基于多变换域特征的雷达辐射源信号融合识别

Fusion recognition of radar emitter signals based on multiple transform domain features

摘要:

针对现有复杂体制雷达辐射源信号识别方法信息利用率低、抗噪性能差等问题,提出一种融合雷达辐射源信号多变换域特征的集成深度神经网络识别方法。首先基于双谱估计、模糊函数、希尔伯特黄变换（HHT）3种变换域方法对辐射源信号展开处理,分别提取信号矩形积分双谱特征、模糊函数正交切片特征和希尔伯特边际谱特征,并转换为表达能力和可解释性更优的二维特征图像；其次,构建基于ResNet18+多层感知机（MLP）的融合识别模型框架,以多个ResNet18网络作为基学习器,分别对3类变换域特征数据集进行初级识别,获得以概率表征的特征向量；最后,通过MLP对特征向量进行融合学习,输出最终的信号类别信息。实验结果表明,该方法在信噪比为 0 dB时,对6类雷达辐射源信号的整体平均识别率均保持在99.23%以上,即使是在-4 dB低信噪比环境中,识别率也稳定在96.54%以上,验证了所提方法的有效性和较好性能。

Abstract:

In response to the problems of low information utilization and poor anti-noise performance in existing recognition methods for radar emitter signals of complex systems, we proposed an ensemble deep neural network recognition method integrating multiple transform domain features of radar emitter signals. Firstly, based on the three transform domain methods of bispectrum estimation, ambiguity function (AF), and Hilbert-Huang transform (HHT), we processed the emitter signals, extracted, and transformed the signal’s rectangular integral bispectrum feature, AF orthogonal slice feature, and Hilbert marginal spectrum feature into two-dimensional feature images with stronger expressiveness and interpretability. Then, we constructed a fusion recognition model framework based on ResNet18 + multilayer perceptron (MLP), took multiple ResNet18 as base learners to perform primary recognition on the datasets of three transform domain features, and obtained feature vectors represented by probabilities. Finally, we conducted fusion learning on the feature vectors via the MLP and output the final signal category information. The experimental results show that the proposed method maintains an overall average recognition rate of above 99.23% for six classes of radar emitter signals at a signal-to-noise ratio (SNR) of 0 dB. Even in the low SNR environment of -4 dB, the recognition rate remains stable at above 96.54%. The results verify the effectiveness and better performance of the proposed method.