Research On Improved DBSCAN Algorithm And Its Application

In the era of big data,clustering is an unsupervised learning method commonly used in data mining.It divides the data set into different clusters according to the similarity of the data,so that the elements in the same cluster are as similar as possible and the elements in different clusters are different.In particular,the DBSCAN clustering algorithm can find clusters of different shapes and sizes,and identify noise or outliers.How to improve the accuracy of DBSCAN clustering results and the practicability of the algorithm is a direction worth studying.This paper studies the DBSCAN algorithm and does the following:(1)Aiming at the problem that the traditional DBSCAN algorithm has poor clustering effect on high-dimensional data sets and the selection of parameters is sensitive,a GS-DBSCAN algorithm based on similarity measurement is proposed.The algorithm constructs a similarity matrix between geodesic distance and shared nearest neighbor data points,overcomes the limitations of Euclidean distance on high-dimensional data,better characterizes the real situation of the data set,and can also analyze the distribution characteristics of data Determine Eps and MinPts parameters from adaptation.The experimental results show that the GS-DBSCAN algorithm can effectively cluster complex distributed data,and the clustering accuracy in high-dimensional data is higher than that of the comparison algorithm.(2)Aiming at the problem of inaccurate segmentation caused by simple linear iterative clustering(SLIC)that only considers color and space features,an improved image segmentation method of SLIC and DBSCAN is proposed.First use bilateral filtering to perform image enhancement,remove noise in the image and protect edge features.Then the SLIC algorithm combined with the adaptive multi-threshold LBP texture feature is used to segment the color image into superpixel blocks,and finally the DBSCAN algorithm is used for clustering and merging to obtain the segmented result map.Experimental results show that the proposed algorithm can accurately segment image boundaries.Compared with other existing image segmentation methods,the accuracy and quality of segmentation have been significantly improved.