具有全局特征的空间注意力机制

张连超; 乔瑞萍; 党祺玮; 翟沛源; 孙红帅

doi:10.7652/xjtuxb202011016

您当前的位置：

首页 >

文章列表页 >

具有全局特征的空间注意力机制

更新时间：2025-07-09

- 具有全局特征的空间注意力机制
- Spatial Attention Mechanism with Global Characteristics
- 西安交通大学学报 2020年54卷第11期页码：129-138
- 作者机构：
  
  西安交通大学信息与通信工程学院,西安,710049
- 作者简介：
- 基金信息：
  
  : 2020-01-06。作者简介: 张连超(1993—),男,硕士生;乔瑞萍(通信作者),女,副教授。基金项目: 陕西省重点研发计划资助项目(2020GY-074)
- DOI：10.7652/xjtuxb202011016
  中图分类号： TP391.4
- 网络首发：2020-11-10，
  
  纸质出版：2020
- 稿件说明：
移动端阅览
张连超, 乔瑞萍, 党祺玮, 等. 具有全局特征的空间注意力机制[J]. 西安交通大学学报, 2020,54(11):129-138.

Spatial Attention Mechanism with Global Characteristics[J]. 2020, 54(11): 129-138.
张连超, 乔瑞萍, 党祺玮, 等. 具有全局特征的空间注意力机制[J]. 西安交通大学学报, 2020,54(11):129-138. DOI： 10.7652/xjtuxb202011016.

Spatial Attention Mechanism with Global Characteristics[J]. 2020, 54(11): 129-138. DOI： 10.7652/xjtuxb202011016.

摘要

为了获得卷积神经网络特征图中不同特征点之间的长距离依赖关系

使卷积神经网络更好地区分前景目标和背景信息

提出了一种具有全局特征的空间注意力机制。通过通道融合层将多通道的原始特征图组合成单通道的特征融合图

消除了通道间信息分布对获取空间注意力权重的影响; 将特征融合图经过全局特征获取处理

获得能够反映特征融合图中某特征点与特征融合图中所有点之间相关性的全局特征图; 全局特征图与初始值为0的可学习变量相乘

并且在通道域复制自身

扩展为原始特征图大小

将扩展后的全局特征图与原始特征图对应元素相加

获得具有注意力机制的特征图。在不同卷积神经网络中加入具有全局特征的空间注意力机制进行实验

结果表明:在脑电波二分类任务中

所提注意力机制的分类准确率最高提升了0.839%; 在CIFAR-10数据集多分类任务中

所提注意力机制的分类准确率最高提升了0.484%; 在夜间车辆单类别检测中

在交并比阈值大于0.5的平均精度评判标准下

所提注意力机制最高提升了3.860%

在交并比阈值大于0.75的平均精度评判标准下

所提注意力机制最高提升了11.726%; 在voc2007数据集多类别检测中

在交并比阈值大于0.5的平均精度评判标准下

所提注意力机制最高提升了0.778%

在交并比阈值大于0.75的平均精度评判标准下

所提注意力机制最高提升了1.232%。

Abstract

To obtain the long-distance dependence between different feature points in the feature map of the convolutional neural network

so that the convolutional neural network can better distinguish the foreground target and background information

a spatial attention mechanism with global features is proposed. Combining the multi-channel original feature map into a single-channel feature fusion map through the channel fusion layer

the influence of the information distribution between channels on obtaining spatial attention weights is eliminated; the feature fusion map through global feature acquisition processing to obtain the global feature map

which represents the correlation between a feature point and all points in the feature fusion map; the global feature map is multiplied by the learnable variable with an initial value of 0

and copies itself to the size of the original feature map in channel domain. The expanded global feature map is added to the original feature map to obtain a feature map with attention mechanism. After adding the spatial attention mechanism with global features to different convolutional neural networks

the experimental results show that in the brain wave two-classification task

the algorithm classification accuracy is heightenedby the maximum of 0.839%; in the CIFAR-10 data set multi-classification task

the classification accuracy of the algorithm is improvedby the maximum of 0.484%; in the single-category detection of night vehicles

the algorithm is heightenedby the maximum of 3.860% with the evaluation standard of the average accuracy value of Intersection over union(IoU)greater than 0.5

and the algorithm is improvedby the maximum of 11.726% with the evaluation standard of the average accuracy value of IoU greater than 0.75; in the multi-category detection of the voc2007 data set

the algorithm has improved by the maximum of 0.778% with the evaluation criterion of the average precision value of IoU greater than 0.5

and the algorithm has improvedby the maximum of 1.232% with the evaluation criterion of the average precision value of IoU greater than 0.75.

