摘要
智慧课堂中学生学习表情的精准识别对于提升课堂效率,掌握学生实时学习状态具有重要的作用。传统单一性的卷积神经网络(CNN)算法对学生学习表情的特征抓取精度不够,尤其是学生微表情状态下的抓取效果更不理想。基于此,本文创新性地提出基于CNN与长短期记忆网络(LSTM)的混合表情特征提取识别算法,充分利用CNN提取学生学习表情中的空域特征并存储,对应的时域特征提取层面充分利用LSTM对学生学习表情视频序列特征进行挖掘抓取,将学生学习表情的时域特征与空域特征进行平均化处理,从而构建一套完整的学生学习表情识别算法,最后基于该算法网络进行深度学习训练。基于本文的实验结果,对应的愉悦、困惑、惊讶、中性和疲倦这五种学习情感状态识别率得到了大幅提升,对应的识别准确率最高可达71.9%,识别性能大幅提升。
The accurate recognition of students' learning expressions in smart classroom plays an important role in improving classroom efficiency and mastering students' real-time learning status.The traditional simplex convolutional neural network(CNN)algorithm is not accurate enough to capture the features of students' learning expressions,especially the capture effect of students' micro expressions.Based on this,this paper creatively proposes a hybrid expression feature extraction algorithm based on CNN and long-term memory network(LSTM),which makes full use of convolutional neural network to extract and store the spatial features of students' learning expressions,and the corresponding time-domain feature extraction layer makes full use of long-term memory network to mine and capture the video sequence features of students' learning expressions.The time-domain features and spatial features of students' learning expressions are averaged to construct a complete set of students' learning expression recognition algorithm.Finally,deep learning training is carried out based on the algorithm network.Based on the experimental results of this paper,the recognition rate of the corresponding five learning emotional states of pleasure,confusion,surprise,neutral and fatigue has been greatly improved,and the corresponding recognition accuracy can reach up to 71.9%,and the recognition performance has been greatly improved.
作者
周江
蔡臻
ZHOU Jiang;CAI Zhen(Guangdong Communication Polytechnic,Guangzhou 510650,Guangdong,China)
出处
《广东交通职业技术学院学报》
2023年第1期48-52,共5页
Journal of Guangdong Communication Polytechnic
基金
2020年广东交通职业技术学院校级科研项目(项目编号:GDCP-ZX-2020-006-N1)。
