基于深度学习的图像检索研究进展

Advances on Image Retrieval Based on Deep Learning

【目的/意义】分析提炼国内外基于深度学习的图像检索的最新研究热点及研究方法,揭示该领域未来研究趋势。【方法/过程】文章以国际顶级人工智能会议集(CVPR/AAAI/IJCAI/NIPS/SIGIR)、中国知网(CNKI)数据库、Web of Science核心合集及作为数据来源,对相关文献进行词频统计,生成词云图对该领域研究热点进行分析;利用Citespace可视化工具对研究主题演化进行定量分析,并从研究主题、算法、模型及应用实践四个维度进行详细评述。【结果/结论】现有的图像检索主要集中于卷积神经网络在图像表示和分类方面的应用,未来需要在图像分类的实时性、深度哈希算法的特征融合、图像检索的智能应用等方面进一步研究。【创新/局限】本文的创新性在于系统的总结了基于深度学习的图像检索的应用现状及未来发展趋势。此外,在未来研究中还应进一步扩展和丰富研究资料,从而更加全面准确的对本研究主题进行深入论述与分析。

【Purpose/significance】Analyze and refine the latest research hotspots and research methods based on deep learning image retrieval locally and internationally,and reveal future research trends in this field.【Method/process】The paper uses International toplevel artificial intelligence conference set(CVPR/AAAI/IJCAI/NIPS/SIGIR),CNKI and Web of Science core collection,conducts word frequency statistics on related documents,and generates word cloud maps to analyze the research hotspots in the field;uses the Citespace visualization tool to quantify the evolution of research topics.It is reviewed in detail from four dimensions:research topic,algorithm,model and application practice.【Result/conclusion】The existing image retrieval mainly focuses on the application of convolutional neural networks in image representation and classification.In the future,further research is needed on the real-time of image classification,feature fusion of deep hash algorithm,and intelligent application of image retrieval.【Innovation/limitation】The innovation of this article is to systematically summarize the application status and future development trends of image retrieval based on deep learning.In addition,in future research,the research data should be further expanded and enriched,so as to conduct in-depth discussion and analysis of this research topic more comprehensively and accurately.