基于残差网络注意力机制的人脸表情识别

Facial expression recognition based onresidual network attention mechanism

提出一种基于残差网络的人脸表情识别方法。以残差网络为基础,加入裁剪掩码对图像任意区域任意大小遮掩,通过通道注意力机制对重要程度不同的通道分配不同权重,以增加抓取关键信息能力,将多尺度特征与空间注意力机制相结合,以不同感受野提取信息,提高网络提取能力,使用联合损失函数增加类外距离,减小类内距离。将此网络运用到FER2013,CK+数据集中。实验结果表明,识别率分别为64.81%,96.86%,参数量为5.21 M。

In this paper,a method of facial expression recognition based on residual network is proposed.Based on residual network,cutout is added to mask any area of the image,and channel attention mechanism is used to assign different weights to channels with different importance to increase the ability of grasping key information,by combining multi-scale features with spatial attention mechanism,information can be extracted from different receptive fields to improve the ability of network extraction,and the joint loss function is used to increase the out-of-class distance and reduce the in-class distance.Apply this network to FER2013,CK+data set.The experimental results show that the recognition rate was 64.81%,96.86%and the parameter was 5.21 M respectively.