基于注意力改进残差网络结构的表情识别方法
FACIAL EXPRESSION RECOGNITION METHOD BASED ON MULTI-CHANNEL RESIDUAL NETWORK
张智 1魏蘅2
作者信息
- 1. 武汉科技大学计算机科学与技术学院 湖北武汉 430065
- 2. 武汉科技大学智能信息处理与实时工业系统湖北省重点实验室 湖北武汉 430065
- 折叠
摘要
针对目前CNN在复杂图像中特征提取不充分的问题,提出一种基于注意力的改进残差网络的表情识别网络.设计一个双流网络在完成粗特征表情识别的同时检测关键点,并使用注意力机制增大关键点周边特征的权重.随后以残差网络为基础模型,改进残差块之间的跳跃连接方式,并将残差块中的普通卷积改进为分组卷积来强化特征提取能力.最后联合两个表情识别网络进行分类,实验结果验证了该模型方案有着更卓越的性能.
Abstract
To solve the problem of insufficient feature extraction of CNN in complex images,an improved residual network based on attention is proposed for facial expression recognition.A dual stream network was designed to detect the key points while completing the coarse feature facial expression recognition,and the attention mechanism was used to increase the weight of the features around the key points.Based on the residual network model,the jump connection between residual blocks was improved,and the ordinary convolution in residual blocks was improved to block convolution to enhance the feature extraction ability.Two facial expression recognition networks were combined for classification.The experimental results show that the model scheme has better performance.
关键词
人脸表情识别/残差网络/注意力机制/分组卷积Key words
Facial expression recognition/Residual network/Attention mechanism/Group convolution引用本文复制引用
基金项目
国家自然科学基金项目(61673304)
国家社会科学基金重大计划项目(11&ZD189)
出版年
2024
国家科技期刊平台
NSTL
万方数据