Facial Expression Recognition Based On Deep Learning

Since facial expressions are the most intuitive display platform for human emotions,facial expression recognition has become an indispensable link in the development of artificial intelligence,and it is also one of the challenging topics in the field of computer vision research.Traditional facial expression recognition includes three steps:facial image preprocessing,facial feature extraction,and feature classification.With the birth of deep learning,facial expression recognition get a new way of thinking,which simultaneously implements feature extraction and fuzzy classification through deep networks.The main purpose of this article is to build a complete facial expression recognition system.On the premise of focusing on recognition accuracy,pay attention to the system's space and the time cost of calculation,and make a feasible plan for the mobile terminal.The specific work of this paper is as follows:(1)In the preprocessing stage,a simple face alignment and cropping method was designed to remove useless background areas to extract more useful facial features,and a complete expression recognition system constructed by a simple convolutional neural network,To verify the superiority of the pretreatment method.(2)Based on the classic residual network,a shallow convolutional neural network is designed,and its model is pruned to search for the best lightweight expression recognition model,making it more real-time and fast in mobile applications.Different from the pruning steps of training,pruning,and fine-tuning,this article is pruning while training layer by layer.The feature map of the next layer is determined by the feature map of the previous layer and the convolution kernel to be retained in this layer.The feature map is directly related to the pruned front layer,reducing the layer-by-layer accumulation error.Finally,the recognition accuracy rates of 96.15%,99.05%,and 69.24%were obtained on the three data sets of CK+,JAFFE,and FER-2013,respectively.