Research of Facial Expression Recognition Based on Deep Learning

This paper proposes a convolutional neural network for facial expression recognition (FER) based on deep learning, named FERNet. FERNet contains 4 residual depth-wise separable convolution modules., each of which includes 3 depthwise separable convolution layers and 1 standard convolution layer. It is a fully convolutional neural network that replaces the fully connected layer with global average pool (GAP) layer. The results show that the average accuracy of FERNet in the KDEF dataset is 93.7%, and the average accuracy of the RAF dataset is 71.9%. Compared with other networks and methods, FERNet has a better performance in facial expression recognition.