Student Behavior Recognition in Classroom Based on Improved YOLOv7
Huayong Liu, Ming Yue · 2023
Student behavior recognition in the classroom through deep learning is of great significance for smart education. However, there is currently a lack of public available datasets in this area, and the recognition accuracy is still low. A behavior recognition method based on an improved Yolov7 is proposed. The ConvNeXt module is introduced to improve the extraction of image features. In addition, the channel attention module and the spatial attention module are incorporated into the upsampling and downsampling of the feature pyramid network, which obtains more geometric and semantic information. Furthermore, Wise IoU is used to effectively streamline the learning process for objects of different sizes. A dataset of students’ classroom behavior covering nine types is constructed for experiments. The experimental results show that the improved network achieves 80.29% in the accuracy of student behavior recognition, an improvement of 4.78% over the original YOLOv7, and reduces the missed recognition rate.