Research On Image Retrieval Based On Deep Convolutional Neural Network

In recent years,with the popularity of smart phones and other mobile terminals,as well as the rise of social networking and shopping sites,images,video and other multimedia data on the Internet showing explosive growth.How to retrieve the relevant information from the complex image data quickly and accurately has been a hotspot and difficulty in the field of computer vision and image retrieval.The content based image retrieval mainly includes two key technologies: feature extraction and similarity measurement.The traditional image retrieval methods are mostly based on the features of artificial extraction,due to the "semantic gap" between the low-level image content and the high-level semantics,the current image retrieval efficiency still needs to be further improved.This paper applies the method of deep learning to image retrieval,a depth image retrieval algorithm based on convolutional neural network was proposed.Firstly,the image data is preprocessed.Then,the image data set is trained by convolution neural network.When the network tends to converge,the training network model is saved,and the training model is used to extract the depth image feature and visualize the image feature.Finally,the method of measuring distance by image retrieval results.In the analysis of the experimental results,the relationship between the number of iterations,the loss and the accuracy was compared in detail.The results of image retrieval were analyzed by using the mean average precision.The validity and superiority of the image retrieval method based on convolution neural network were verified by experiments.Compared with the traditional methods,the experimental results show that the algorithm has better retrieval performance,and it has strong universality for different image data.