Research On Intelligent Perception And Recognition Method For Quantification Of Teachers' Credibility

Classroom education management system has been developing into a new direction.For the evaluation of teachers,the traditional way is no longer suitable which is based on the results of student examinations.The evaluation for teachers' literacy now focuses on comprehensive variables based on big data,machine learning and other new technologies.Most of the existing instruction and learning intelligent management system researches have paid much attention on the analysis of students' learning behaviors and engagement,while there is no intelligent analysis and management system research on teachers'instructional behaviors.A teacher with high credibility has more obvious influence on students,which can promote students' cognitive and emotional learning.At the same time,teachers' credibility is also one of the important components to measure the basic professional skills.Therefore,the quantification of teachers' credibility plays a vital role in the instructional activities through ways of providing teachers and administrators with visualized reflection and evaluation results and promote teachers' comprehensive ability.So far,the evaluation of teachers' credibility is mainly based on the results and analysis of questionnaires.This method consumes a lot of time and efforts with a low efficiency.At the same time,using the Likert seven-point scale to measure the credibility is subjective,without timely feedback,which is not convenient for teachers to adjust instructional behaviors and strategies in time.Previous studies on education management systems generally adopt visual sensor devices to obtain behavioral data.However,in classroom teaching scenarios,teachers' behavior recognition through recorded video also faces its unique technical problems,such as:lighting changes,blurring,complicated background,low resolution of video images,occlusion,and the low facial expression recognition because of frequent head pose changes.In the instruction process analysis,there is a high-performance requirement for classroom teaching behavior recognition,especially for the accuracy and stability in the intelligent teaching environment.The main research goal of this dissertation is to accurately identify teachers' facial expressions and gestures from the video image sequences,and develop an intelligent teacher crebility evaluation management system based on teacher facial expression and gesture recognition.This research aims to support teacher evaluation by providing procedural results in the era of big data.The proposed evaluation system provides objective data,which is the first to carry out the research on teacher facial expression recognition and gesture recognition based on classroom recorded videos,and develops a teacher credibility evaluation system based on teacher behavior perception quantification.The main innovations of this paper mainly include the following four aspects:(1)We propose a method for improving the quality of classroom recorded videos based on the Super Laplace a priori constraint.In this chapter,the image gradient feature statistics of a large number of high-resolution classroom teaching video images is calculated,which shows that the high-resolution video image statistical distribution is in line with the distribution of the super Laplacian prior.It is proposed to use the power function ?·?0.6 to fit the video data.In this paper,the method is verified in the simulation and actual instructional scenarios.The experiments indicate that the proposed method has a better image quality improvement than the existing state-of-the-art,which prepares for facial expression and gesture recognition next using videos image sequences.(2)We propose a teacher facial expression recognition method integrated with image generation,which is an end-to-end learning meathod for facial expression recognition.Combined with 68 facial landmarks,a triangular consistency loss function is proposed to generate the teacher facial images with arbitrary expressions and postures which is used for the training set amplification to improve the expression recognition accuracy rate.The proposed model is tested on multiple standard datasets.At the same time,a depthwise separable convolution network is added to the expression recognition network to replace the traditional 2D convolutional layer.The global average pooling is adopted to replace the fully connected layer with a Softmax for classification task.This method aims to reduce network parameters and improve the running speed.Finally,facial expressions classification tasks are performed on multiple public facial expression data sets.The proposed algorithm achieves an average recognition accuracy of about 80%.(3)A method is proposed based on deep learning to identify teachers'gestures when teaching.A deep neural network model is constructed for automatically identifying teacher gesture categories.We propose to integrate into a SCF algorithm to accurately identify the human skeleton through a bottom-up approach.The proposed method has a good recognition result in the case of occlusion.On this basis,the teacher's gesture is identified by a supervised learning model,thereby identifying the gestures of discourse assisting gesture and attention guiding gesture with the whiteboard detection.If the teacher's hand is located inside the whiteboard,it is recognized as the attention guiding gesture.If outside the frame of the whiteboard,it is a discourse assisting gesture.The model is tested with a recognition accuracy over 83%.(4)We construct a teacher credibility quantification evaluation model.Based on the constructed model,facial expression recognition and gesture recognition are integrated in the intelligent system.The teacher credibility evaluation management system also integrates the teacher's eye gaze module and the use of the cloud platform interaction module to comprehensively evaluate teachers' credibility from the perspective of non-verbal behavior.This research makes a comparative analysis with the results of expert evaluation,student evaluation,and teacher self-evaluation,which proves that the intelligent management system can objectively and effectively quantify and analyze the teacher's credibility in the classroom.