AUTHOR=Lu Mimi , Li Dai , Xu Feng 

TITLE=Recognition of students’ abnormal behaviors in English learning and analysis of psychological stress based on deep learning

JOURNAL=Frontiers in Psychology

VOLUME=Volume 13 - 2022

YEAR=2022

URL=https://www.frontiersin.org/journals/psychology/articles/10.3389/fpsyg.2022.1025304

DOI=10.3389/fpsyg.2022.1025304

ISSN=1664-1078

ABSTRACT=In order to grasp the changes of students' psychological characteristics timely and prevent students from abnormal behaviors and dangerous actions, a deep learning model for automatic recognition of students' behaviors in English class is proposed. Students' English classroom videos are collected as the research data. Combined with pedagogical theories and observation of the rules of classroom videos, the categories of students' behaviors are defined. There are six categories of English classroom behaviors mainly investigated, including listening, writing, sitting at the desk, raising hands, standing and looking left and right. The data are preprocessed. The preprocessing operation mainly includes dividing the collected English classroom video into frames to form picture data, obtaining the position information of the students in the picture data through the target detection algorithm, cutting out the irrelevant background according to the position information and retaining only the target area, obtaining image data of students' English classroom behavior, and using several common data augmentation methods to augment the training set. In the research, ResNet50 is selected as the basic classification network. First, pre-training is performed on the public data set. After the network learns the relevant features, it is retrained by parameter transfer to obtain the final model for students' behaviors classification. For the design and implementation of visualization system, based on the trained model, a visual display system of behavior recognition results is quickly constructed by mainly using Python's Tkinter library. The positive and negative behaviors of students identified by the machine are not much different from the real data. In order to verify the recognition effect of the deep learning model in the real English classroom environment, the statistical results of the 100 recognition result maps are compared with the results of manual marking, and finally the average recognition accuracy of the 100 recognition result maps is taken, and the result is 87.33%. It can be concluded that the behavior recognition model has an accuracy of 87.33% for the students' behaviors recognition in the real English classroom environment.