基于迁移学习的雷达信号类型自动识别方法

Automatic Recognition of Radar Signal Type Based on Transfer Learning

摘要: 针对雷达辐射源信号类型的自动识别方法进行研究，引入迁移学习（transfer learning，TL）思想，将 GoogleNet 预训练模型迁移到雷达信号数据集中实现信号特征提取，在全连接层中采用极限学习机（extreme learning machine，ELM）作为分类器，完成对雷达信号的自动识别。仿真结果表明，针对 9 类雷达信号，在信噪比为 0 dB 的情况下，基于 GoogleNet-ELM 的识别算法具有很好的识别性能，得到 95.7%的正确识别率，验证了所提算法在电子侦察领域应用的有效性。

Abstract: The automatic recognition method of radar emitter signal type is researched. In the complex environment of small sample training data set and limited training time. The transfer learning （TL）concept is introduced and the pre-training GoogleNet models transferred to the radar signal data set to realize the automatic feature extraction. Then, in the full connection layer, Extreme Learning Machine （ELM）is used as classifier to complete the automatic recognition of radar signals. The simulation results show that, for nine kinds of radar signals, the GoogleNet-ELM-based recognition algorithm has good recognition performance when the signal-to-noise ratio is 0 dB, and the correct recognition rate is 95.7%, the effectiveness of the proposed algorithm in the field of electronic reconnaissance is verified.