A Light-Weight Convolutional Neural Network for Facial Expression Recognition using Mini-Xception Neural Networks

This paper builds a convolutional neural network model based on Xception; we delete the fully connected layer in the traditional convolutional neural network model and use four depthwise separable convolutions to replace the convolution layer in convolutional neural network; we use batch normalization to process the output data after each convolution operation, and use the ReLU activation function to add nonlinear factors to the output data, and finally use the SoftMax function for final result classification. Our model achieved an accuracy rate of 73% on the FER2013 dataset, which is a particular improvement compared to the original model Xception. We design and implement a facial recognition system that can be used for static images and real-time recognition, which can quickly and accurately recognize authentic facial expressions.