Facial Expression Recognition Based on Pruning Optimization Technology

Abstract

Facial expression recognition is based on deep learning, which has become one of the main topics in artificial intelligence. Nevertheless, the complex network structure causes numerous parameter redundancy, which leads to low recognition efficiency. Therefore, researching how to reduce parameter redundancy makes sense. In addition, today's machine expression recognition has some limitations and constraints, while the generalization ability is poor. Aiming at addressing the problems that excessive parameter redundancy brings, for example, low efficiency and massive limitations. This paper will put forward a facial expression recognition technology that is based on pruning optimization to improve existing models. Based on VGGNet, ResNet and SENet, these three convolutional neural networks modify the convolution layer of the network, then add the attention module and pruning optimization. Through face detection, face area feature representation, expression recognition, and other steps. Realize the recognition of multiple facial expressions. The CK+ dataset and JAFFE dataset are independently divided into 4/5 for training and 1/5 for testing. Experimental results show that after inserting CBAM, except ResNet18, the accuracy of the other two networks all improved, SENet18 improved the most significantly. After pruning, the accuracy and speed of VGG16 and ResNet18 improved, but the accuracy of SENet18 decreased. In summary, the optimized model has good applicability and generalization ability.