[1]韩刚涛,马瑞鹏,吴 迪.基于时频图切割的宽带信号智能检测与识别[J].郑州大学学报(工学版),2023,44(03):44-51.[doi:10.13705/j.issn.1671-6833.2023.03.008]
　HAN Gangtao,MA uipeng,WU Di.Intelligent Detection and Identification of Broadband Signals Based on Time-frequency Map Cutting[J].Journal of Zhengzhou University (Engineering Science),2023,44(03):44-51.[doi:10.13705/j.issn.1671-6833.2023.03.008]

点击复制

基于时频图切割的宽带信号智能检测与识别()

分享到：

《郑州大学学报(工学版)》[ISSN:1671-6833/CN:41-1339/T]

卷:

44卷

期数:

2023年03期

页码:

44-51

栏目:

出版日期:

2023-04-30

文章信息/Info

Title:

Intelligent Detection and Identification of Broadband Signals Based on Time-frequency Map Cutting

作者:

韩刚涛; 马瑞鹏; 吴 迪

郑州大学 电气与信息工程学院,河南 郑州 450001, 郑州大学 网络空间安全学院,河南 郑州 450002, 中国人民解放军战略支援部队信息工程大学 数据与目标工程学院,河南 郑州 450001

Author(s):

HAN Gangtao; MA Ｒuipeng; WU Di

School of Electrical and Information Engineering, Zhengzhou University, 450001, Zhengzhou, Henan, School of Network Space Security, Zhengzhou University, 450002, Zhengzhou, Henan, School of Strategic Support for the Chinese People’s Liberation Army, School of Information Engineering Engineering Engineering, Henan Zhengzhou 450001

关键词:

信号检测;  深度学习;  大带宽;  轻量型;  Focal-EIOU

Keywords:

signal detection;  deep learning;  large bandwidth;  lightweight;  Focal-EIOU

DOI:

10.13705/j.issn.1671-6833.2023.03.008

文献标志码:

A

摘要:

针对大带宽场景中非合作通信信号的高效检测问题，结合机器学习技术提出一种基于时频图切割的宽带 信号智能检测与识别方法。该方法采用 Mobilenet 网络替代 YOLOv4 当中的 CSPdarknet53 网络进行特征提取，构建 了一种轻量型的 YOLOv4 模型。同时，模型引入 Focal-EIOU 代价函数和一种改进的加权盒融合算法( WBF) ，有效 提高了训练效率以及检测与识别精度。实验结果表明: 本文方法可以快速地检测出大带宽下通信采集数据中的连 续信号和突发信号，以及信号的出现时刻、频率范围、调制方式等相关参数，其性能优于传统的能量检测方法。与 其他同类方法相比，本文方法的平均检测精度( mAP) 均大于 81%，其中，采用 YOLOv4-MobilenetV1 模型的检测速度 达到 77. 60 帧/s，较好地兼顾了检测精度与实时性要求，更利于工程部署。

Abstract:

To address the problem of efficient detection of noncooperative communication signals in large bandwidth scenarios，an intelligent detection and identification method for broadband signals based on time-frequency map cutting was proposed in combination with machine learning techniques． In this study，the method adopted Mobilenet network instead of CSPdarknet53 network in YOLOv4 for feature extraction，and constructed a lightweight YOLOv4 model． At the same time，the model introduced the Focal-EIOU cost function and an improved weighted box fusion algorithm ( WBF) ，which could effectively improve the training efficiency and the detection and recognition accuracy． The experimental results showed that the method in this study could quickly detect the continuous and burst signals in the communication acquisition data with large bandwidth，as well as the moment of appearance，frequency range，modulation mode and other related parameters，and its performance is better than the traditional energy detection methods． Compared with other similar methods，the average detection accuracy ( mAP) of this study was greater than 81%，and the detection speed of YOLOv4-MobilenetV1 model reached 77. 60 frames/s，which was better for both detection accuracy and real-time requirements and was more conducive to engineering deployment.

常用功能