Research on the Innovation and Development of Ideological and Political Education Based on Artificial Intelligence Technology

In education digital transformation, classroom behavior analysis is core of educational evaluation innovation; traditional teaching lack comprehensive learning feedback, deep learning provide new solutions. This paper address YOLO's insufficient accuracy, efficiency and real-time performance in complex classroom scenarios to support accurate monitoring system construction. Due to dense/small targets and occlusion causing low accuracy, accuracy-efficiency imbalance, and real-time shortage leading to missed/false judgments, this paper propose three algorithms: MRD-YOLO (high precision, MFFN module) improves mAP50/mAP50-95 by 3.2%/6.1% on POCO and 3.2%/4.8% on SCB-Dataset3 vs YOLOv8n; DAM-YOLO (lightweight, DGCST module) balances accuracy-efficiency, with slightly lower accuracy than MRD-YOLO but 1.1M fewer parameters, 2.3B less FLOPs and 2.2MB smaller size; FME-YOLO (lightweight/real-time, FEAM module) optimizes accuracy and lightweight vs the two, achieving 178.3/177.6fps; Web first loading <3s, single frame <1s, meeting real-time requirements.