关键词

Keywords

references

GEHRING J, AULI M, GRANGIER D, et al. A convolutional encoder model for neural machine translation [C]∥Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics: Volume 1 Long Papers. Stroudsburg, PA, USA: Association for Computational Linguistics, 2017: 123-135.

GEHRING J, AULI M, GRANGIER D, et al. Convolutional sequence to sequence learning [C]∥Proceedings of the 34th International Conference on Machine Learning. Princeton, NJ, USA: International Machine Learning Society(IMLS), 2017: 2029-2042.

VASWANI A, SHAZEER N, PARMAR N, et al. Attention is all you need [C]∥Proceedings of the 31st Annual Conference on Neural Information Processing Systems. Vancouver, Canada: NIPS, 2017: 5999-6009.

WANG F, JIANG M Q, QIAN C, et al. Residual attention network for image classification [C]∥Proceedings of the 2017 IEEE Conference on Computer Vision and Pattern Recognition(CVPR). Piscataway, NJ, USA: IEEE, 2017: 6450-6458.

HU H, GU J Y, ZHANG Z, et al. Relation networks for object detection [C]∥Proceedings of the 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway, NJ, USA: IEEE, 2018: 3588-3597.

HUANG Z L, WANG X G, HUANG L C, et al. CCNet: criss-cross attention for semantic segmentation [C]∥Proceedings of the 2019 IEEE/CVF International Conference on Computer Vision and Pattern Recognition. Piscataway, NJ, USA: IEEE, 2019: 603-612.

YUAN Yuhui, WANG Jingdong. OCNet: object context network for scene parsing [EB/OL].(2019-01-22)[2019-12-06].https:∥arxiv.org/abs/1809.00916.

HU J, SHEN L, SUN G. Squeeze-and-excitation networks [C]∥Proceedings of the 2018 IEEE/CVF Con-ference on Computer Vision and Pattern Recognition. Piscataway, NJ, USA: IEEE, 2018: 7132-7141.

CAO Yue, XU Jiarui, LIN S, et al. GCNet: non-local networks meet squeeze-excitation networks and beyond [EB/OL].(2019-04-25)[2019-12-06]. https: ∥arxiv.org/abs/1904.11492.

WANG X L, GIRSHICK R, GUPTA A, et al. Non-local neural networks [C]∥Proceedings of the 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway, NJ, USA: IEEE, 2018: 7794-7803.

WOO S, PARK J, LEE J Y, et al. CBAM: convolutional block attention module [C]∥Proceedings of the 15th European Conference on Computer Vision. Cham, Germany: Springer, 2018: 3-19.

ROY A G, NAVAB N, WACHINGER C. Concurrent spatial and channel ‘squeeze excitation’ in fully convolutional networks [C]∥Proceedings of the 21st International Conference on Medical Image Computing and Computer Assisted Intervention. Cham, Germany: Springer, 2018: 421-429.

BUADES A, COLL B, MOREL J M. A non-local algorithm for image denoising [C]∥Proceedings of the 2005 IEEE Computer Society Conference on Computer Vision and Pattern Recognition(CVPR). Piscataway, NJ, USA: IEEE, 2005: 60-65.

KRIZHEVSKY A, SUTSKEVER I, HINTON G E. ImageNet classification with deep convolutional neural networks [J]. Communications of the ACM, 2017, 60(6): 84-90.

SIMONYAN K, ZISSERMAN. Very deep convolutional networks for large-scale image recognition [EB/OL].(2015-04-10)[2019-12-06]. https: ∥arxiv.org/abs/1409.1556.

REDMON J, FARHADI A. Yolov3: an incremental improvement [EB/OL].(2018-04-08)[2019-12-06]. https: ∥arxiv.org/abs/1804.02767.

REN Shaoqing, HE Kaiming, GIRSHICK R, et al. Faster R-CNN: towards real-time object detection with region proposal networks [C]∥Proceedings of the 29th Annual Conference on Neural Information Processing Systems. Vancouver, Canada: NIPS, 2015: 91-95.

浏览量

下载量

CSCD

文章被引用时，请邮件提醒。

提交

工具集

关联资源

采用感受野优化与渐进特征融合的图像超分辨率算法

采用双树复小波变换的可解释机械故障诊断方法

多尺度特征融合的铁轨异物入侵检测研究

调频连续波雷达的二维截断统计量恒虚警检测方法

多特征融合的图像格贴近度匹配方法