Research On Facial Expression Detection Algorithm Based On Deep Learning

Facial expression detection is a very important topic in computer vision.It is widely used in human-computer interaction,psychological medicine,health care and intelligent driving.Facial expression has been studied for more than 50 years.The traditional facial expression detection algorithm relies on manual feature extraction,which has high cost and poor robustness.With the development of science and technology and the diversification of application fields,traditional methods are difficult to meet the actual needs.In view of the shortcomings and limitations of traditional methods,combined with deep learning and image processing technology,the research on facial expression detection is carried out.In order to improve the accuracy of facial expression detection based on the common target detection algorithm,the facial expression detection system is designed based on the detection algorithm.The main research contents are as follows:The theory and structure of deep learning are introduced in detail,such as neural network,convolution kernel,pooling operation and activation function,etc.This paper introduces the application of deep learning in facial expression detection,and summarizes the commonly used facial expression data sets.Combined with deep learning technology,aiming at the problem of facial expression detection in natural environment,a facial expression detection algorithm based on Faster Region-CNN(Faster RCNN)is proposed.The basic principle of Faster RCNN detection network is analyzed in three parts: feature extraction module,Region ProposalNetwork(RPN)module and detection module,and the shortcomings of the network are improved.DenseNet is used to replace the original feature extraction module,and the original candidate frame Non Maximum Suppression(NMS)are replaced by Soft-NMS.Aiming at the problem that the existing expression data sets are not suitable for network training,facial expression data sets in natural environment are independently collected and made,and the relevant image processing technology is used to expand and gray the data set.Through experimental comparison,the detection accuracy is improved by 5%,which proves the effectiveness of the method.In order to solve the real-time problem of facial expression detection,a facial expression detection algorithm based on YOLOv3 is proposed.The detection principle of YOLOv3 is analyzed,and K-means clustering is used to analyze the data set,which makes the network more suitable for facial expression detection.By adding Spatial Pyramid Pooling(SPP)module,more feature information can be obtained by fusing feature maps of different sizes,so as to further improve the accuracy of YOLOv3 network.Experimental results show that the detection accuracy of the improved YOLOv3 is improved by 3%compared with the original version.By comparing the experimental results of Faster RCNN and YOLOv3,the advantages and application scenarios of the two kinds of detection networks are analyzed.By analyzing the actual needs of facial expression detection,a facial expression detection system is implemented.Design system interface to convenient operation,after analyzing the requirements of two facial expression detection methods based on image or video,the trained Faster RCNN and YOLOv3 network models are embedded to detect the expression of images and videos respectively.Analyze the modules of the system and add the data preprocessing function.The system has the characteristics of simple operation and convenient use,and has certain practical value.