Research On Overlap Similarity-based Hierarchical Clustering Algorithms And Its Application

With development of information techinology, it's critical to extract relationships and connotative information from a large amount of data. So Data Mining is proposed to resolve this problem and comprised statistics, database, artificial intelligence, machine learning and so on. Clustering analysis is an important study field in data mining. It has found many applications to data classification and plays a key role in assessing relationships among patterns of data.The objective of clustering analysis is to partition a given data into some groups. Selecting the appropriate number of clusters and distinguishing partially overlapping and irregular data are two important problems in clustering. The appropriate number of clusters often requires to be specified in advance. Unfortunately, in most actual situations, it's not known and sometimes it is difficult to specify any desired number of clusters. The other problem is distinguishing partially overlapping clusters of data. Sometimes, clusters are similar with each other. We consider them as overlapping clusters. Hierarchical clustering provides a good solution to them, but how to control the iterative process of it is the key. Similarity measure is the key of controlling the iterative process of hierarchical clustering. In this paper, we studid on definitions of overlap similarity measure and proposed some hierarchical clustering algorithms based on them. These definitions of overlap similarity measure were based on gaussian mixture model or fuzzy logic. Without specified number of clusters in advance, appropriate value can be decided in the iterative process. These hierarchical algorithms stop clustering according to the overlap similarity between clusters. After discussing some related topics, based on these hierarchical clustering above, two object extractors were designed to detect ships from SAR images and high resolution remote sensing images.