Research Of Expression Recognition Algorithm Based On Deep Learning

With development of computer technology,people's demand for intelligent human-computer interaction is increasingly urgent.In order to satisfy human needs for the future development of computer intelligence,we must learn to understand human feelings firstly.Face expressions have rich emotional information,so facial expression recognition has become an important research topic in the field of artificial intelligence.In recent years,deep learning technology has developed rapidly and achieved remarkable achievements in the field of image recognition.However,there are still many difficulties and challenges in the field of expression recognition.For example,there is a lack of facial expression data,and only small data sets under laboratory conditions exist.Moreover,head posture,scarves,hair and other obstacles have a great influence on the accuracy of expression recognition.In addition,there are inconsistencies in the categories among the expression databases,which leads to a limited scope of application of the learned classification model,and the generalization ability is not high.This paper proposes two different algorithms of expression recognition based on deep learning.The main contributions are as follows:(1)For the problem that the deep model is difficult to train with a small number of labeled samples,this paper proposes a novel algorithm of facial expression recognition by feature transfer learning based on small datasets.The innovation of the algorithm is to make full use of the inherent correlation between the existing rich face data and facial expression data,and then transfer the knowledge learned by the face recognition model to the task of expression recognition.The algorithm adopts the shallow target expression recognition model by using the feature-based and model-based transfer learning methods.(2)For the problem that facial expression recognition results are affected by gestures,obstacles,and inconsistency of expression categories,this paper proposes a recursive neural network expression recognition network based on feature fusion in the time dimension(DTAGR),which can capture dynamic information of multiple features between video sequences.Firstly,the algorithm encodes the expression image and the face key points into the appearance texture feature vector and the geometric feature vector through the convolutional neural network coding structure.And then splice and merge the two feature vectors to enrich the expression features in the time dimension.Finally,the mixed features are sent to the recurrent neural network structure to capture spatial and temporal information between video sequences.The experimental results show that the proposed expression recognition algorithm based on transfer learning has the characteristics of fast convergence and good stability.The algorithm has obvious advantages for the expression samples with clear emotions.And the network DTAGR proposed in this paper can significantly promote the expression recognition task under the influence of attitude and obstacle.