HARM-YOLO11n: A Lightweight Real-Time Object Detection Framework for Robust Classroom Behavior Monitoring

Area of Science:

• Computer Science
• Artificial Intelligence
• Educational Technology

Background:

• Real-time classroom behavior monitoring is crucial for intelligent education systems.
• Challenges include dense student layouts, occlusions, and small behavior targets.
• Existing methods struggle with efficiency and accuracy in complex classroom environments.

Purpose of the Study:

• To propose a lightweight object detection framework for effective classroom behavior analysis.
• To enhance feature representation, multi-scale interaction, and optimization for improved detection accuracy.
• To provide a practical solution for behavior detection in resource-constrained educational settings.

Main Methods:

• Introduced Hierarchical Adaptive Re-parameterized Multi-scale YOLO (HARM-YOLO).
• Integrated Structural Re-parameterized Feature Enhancement (SRFE) for improved feature extraction.
• Incorporated Spatial-Channel Adaptive Module (SCAM), Position-Adaptive Multi-scale Fusion (PAMSF), and IoU-Adaptive Soft Weight (IASW).

Main Results:

• HARM-YOLO demonstrated superior detection performance compared to existing lightweight detectors.
• Achieved real-time inference efficiency crucial for practical applications.
• Showcased stable performance across diverse datasets (POCO, SCB-Dataset3, STBD-08).

Conclusions:

• HARM-YOLO offers a practical and efficient approach for real-time classroom behavior detection.
• The framework effectively addresses challenges like occlusions and dense student layouts.
• Provides a viable solution for intelligent education systems in resource-limited environments.