[1]白瑜颖,刘宁钟,姜晓通.结合注意力混合裁剪的细粒度分类网络[J].计算机技术与发展,2021,31(10):38-42.[doi:10. 3969 / j. issn. 1673-629X. 2021. 10. 007]
　BAI Yu-ying,LIU Ning-zhong,JIANG Xiao-tong.Fine Grained Image Classification Network Combined with Attention CutMix[J].,2021,31(10):38-42.[doi:10. 3969 / j. issn. 1673-629X. 2021. 10. 007]
点击复制

结合注意力混合裁剪的细粒度分类网络()
分享到：

《计算机技术与发展》[ISSN:1006-6977/CN:61-1281/TN]

卷:
31
期数:
2021年10期
页码:
38-42
栏目:
图形与图像
出版日期:
2021-10-10

文章信息/Info

Title:
Fine Grained Image Classification Network Combined with Attention CutMix
文章编号:
1673-629X(2021)10-0038-05
作者:
白瑜颖; 刘宁钟; 姜晓通
南京航空航天大学 计算机科学与技术学院,江苏 南京 211106
Author(s):
BAI Yu-ying; LIU Ning-zhong; JIANG Xiao-tong
School of Computer Science and Technology,Nanjing University of Aeronautics and Astronautics,Nanjing 211106,China
关键词:
细粒度; 卷积神经网络; 弱监督; 注意力机制; 混合裁剪; 数据增强
Keywords:
fine-grained; convolutional neural network; weak supervision; attention mechanism; CutMix; data augmentation
分类号:
TP391.4
DOI:
10. 3969 / j. issn. 1673-629X. 2021. 10. 007
摘要:
细粒度图像识别旨在区分同属某一大类下更为精细的子类,具有类间差距小和类内差距大的特点。 同时细粒度数据集往往种类多,而数据量较少,容易产生训练时的过拟合。 针对上述问题,文中提出了一种结合注意力混合裁剪的细粒度分类网络,利用注意力机制指导改进的混合裁剪数据增强。 首先使用 ResNet50 作为基础网络提取图像特征,之后利用 1*1 卷积获取注意力图,再通过双线性注意力池化操作将特征图与注意力融合拼接成特征矩阵,最后利用注意力图进行改进的混合裁剪数据增强。 其中改进的混合裁剪数据增强是交换两张图片的注意力高峰区域,同时交换两张图片的标注信息,之后再将两张图片重新送入网络再次进行学习,以达到强化局部特征学习和丰富训练集背景的效果。 实验在 4个通用细粒度数据集上与弱监督数据增强网络(WS-DAN) 和目前主流先进方法进行了比较,取得了具有竞争力的效果,相比 WS-DAN 分别提升了 0. 5% ( 鸟类) 、0. 4% ( 车型) 、0. 6% ( 狗类)、0. 4% ( 飞机) ,验证了方法的有效性。
Abstract:
Fine-grained image recognition aims to distinguish the finer sub classes belonging to a large category,and has the characteristics of small inter-class gap and large intra-class gap. At the same time,fine-grained data sets tend to have more types and less data,which is easy to cause over fitting during the training process. To solve the above problems,we propose a fine - grained image classification network combined with attention CutMix,which uses attention mechanism to guide the improved CutMix data - augmentation. Firstly,ResNet50 is used as the backbone to extract image features,and then multiple 1*1 convolution kernels are used to obtain attention maps.Then,bilinear attention pooling operation is used to fuse the feature map and attention into a feature matrix. Finally,the improved CutMixis performed by using the attention map. The improved attention-CutMix is to exchange the attention peak regions of two images,and exchange the annotation information of the two images at the same time,and then send the two images back to the network for learning again,so as to achieve the effect of strengthening local feature learning and enriching the training set background. Experiments on four general fine - grained data sets are carried out with the weak supervised data enhancement network ( WS-DAN ) and the current mainstream advanced methods. Compared with WS-DAN,the proposed method improves by 0. 5% ( cub200-2011) ,0. 4% ( Stanford cars) ,0. 6% ( Stanford dogs) ,and 0. 4% ( FGVC aircraft) ,respectively,which verified the effectiveness of the proposed method.

