The Study Of Graph Matching Based On Graph Kernels And The Application In Architectural Space

With the improving kernel method theory and its wide application in real life,people are increasingly optimistic about the development of kernel methods. The mostimportant is that the kernel method can be applied not only in the feature vector ofstatistical pattern recognition, and also very good application on a structuredrepresentation of the structure pattern recognition. Therefore, to solve graph matchingproblems, the introduction of kernel method has become a new research direction. Withthe graph structure data continuously generated, graph learning and data miningapplications become more and more challenging. The present kernel methods have leftout the importance of topology information, had a significant amount of running timeand not been calculated for large graphs, while graph kernels have followed graphtopology, node and edge lables represent information. In recent years, kernel methodhas transformed structure recognition problems to statistica recognition problems. Thereare several graph kernels such as random walk kernel, shortest path kernel and diffusionkernel. This paper mainly studies the graph matching method of graph kernel and itsapplication in the construction sector property.First of all, this paper expounds the present random walk kernel and shortest pathkernel, discusses the applications of graph kernels, discuss the basic operating principleof support vector machines classification algorithm and superiority.Secondly, this paper does a deep research of random walk kernel, derives andachieve the quickly calculate random walk kernel. By synthetizing random walk kernelwith radial basis kernel function,we have improved the accuracy of the random walk ofkernel. And combining the support vector machine, we have performed graph matching.The experiments show that the synthetized random walk kernel compared to the above algorithm has higher classification accuracy.Finally, this paper analyses and achieves the shortest path kernel, furtherly expandssynthesis sensor shortest path kernel and achieves it. The experiment shows that it hashigher prediction accuracy that can indicate the superiority of algorithm.