Facial Expression Recognition Technology Based On Deep Learning

Facial expression recognition mainly refers to the use of computer to analyze and extract features of the detected facial expression information and recognize facial expression information,so that the computer can recognize facial expressions according to the psychological thinking of normal people and corresponding analysis,understanding and synthesis processing to meet people's needs in different application scenarios and establish a more intelligent human-computer interaction environment.Since the 21st century,computers and artificial intelligence,as well as their related equipment and technologies,have developed rapidly,and humans have become more and more aware and demanding of computers.Therefore,people not only eagerly hope that computers can listen,speak,and watch like humans,but also eagerly hope that computers can better understand and express some of their own ideas,and be more intelligent,so as to help people complete their work more quickly and conveniently.In this paper,the main research goal is to further improve the accuracy of facial expression recognition.A series of explorations and researches on facial expression recognition algorithms based on deep learning methods are carried out to provide theoretical basis analysis and corresponding technical basis support for the construction of more effective visual feature services,and built a facial expression recognition system.The main research work of this article is as follows:1.This paper proposes an improved SSD facial expression recognition algorithm based on network reconstruction.On the basis of the original network model SSD target detection algorithm,in order to improve the ability of SSD target detection algorithm to extract the features of facial expression,the SSD network detection model is reconstructed,the convolutional layer with better feature extraction effect is retained.The feature visualization analysis technology of convolution neural network is used to analyze the feature visualization of the convolution layer of SSD network model.For a given image input,it shows the feature visualization map of each convolution layer in the network model,and verifies the effectiveness of the facial expression recognition algorithm by analyzing the results through experiments.2.This paper further proposes an improved SSD facial expression recognition algorithm based on the Central Loss function.For the Softmax Loss in the loss function of the SSD target detection algorithm,it can only restrict the feature classification extracted from different expression categories,but does not specifically restrict the features extracted from the same expression category.The design in Chapter 3 of this article is based on the improved SSD facial expression recognition algorithm,Center Loss function is added to constrain the classification of similar facial expression features,thereby improving the distribution of facial expression features,making it easier to aggregate facial expression features of the same category,and separate facial expression features of the different category,so as to improve the accuracy of facial expression discrimination,and through experiments and comparative analysis to effectively verify the effectiveness of facial expression recognition algorithm.3.This paper implements a facial expression recognition system The system is based on the model generated by the facial expression recognition algorithm training proposed in this paper to classify facial expressions.We use PyCharm and Qt designer to design a facial expression recognition UI interface.The system interface can display facial expression recognition results and detection time-consuming in real time,which can satisfy the real-time recognition of the facial expression.