM)的残差网络(ResNet-Peca),该网络可同时获得通道和位置维度特征权重,提升网络的特征学习能力。研究结果表明,融合注意力机制的残差网络(ResNet-Peca)整体效果,相较ResNet-ECA方法识别准确率提升了约1.5%,较ResNet-PAM方法识别准确率提升了约1.6%,较CNN方法识别准确率提升了约3.8%。

In order to solve the problem that the traditional signal sorting method has poor modulation signal recognition effect when the signal-to-noise ratio is too low,a residual network(ResNet-Peca)integrating efficient channel attention mechanism(ECA)and position attention mechanism(PAM)is proposed.This network can simultaneously obtain the feature weights of channel and position dimensions to improve the feature learning ability of the network.The study results show that the cognition accuracy of the residual network with attention mechanism(ResNet-Peca)is about 1.5%higher than that of the ResNet-ECA method,about 1.6%higher than that of the ResNet-PAM method,and about 3.8%higher than that of the CNN method.