基于有监督对比学习的哈希图像检索方法  

A supervised contrastive learning based hashing image retrieval method

作　　者：江祥奎 呼飞 JIANG Xiangkui;HU Fei(School of Automation,Xi’an University of Posts and Telecommunications,Xi’an 710121,China)

机构地区：[1]西安邮电大学自动化学院,陕西西安710121

出　　处：《西安邮电大学学报》2024年第2期94-102,共9页Journal of Xi’an University of Posts and Telecommunications

基　　金：陕西省重点研发计划项目(2022NY-087,2024GX-YBXM-300);陕西省社科联/陕西省应急管理厅项目(2021HZ1121);陕西省青年托举项目(20220129);陕西省教育厅科学研究计划项目(22JK0565)。

摘　　要：针对哈希图像检索方法检索精度低的问题,提出一种有监督对比学习的哈希图像检索方法。在残差网络50(Residual Network 50,ResNet50)中嵌入协调注意力模块,提取图像的关键信息,优化网络的特征提取能力,并采用有监督对比学习方法进行训练,增强网络的类区分能力。在目标函数中引入量化约束减小误差,保持生成哈希码间的平衡性和相似性,提升哈希码的质量。实验结果表明,所提方法优于其他基于哈希的图像检索方法,能够实现较高的检索精度。A supervised contrastive learning hash image retrieval method is proposed to address the issue of low retrieval accuracy in hash-based image retrieval.By embedding a coordinated attention module into residual network 50(ResNet50),it can extract key information from images,optimize the network’s feature extraction ability,and enhance its class discrimination ability by employing supervised contrastive learning methods for training.Quantization constraints are introduced into the objective function to reduce errors,and maintain balance and similarity among generated hash codes,thus improving the quality of hash codes.Experiment results show that the proposed method outperforms the other hash-based image retrieval methods,and can achieve high retrieval accuracy.

关 键 词：图像检索 残差网络 注意力机制 有监督对比学习 量化误差 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]