相似文献/References:

[1]肖威 程文青 许炜.基于Web的工作流细粒度授权框架[J].计算机技术与发展,2006,(12):240.
　XIAO Wei,CHENG Wen-qing,XU Wei.A Web-Based Workflow Authorization Architecture with Fine Granularity[J].,2006,(10):240.
[2]郝小龙. 改进的RBAC模型在电网视频监控平台中的应用[J].计算机技术与发展,2014,24(12):212.
　HAO Xiao-long. Application of Improved RBAC Model in Grid Video Monitoring Platform[J].,2014,24(10):212.
[3]崔凤焦.表情识别算法研究进展与性能比较[J].计算机技术与发展,2018,28(02):145.[doi:10．3969/j．issn．1673－629X．2018．02．031]
　CUI Feng-jiao.Ｒesearch and Performance Comparison of Facial Expression Ｒecognition Algorithm[J].,2018,28(10):145.[doi:10．3969/j．issn．1673－629X．2018．02．031]
[4]张丹丹,李雷. 基于PCANet-RF的人脸检测系统[J].计算机技术与发展,2016,26(02):31.
　ZHANG Dan-dan,LI Lei. Face Detection System Based on PCANet-RF[J].,2016,26(10):31.
[5]陈强锐,谢世朋.基于深度学习的肺部肿瘤检测方法[J].计算机技术与发展,2018,28(04):201.[doi:10.3969/ j. issn.1673-629X.2018.04.043]
　CHEN Qiang-rui,XIE Shi-peng.Lung Cancer Detection Method Based on Deep Learning[J].,2018,28(10):201.[doi:10.3969/ j. issn.1673-629X.2018.04.043]
[6]郭子琰,舒心,刘常燕,等.基于ReLU 函数的卷积神经网络的花卉识别算法[J].计算机技术与发展,2018,28(05):154.[doi:10．3969/j．issn．1673－629X．2018．05．035]
　GUO Ziyan,SHU Xin,LIU Changyan,et al.A Recognition Algorithm of Flower Based on Convolution Neural Network with ReLU Function[J].,2018,28(10):154.[doi:10．3969/j．issn．1673－629X．2018．05．035]
[7]缪宇杰,吴智钧,宫 婧.基于3D 卷积的视频错帧筛选方法[J].计算机技术与发展,2018,28(05):179.[doi:10.3969/ j. issn.1673-629X.2018.05.040]
　MIAO Yu-jie,WU Zhi-jun,GONG Jing.A Wrong Temporal-order Frames Identification Method Based on 3D Convolution[J].,2018,28(10):179.[doi:10.3969/ j. issn.1673-629X.2018.05.040]
[8]吴玉枝,吴志红,熊运余.基于卷积神经网络的小样本车辆检测与识别[J].计算机技术与发展,2018,28(06):1.[doi:10.3969/ j. issn.1673-629X.2018.06.001]
　WU Yu-zhi,WU Zhi-hong,XIONG Yun-yu.Vehicle Detection and Recognition of a Few Samples Based on Convolutional Neural Network[J].,2018,28(10):1.[doi:10.3969/ j. issn.1673-629X.2018.06.001]
[9]李相桥,李晨,田丽华,等.卷积神经网络并行训练的优化研究[J].计算机技术与发展,2018,28(08):12.[doi:10.3969/ j. issn.1673-629X.2018.08.003]
　LI Xiang-qiao,LI Chen,TIAN Li-hua,et al.Research on Optimization of Parallel Training for Convolution Neural Network[J].,2018,28(10):12.[doi:10.3969/ j. issn.1673-629X.2018.08.003]
[10]邓宗平,赵启军,陈虎. 基于深度学习的人脸姿态分类方法[J].计算机技术与发展,2016,26(07):11.
　DEND Zong-ping,ZHAO Qi-jun,CHEN Hu. Face Pose Classification Method Based on Deep Learning[J].,2016,26(10):11.

更新日期/Last Update: 2021-10-10