Evaluation of Physical Electrical Experiment Operation Process Based on YOLOv5 and ResNeXt Cascade Networks

Automatic scoring of students’ physical experimental operation process plays a positive role in understanding and improving the teaching effectiveness. In order to realize the automatic scoring of experimental operation, a method to infer the experimental operation process by detecting the state changes of the experimental equipment is proposed. In addition, simultaneously detecting all the experimental equipment and accurately identifying different states of each experimental equipment is a vital but challenging task. To solve the two issues mentioned above, YOLOv5 and ResNeXt cascade networks are proposed, which are compared with other models on the self-made data set, and a good matching result is obtained. Based on the cascade networks, the method of continuous state maintenance is added to reduce missed detection. Finally, 8 groups of circuit states are designed as automatic scoring points for the experiment of "Measuring the resistance of small bulb by voltammetry". On 10 videos of students doing experiments and obtaining different evaluation grades, the scoring results of the proposed method are compared with those of six experimental teachers. The results show that although there is a gap between the automatic scoring and the teacher’s scoring due to different evaluation methods, the evaluation grade of automatic scoring is consistent with the teacher’s evaluation grade, which can reflect the students’ experimental operation level. We hope this report can provide useful experience for a reference for automatic scoring in experimental teaching.