基于HOS的雷达目标分类新算法

A New Radar Target Classification Method Based on High Order Statistics (HOS)

提出一种基于高阶谱 ( HOS)的雷达目标分类方法 ,将双谱概念从频域推广到时域和距离域 ,并给出了将双谱和双相干函数的“均值”及“重心”作为特征量的特征提取新算法 ,在目标散射信号中含有加性噪声和指数噪声的情况下 ,进行了模拟 ,同时对 BSB( Brain State in Box)模型的人工神经网络进行了学习训练和分类仿真 。

Existing algorithms based on high order statistics (HOS) for recognizing radar targets appear to suffer from three shortcomings: (1) computational efficiency is low; (2) implementation of algorithm with hardware is not easy; (3) most of the algorithms require the prior information about statistical properties of the measured data. We regard bispectrum as image in plane and the amplitude of bispectrum as the gray scale of the image. We use the mean and coherence of bispectrum as two feature parameters and we define them as two feature vectors for target classification. In order to improve recognition performance, we also use the concepts of birange and bitime. On the basis of the two feature vectors we propose, we use BSB (brain state in box) artificial neural network model to classify target signals. We obtained simulation results of the performance of classification of radar target signals in the presence of additive Gaussian noise (GN) or exponential noise (non Gaussian noise,NGN) as shown in Tables 1 and 2. These results show preliminarily that classification accuracy is high.