Research On Data Science Oriented Intelligent Image Retrieval Methods

With the rapid development of data science and the widespread application of network in daily life,the role of information search in the development of the network has become more and more prominent.Image retrieval technology is a hot issue in the field of information processing and artificial intelligence.Image retrieval technology is getting more and more attention from researchers.At present,the popular retrieval methods are mainly content-based image retrieval methods,that is,starting from the underlying visual characteristics of the image,the image features are retrieved using the content characteristics of the image.In recent years,with the successful application of convolutional neural networks in the field of computer vision,capsule network with better structure has come into being.In this paper,the image retrieval technology is studied from two aspects: the underlying visual features and the capsule network.The main work includes:(1)A two-stage sequential search algorithm based on feature point matching is proposed.First,the image is converted from RGB space to HSV space,and the color histogram of the image is obtained.Then,the color feature is used to form an image set containing images of similar color distributions to that of the query image.Secondly,the query image is converted into gray-scale image first,then the gray-scale image is denoised by using the sym5 wavelet function,and then the gray-scale image is reconstructed by using the low-frequency information in the image wavelet transform domain.SIFT feature points are extracted from the reconstructed image,and the feature points between the query image and each image in the primary image set are matched,and these images more similar to the query image are searched from the primary image set.Finally,experiments are performed on the ZuBuD dataset and the UKBench dataset,and the experimental results verified the effectiveness of the algorithm in this paper.(2)An image retrieval algorithm based on capsule network is proposed.First of all,combined with the idea of jumping connections in ResNet and DenseNet networks,a capsule network with jumping connection is constructed by stacking multiple full connection modules.Then,the parameters obtained from the last layer of the network are taken as the feature of the query image,which use to predict the class of the query image.The distance similarity of feature between the query image and each image in the class is calculated.The retrieval sequence is carried out according to the distance similarity,so that the performance of image retrieval can be optimized by image classification.Finally,the CIFAR-10 dataset is used to conduct experiments on the Keras deep learning framework,and the effect of the ensemble network model composed of multiple fine-tuning network models is verified.Experiments show that the capsule network as a feature extractor can effectively improve the expressive ability of image features,and the search effect is better.