Large-scale Hash image retrieval method based on deep representation learning

Abstract

The invention discloses a large-scale Hash image retrieval method based on deep representation learning. The method comprises the steps of data preprocessing, construction of a convolutional neural network layer based on semi-supervision, design of a loss function based on paired Hash loss functions, optimization learning of the loss function and post-processing of a search result. According to the invention, the deep convolutional neural network layer and the full connection layer are adopted to carry out image feature extraction and hash function learning. A combined loss function composed of a cross entropy loss item with a label. A triple loss item and a pseudo label loss item is designed. A stochastic gradient descent method of a drive quantity is adopted for optimizing and solving. High calculation efficiency is achieved, and finally image retrieval performance with uniform precision and speed is achieved. Under the current situation of rapid development of the image search field, efficient modeling is carried out on the image search problem on the basis of an image data structure and a label, and the model accuracy and the more optimized query speed are effectively improved.

Description

technical field [0001] The invention belongs to the field of image retrieval and relates to a large-scale image retrieval method. Background technique [0002] With the rapid development of big data technology, mobile Internet technology, Internet of Things technology and other technologies, the collection, aggregation and storage of multimedia resources such as images and videos are becoming more and more convenient. In practical applications, about 80% of data is unstructured data stored in the form of documents, images, videos, audios, etc., and the same type of unstructured data grows exponentially by about three-fifths every year. In unstructured data, images are the most important and occupy a large proportion, but at the same time they also contain important information. Computer vision is primarily concerned with the study of image data, so it is only natural that it is one of the hottest areas of research in machine learning and artificial intelligence techniques. ...

AI Technical Summary

Problems solved by technology

[0007] In order to overcome the shortcomings of the low accuracy of existing large-scale image retrieval methods, the present invention provides a large-scale hash image retrieval method based on deep representation learning with high accuracy

Images