Multi-Scale Attention-based Wireless Network Algorithm for Enhancing Language Learning Outcomes

Identifying students' behaviors in language learning classrooms can serve as a criterion for evaluating the efficacy of instructional methods. This research introduces an algorithm for detecting language learning classroom behavior with an enhanced object detection framework (i.e., YOLOv5) through wireless networks. The feature pyramidal framework in the cervical network of the initial YOLOv5 system is integrated with a balanced bidirectional feature pyramidal network. Their subsequent processing involves feature fusion across several object sizes to extract fine-grained characteristics of distinct actions. A spatial and channel Multi-Scale Attention Method (MSAM) is incorporated between the neck and prediction networks to enhance the model's concentration on object data, improving the detection precision. The initial minimal suppression is enhanced by employing the distance-based intersecting ratio to augment the differentiation of occluded items. Several studies on the newly established database encompassed four behaviors: listening, gazingdown, lying down, and rising. The findings indicated that the algorithm introduced in this work can precisely identify diverse student actions, exhibiting greater accuracy than the YOLOv5 framework. Upon evaluating the impact of student behavior recognition across several scenarios, the enhanced system demonstrated a mean precision of 89.2% and a recall of 91.3%, surpassing the performance of the systems under comparison.

Hali tarjima qilinmagan