基于时频图像与迁移学习的雷达信号调制类型识别算法
Radar Signal Modulation Type Recognition Algorithm Based on Time-Frequency Image and Transfer Learning

针对雷达信号调制类型识别领域中基于人工提取脉内特征的算法存在的特征提取步骤繁琐、泛化能力弱的缺点，本文提出了一种基于同步压缩短时傅里叶变换(the STFT-based synchrosqueezing transform, FSST)与迁移深度学习的调制类型识别算法。该算法借助深度学习在图像识别领域的优势，使用FSST将信号转换为时频图像并做预处理后作为数据集用于ResNet101网络的训练，同时还借助迁移学习的方法优化了网络的特征提取的能力、加速了模型的训练。考虑到复杂相位编码雷达信号的时频图像较为复杂，具有一定的代表性，本文以多相编码与多时编码信号为研究对象。实验结果表明，本文算法能够有效区分不同调制类型以及相同调制类型不同调制参数的相位编码信号。
Aiming at the shortcomings of cumbersome feature extraction steps and weak generalization ability of algorithms based on artificially extracting intra-pulse features in the field of radar signal modulation type recognition, this paper proposes a synchronous compression based short-time Fourier transform and transfer deep learning modulation type recognition algorithm. The algorithm uses the advantages of deep learning in the field of image recognition, uses FSST to convert the signal into a time-frequency image and preprocesses it as a data set for the training of the ResNet101 network. At the same time, it also optimizes the feature extraction of the network by means of transfer learning. Ability, accelerate the training of the model, and reduce the training requirements for the size of the data set. Considering that the time-frequency images of complex phase-coded radar signals are more complex and representative, this paper takes polyphase-coded and multi-time-coded signals as the research object. Experimental results show that the proposed algorithm can effectively distinguish phase-encoded signals with different modulation types and the same modulation type with different modulation parameters.