Precise PIV Measurement in Low SNR Environments Using a Multi-Task Convolutional Neural Network

Particle Image Velocimetry (PIV) is essential in experimental fluid mechanics, providing nonintrusive flow field measurements. Among the recent advances in PIV algorithms, deep-learning-based optical flow estimation is distinguished by its high spatial and temporal resolution, as well as remarkable efficiency, especially RAFT-PIV, which is based on Recurrent All-Pairs Field Transforms (RAFT). However, RAFT-PIV is extremely susceptible to experimental conditions characterized by low signal-to-noise ratios (SNR), leading to unacceptable errors. This study proposes PIV-RAFT-EN, an enhanced RAFT-based algorithm integrating image denoising, enhancement, and optical flow estimation via a Multi-Task Convolutional Neural Network (MTCNN). Evaluations on synthetic and real-world low-SNR data demonstrate its superior accuracy and efficiency. PIV-RAFT-EN offers a reliable solution for precise PIV measurements in challenging environments, including practical applications like vehicle water entry.