基于改进YOLOv4的通信信号检测模型

Communication Signal Detection Model Based on Improved YOLOv4

在军事、国防、工业应用等领域,信号检测被广泛应用,其中精确高效地检测出通信信号出现的时段及频段是非常重要的。因此,提出了一种基于改进YOLOv4的通信信号检测模型。该模型革新了传统的信号检测理念,将信号转化为时频图,使用YOLOv4目标检测算法对时频信号进行检测。针对YOLOv4检测信号时效率较低、精度不足的问题,对YOLOv4进行了针对性改进。首先,为了使模型更加轻量化,将YOLOv4主干网络修改为MobileNetv3;其次,对空间金字塔池化(Spatial Pyramid Pooling,SPP)模块的池化核参数进行了修改,使其更加匹配信号数据;最后,在YOLOv4的颈部网络中加入坐标注意力(Coordinate Attention,CA)模块,增强模型的广域特征提取能力。实验结果表明,相比于YOLOv4,改进后的YOLOv4参数量降低了50.53×10^(6),且检测准确率有所提高。

In fields such as military,national defense,and industrial applications,signal detection finds widespread use,where it is important to accurately and efficiently detect the time periods and frequency bands in which communication signals appear.Therefore,this paper proposes a communication signal detection model based on improved YOLOv4.The model breaks away from traditional signal detection concepts by transforming signals into time-frequency graphs and employing the YOLOv4 object detection algorithm for detecting time-frequency signals.In order to solve the problem of low efficiency and lack of accuracy when detecting signals in YOLOv4,it makes targeted improvements to YOLOv4.First,in order to make the model more lightweight,the YOLOv4 backbone network is modified to MobileNetv3.Then,the pooling kernel parameters of the SPP (Spatial Pyramid Pooling) module are modified to better match the signal data.Finally,the CA (Coordinate Attention) module is incorporated into the neck network of YOLOv4 to enhance the model’s wide-range feature extraction capability.Experimental results demonstrate that compared to YOLOv4,the improved YOLOv4model parameters volume is reduced by 50.53×10^(6) and its detection accuracy is improved.