Laboratory Behavior Detection Method Based on Improved Yolov5 Model

• Published in: 2021 International Conference on Cyber-Physical Social Intelligence (ICCSI) pp. 1 - 6
• Main Authors: Zhang, Zhaofeng, Ao, Daiqin, Zhou, Luoyu, Yuan, Xiaolong, Luo, Mingzhang
• Format: Conference Proceeding
• Language: English
• Published: IEEE 18.12.2021
• Subjects: Adaptation models; Big Data; Computer vision; Deep learning; Education; laboratory behavior detection; object detection; Remote laboratories; Surveillance; yolov5
• DOI: 10.1109/ICCSI53130.2021.9736251

With the development of deep learning and big data, behavior detection has become a hot spot in computer vision. Laboratory is an important place for teaching or scientific research. As the subject of the laboratory, laboratory behavior of students determines the quality of experimental teaching. Therefore, this paper took the laboratory as the research scene and proposed a laboratory behavior detection method based on deep learning. Firstly, the common categories of laboratory behaviors were defined and a dataset of laboratory behaviors was established. Then, YOLOv5 model was improved and a laboratory behavior detection method was proposed based on the improved YOLOv5. Lastly, the proposed method was trained and tested based on the laboratory behavior dataset. The experimental results have shown that the improved YOLOv5 model can be well applied to laboratory behavior detection of students. Compared with the original YOLOv5 model, the improved model can better adapt to the data characteristics of the laboratory behavior. Its precision and recall are significantly improved, and mAP (mean average precision) is increased by 2.1%. The proposed laboratory behavior detection method can not only be used to analyze laboratory behavior of students and optimize the experimental teaching. Moreover, it can be extended to remote laboratory surveillance and improve the quality of remote laboratory.