Facial Expression Recognition Based On Deep Learning

Facial expressions are one of the main ways that humans convey their emotions in the communication process,and have important values in interpersonal communication or human-computer interaction.Facial expression recognition mainly uses computer feature extraction and classification of facial expressions in images to realize the task of facial expression recognition.This technology enables computers to understand the mood of human beings,and has a great role in promoting the establishment of a more intelligent human-computer interaction platform.The current is a new era full of information.With the rapid development of science and technology,the performance of computers is rapidly improving,which is of great help to the research of artificial intelligence technology.As an important part of artificial intelligence,computer vision has attracted many researchers to explore and pay great attention.This article mainly focuses on the subject of facial expression recognition.Based on the current application of deep neural network facial expression recognition and the actual problems faced,explore the neural network feature extraction theory and network structure as well as the impact of image processing on the recognition results.The main content includes the following aspects:(1)This article studies three light balance algorithms for the impact of light factors,which are based on the gray-scale world algorithm in white balance,gray-scale histogram equalization and gamma correction algorithms.Subsequently,this paper uses the convolutional neural network VGG16 to verify the performance of the lighting algorithm,and select the best algorithm through the analysis of the experimental results,and use this algorithm as an important part of image preprocessing.(2)Through the study of the classic convolutional neural network structure,this paper proposes a basic neural network model based on the depth of the separable residual module,and verifies the effectiveness of the model through comparative experiments,and further optimizes the structure in the following text.(3)In this paper,the activation heat map(Grad-CAM)of neural network visualization technology is used to explore the feature extraction of facial expressions in the network layer at different depths in the model.The effective feature extraction scheme of the model is analyzed through the heat map,and a convolutional neural network of multi-feature fusion is proposed,and the superiority of the optimized model is verified in the comparison and demonstration under the same data set.(4)This paper designs a facial expression recognition system based on multi-feature fusion convolutional neural network.Experimental tests are carried out on static pictures and dynamic video images,and the application value of the neural network model proposed in this paper in actual scenes is further demonstrated through experiments.