Graph Matching Algorithm Based On Neural Network And Its Application

Graph matching is a technology for efficient query in data based on graph structure,which is widely used in bioinformatics,knowledge mapping,computer vision and other fields.Compared with the traditional graph matching algorithm,the method based on machine learning and deep learning describes the matching task as the problem of minimizing the energy function,and uses the constructed model to adaptively learn the characteristics of the graph and modify the parameters,so that the whole process is more intelligent and efficient.However,the existing unsupervised learning matching algorithm has the problem of strong randomness,which leads to the instability of the results,and the possibility of inaccuracy of the artificially constructed features in the supervised learning algorithm,which leads to the low accuracy of the matching results.Based on this,this paper studies the adaptive feature learning matching algorithm based on neural network,and verifies the accuracy and effectiveness of the proposed algorithm with the application background of protein complex recognition and feature point matching.The main research work of this paper includes:(1)Aiming at the problems of unsupervised learning matching algorithm and supervised learning matching algorithm,a matching algorithm based on graph embedding and topology information is proposed.The algorithm uses graph embedding neural network model to adaptively learn the characteristics of graph,and uses nonlinear classifier to get the matching results,which effectively solves the problems of existing methods and improves the matching accuracy.(2)In view of the neural network model in the research content(1),the root subgraph is generated by random walk,and it will be redundant as the basic element of learning graph embedding,which leads to the problem that the representative of the final image embedding feature is not strong.The more targeted graph neural network comes from adaptive learning features,and the space conversion network is added to the network model.It is used to enhance the learning of the important structure in the graph,so as to improve the accuracy and effectiveness of the algorithm.(3)In order to verify the accuracy and effectiveness of the above algorithm,it is applied to the field of bioinformatics and computer vision to solve the problem of protein complex recognition and feature point matching respectively.Experimental results show that,compared with other algorithms,the proposed algorithm has good performance in F-measure,accuracy and sensitivity.