Performance of deep learning in moiré fringe analysis with different intensities

Moiré deflectometry plays a crucial role in the measurement of flow field, but the stability of deep learning applied to moiré fringe analysis could be affected by many physical factors. In this paper, fringe intensity is considered an influential factor affecting deep learning performance, and it is explored thoroughly. In our experiment, nine groups of moiré fringes with different intensities are obtained, each group with 1100 frames, which means 9900 frames are used for further model training and prediction. Root-mean-square error (RMSE) and structural similarity index measure (SSIM) are used to verify the model performance with different fringe intensities. The results show that the U-net++ and ResUnet models perform best when root-mean-square (RMS) contrast is approximately 25.94, which means the models achieve better performance in the moderate fringe intensity. The relevant conclusions of this paper can improve the performance of deep learning in dynamic flow field fringe analysis and contribute to the advancement of the intelligence of moiré deflectometry.