Research on Student Behavior Recognition Based on Improved YOLOV8

This study utilizes Labelme software to annotate 1500 student behavior images, constructing a dataset that includes key student behaviors such as Think, Write, Mobile, Pass notes, Glance, and Talk. To enhance the diversity and coverage of the dataset, traditional digital image processing techniques, including image flipping, rotation, brightness adjustment, and the addition of Gaussian noise, were employed to expand the initial dataset, resulting in a small dataset of 2000 images. In terms of model training, this paper embeds the CBAM (Convolutional Block Attention Module) attention mechanism into the YOLOV8 network structure to improve the accuracy of behavior recognition. After 100 rounds of training, the average mAP0.5 (mean Average Precision) for all behavior categories reached 63.2<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">%</sup>, which is an approximately 7% increase in recognition accuracy compared to the original YOLOV8 model.