The classroom atmosphere is an important factor affecting classroom learning experience, participation and teaching effectiveness. Accurate recognition of classroom atmosphere is vital for optimizing classroom teaching and promoting cognitive development. However, current classroom atmosphere recognition suffers from limitations such as one-sided representational dimensions and single data modality, which makes the existing research results difficult to directly serve classroom teaching practice. Therefore, a bimodal classroom atmosphere recognition method based on the classroom assessment scoring system was proposed in this paper: firstly, label according to classroom atmosphere indicators; then, use temporal-aware bi-directional multi-scale network integrated with multi-head self-attention and the video sliding window transformer model to conduct single-modal classroom atmosphere recognition based on audio data and based on video data, respectively; finally, adopt the random forest algorithm for dual-modal classroom atmosphere recognition to determine the classroom atmosphere level. Through a series of comparative experiments, this paper found out that the performance of classroom atmosphere recognition by integrating bimodal data was superior to that of single-modal data, and the disciplinary characteristic was an important influencing factor for classroom atmosphere recognition, and concluded that a multi-dimensional structure can more precisely represent classroom atmosphere, and bimodal data support can more accurately predict classroom atmosphere. The research in this paper can provide effective technical support for precise teaching diagnosis, and hold significant value for optimizing classroom teaching practice.

Cognitive engagement is an important factor influencing the quality of classroom learning, with significant value in teaching interventions, learning monitoring, and classroom assessment. However, cognitive engagement itself presents the characteristics such as complexity, implicitness, and dynamism, making precise assessment of cognitive engagement in real complex classroom situations extremely challenging. Moreover, traditional assessment techniques are often subjective, while the accuracy and reliability of automated assessments are difficult to guarantee. Therefore, from the perspective of complex dynamic systems theory and based on the Interactive Constructive Active Passive (ICAP) framework, Pekrun model, and Bloom taxonomy, this paper established a three-dimensional explicit representation index system for cognitive engagement, designed a dual-mode cognitive engagement recognition method, and formed a hierarchical model for classroom cognitive engagement assessment. At the same time, this model was applied to the analyis of real classrooms. Results verified the effectiveness of this model and showed that cognitive engagement had multidimensionality, dynamism and individual differences. Through research, this paper was expected to provide theoretical guidance and technical support for perceiving the cognitive engagement status in classrooms, understanding the essence of cognitive engagement, and revealing the evolutionary patterns of cognitive engagement.