With the widespread use of spatial data technologies, enormous and complex spatial data have been accumulated, thus
the traditional GIS (Geographic Information Systems) spatial analysis methods are confronted with great challenges.
Therefore, we need a new spatial analysis method of data-driven rather than model-driven, exploratory rather than
reasoning. Spatial clustering, which groups similar spatial objects into classes such that the intra-cluster similarity is
maximized and the inter-cluster similarity is minimized, is an important method of spatial data mining .
In the article, by virtue of Delaunay diagram, we propose a spatial clustering algorithm, which incorporates spatial
relationships with non-spatial attribute. Then the objects whose characters are less obvious are classified into clusters
that are more obvious and the precondition is that they are neighbouring, namely they must share the same Delaunay
edge. Along with rescaling, the same spatial object presents different states of distribution. We show via experiment of a
synthetic data set that our algorithm can integrate spatial relationships and non-spatial attribute. The obtained clustering
result is highly consistent with that perceived by human eyes and is capable of recognizing clusters of arbitrary shape.