Study On Online Learning Expression Recognition Of Pupils Based On Feature Fusion

The rise of a new generation of information technology has promoted the rapid development of online learning.However,due to lack of emotion and other reasons,the online learning participation rate is very low.In order to eliminate the negative impact of lack of emotion on online learning,it is necessary for the online learning platform to be able to perceive the learner's emotional state of learning in time.Facial expression recognition is one of the important means of real-time perception of online learning emotional state.At present,there are a series of problems in online learning expression recognition,such as imperfect expression classification system,lack of open data sets,and poor robustness of learning expression recognition algorithms,etc.In response to the above problems,this thesis has conducted in-depth research on online learning expression recognition,and the main work is as follows:First,construct an online learning expression database for primary school students.The six common emotional states of learning:understanding,listening,curiosity,doubt,exhaustion,and distraction are determined by analyzing student expressions in online learning situations.Based on above,an online learning expression database was constructed,which contains a total of 1447 images from 115 students.Second,it proposes to use transfer learning-based methods to extract learning facial expression features.In this thesis,the network weights of the VGG16 convolutional neural network model trained on the large image data set ImageNet are used as the initial weights.On this basis,the learning expression images collected in this thesis are used for transfer learning,and the network weights of the VGG16 convolutional neural network model are updated.Finally,the VGG16 convolutional neural network model obtained by transfer learning is used to extract deep features of learning expression images.Third,online learning expression recognition based on feature fusion.In this thesis,CNN depth features,HOG texture features and SIFT features are fused,and SVM is the classifier that realizes the classification of learning expressions.This thesis explores the effect of different features on the recognition effect of learning expressions through a large number of experiments,and verifies the effectiveness of this method.This thesis focuses on the expression recognition of primary school students in the online learning environment,and proposes to apply feature fusion to online learning expression recognition to realize the perception of students' online learning emotion.The research can provide technical support for improving the level of online learning intervention.