Research On Density-based Distributed Clustering Algorithms

With the development of information technology and the rapid accumulation of data in the modem society,unsupervised learning algorithms are increasingly attracting people's attention.Clustering algorithm is not only an important branch of machine learning algorithm,but also an important representative of unsupervised learning.It is widely used in image pattern recognition,social network analysis,network security and other fields.The traditional clustering algorithm and engineering practice are mainly based on single site,but this approach will be constrained in the data distribution,computing resources,storage space and other aspects with the growing size of the data.Distributed clustering algorithms solve the problem that traditional clustering algorithms cannot be applied in distributed environments and ensure the scalability of computing resources and storage space at the same time,so that clustering algorithms are no longer constrained by data size and computing resources.In this paper,we propose a new distributed clustering algorithm,named REstore MOdel with Local Density estimation(REMOLD).REMOLD using Density Peaks algorithm as local clustering algorithm to generate atom clusters.The density distribution of the atom clusters is represented by models,and all models are collected to the host site,and the host simulates the global density through the density distribution models and performs clustering again.REMOLD inherits the advantages of density clustering,while greatly reducing the network cost on the central node.The experimental results show that the REMOLD algorithm can achieve good results in terms of computing time,network transmission and clustering indexes.The REMOLD algorithm has satisfactory performance compared with the three classical density-based distributed clustering algorithms,DBDC,Basic-DDP,LSH-DDP.And the scalability of the algorithm is verified through experiments.Hyperspectral images clustering has always been one of the most important methods for analyzing hyperspectral images.Hyperspectral images have attracted a great deal of attention due to their huge potential value in many areas,such as resource exploration,military reconnaissance and land surveys because they contain a large amount of ground object information.The traditional hyperspectral image clustering method is limited by the spatial resolution of hyperspectral image and the dimension of the band,and it is not scalable.In this paper,the method of density-based distributed clustering is used to cluster the hyperspectral images.Input is the original hyperspectral images,and output is the visualized clustered images.Experimental results show that the clustering result of this method has good computational efficiency and high performance and can be applied to hyperspectral image clustering in big data scenarios.