Interference fringe pattern direct defect detection analysis based on deep learning

DHSPI is a full-field quantitative structural diagnosis technique. The quantifiable results are based on a stepwise investigation procedure implementing phase-shifting to calculate the wrapped phase. The process it is susceptible to external disturbances encountered in investigation campaigns for Cultural Heritage Monuments usually performed on scaffolds and noisy conditions. It is being proposed for application in structural diagnosis via DHSPI a single neural network architecture that is efficiently to combine denoise and phase prediction. The method not implies a conventional U-Net but an asymmetric network architecture with a simplified decoder, allowing to use state-of-the-art Convolutional Neural Network (CNN) classifiers as the encoder without significantly increasing the total number of learnable parameters. A dataset consisting of simulated and real interferograms data taken from real case studies of CH is used to train and validate the results presented in this research work.