论文部分内容阅读
针对在信号特征提取与识别中使用双谱估计数据量大、维度高的问题,本文提出了双谱对角切片与广义维数相结合的识别方法。通过提取信号双谱对角切片(Bispectra Diagonal Slice,BDS)减少数据量,并利用多重分形理论中的广义维数(Generalized Dimension,GD)降低数据维度,对切片内部特性进行细微描述,基于距离测度提出特征评价指标,从而选出最具有区分度的三个阶数q对应的广义维数Dq作为特征向量,最后输入到最小二乘支持向量机中进行分类识别。使用四种低截获概率雷达信号作为待识别信号,其仿真结果表明,算法提取的信号特征在特征空间中有良好的聚集性和离散性,在0d B信噪比下,识别准确率能达到92.2%,与选取的其他算法对比说明其具有很好的识别性能。
Aiming at the problem of large amount of data and high dimension using bispectrum in signal feature extraction and recognition, this paper presents a method of identifying bispectrum diagonal slices combined with generalized dimension. The data volume was reduced by extracting Bispectra Diagonal Slice (BDS) signals and the data dimension was reduced by the Generalized Dimension (GD) in the multifractal theory to describe the internal characteristic of the slice subtly. Based on the distance measure Then the feature evaluation index is proposed to select the generalized dimension Dq corresponding to the three degree q which has the highest degree of discrimination as the eigenvector and then input to least square support vector machine for classification and recognition. Four kinds of low-probability-of-arrival radar signals are used as signal to be identified. The simulation results show that the signal features extracted by the algorithm have good clustering and dispersion in the feature space, and the recognition accuracy can reach 92.2 %, Compared with other algorithms selected to show that it has good recognition performance.