YOLOv3在教室内人数识别中的应用

针对教室内人数的自动识别场景,研究了深度学习目标检测架构YOLOv3的识别模型训练及应用。通过分析YOLOv3神经网络的全卷积架构,标注训练数据的坐标与类型,调整单一目标识别网络设置,迭代优化神经网络参数,训练出损失收敛的神经网络识别模型。经数据集验证并统计结果,分析了模型识别的精确率、召回率和准确率。实验结果表明,在100张教室截图验证测试中,其人数识别的平均准确率达到93.58%,可为学校教学管理提供相关参考数据。

Aiming at the scene of automatic recognition of the number of people in the classroom,the recognition model training and application of the deep learning object detection architecture YOLOv3 are studied.By analyzing the full convolution architecture of YOLOv3 neural network,labeling the coordinates and types of training data,adjusting the single target recognition network settings,iteratively optimizing the neural network parameters,the neural network recognition model with loss convergence is trained.The precision rate,recall rate and accuracy rate of the model recognition are analyzed by data set verification and result statistics.The experimental results show that the average accuracy of the number recognition is 93.58%in 100 classroom screenshots verification test,which can provide relevant reference data for school teaching management.