Improvement Of Spectral Clustering Algorithm And Application In Landslide Field

China is located in the area where the Himalayan plate and the Pacific Rim plate are connected.The mountain area accounts for about 70% of the country's land area.It is a country with frequent geological disasters,especially landslide disasters.The occurrence of landslide disasters not only bring huge casualties and economic losses to the country and the people,but also cause environmental damage and resource loss to a certain extent.Therefore,how to obtain information on landslide hazard and excavate the occurrence rule of landslide disasters,so that it is an urgent problem to detect landslide disasters early.With the development of computer technology,the data storage processing capability has also been improved,and computer technology has been widely applied to various industries.Using computer's ability to analyze and process data,combined with data mining algorithms to analyze and process landslide hazard data,has become one of methods landslide data analysis.As an important branch of data mining,cluster analysis is a common method used in data analysis and processing.Spectral clustering algorithm is a classical data analysis method of clustering analysis algorithm.It is developed on the basis of spectral graph theory.It can realize clustering on arbitrary non-convex data sets and is not easy to fall into local optimum.However,there are certain defects.In view of its defects,this thesis proposes an improvement of related algorithms and applies it to the Kash Taskuergan area of Xinjiang.The main work of this thesis is as follows:(1)Briefly describes the relevant theoretical knowledge of spectral clustering algorithm and spectral graph theory,introduces the research status and research hotspot of spectral clustering algorithm in detail,and proposes our own improvement ideas for the problems existing in current spectral clustering algorithm.(2)An adaptive spectral clustering algorithm based on Artificial Bee Colony(ABC)algorithm is proposed.For the problem that the initial clustering number of the spectral clustering algorithm cannot be determined,this thesis uses the idea of intrinsic gap to obtain the initial cluster number K value by constructing the eigengap sequence.To further improve the global optimization ability of the algorithm,this thesis adds the ABC algorithm idea to the post-clustering process,and improves the position update formula of the ABC algorithm,thus improving the global optimization ability of the algorithm.Finally,the standard benchmark functions are tested and the Iris,Wine and Glass data sets in the UCI data set are selected to simulate the improved algorithm,and the results are compared with different algorithms.(3)Instance verification.The remote sensing images of the Kash Taskuergan area in Xinjiang were selected as the experimental verification area of this thesis.Using the improved spectral clustering algorithm to detect the change of remote sensing image in the study area,and the change information in the non-local difference map of the study area is extracted.In the experiment of construction accuracy of difference map,in this thesis,the number of error detection in the difference graph construction method is 3741,which is less than the difference method of 12177 and the ratio method 7052.In the experiment of quantitative analysis of the effect of comparison algorithm on the extraction of differential graph,the detection rate of the improved algorithm is 96.437% in PCC,which is higher than that of PCA-K and Otsu's.Therefore,the improved algorithm in this thesis has strong ability to analyze and extract data,and the implementation effect of the algorithm is better.