Article Preview
TopIntroduction
Classroom observation is a more common way to evaluate teaching and learning in higher education (Byeon et al., 2021). Detecting and analyzing students’ different behaviors in the classroom through digital technology can not only remind students to adjust their own behaviors, but also reflect the degree of classroom activities and help teachers improve teaching methods (Chonggao, 2021). The students’ behavior and posture in class reflect not only the students’ participation in learning, but also the teachers’ teaching level to some extent. Various interactive analysis systems have emerged in the existing classroom behavior analysis research, such as Student-Teacher classroom teaching analysis and analysis methods based on information technology (Chen et al., 2020). All these methods are based on manual observation to record students’ learning behavior (Cheng, Wei et al., 2022). However, this method wastes more time and energy and is inefficient. Teachers cannot always pay attention to every student (Cheng, Ma et al., 2022). With the development of artificial intelligence technology and machine learning algorithm, new research methods have emerged in student behavior analysis.
In recent years, some intelligent video recording and broadcasting systems have combined computer vision technology, which can reduce manual intervention and greatly improve the recognition effect and efficiency. The intelligent video recording and broadcasting system does not need manual participation in the recording process of the whole classroom teaching activities. It only needs to install one or more cameras at the fixed position of the classroom in advance and simulate the operation of manual lens switching through computational vision technology, so that the camera can automatically record classroom teaching, which greatly liberates manpower and ensures the quality of shooting. It is a major breakthrough in the development of video recording and broadcasting.
Based on this background, in this paper the author investigates student behavior recognition in college classrooms based on improved deep learning algorithms. The paper is divided into four parts. The first part is a brief introduction about student behavior analysis in college classrooms and the arrangement of this study. The second part introduces domestic and foreign algorithms about behavior analysis and deep learning algorithms and summarizes the shortcomings of current research. The third part provides a framework for student behavior recognition in college classrooms. then, it proposes an adaptive normalization algorithm based on deep learning filters distracting information and addresses the shortcomings of deep learning algorithms. Migrating behavioral features are added to the model to meet the demand for online real-time updates, while empirical memory prior knowledge is screened to learn new knowledge. The fourth part offers a simulation and analysis of the student behavior analysis model of college classroom based on the improved deep learning algorithm the author constructed in this paper; subsequently, it provides an evaluation of the algorithm performance by the accuracy of behavior analysis. The experimental results show that the proposed algorithm has better classification recognition accuracy, compared with the existing behavior analysis algorithms.
The innovation of this paper lies in the improved learning algorithm. Considering that the identification of the same type of student behavior in the classroom environment requires the extraction of higher-order spatio-temporal features, the author proposes the normalization method based on deep learning, and filters the noise information through the fusion method. The author establishes an adaptive learning model, which incorporates a domain judge and an attention learning model to achieve migration behavior analysis. Then, the researcher introduces the continuous learning mechanism of empirical memory to enable the analysis of new data for action recognition online.