Research On The Algorithm Of Class Heading Rate Based On Dual-branch Network

It is very necessary to carry out effective basic data statistics and analysis on students’ classroom learning status.As an important indicator to measure the quality of students’ classroom learning,"class head-up rate" can provide real-time feedback on students’ learning status in the classroom;it can provide data support for teachers or schools to improve management and improve teaching quality.At the same time,with the construction of informatization in various schools and the development of artificial intelligence,a large amount of video data recorded by camera equipment in classrooms has laid the foundation for deep learning research.Aiming at the problems that students’ face images are too small in scale,change in size,have occlusions,and are easily negatively affected by the environment in classroom scenarios,resulting in complex data features and inability to form a reasonable and effective mathematical model,this paper uses deep learning technology to develop Research on the algorithm of classroom student head-up rate based on dual-branch network.The main research contents of the class head-up rate algorithm research based on dualbranch network are divided into the following three parts:(1)Research the face detection model based on improved DSDF.In order to improve the performance and generalization ability of the model,the public data set WIDER FACE has been expanded in a targeted manner,and the data is derived from real classroom surveillance videos.Secondly,in view of the problems such as large change in size and too small size of students’ face images in classroom surveillance videos,a dual-branch multiscale feature extraction network is obtained by using the feature enhancement module,which further improves the ability to obtain face feature information of different sizes.,the students’ face images in the classroom scene also occlude each other,so the model adopts the Anchor based on the context information of the face,so that the model can obtain the feature information around the face from the upper-level feature map with a larger receptive field as a judgment.Whether it is the basis of the face.The final generated face detection model improves the accuracy of face detection in unconstrained environments such as partial occlusion and small size,and is 1.1 higher than the original model on three subsets of different difficulties in the expanded WIDER FACE dataset.3.2%,3.6% and 2.4%.(2)A dual-branch network-based head-up gesture recognition network model is proposed.In order to identify the student’s head pose through a single two-dimensional RGB face image,the head pose feature information is extracted in three dimensions by using a compact two-branch network,which better expresses the two-dimensional face image in the three-dimensional head pose space.mapping relationship.Students’ face images in classroom surveillance videos are affected by unfavorable factors such as multiple angles,expressions,lighting,background and distance,resulting in different color,shape,texture and other information of the image data presented.Therefore,the two sub-networks of the dual-branch network adopt different feature extraction strategies to ensure the model’s ability to perceive different feature information;at the same time,in order to enhance the robustness of the model,the feature evaluation function is improved to suppress other redundant information.,to obtain more required face pose information.The final generated head pose recognition classification model has a recognition accuracy of 97.1% in the test set of the CAS-PEAL-R1 dataset with multi-angle,expression,illumination,background,distance and other feature changes in face images,and the number of frames processed per second is97.1%.reached 51.8.(3)Design and implement the WEB classroom head-up rate analysis system.The system integrates the algorithm research results of the face detection model and the head gesture recognition model in this paper.In order to reduce the influence of the false detection rate of the algorithm on the analysis of the class head-up rate,a class head-up rate method is proposed based on the face detection algorithm in this paper to obtain the face position information in the image.The statistical error of the number of students in the classroom is less than 9%.Finally,the system experiment was carried out and the feasibility of the data analysis of the student’s head-up rate in the system sample was evaluated.The experimental results showed that the algorithm research results of this paper basically achieved the expected research goals.