一种基于CNN-RNN模型的图像检索技术  

An Image Retrieval Technology Based on CNN-RNN Model

作　　者：汤永斌[1] TANG Yongbin(Nanchong Vocational and Technical College,Nanchong Sichuan 637131,China)

机构地区：[1]南充职业技术学院,四川南充637131

出　　处：《信息与电脑》2023年第9期182-184,共3页Information & Computer

摘　　要：图像检索是一项重要的研究课题,涉及如何快速、准确地检索和管理海量的图像数据。传统的图像检索技术主要依赖图像的视觉特征或文本描述进行匹配,但是难以充分理解图像的语义信息,对复杂场景的适应性较差。针对这一问题,文章提出了一种基于卷积神经网络-循环神经网络(Convolutional Neural Networks-Recurrent Neural Network,CNN-RNN)模型的图像检索技术。该技术将CNN和RNN相结合,构建了一个统一的深度学习框架。其中,CNN模型用于从图像中提取全局特征,RNN模型用于学习图像与标签之间的语义关联和共现依赖。文章通过将CNN输出的特征序列输入到RNN模型中,实现了对图像全局语义信息的捕获。将设计系统在多个数据集上进行实验,结果表明,设计的方法能够有效提高图像检索的效率和准确性。Image retrieval is an important research topic that involves how to quickly and accurately retrieve and manage massive amounts of image data.Traditional image retrieval technologies mainly rely on the visual features or text descriptions of images to match,but these technologies are difficult to fully understand the Semantic information of images and have poor adaptability to complex scenes.In response to this issue,this article proposes an image retrieval technology based on the Convolutional Neural Networks-Recurrent Neural Networks(CNN-RNN)model,which combines CNN and RNN to construct a unified deep learning framework.The CNN model is used to extract global features from images,while the RNN model is used to learn semantic associations and co-occurrence dependencies between images and labels.By inputting the feature sequence output from CNN into the RNN model,this paper realizes the capture of the global Semantic information of the image.This article conducted experiments on multiple datasets,and the results showed that the method proposed in this paper can effectively improve the efficiency and accuracy of image retrieval.

关 键 词：图像检索 循环神经网络(RNN)模型 卷积神经网络(CNN)模型 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]