Image Retrieval Based On Deep Learning And Relevance Feedback

With the advent of the era of big data on the Internet,image data as one of the important data forms has been growing on a large scale.In the face of such a huge image data,how to retrieve it quickly and accurately to meet the need of users for fast browsing and querying interested images is still an urgent problem to be solved.Deep learning is the product of the era of big data,which is an essential key technology in the field of artificial intelligence and can effectively learn features from big data.At the same time,image classification technology can analyze and understand the image data automatically to a certain extent,and which is consistent with the user's cognition.The rapid development and improvement of deep learning and image classification technology provide important technical support for the optimization of image retrieval technology.Aiming at the problems with content-based image retrieval of low-level visual features are inconsistent with the high-level semantic meaning of the user's understanding of the image,the low accuracy of image retrieval,and the traditional classification method is insignificantly accurate in image classification.An image retrieval method based on deep learning and relevance feedback is proposed.Firstly,image features are extracted from the original image data by using the convolution neural network with good self-learning ability in the deep learning framework,and through supervised learning of multi-layer neural network,abundant image features are extracted from original image data.The support vector machine is used as the basic classifier to classify the extracted depth features.Before classifying,the original dataset needs to be divided into training and test sets.Secondly,in order to improve the classification accuracy and speed,Top-K sorting method combined with multiple distances is used to select training sets reasonably.Finally,the optimal hyperplane is trained according to the screened training sets,and the retrieval results are sorted according to the distance between the test sample data and the classified hyperplane.The scheme is compared with other retrieval methods in Corel5 K dataset and UC Merced Land Use dataset.The results show that the proposed method is superior to the comparison method under various accuracy evaluation indexes,and has important significance for improving retrieval performance.