Application of Improved YOLOv8 in Classroom Attention Analysis for Agricultural Education: A Pathway to Professional Identity Cultivation

Real-time monitoring of classroom attention is of great significance for improving the quality of engineering education. However, existing studies mostly focus on general classroom scenarios, and few have transformed detection results into quantified indicators with pedagogical meaning. To address these deficiencies, an improved YOLOv8 model is proposed for agricultural engineering classroom scenarios, which integrates a Swin Transformer backbone, a Bi-Level Routing Attention (BAR) mechanism, a Shape IoU loss function, and a lightweight Neck-DSC structure. On the self-constructed Agricultural Engineering Classroom Attention Dataset (AECAD), the improved model achieves an mAP@0.5 of 95.30%, with only 3.3 M parameters and an inference speed of 127.9 FPS, meeting the requirements for real-time deployment on edge devices. Through a quasi-experimental design spanning one semester with 106 agricultural engineering students, the study verifies the significant enhancement effect of attention monitoring feedback on classroom engagement and professional identity (Cohen's d = 0.68, p < 0.01), and establishes a quantitative analysis framework between the Attention Index and professional identity, providing empirical evidence for data-driven instructional intervention in engineering education.