基于数据场联合决策图改进的GMM聚类

针对传统聚类方法在处理复杂电磁环境下的雷达信号时存在的聚类质量低、参数需要人为设置、易受孤立噪声脉冲干扰等问题, 提出一种基于数据场联合决策图改进的高斯混合模型(Gaussian mixture model, GMM)聚类算法。将数据场理论应用于数据对象密集程度的表征, 生成势能距离决策图, 进而自动实现聚类数目和中心点的选择, 最后结合GMM聚类实现对数据对象的聚类划分。仿真实验结果表明, 在脉冲到达角、脉宽、载频等参数存在较大抖动, 测量误差以及存在孤立噪声脉冲干扰和脉冲丢失时, 本文方法相较于现有典型分类方法具有更好的分选效果。

An improved GMM clustering based on data field and decision graph

Aiming at the problems of traditional clustering methods in processing radar signals in complex electromagnetic environment, such as low clustering quality, manual parameter setting and poor tolerance to isolated noise pulses, an improved Gaussian mixture model (GMM) clustering algorithm based on data field and decision graph is proposed. The data field theory is applied to the representation of the density of data objects, the potential energy distance decision graph is generated, and then the selection of cluster number and center point is realized automatically. Finally, the clustering division of data objects is realized combined with GMM. The simulation results show that this method has better sorting effect than the existing typical classification methods when there are significant jitter and measurement error in direction of arrival, pulse width and radio frequency, with the isolated noise pulses interference and pulse loss in the meantime.