Hybrid artificial intelligence − enhanced single-shot transport of intensity equation: Algorithms and applications for real-time opto-mechanical characterization of polymeric fibers

Abstract

This study presents a novel approach to optical phase demodulation by combining the Transport of Intensity Equation (TIE) technique with deep learning, enabling real-time characterization of polymeric fibers under mechanical stress. Traditional TIE methods, while effective, require multiple defocused images, limiting their application in dynamic systems. We developed a convolutional neural network architecture that performs phase demodulation using only single focused intensity images, trained on a comprehensive dataset of 672 image sets captured at various wavelengths (550–602 nm). The network achieved remarkable accuracy with a final validation RMS error of 0.0428, demonstrating 99.91 % error reduction during training. The method’s efficacy was validated through in-situ opto-mechanical characterization of polypropylene (PP) fibers under varying draw ratios. Real-time measurements revealed critical insights into the fiber’s structural evolution, including the refractive index and birefringence. Also, this study introduces an innovative AI-enhanced single-shot TIE method integrated with filtered back projection (FBP) algorithm for real-time 3D morphological analysis of PP fibers. The proposed technique enables unprecedented temporal resolution in studying dynamic material behavior, overcoming key limitations of conventional multi-image TIE methods.