Reactor level detection method based on IYOLOv5s

• Main Authors: Dang, Zheng-Jun, Li, Kun, Li, Meng-Nan, Wang, Zhi-Qiang
• Format: Conference Proceeding
• Language: English
• Subjects: Accuracy; attention mechanism; Attention mechanisms; Data visualization; improved bounding box regression loss function; level detection; Liquids; Location awareness; new detection layer; Production; YOLO; Yolov5s
• ISSN: 1934-1768
• DOI: 10.23919/CCC63176.2024.10661474

Liquid-level detection in industrial reactors is an important part of chemical production. To accurately detect and visualize the liquid level in industrial production reactors, a liquid-level detection method with an improved YOLOv5s model is proposed. Firstly, the liquid-level image dataset is constructed by collecting the reactor liquid-level data in the actual production process at the industrial site, and then preprocessing and enhancing the data. Secondly, the traditional YOLOv5s algorithm has been improved in three aspects: the CBAM attention mechanism has been introduced into the backbone feature extraction network to improve the accuracy of localization of the target; a new detection layer was added to the backbone feature fusion network to extract small features from the target; and the loss function of bounding box regression was improved. CIOU to improve the accuracy of the prediction frame. Experimental results show that the improved YOLOv5s model obtains better performance, and the mean average precision value of level detection reaches 95.67%, which is 7.14% higher than the traditional YOLOv5s model, which improves effectively the accuracy of liquid level recognition and provides a new idea for further research on liquid level detection.