Research On Image Retrieval Method Based On Convolutional Neural Network

With the rapid development of the Internet,the iterative update of storage devices,the maturity of cloud technologies,and the storage of multimedia information such as images,audio,and texts are becoming more and more convenient.Currently,the era of big data in the true sense is ushered in.Under the premise of possessing massive image data,how to quickly and accurately retrieve the images required by users in massive databases has become a difficult and hot topic in the field of image retrieval.In recent years,Hash technology and single feature convolutional networks have been widely used in image retrieval,but there are still many shortcomings.For the hashing technique,the process of hashing and hashing in the supervised hash learning method is mostly separated.There are two problems in this method,namely,learning the suboptimal hash code and generating lower distinguishability.The characteristic representation of the single feature convolution network has the following problems: the single feature convolutional neural network can only capture high-level abstraction information and lose local structure information,thus affecting the performance of the retrieval model.In view of the above image retrieval problems,this paper summarizes the research and development of image retrieval technology at home and abroad,and proposes an image retrieval method based on deep hash convolutional neural network for image retrieval tasks,based on the depth of multi-level multi-scale feature fusion.Hash image retrieval party.The main contents of this paper are as follows:(1)Image retrieval method based on deep hash convolutional neural network(DHCQ)Aiming at the problem of feature extraction and hash code step-by-step learning,an improved network model is proposed to learn the feature representation of images and hash codes simultaneously in convolutional neural networks,and then use the supervised information of markers to learn network differences.The feature representation is directly used to generate the hash code and the predicted image tag,and the learning of the network is adjusted according to the associated quantization error and the prediction error,and finally the query image is quickly retrieved in a low-dimensional space.Experiments show that it is feasible to extract image features and hash code learning in the same convolutional neural network model.(2)Deep hash image retrieval method based on multi-layer multi-scale feature fusion(MLSF)In the image retrieval task,the high-level abstract information captured by the model is rich and the local detail information is missing.The result of the model retrieval can not meet the user's detailed requirements for the retrieval of images,and a multi-scale multi-scale feature fusion method is proposed.The feature maps obtained by different layers in the model are fused.The features of the deep hash image retrieval model based on multi-layer multi-scale feature fusion designed in this paper include high-level abstract information and underlying local information.The high-level abstract information is responsible for capturing the abstract structure,the underlying local information captures the image details,and the model reference Similar to the residual network method,the model training speed has been improved.In this paper,an image retrieval method based on deep hash convolutional neural network and a deep hash image retrieval method based on multi-level multi-scale feature fusion are proposed.The hash retrieval and multi-scale features are used to mitigate image retrieval speed and accuracy.problem.The experimental results show that compared with the existing mainstream methods,the proposed method has the characteristics of high retrieval precision and high speed.