| 摘要: |
| 针对跳频信号分选存在人工提取参数特征具有复杂性的问题，提出了一种基于深度学习的识别方法。首先对跳频信号进行短时傅里叶变换，得到二维的时频矩阵；接着提取信号的轮廓特征，构造三维矩阵作等高线图，并对等高线图进行预处理；最后把预处理后的等高线图输入到卷积神经网络中进行训练、测试，进而实现分类识别。仿真结果表明，在不需要复杂的人工提取参数特征的基础上，在分选率为100〖WT《Times New Roman》〗%〖WTBZ〗时，所提方法经裁剪处理下的信噪比为-15 dB，比支持向量机和传统K-Means聚类算法都低10 dB。实测数据的算法验证表明，所提方法能够将大疆精灵4Pro、hm无人机、司马航模X8HW以及大疆悟2这四类无人机正确分类。 |
| 关键词: 跳频信号 分类识别 深度学习 图像预处理 卷积神经网络 |
| DOI： |
|
| 基金项目:国家自然科学基金资助项目（6146105）；广西科技重大专项资助项目（AA17202022）；桂林电子科技大学研究生优秀学位论文培育资助项目（17YJPYSS10） |
|
| Recognition of Frequency Hopping Signal Based on Deep Learning |
| LYU Guopei,XIE Yuelei |
| (Ministry of Education Key Laboratory of Cognitive Radio and Information Processing,Guilin University of Electronic Technology,Guilin 541004,China) |
| Abstract: |
| For the problem of the complexity of manually extracting features in the frequency hopping(FH) signal sorting,a recognition method based on deep learning is proposed.First,short-time Fourier transform is performed for the FH signal to obtain a two-dimensional time-frequency matrix.Then the contours features of the signal are extracted,the three-dimensional matrix is constructed,so that the contour map is drawn and then preprocessed.Finally,the processed contour map is input to a convolutional neural network(CNN) for training and testing and then classification recognition is realized.Simulation results show that the proposed method requires no complicated manual extraction of parameter features.When the sorting rate is 100〖WT《Times New Roman》〗%〖WTBZ〗,the signal-to-noise ratio(SNR) of the proposed method after clipping processing is -15 dB,which is 10 dB lower than that of the support vector machine(SVM) and traditional K-Means clustering algorithm.The algorithm verification of the measured data shows that the method can correctly classify the four types of drones:Phantom 4 Pro,hm UAV,SYMA_X8HW,and Inspire 2. |
| Key words: frequency hopping signal classification and recognition deep learning image preprocessing convolutional neural network |
《电讯技术》编辑部 