FPN-MSTCN模型在学生专注度评价中的应用

The application of FPN-MSTCN model in student concentration evaluation

为了提高智慧教育场景下的学生专注度评价准确率,针对小样本类别难以识别的问题,提出一种FPN-MSTCN模型进行专注度评价,该模型通过改进的FPN网络对单帧人脸进行多尺度的特征提取,解决了在图像中人脸特征无法完整提取的问题。然后,通过融合了SimGNN模块的MSTCN网络对图像序列进行分类,并通过SimGNN模块解决了图像标签与视频标签不一致的问题。采用DAiSEE和EmotiW数据集进行实验。由于DAiSEE和EmotiW数据集的分布严重不均衡,使用代价敏感损失函数作为该模型的损失函数,解决了过拟合问题,测试集准确率分别提高了3.8%和3.1%。

In order to improve the evaluation accuracy of students’concentration under the scenario of intelligent education,as well as the difficulty of the identification towards the small sample categories.This paper proposes a FPN-MSTCN model to evaluate the concentration.The model uses improved FPN network to extract multi-scale features from a single frame of face,which solves the problem that face features cannot be completely extracted from images.Then,the image sequence is classified by MSTCN network integrated with SimGNN module,and the problem of inconsistency between image label and video label is solved by SimGNN module.Finally,DAiSEE and EmotiW data sets are used for experiments.Since the distribution of DAiSEE and EmotiW data sets is seriously unbalanced,the cost sensitive loss function is used as the loss function of the model to solve the over-fitting problem,and the test set accuracy is improved by 3.8%and 3.1%,respectively.