基于YOLOv5的多信道信号样式识别系统的设计与实现

Design and implementation of multi-channel signal pattern recognition system based on YOLOv5

基于YOLOv5的目标检测算法,致力于解决战场通信侦察对抗系统中复杂通信信号识别问题。随着通信信号体制的快速发展,非合作接收条件下的通信信号检测、调制识别及信号辐射源个体识别变得更加困难。针对此类问题,提出了信号高阶图谱分析和深度神经网络的多信道信号样式识别方法,解决了当辐射源为N个时,对AM-SSB-SC、AM-DSB-SC、FM、GMSK 4种调制信号的快速识别问题。首先,利用信号高阶谱分析将4种调制信号变换为信号高阶谱图,再将标注后的图片输入到YOLOv5神经网络模型中进行深度学习;然后,通过优化YOLOv5网络结构,使其能并行处理N幅图片;最后将训练好的网络嵌入到软件无线电系统中进行测试,对辐射源产生的调制信号进行识别分类。完成了4种调制信号的快速识别系统设计,实现了非合作接受条件下多信道信号样式的检测和识别。

YOLOv5-based target detection algorithm aimed at addressing the complexity of communication signal reconnaissance in adversarial systems on the battlefield.With the rapid evolution of communication signal systems,detecting,modulating,and identifying signals from individual radiation sources under non-cooperative reception conditions have become increasingly challenging.In response to such challenges,a multi-channel signal pattern recognition method employing high-order spectral analysis and deep neural networks is proposed.This method resolves the rapid identification problem of AM-SSB-SC,AM-DSB-SC,FM,and GMSK modulation signals when there are 1-N radiation sources.Initially,the four modulation signals are transformed into high-order spectrograms using high-order spectral analysis.These annotated images are then input into the YOLOv5 neural network model for deep learning.Subsequently,the YOLOv5 network structure is optimized to enable parallel processing of 1-N images.Finally,the trained network is embedded into a software-defined radio system for test-ing,facilitating the identification and classification of modulation signals generated by radiation sources.The pro-posed system design accomplishes rapid recognition of the four modulation signals under non-cooperative recep-tion conditions,providing a solution for the detection and identification of multi-channel signal patterns.