Classroom Student Detection And Behavior Recognition Based On YOLOv5 Deep Learning

Teaching quality evaluation is an important part of the teaching process.The existing teaching evaluation methods are mainly random lectures,and the monitoring of teaching process cannot be carried out normally.At the same time,many classrooms are not used,lighting,air conditioning,fans and other equipment still work normally,resulting in a lot of energy waste.Based on YOLOv5 deep learning theory,this paper carries out a vision-based student number detection and behavior recognition research in the classroom environment,and develops a complete detection system.The specific research contents are as follows:(1)A dataset of deep learning algorithms for classroom student object detection and behavior recognition is established.Considering the influence of factors such as illumination,personnel density,classroom space and classroom seat distribution,a comprehensive data set of 4 aspects and 16 dimensions was constructed,and Labelimg was used to label and classify the data set,laying a foundation for the research of target detection and recognition algorithm.(2)A deep learning algorithm for classroom student object detection and behavior recognition based on YOLOv5 is proposed.The algorithm first detects the student target.On this basis,the Deep-sort multi-target tracking algorithm is integrated to track the student target in the classroom and solve the problem of target loss caused by occlusion in the video data.The experimental results show that: In the self-built data test set,the accuracy rate of student target detection based on YOLOv5 was 97.0% and the accuracy rate of behavior recognition was 75.7%.Compared with Faster R-CNN,SSD and YOLOv3,the accuracy rate of student target detection was improved by 24.35%,13.54% and 11.99% respectively.The accuracy of comprehensive behavior recognition improved by 7.91%,34.13% and 32.21%,respectively.(3)An improved YOLOv5 algorithm combining CA attention mechanism,Ghostnet lightweight neural network and Bi FPN is proposed.The algorithm optimizes the YOLOv5 model.Firstly,CA attention mechanism is added in Backbone layer to make the model obtain more interesting target areas and reduce the interference of classroom background factors.Secondly,Ghostnet lightweight neural network is integrated in the Neck part,which reduces the calculation amount and improves the detection speed of the model through the operation of separable volume integration stage.Finally,the PANet layer is changed to Bi FPN in the Neck part,and the recognition effect of the model on small targets is improved through efficient bidirectional cross-scale connection and weighted feature map fusion,and the missing detection of the model is effectively solved.Experimental results show that the target detection accuracy of the algorithm is 97.9%,and the behavior recognition accuracy is 80.4%.Compared with the unimproved YOLOv5 model and YOLOv7 model,the accuracy of student object detection was increased by 0.9% and 5.1%,and the accuracy of comprehensive behavior recognition was increased by 4.7% and 21.1%,respectively.(4)Developed a Python based student object detection and behavior recognition system in the classroom,and realized the classroom student object detection and behavior recognition functions of static classroom pictures,dynamic classroom video and real-time video stream.Experiments show that the improved YOLOv5 algorithm proposed in the paper has high detection accuracy and detection efficiency,can effectively realize the number of students detection and behavior recognition,and provides an effective means for teaching process supervision and teaching resource management,which has very important theoretical and practical value.