摘要
针对细粒度图像分类中数据分布具有小型、非均匀和不易察觉类间差异的特征,提出一种基于注意力机制的细粒度图像分类模型.首先通过引入双路通道注意力与残差网络融合对图像进行初步特征提取,然后应用多头自注意力机制,达到提取深度特征数据之间细粒度关系的目的,再结合交叉熵损失和中心损失设计损失函数度量模型的训练.实验结果表明,该模型在两个标准数据集102 Category Flower和CUB200-2011上的测试准确率分别达94.42%和89.43%,与其他主流分类模型相比分类效果更好.
Aiming at the characteristics of subtle,uneven,imperceptible inter-class differences between classes and real-world data distribution in fine-grained image classification,we proposed a fine-grained image classification model based on attention mechanism.Firstly,the preliminary feature extraction of the image was carried out by introducing the fusion of a two-way channel attention and residual network.Secondly,the multi-head self-attention mechanism was applied to extract fine-grained relationships between deep feature data.Thirdly,the training of loss function measurement system was designed by combining cross entropy loss and center loss.The experimental results show that the test accuracy of the model on two standard datasets 102 Category Flower and CUB200-2011 is 94.42%and 89.43%,respectively.Compared with other mainstream classification models,the classification effect is better.
作者
朱丽
王新鹏
付海涛
冯宇轩
张竞吉
ZHU Li;WANG Xinpeng;FU Haitao;FENG Yuxuan;ZHANG Jingji(College of Information Technology,Jilin Agricultural University,Changchun 130118,China)
出处
《吉林大学学报（理学版）》
CAS
北大核心
2023年第2期371-376,共6页
Journal of Jilin University:Science Edition
基金
吉林省教育厅科学技术研究项目(批准号:JJKH20220331KJ,JJKH20220332KJ,JJKH20220333KJ)
吉林省发展和改革委员会省预算内基本建设基金(批准号:2020C037-7).
关键词
细粒度图像分类
注意力机制
残差网络
fine-grained image classification
attention mechanism
residual network
