Research On Image Retrieval Technology Based On Multi-source Big Data Analysis

In recent years,information science and technology are developing at a high speed and at the same time changing the human lifestyle.People share various images on the Internet and hundreds of millions of images are uploaded to the Internet every day.How to quickly and accurately retrieve the images the user needs from these massive images is a major research topic.As an important part of image retrieval,image annotation plays an important role in the image retrieval system.We study the development and latest achievements of image retrieval technology,focuses on the technical characteristics of text-based image retrieval systems and the deficiencies of the current mainstream Internet retrieval systems.Based on this,it proposes a method based on multi-source big data fusion.Image annotation framework,the framework first makes use of multi-source information fusion to make the annotation of the image more complete.Secondly,the scene is annotated with deep learning.Finally,the two annotations are merged to solve the problem that the depth and richness of the image annotation.Firstly,an annotation model based on multi-source big data fusion is proposed.The model obtains similar images by analyzing related multi-source information of the images,and labels the images by using the annotations of similar images.This model solves the problem that Internet image text cannot be retrieved when it is insufficient.In the following text,the specific analysis method for multi-source information of images will be introduced.Finally,it is verified in the specific scene of Weibo image that the model can well enhance the depth and richness of the image annotation.Secondly,an image scene annotation model based on deep learning is proposed.The model identifies the scene through the key objects and background features in the scene image.By improving on the fast RCNN network,an image scene model based on deep learning is realized.After a large number of data sets are trained,the model can accurately animate the scene.The detailed framework,implementation and training of the model will be described in detail later,and the model has been experimentally verified on the standard scene identification data set to have excellent accuracy for scene recognition.Finally,the two annotation models proposed in the previous text are combined,and a web-based image retrieval system is designed and implemented based on crawler technology and distributed storage technology.