An Improved Density Peak Clustering Algorithm And Its Application Research

Clustering by fast search and find of density peaks(abbreviated as DPC)is a new clustering algorithm proposed in Science magazine in 2014.Compared with traditional algorithms,DPC has many advantages,such as fewer parameters to be set artificially,no iteration in clustering process,and can discover clusters of arbitrary shape.Therefore,this algorithm has become a research hotspot of clustering algorithms in recent years and has been widely used in various fields.Although DPC algorithm performs well on most datasets,it still has some shortcomings:(1)the algorithm needs manual intervention to select the center points,which results in the subjectivity of clustering results;(2)the algorithm defaults that there is only one center point in a cluster,and it is easy to misclassify clusters with multiple centers.Therefore,in order to solve these problems,this paper proposes a linear density peak clustering algorithm(LDPC)based on linear fitting,which improves the DPC algorithm in two aspects:(1)LDPC adopts the method of linear fitting to automatically determine the cluster center;(2)LDPC combines multiple central points in the same cluster according to the principle that the density of data points in the same cluster is reachable.The core idea of LDPC algorithm is to automatically determine clustering centers.In order to verify the effectiveness of LDPC algorithm,this paper uses classical K-Means algorithm to verify whether LDPC algorithm can automatically and accurately determine clustering centers.LDPC algorithm is used to initialize K-Means.In this paper,the improved K-Means algorithm initialized by LDPC is named LDKM algorithm(Linear fitting Density Peaks K-Means algorithm).The algorithm results show that LDPC can accurately determine the clustering center,that is,the algorithm verifies the effectiveness of LDPC.In this paper,five artificial data sets and three UCI data sets are used to experiment on LDPC and LDKM algorithms.The effectiveness of the algorithm is verified and compared with K-Means,FCM,DBSCAN and DPC algorithms.Experiments show that the LDPC algorithm performs best on most data sets,and the evaluation index is better than other algorithms.Compared with traditional algorithms,LDPC has a wider range of data types and can handle arbitrary shape cluster data.Compared with DPC algorithm,it does not need manual interference processing,and the experimental results are better.LDKM algorithm is better than traditional K-Means algorithm.Compared with the traditional K-Means algorithm,LDKM does not need incoming parameters and has better clustering effect.Finally,the LDPC and LDKM algorithms are applied to the field of image segmentation.Compared with other algorithms,the experimental results have clear outline and less noise.Finally,LDPC algorithm are applied to the classification and recognition of leukocyte.The background noise of the blood cell image is eliminated,and only leukocyte are retained,which is convenient for subsequent processing.Experiments show that LDPC algorithm can extract leukocyte completely,and the extraction results are pure.