Liquid position prediction study of Enamel reaction kettle based on modified YOLOV7-Tiny

• Main Authors: Gao, Hong-Yu, Li, Kun, Wang, Zhi-Qiang, Li, Meng-Nan
• Format: Conference Proceeding
• Language: English
• Subjects: Accuracy; CA attention; Enamel reactor; involution Operator; Liquid position prediction; Liquids; Performance evaluation; Prediction algorithms; Predictive models; Production; Shape; two scales
• ISSN: 1934-1768
• DOI: 10.23919/CCC63176.2024.10662254

For a chemical company, the ultrasonic detection method cannot meet the industrial accuracy requirements in the internal liquid level detection of the enamel reactor.By identifying the internal agitation equipment of the enamel reactor, the height of the agitation equipment, so that the internal liquid level volume of the reactor is obtained indirectly. The method is non-contact detection and can reduce the influence of bubble attachment and gas interference on the accuracy of reaction kettle liquid level detection. The algorithm was improved on YOLOV7-Tiny, proposing the YOLO-ICT algorithm, EIoU is used instead of CIoU in the bounding box loss function, the involution Operator is introduced on Backbone, CA attention module is added on Neck, and the prediction is reduced from three scales to two scales prediction on Head. In the prediction box, the liquid level height is mapped to obtain the liquid level volume. According to the ablation experiment, the mAP value is 4.3 percentage points higher than the traditional YOLOv7-tiny algorithm, the parameter quantity is reduced to two-thirds of the original, and the computation is reduced to three-fifths of the original. The experimental results achieve a good balance between detection accuracy and model complexity, which is conducive for deployment in the production process.