Research On Action Recognition Of Student In Classroom Based On Convolutional Neural Network

Classroom action recognition always has been the emphasis and difficulty of education.In traditional classrooms,teachers judge students' learning status by observing their actions,and then form teaching feedback.However,this method is time-consuming and laborious,and cannot meet the needs of large-scale classroom analysis in today's intelligent learning environment.Therefore,it is of great significance to explore the use of artificial intelligence technology to automatically identify students' classroom actions.This thesis focuses on the recognition of students' classroom action in a complex classroom environment,and proposes the application of deep learning methods with classroom action recognition,achieving the monitoring and analysis of students' daily classroom action.The research has important guiding significance for improving the quality of classroom teaching,helping educational decision-making and teaching management.This study aims at the problems of manual observation,tedious and time-consuming in the recognition of classroom actions,and uses the powerful feature learning ability of convolutional neural networks to explore effective solutions,which aim to solve the automatic recognition of students' classroom actions.The main research work of the thesis is as follows:(1)Building a database of students' classroom actions.There is currently no public database on student classroom actions.Therefore,this thesis collected a large amount of student classroom data and finally selected 3630 images of five types of actions to build a database of student classroom actions.Specific actions include:raising hands,writing,listening,standing and reading.(2)Design student classroom action recognition method basing on deep learning.Due to the complex background in the classroom,the few individual pixels of students,the large number of students,and the serious occlusion among students,it is difficult for students to recognize classroom action.Therefore,this thesis first proposes to train the YOLOV3 model with good detection effect and fast speed of small targets to achieve the detection of individual students.Secondly,the well-trained Xception network is used as a pre-training model on ImageNet,and the trained-obtained model is applied to the recognition of students' classroom actions by using transfer learning.(3)Analysis of student action recognition results.In order to facilitate intuitive analysis and evaluation of student action,this thesis uses the Python library Streamlit to quickly build user interaction tools for student action data analysis,which can achieve visual output of student action detection and recognition results to judge the student's overall learning status.