![]() ![]() Therefore, detection accuracy largely varies depending on the skillfulness of the researcher, and the overall processing cost including human resources is very high. ![]() Most importantly, conventional methods require the optimization of coefficients or thresholds by trial-and-error, and they are not universally applicable to various types of bubbly flow. While addressing this issue, our group established a reliable framework to detect bubbles in different types of bubbly flows with volume void fractions as high as 2% by rigorously synthesizing digital image processing algorithms 5, 8 however, the limitation of this approach still exists. Even in a single image of a bubbly flow, bubble images have different characteristics that cannot be readily distinguished using a single process (criteria). These methods have proved to be useful, but the application of a simple image processing filter is insufficient to process all images of bubbles with various geometrical features, because of the wide scatter of the flow conditions and optical settings of each study. Numerous image processing techniques have been proposed as tools for effective bubble detection, such as the Hough transform 10, 11, breakpoint method 12, 13, and Watershed transform 9, 14, 15. While detecting bubbles using optical visualization, the major obstacle is to identify and track individual bubbles (and statistics including the size and velocity) from the overlapped bubble cluster. In dealing with a gas–liquid two-phase (bubbly) flow, in particular, it is critical to measure the spatiotemporal variation of the interfacial shape accurately for the purpose of analyzing the transport phenomena between phases 5, 6, 7, 8, 9. This is also true while studying multiphase flows, where the simultaneous measurement of individual phases over a large region of interest, without disturbing the flow, is advantageous (compared to intrusive measurement methods) in understanding the interaction between phases. Measurement techniques based on optical visualization are ubiquitous approaches that are now being adopted in the experimental investigations of diverse problems from biological (small scale) to industrial (large scale) phenomena 1, 2, 3, 4. The present bubble detection and mask extraction tool is available online ( ). The pure processing speed for mask extraction is more than twice as fast as conventional approaches, even without counting the time required for tedious threshold parameter tuning. Even while testing with bubble-swarm flows not included in the training set, the model detects more than 95% of the bubbles, which is equivalent or superior to conventional image processing methods. Validation with different bubbly flows yields promising results, with AP 50 reaching 98%. The range of detectable bubble size (particularly smaller bubbles) could be extended using a customized weighted loss function. The training dataset was rigorously optimized to improve the model performance and delay overfitting with a finite amount of data. Motivated by the remarkable improvements in deep learning-based image processing, we trained the Mask R-CNN to develop an automated bubble detection and mask extraction tool that works universally in gas–liquid two-phase flows. However, they require trial-and-error optimization of thresholding parameters, which are not universal for all experimental conditions thus, their accuracy is highly dependent on human experience, and the overall processing cost is high. For this, numerous image processing techniques have been proposed, showing good performance. While investigating multiphase flows experimentally, the spatiotemporal variation in the interfacial shape between different phases must be measured to analyze the transport phenomena. ![]()
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |