基于圆周积分样本的Resnet1D雷达工作模式识别

Resnet1D radar mode recognition of operation based on circular integral samples

摘要:

针对传统的机器学习算法对脉冲描述字表征的雷达工作模式进行学习与识别时，存在对脉冲描述字捕获的准确率具有较高的依赖性以及识别准确率有限的问题。提出了一种基于圆周积分样本的Resnet1D模型的雷达工作模式识别方法。该方法采用积分双谱提取雷达工作模式的原始电磁信号高维表征样本特征，在保留电磁信号相位和幅度信息的同时，也实现了数据维度从二维降至一维的降低。降低计算复杂度的同时不会丢失雷达工作模式携带的电磁信号特征信息。通过对比积分双谱特征，计算机仿真表明圆周积分特征具有较好的识别准确率，在信噪比0dB条件下识别准确率超过95%。

Abstract:

Aiming at the traditional machine learning algorithms for learning and recognizing the radar operating modes characterized by pulse descriptors, there are problems of high dependence on the accuracy of pulse descriptor capturing and limited recognition accuracy. A radar operating mode recognition method based on the Resnet1D model with circular integral samples is proposed. The method uses integral bispectrum to extract the high-dimensional characterization sample features of the original electromagnetic signal of the radar operating modes, which also achieves the reduction of the data dimensions from two-dimensional to one-dimensional while retaining the phase and amplitude information of the electromagnetic signal. Reducing the computational complexity will not lose the electromagnetic signal feature information carried by the radar operating modes at the same time. By comparing the integral bispectral features, the computer simulation shows that the circumferential integral features have a good recognition accuracy, which is more than 95% under the condition of 0 dB.