Within the framework of educational methodology innovation and classroom management efficience optimization, automatic recognition and classification of learner postures have become increasingly essential to assess levels of attention, interaction, and learning states. This study aims to introduce a learner posture classification system based on state-of-art deep learning models. Specifically, YOLOv11-Pose is utilized to extract key points with two geometric features calculated through algorithms to determine angles between student’s body parts. These features take a role as inputs for a MLP artificial neural network with a three-hidden-layer architecture to classify. The system is trained and validated on a real-world dataset including 8,400 images that cover four typical postures (sleeping, hand raising, writing, and looking at the board). The proposed model achieved an accuracy rate of 99.048% after 160 training epochs, indicates that our approach outperforms traditional classifiers, and provide robust solutions for intelligent classroom monitoring and assistance systems.