Recognition And Application Of Head-down And Head-up Behavior In Classroom Based On Deep Learning

In recent years,deep learning algorithms based on convolutional neural networks have achieved great success in many fields,such as text processing,speech recognition,image,video analysis,computer vision and so on.Especially in the field of image recognition,image classification and detection technology based on deep learning has begun to play an important role in specific and professional research scenarios.For example,in the field of botany,relevant researchers use regional selection and feature learning to classify flowers and detect vehicles and pedestrians in various fields,such as automatic security,intelligent transportation and automatic driving.The use of deep learning technology to recognize human behavior in video is one of the research hot spots in the field of computer vision.In recent years,it has received extensive attention from academic and engineering.Human behavior recognition technology has broad application prospects and potential economic value in intelligent monitoring,human-computer interaction,virtual reality and content-based video retrieval and interpretation.Economic value.At the same time,in the classroom scene,identifying students' classroom behavior has important reference significance for evaluating students' learning status,learning quality and teachers' teaching effect.In this paper,deep learning technology is applied to college classroom scenes.From the perspective of students' head down and head up,students are tested and their behavior status is identified.The main work of this paper includes the following aspects:Firstly,this paper sorts out the domestic and international related technologies of deep learning and related theories of behavior recognition,combining with the characteristics of specific scenarios in university classroom and the advantages and disadvantages of various algorithms,eventually propose to apply deep learning technology in behavior recognition of university classroom scenarios.By introducing the related theories of deep learning in detail,it lays a foundation for the application of deep learning technology in specific scenarios.Secondly,from the perspective of classroom video,we can identify the behavior of all students in the classroom and get the rate of each class.The specific method is to select the Faster R-CNN algorithm with the best detection effect,and compare it with the algorithm after modifying the size of Anchor according to the characteristics of the data set.After extracting the video frames,the head-down and head-up status of the classroom is recognized.The experimental results show that the Faster R-CNN algorithm with modified parameters converges faster and can accurately detect students and recognize their head-down and head-up states.On the basis of this study,we can get the change of the rate of rise in each class and the average rate of rise in a period of time.Thirdly,from the perspective of each student in the classroom,identify each student's specific behavior of bowing his head and raising his head.Specific method is:first,define the categories of each student's behavior state and make relevant data sets.Then,select Yolo-v3 algorithm with ideal detection effect and recognition speed to extract the sequence of students,and use ResNet algorithm to classify the extracted students' behavior state.The results show that,in the university classroom scene,most of the students' behavior sequences can be extracted and identified in the video,so that the changes of the students' behavior state can be obtained.Finally,on the basis of the previous research,we combine the whole class head-down behavior recognition with each student's behavior status recognition,calculate the time needed for each algorithm to recognize,and complete the class head-down behavior status recognition and application system based on deep learning through interface visualization,in the end,summarize the relevant methods and problems.