| Multi-view can comprehensively analyze data from multiple perspectives,and can also effectively use the relevant information and complementary information between various views.Therefore,multi-view clustering analysis has become one of the research hot spots in the fields of machine learning and pattern recognition.However,in multi-view integrated clustering analysis,the base clustering cluster is a cluster in the base clustering,which contains several similar data objects.Its density can only reflect the distribution characteristics of the data itself,and cannot reflect the quality of the base clustering cluster,which affects the integrated clustering effect.In this thesis,the joint entropy is used to evaluate the uncertainty and quality of the base clustering,and the reconstruction of the co-association matrix.The multi-view ensemble clustering analysis method and its application are studied in depth,which effectively reflects the distribution characteristics of the multi-view data itself,and deletes the negative evidence in the co-association matrix,and improves the multi-view ensemble clustering performance.The main work is as follows :(1)A multi-view ensemble clustering analysis method based on joint entropy is proposed.Firstly,the quality of base clustering is evaluated by joint entropy,and an uncertainty index of base clustering is defined to effectively reflect the importance and quality of base clustering.Secondly,a weighted co-association matrix is constructed by using the uncertainty index of base clustering cluster,and a multi-view ensemble clustering algorithm is proposed,which effectively improves the performance of multi-view ensemble clustering analysis.Finally,on UCI and artificial data sets,experiments verify the importance of cluster weight in multi-view ensemble clustering analysis,which can effectively improve its ensemble clustering performance.(2)A co-association matrix and multi-view integrated clustering analysis method for negative evidence deletion is proposed.Firstly,the original co-association matrix of the data is generated by the evidence accumulation model,and the original co-association matrix is reconstructed by Visual assessment of cluster tendency(VAT).Secondly,the Ncut algorithm is used to define the low frequency part of the co-association matrix as negative evidence and delete it from the co-association matrix.Then,using the Ncut algorithm,the candidate cluster is obtained from the co-association matrix,and on this basis,the best cluster is selected from the candidate cluster by using the maximum and minimum similarity.Finally,using UCI and artificial data sets,experiments show that the method can obtain better multi-view clustering results.(3)The prototype system of multi-view ancient building data integration clustering is designed and implemented.Firstly,the functional module diagram of the prototype system is described,as well as data import,data processing,clustering data mining and other functional modules.Secondly,on the Pycharm platform,the prototype system is designed and implemented by using Python language.Finally,the running results on the multi-view data-set of Jinguang Temple show that the prototype system can provide an effective feature matching approach or the three-dimensional reconstruction of ancient buildings. |
