IMRMB-Net: A lightweight student behavior recognition model for complex classroom scenarios.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Educational Technology

Background:

• Classroom behavior analysis is crucial for improving education quality.
• Existing methods struggle with occlusion, small objects, and environmental interference, leading to low accuracy.
• There is a need for accurate and computationally efficient student behavior recognition models.

Purpose of the Study:

• To propose a lightweight student behavior recognition model, IMRMB-Net, to address challenges in accuracy and performance.
• To improve the recognition of occluded and small objects in classroom settings.
• To enhance the overall robustness and efficiency of student behavior analysis.

Main Methods:

• Developed a lightweight feature extraction module, Inverted Residual Mobile Block (IMRMB).
• Implemented DySample in the neck network to improve small object recognition.
• Designed a novel Focaler-ShapeIoU loss function to enhance model robustness and occlusion handling.

Main Results:

• IMRMB-Net achieved high accuracy (mAP@50=93.3%, mAP@50:95=78.7%) and lightweight performance (FPS=60.37, Params=7.32MB).
• Effectively addressed occlusion problems in classroom scenarios on UK_Dataset and SCB_Dataset.
• Demonstrated strong generalization and small target recognition capabilities on the VisDrone2021 dataset.

Conclusions:

• IMRMB-Net offers a promising solution for accurate and efficient student behavior recognition in educational settings.
• The model's design effectively tackles key challenges like occlusion and small object detection.
• This research contributes to advancing educational technology through improved classroom behavior analysis.