Classroom Teaching Behavior Analysis System Based on Multi-Feature Behavior Attitude Recognition Algorithm

To address the limitations of traditional and single-modality methods in classroom teaching analysis, this study proposes a Classroom Teaching Behavior Analysis (CTBA) system based on a multi-feature behavior posture recognition algorithm. The system initially employs object detection and pose estimation to locate key body points of teachers and students. A multi-stream neural network then integrates skeleton spatiotemporal features (via a spatial–temporal graph convolutional network) with RGB appearance features (via a dilated 3D convolutional network). An attention-based mechanism adaptively fuses these feature streams before classification. Experiments on a self-built dataset demonstrate that the system achieves over 93.5% accuracy and F1-score, with a single inference time of 62.2 ms. It effectively quantifies behavioral pattern differences across classrooms, offering a robust, data-driven support tool for teaching evaluation and reflection, with significant theoretical and practical value.