智慧教室课堂教学中学生抬头率检测与分析

Detection and analysis of students'head-up rate in smart classroom teaching

智慧教室建设和使用是教育信息化的重要工作。智慧教室信息使用是其中的难点。文章尝试将深度学习方法应用于智慧教室课堂教学视频数据处理,选择录制清晰的教室全景视频,采用目标检测方式对上课的学生进行抬头率检测。通过融合BiFPN模块改进YOLOv5目标检测算法使其更适应教学场景,使用迁移学习从预训练权重开始训练,能够快速获取最佳模型进行预测。实验结果表明,模型的平均准确率为95.8%,比YOLOv5s模型提高了2.1%,在测试集上进行人工核验时准确率达到了98.42%。研究结果表明,在此提出的抬头率检测模型具有较强的检测能力和泛化能力,可为学校进行教学评价和学生选课提供客观依据。

Construction and use of smart classrooms are important for education informatization.The efficient use of information produced in the smart classroom is the difficult problem.This paper attempts to apply deep learning methods to processing teaching video.The object detection method is used to detect the head-up rate of students in class with a clear panoramic video in the classroom.The YOLOv5 object detection algorithm is improved by fusing BiFPN module to make it more suitable for teaching scene,and use transfer learning method to train model from pre-training weights to make it get the best model faster.The experimental results show that the average accuracy of this model is 95.8%,which is 2.1%better than YOLOv5s model,and the accuracy reaches 98.42%when the manual verification is performed on the test set.The research results show that the head-up rate detection model proposed in this paper has strong detection and generalization capabilities,which can provide an objective basis for school teaching evaluation and student course selection.