Research On Image Retrieval Based On Deep Learning Model

As the number of images increases dramatically,it becomes more and more difficult to retrieve the required information from a large number of images.Therefore,image retrieval technology has emerged.Image retrieval techniques can effectively utilize the features of an image to retrieve a desired image.As the complexity of the image content increases,the performance of the Image retrieval based on image underlying features is still limited.With the development of deep learning,image retrieval based on deep learning has become a research hotspot.In order to solve the problem that the current hashing retrieval method can not keep the semantic similarity of images well,the generated hashing codes have information redundancy and so on,a method based on deep convolutional neural network to learn binary hashing coding is proposed.Firstly,the deep convolutional neural network is used to extract the feature representation of the image.Secondly,the image's feature representation from two fully connected layers is entered into the hash layer,and the classification error and threshold error are added to the loss function for training.Finally,the query image is entered into the model to obtain the corresponding hash codes.The experimental results on CIFAR-10 and NUS-WIDE datasets show that the proposed method is superior to other methods in mean Average Precision and can effectively improve retrieval performance.In order to further alleviate the existing problem of the semantic gap and improve the retrieval performance of multi-target images,a multi-objective image retrieval model based on deep convolutional neural network is proposed.Firstly,the candidate regions are generated by using the region proposal network,and then the candidate regions are mapped onto the feature map of the image,and the input of different scales are mapped into the fixed-scale feature vector through the ROI pooling layer.Then,image features are extracted by the fully connected layer in the deep convolutional neural network.Finally,the image feature representations are mapped as the hash codes that are easy to retrieve.The experimental results on the VOC2012 and NUS-WIDE datasets show that the proposed method has a certain improvement in average cumulative gain and weighted mean Average Precision,which can effectively improve the retrieval performance of multi-target images.