综合线面特征分布的点目标多尺度聚类方法

考虑空间数据分布的复杂性与不连续性,提出了一种点目标聚类方法。算法利用全要素Voronoi图准确识别与表达点目标与线面实体的空间相关性;根据点目标位置分布特征计算面积阈值来控制聚类的粒度,同时以空间尺度变化下面积阈值的恒定作为判断尺度收敛的条件,实现点目标的多尺度划分,时间复杂度为O(nlogn)。经试验验证,聚类尺度随点目标分布特征自适应收敛,算法无须自定义参数,能够有效地发现受线面目标约束的任意形态点目标集群,对异常值处理稳健。

Multi-scale Clustering of Points Synthetically Considering Lines and Polygons Di stri buti on

Considering the complexity and discontinuity of spatial data distribution ,a clustering algorithm of points was proposed .To accurately identify and express the spati al correl ation among points ,lines and polygons ,a Voronoi di agram that is generated by all spati al features is introduced .According to the distribution characteristics of point’s position ,an area threshold used to control clustering granularity was calculated .Meanwhile ,judging scale convergence by constant area threshold ,the algorithm classifies spatial features based on multi‐scale ,with an O(nlogn ) running time .Results indicate that spatial scale converges self‐adaptively according with distribution of points .Without the custom parameters , the algorithm capable to discover arbitrary shape clusters which be bound by lines and polygons ,and is robust for outli ers .