Performance of a four deep neural network structure in moiré fringe intelligent analysis

Moiré fringe analysis holds significant importance in the assessment of flow fields, while deep learning is advancing rapidly and is frequently employed to handle complex computational challenges. In this paper, the performance of U-net, Unet++, Unet 3+, and ResUnet in moiré fringe analysis is compared and analyzed. The results indicate significant differences among the four neural networks in terms of the training time, prediction speed, prediction accuracy, and generalization capability regarding positional variations. Among the four neural networks, ResUnet exhibits the shortest training times, highest accuracy, and fastest prediction speeds while also exhibiting good generalization capability. Consequently, in practical applications, using ResUnet for moiré fringe analysis is an excellent choice.