Bubble segmentation algorithm for gas-liquid two-phase flow based on deep learning

Aiming at the problem of difficult detection and segmentation of overlapping bubbles in gas-liquid two-phase flow，based on a deep learning framework，and on the basis of the YOLACT (you only look at coefficients) modeling algorithm，by introducing efficient channel attention (ECA) and void convolution into the feature extraction network，an ECA-YOLACT bubble detection and segmentation algorithm was proposed to increase the edge extraction ability of overlapping bubbles．To obtain the bubble dataset，based on bubble generation adversarial network (BubGAN)，the dataset was generated and meanwhile the gas-liquid two-phase flow test work was carried out，which was expanded by using data enhancement to complete the training of the network model．To verify the feasibility of the algorithm，modeling algorithm experiments were carried out for different void fractions．Experimental results show that based on the improved YOLACT bubble detection and segmentation algorithm，which is validated on the test set，the accuracy is 89.49%，and the recall is 97.51%，with the average accuracy of 96.80%． © 2023 Huazhong University of Science and Technology. All rights reserved.