一种基于改进VGG网络的表情识别算法

A facial expression recognition algorithm based on improved VGG network

针对深度学习中的卷积神经网络存在对人脸表情特征提取不充分的问题,文中提出一种改进的VGG16网络模型,以更充分地提取人脸表情特征,从而更好地进行人脸表情识别。首先,在VGG16网络的每个卷积层前加入一个GCT通道注意力,用于增强人脸表情的特征提取;然后,将VGG16网络中相同通道数的卷积层分为一个Block,在每个Block后使用迭代式的特征融合,将浅层网络提取的特征与深层网络提取的特征进行融合,以丰富对人脸表情特征的提取。另外,去掉VGG16网络的3个全连接层,改为一个全连接层直接输出分类结果,不仅可以减少参数量还能够保证识别精度。实验结果表明,改进后的VGG网络在人脸表情数据集RAF-DB和SFEW上的识别率分别达到87.842%和56.881%,较原网络有显著提升。

In allusion to the insufficient extraction of facial expression features in convolutional neural networks in deep learning,an improved VGG16 network model is proposed,which can more fully extract facial expression features,so as to better perform facial expressions recognition. A GCT channel attention is added in front of each convolutional layer of VGG16network to enhance the feature extraction of facial expressions. The convolutional layers with the same number of channels in VGG16 network are divided into a block. After feature extraction of each block,the iterative feature fusion is used to fuse the features extracted from the shallow network with the features extracted from the deep network,so as to enrich the extraction of facial expression features. In addition,three full connections of the VGG16 network are removed and changed to one full connection layer to directly output the classification results,which can not only reduce the amount of parameters but also ensure the recognition accuracy. The experimental results show that the recognition rates of the improved VGG network on facial expression data sets RAF-DB and SFEW are 87.842% and 56.881% respectively,which is significantly higher than the original network.