Research And Implementation Of Logo Recognition Method For Web Applications For Smartphones

With the rapid development of Internet technologies and mobile devices,smartphone web applications have been widely welcomed and valued due to their natural cross-platform nature and the characteristics of no need to download and user-friendly.As a basic technology and important step in many smartphone web applications,logo recognition technology has also received the attention and development of developers.So far,the traditional logo recognition technologies basically use the method of feature matching by artificially selected features or contour extraction to achieve the purpose of recognition.However,these traditional methods have many shortcomings when applied to smartphone web application scenarios due to their limitations.In recent years,convolutional neural network has achieved great success in the field of image recognition with the rapid development of deep learning,which provides new ideas for the solution of logo recognition problems in smartphone web applications.Convolutional neural network generally performs well in solving multi-classification problems of massive data.When directly applied to solve the problem of logo recognition of a single brand,the network performance is not very good due to simple logo structure and limited size of data set.In order to utilize the excellent characteristics of convolutional neural network to solve the problem of logo recognition of a single brand effectively,this paper proposes a sparring mechanism and a threshold judgment method.This paper studies the application of several classical convolutional neural networks in logo recognition and explores the influence of the number of classification categories on the performance of convolutional neural networks.The concept of sparring data sets is proposed and the motivation of introducing sparring mechanism is analyzed.Then,a series of experiments were designed to prove the effectiveness of the sparring mechanism to improve the recognition ability of the convolutional neural network through the data enhancement principle,and then the strategy of finding the optimal sparring set quickly and accurately by means of shape context algorithm and transfer learning is proposed and verified by experiments.Then,in order to further optimize the convolutional neural network structure for single logo recognition,a novel threshold-based judgment method is proposed.This method are compared with some excellent traditional logo recognition methods by testing the performances such as recall rate,accuracy rate and recognition time.The experimental results show that the convolutional neural network model proposed in this paper can solve the problem of single logo recognition for smartphone web applications more effectively than traditional methods.At the end of the thesis,the augmented reality game is taken as an example to realize the smart phone web application,and the application is tested in the actual scene,which proves that the application has strong practicability.