针对非合作目标的自适应网格聚类算法

An Adaptive Grid-based Clustering Algorithm for Noncooperative Targets

武器系统的探测设备通常面对的是非合作目标,观测样本在特征空间中的分布形式难以预期,噪声、不规则的类簇形状以及差异化的类簇密度给聚类分析带来极大挑战。提出了一种自适应的网格聚类算法,该算法包括基于k-近邻方法的空间分辨率自适应网格化处理方法,以及基于自适应分水岭变换的类簇结构检测与划分方法。实现了对噪声以及密度差异极大类簇的自适应处理,同时保留了网格聚类方法对类簇形状不敏感、不需要类个数作为先验参数等优点。通过雷达、电子侦察以及复杂人造数据集的仿真,证明了该算法的有效性。

The detection equipment of weapon systems is usually used to detect the noncooperative targets,causing the distribution patterns of observed samples to be unpredictable in feature spaces.The irregular cluster shapes,diversified cluster densities and noise bring great challenges to clustering algorithms.A novel adaptive grid-based clustering algorithm,which consists of a k-nearest neighbor methodbased gridding method with spatial resolution adaptability,and an adaptive watershed transform-based method for cluster detection and segmentation in the gridded space are presented.The proposed algorithm could process the clusters with noises and significantly diverse densities,meanwhile keeps the advantages of gird-based clustering,including robustness for cluster shape and no need for cluster number as priori parameter.The effectiveness of the algorithm is tested with simulation and artificial datasets.