Pseudo-image-feature-based identification benchmark for multi-phase flow regimes

Multiphase flow is a prevalent topic in many disciplines, and regime identification an essential foundation multiphase research. Computer vision deep learning have achieved numerous excellent models, but not demonstrated satisfactory performance fundamental research, including identification. This research proposes advanced pseudo-image feature (PIF) as the descriptor benchmark of multiple classifiers. The PIF simulates image format compactly encodes to pseudo-image, which explicitly displays implicit signals. further evaluates three proposed five existing popular provides baseline for applying fully convolutional network (FCN) classifier state-of-the-art performance, testing verification accuracy respectively reached 99.95% 99.54%. suggests that has capability representation, classifiers achieve superior compared Industries can utilize technology obtain greater production efficiency, productivity, financial gain.