一种基于注意力机制的细粒度图像分类方法  

A fine-grained image classification method based on attention mechanism

作　　者：王婷 王新[1] 郑承宇 邓亚萍 尹甜甜 WANG Ting;WANG Xing;ZHENG Cheng-yu;DENG Ya-ping;YIN Tian-tian(School of Mathematics and Computer Science,Yunnan Minzu University,Kunming 650500,China)

机构地区：[1]云南民族大学数学与计算机科学学院,云南昆明650500

出　　处：《云南民族大学学报（自然科学版）》2021年第6期581-586,共6页Journal of Yunnan Minzu University:Natural Sciences Edition

基　　金：国家自然科学基金(61363022);云南民族大学研究生创新基金(SJXY-2021-005).

摘　　要：细粒度图像分类指对大类下的子类进行识别,其实质是挖掘图像中微妙而有区别的特征.三线性注意力抽样网络是一个以注意力机制为基础的细粒度图像分类模型,虽然对图像特征提取及分类性能得以提升,但模型的鲁棒性和泛化能力没有得到体现.在三线性注意力抽样网络基础上注入dropout和随机深度给模型添加噪声,并用数据增强对图像数据做预处理,以提高模型的泛化能力和鲁棒性.实验结果表明,相较于与3种主流的细粒度分类算法,改进后细粒度图像分类的准确率明显提升.Fine-grained image classification refers to the recognition of sub-categories under a broad category,and its essence is to mine subtle and distinguishing features in images.The trilinear attention sampling network is a fine-grained image classification model based on the attention mechanism.Although the image feature extraction and classification performance can be improved,the robustness and generalization ability of the model is not reflected.On the basis of the trilinear attention sampling network,dropout and random depth are injected to add noise to the model,and the image data is preprocessed with data enhancement to improve the generalization ability and robustness of the model.Experimental results show that,compared with the three mainstream fine-grained classification algorithms,the accuracy of the improved fine-grained image classification is significantly improved.

关 键 词：细粒度图像分类 注意力机制 DROPOUT 随机深度 数据增强 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]