Design And Implementation Of Classroom State Analysis System Based On Deep Learning

With the rapid development of artificial intelligence and deep learning in recent years,many colleges and universities have put forward the goal of realizing campus digitization,intelligence and educational informatization.In the whole course of teaching and learning,classroom state is an important reference factor to evaluate students' acceptance and quality of teaching.But at present,the classroom state analysis is mainly carried out by manual method,which not only distracts teachers' attention,but also has low and incomplete accuracy.Therefore,it is of great significance to find a method that can improve the efficiency of classroom state analysis.By using the method of deep learning,this paper analyzes the state of the video recorded in class from two aspects: students' class behavior and attendance.The student behavior recognition part is completed by the improved target detection algorithm.Attendance analysis is realized by face recognition method.The main contents of the thesis are as follows:(1)SSD algorithm improvement.Through analyzing the SSD model structure,the improved SSD model is proposed according to the characteristics of large number of basic network parameters and poor detection effect on small targets.replacing the backbone network with the improved Mobilenet network,using the characteristics of the deep separable convolution network to reduce the network parameters and improve the operation speed;The information in the deeper feature map is fused upward into the shallow layer to improve the accuracy of small target recognition.Finally,the RMSProp optimization algorithm is integrated into the network to realize the network model optimization and accelerate the convergence speed of the model.(2)Collection and preprocessing of student behavior data sets.Through the way of network search and actual shooting,2500 images of five behaviors including raising hands,sitting,writing,sleeping and playing with mobile phones are obtained as the target detection data set.In order to ensure the model effect and not increase the difficulty of model training,all images are preprocessed before training.(3)Status analysis and attendance implementation.Frame extraction of classroom video is carried out by using the method of Open CV library,and it is usedas the image data source of student behavior recognition and face recognition.By using the improved behavior recognition model and the MTCNN-Insightface face recognition method to identify the extracted video frames,the number of each of the five behaviors in the frame,as well as the number of students in class and the list of students in class are obtained.(4)Design and implementation of classroom state analysis system in colleges and universities.A Flask vue framework is used to build the system,and the improved behavior recognition method and face recognition method are integrated into the system to realize offline classroom state analysis.The system is divided into teacher end and management end,teacher end is responsible for uploading course record screen,management end is responsible for random analysis of student status and attendance at any time,the combination of the two forms a convenient and comprehensive classroom state analysis platform.