Research On Facial Expression Recognition Based On Fusion Convolution Neural Network Model

Conventional facial expression recognition methods construct a classifier only by manually selecting expression features.This method not only handles complexity,but also easily loses the feature information of the expression.The deep convolutional neural network can autonomously perform end-to-end learning on a large number of samples,thereby obtaining deeper facial expression feature information.Aiming at the problem that the existing neural network algorithm has low recognition rate in facial expression recognition,a facial expression recognition model of nine-layer convolutional neural network is designed.The deep convolutional neural network model for expression recognition was constructed and simulated by reasonably increasing the number of convolution layers and modifying the convolution kernel size.The results show that compared with the LeNet-5 and VGGNet models,the facial expression recognition rate of the model is increased by 14.93% and 4.83%,respectively.Aiming at the problem of poor robustness of convolutional neural networks due to the influence of illumination intensity and facial pose in practical applications,a facial expression based on global partial binary pattern features and CNN feature fusion is proposed.Firstly,the algorithm uses the AdaBoost algorithm to quickly detect and crop facial images from the image to be recognized.The face image is grayed out,cropped and normalized,CLBP features and convolution features are extracted and feature fusion is performed.The results show that the algorithm improves the robustness and improves the recognition accuracy in the real environment.A facial expression recognition system based on TensorFlow deep learning framework was constructed.The trained model was ported to the NVIDIA Jetson TX2 embedded platform for facial expression recognition experiments.The results show that the method can overcome most of the effects of illumination,attitude and other factors,and realize real-time dynamic recognition of facial expression.