A Semi-supervised Classification Algorithm Based On Figure

In the field of machine learning and data mining, semi-supervised learning has attracted more and more attention by the researchers, has gradually become a research hotspot. Semi-supervised classification using large amounts of unlabeled data auxiliary part of unlabeled data for supervised learning, so as to improve the classification performance. Graph-based semi-supervised classification is method that research more now, the method has good classification performance. However, computational complexity of the graph-based method is relatively high, while the scale is large, the time required and the cost of storage are very large, and graph-based methods are Direct Push, the new data can not be classified. To a certain extent, this limits the use of graph-based methods. In this paper, based on these problems in graph semi-supervised classification, carried out analysis and research.In this paper, we are research based on local and global consistency algorithm, and the basic idea of the algorithm is to create a graph based on the labeled samples and unlabeled samples, to represent the similarity between the samples with edge weights, and then let each sample of the tag information to its adjacent sample transmission iteratively, until it reaches the global stable state, the method is intuitive, flexible, but the computer complexity, and unable to classify new data. The article improves the algorithm aimed at its shortcomings, and applied to the image classification to experimental analysis, compared with other methods. This research work mainly includes the following several aspects:(1)describes the image classification and the basic concepts of the common classification methods, the semi-supervised learning of fundamental theory, current research status at home and abroad, and the basis of the theory involved in graph-based approach in the semi-supervised learning, research status, and the problems and difficulties in graph-based semi-supervised learning methods are reviewed, the main applications are described, and the key is introduced the local and global consistency algorithm.(2)Put forward a graph-based semi-supervised classification algorithm for image, In this algorithm, the measurement method for graph similarity matrix is improved, using geodesic distance to measure of matrix of similarity, which are more accurately to reflects the topological structure of the sample, using the compound nuclear strategy, combines the image spectrum information and spatial information, and then make the algorithm improves the performance of the classification. The experimental results show, using this algorithm compared with the original algorithm, get better classification performance.(3)In order to solve the problem that the computational complexity of graph-based semi-supervised classification algorithm is relatively high and the new data to the classification. This paper presents the classification algorithm combining K means algorithm and graph based semi-supervised algorithm, using the labeled sample label information of graph-based methods, through the K means method to obtain more labeled information, reduce the number of iterations, which reduces the complexity of graph-based methods; the final classification center determined by the algorithm can solve the classification problem of new data, enhance its expandability to new data classification; and through the experiment analysis of the parameters, find out the optimal range of parameters. The experimental results show, the algorithm of graph-based methods puts forward by this chapter improved the computing efficiency and classification performance obviously.