<![CDATA[Gas-liquid two-phase is a situation where the gas phase of a liquid coexists together. The presence of gas that forms a region in the liquid environment causes the formation of bubble flow. The parameters of the bubble flow carry important information about the behavior and characteristics of the bubble. This research was carried out by detecting the size and area of the bubble using YOLOv8-based image processing by comparing the model's performance to produce improvements in inference time, increase accuracy, and reduce computational load. Bubble images were collected by adding 0.4 mm copper wire as a comparison to convert mm to pixels; then, the images were labeled and trained with various YOLOv8 models. Confusion matrix, precision and recall are used as comparative evaluation materials for the YOLOv8 model to obtain good model performance. In this study, the AUC of the Precision and Recall curve closest to the value 1 is the YOLOv8m model of 0.990. The comparison results of the matrix evaluation with the best model are the YOLOv8m model with mAP of 99.00% and F1-score of 96.86%. Microbubble measurements are calculated from the output of the YOLOv8 model by converting pixel units to mm. The model used in bubble measurements is the model with the best evaluation results and the model that gets the smallest radius value by considering measurement uncertainty, namely YOLOv8m with a minimum radius of 0.66 ± 0.04 mm..]]>
Copyrights © 2024
