Study On Detection Of Behavioral Abnormalities In Self-study Class

With the universal application of artificial intelligence in various fields of society,"artificial intelligence education "has gradually become a research hot spot.In the current practice of primary and secondary education,students' autonomous learning ability has become a very important teaching goal.In order to fully cultivate students' independent learning ability,all kinds of schools at all levels can independently complete their learning tasks by setting up self-study classes.Because of the poor self-learning consciousness of primary and middle school students,they are vulnerable to many factors,which leads to the unsatisfactory effect of self-learning.Therefore,when primary and secondary school students carry out autonomous learning,they also need to be supervised by various sources.As a part of the school education and teaching management facilities,the main function of video surveillance is only used for campus safety monitoring,and its application in teaching management has not been fully paid attention to and played.In order to effectively improve the control effect of video surveillance on the improvement of students' autonomous learning ability,the superiority of using video surveillance was pointed out by analyzing the internal and external supervision involved in the process of students' autonomous learning in this paper.The behavior anomaly detection method was studied based on deep self-coding learning in this paper.By transferring a ConvLSTM model with good detection effect in the universal scene to the school self-study class,a U-ConvLSTM model for detecting the behavior anomaly of the student self-study class is constructed in combination with the U-Net network.U-ConvLSTM model is divided into two parts:space encoder and time encoder.In the space encoder,convolution is used to extract the spatial features of the input video frame.The time encoder uses ConvLSTM to extract the features on time dimension.In addition,the model adds jump connections in the process of spatial decoding,and fuses some of the low-level features lost in the coding process into upsampling,which increases the accuracy of model detection students.It is proved by experiments that the accuracy of the model in the training set reconstruction video is 78.07%,and the accuracy of detecting abnormal behavior of students is 66.7%.Compared with the reconstruction accuracy of the ConvLSTM model of 72.30%and the accuracy of anomaly detection of 60%,the model does effectively improve the accuracy of detecting students' abnormal behavior.Finally,some suggestions on how to improve the effectiveness of video supervision in education management are put forward.The effective function of video supervision also needs the improvement of technical means,the attention of educational managers and the rational use of teachers.