APP Similar Icon Retrieval System Based On Deep Learning

With the development of the Internet and the increasing popularity of smart phones,handheld mobile terminals have been greatly developed.Accompanied by the increase in the number of mobile phone terminal software,for users and safety inspection personnel through the mobile phone application icon to quickly search a mobile phone application corresponding or similar is an urgent need to solve.Current image retrieval techniques are mainly based on text retrieval and content-based retrieval.With the development of deep learning,image retrieval technology based on deep learning has become a hot topic.In this paper,the theory of deep learning and image retrieval is studied,and the depth learning technology is used to realize the image retrieval system.The main work of this paper is as follows:First,the related techniques of deep learning are investigated.The structure of convolutional neural network,the training process of network model and the depth learning framework are studied.The traditional image retrieval methods and feature extraction methods are compared and analyzed,and the shortcomings of the traditional model in retrieval results and retrieval effects are analyzed.Secondly,a convolutional neural network model is proposed.The convolutional neural network model can extract the multi-layer features of the image,and use a layer of information which is rich in information to be used as a feature in the retrieval system.Compared with the traditional artificial features,this feature has better robustness,and has better anti-interference for the image rotation,zoom,contrast and so on.In order to facilitate the retrieval of images,hash numbers are hashed and different indexes are established for the features after hashing.The index of features and features is stored in an efficient search system.Then,combining large data platform and local sensitive hashing algorithm,the extracted features are used for fast retrieval in large-scale dataFinally,the results of image retrieval are compared and analyzed on public data sets,and satisfactory results are obtained.