A Facial Expression Recognition Model Based on ResNet50

With the advancement of technology, computers have acquired the ability to recognize human facial expressions from images. This technology can play a crucial role in fields such as psychotherapy, autonomous driving, and public safety. The purpose of this paper is to enhance the accuracy of the computer’s classification of facial expressions by applying a neural network to speed up expression-determinant with a high accuracy. Firstly, the project creates a base model uses ResNet50 to pre-train the data on the input 48*48-pixel images. Then, new convolutional blocks with max pooling layer are added to enhance the feature extraction capability. Finally, a global average pooling layer and a fully connected layer with dropout were appended to solve the problem of overfitting. This model performs a classification with 7 different kinds of emotions. This paper employed a Convolution Neural Network (CNN) model and Convolution Neural Network model with residual blocks. And analysis these two models by comparing loss and accuracy. Since the accuracy of both models performs not well, this paper improved the CNN model by appending a ResNet50 network. Eventually, the model based on ResNet50 accomplished the highest efficiency and accuracy which performed best among all the models.