Research On Teaching Quality Evaluation Strategy Based On Multi-dimensional Pattern Recognition

Teaching is one of the important means to spread and spread knowledge,and classroom quality evaluation is an important means to ensure the quality of teachers’ teaching and cultivate students’ talents.There are some problems and deficiencies in the traditional classroom teaching quality evaluation.With the continuous development of Internet,the coming of Internet plus education era will open the integration of new network technology and traditional education.Recently,researchers at home and abroad have focused more on educational practice and model innovation,and lack of standards and related research on classroom quality evaluation.Moreover,the traditional form of classroom quality evaluation is generally based on the subjective judgment of students and teacher leaders.The traditional classroom quality evaluation has the following problems: first,the evaluation standard is relatively single and lacks certain flexibility and scientificity;Second,the evaluation index has a certain one sidedness;Third,the evaluation is too quantitative.In view of the above problems,this thesis proposes to use the methods of pattern recognition and machine learning in artificial intelligence to objectively and scientifically evaluate the classroom quality by combining the traditional quality evaluation with the emerging science and technology.In this thesis,in the real classroom environment,by setting up the camera and rewriting the camera SDK for autonomous control and movement,so as to obtain the real and available face,posture,attendance and other information in classroom teaching.Two feature fusion methods are proposed.Based on the feature level and the model level,a single-dimensional feature experiment and a multi-dimensional feature fusion experiment are designed to compare,Verify the specific feasibility and efficiency of this method.In the one-dimensional feature experiment,this thesis mainly studies the influencing factors and weights of classroom quality in different dimensions.For facial feature extraction,Gabor wavelet transform is used to extract texture information and improved LBPH algorithm are used as feature extraction methods.Circular LBP Operator is used to maintain the rotation invariance of the image on the original basis,And the perception of texture is more flexible and more robust to illumination.For posture recognition,Alphapose framework is used to obtain the bone posture actions of classroom students.Select the specific bone points to connect and calculate the distance and other parameters to construct the action feature vector.In the multi-dimensional feature fusion experiment,it mainly studies the effect of the fusion methods of feature level and model level on classroom quality evaluation.Firstly,the data pre-processing operations such as standardization and multi-dimensional feature matching will be carried out for the features under different dimensions,and then the weight will be obtained according to the correlation with the classroom quality evaluation,the weight and addition will be carried out,the feature level fusion training will be carried out to obtain the model,and the stacking method will be used to train and summarize the models under different dimensions to obtain the model level fusion.Finally,the feasibility and effectiveness of the multi-dimensional feature fusion method are verified by comparing with the original single-dimensional feature.