Research On Spectral Clustering Algorithm And Its Application In Auroral Classification

Cluster analysis not only occupies an important position in the field of machine learning,but also plays a very useful role in people's understanding and exploration of the internal relations between things.At the same time,it is also one of the main methods.The process has the following five aspects:selection of features,similarity measures,criteria for clustering,clustering algorithms,and validation of results.When the clustering algorithm is applied to data mining and machine learning,its application prospect is very wide.For cluster analysis,one of the important branches is spectral clustering algorithm,which is also an important research object in the field of pattern recognition,machine learning,and data mining.When the spectral clustering algorithm is used to calculate the local similarity between sample point pairs and the spectral structure theory is used to mine the global structure between sample spaces,it is not necessary to use the sample space probability distribution as a premise,and its applicability is compared.It is more extensive than other clustering algorithms.Therefore,it has been widely concerned by scholars.The algorithm consists of two main steps:the first one is to establish a relation graph,and the algorithm establishes a relation graph mainly using the similarity measure between data sample points;the second is to build a correct clustering algorithm and use the built algorithm divides the graph.This paper is based on the above two main steps of spectral clustering algorithm as a starting point.For traditional spectral clustering algorithms,Euclidean distance is often used as a measure of similarity,and it can only be used for polynomials with local consistency.The class structure makes reflections and cannot reflect the clustering structure with global consistency characteristics,and when it encounters problems similar to the real world,most of the data that needs to be processed have multiple scales.The characteristics of the traditional spectral clustering algorithm are not suitable for solving these multi-scale clustering problems,and in the traditional algorithm,scholars often use the K-means algorithm in the final step,but due to K-means is to set the initial clustering center randomly,which will generate some very unstable clustering results.Therefore,this paper studies a similarity measurement method that can solve the problem of global consistency characteristics and a method to optimize the initial clustering center.The proposed algorithm is compared with some other clustering algorithms,and compared with the results obtained by the artificial data set and the UCI real data set.The final experimental results showing that the proposed clustering algorithm solves the problem.Both global consistency features and non-convex datasets have relatively good clustering results.After verifying the results obtained by the above two data sets,this paper will continue to obtain better results for the verification of the algorithm again.Therefore,the clustering algorithm is used again in the actual data.Aurora is a very beautiful phenomenon of luminescence,and it is also a gorgeous natural phenomenon.Its production is caused by the ionization or excitation of atmospheric molecules and atoms located in the upper layers.The classification of auroral images has always been a complicated process.One of the most important prerequisites for studying the evolution of auroral images is the classification of auroral images.It is also a problem that is highly valued and concerned by the scientific community and related scientific research fields.Therefore,in this paper,Aurora images are selected to cluster the aurora pictures.The results showing that the clustering algorithm proposed in this paper is also a good application in the clustering of auroral pictures.