Aiming at problem that density-based spatial clustering of application with noise（ DBSCAN） clustering algorithm uses fixed parameters Eps and Minpts

which makes clustering effect for multi-density data non ideal. An improved DBSCAN algorithm for multi-density is proposed. The algorithm processes data

and identify the density around each data target. According to this density

density threshold suitable for the region density is automatically generated. Extended clustering is performed with this density threshold before the end of this clustering. When the next cluster is clustering

the density threshold suitable for the region is automatically generated

so continue like this until the clustering condition is not reached. A large number of experiments show that the algorithm can effectively cluster the multi-density data with arbitrary shape.