Stabilization system for solar loading thermography applied on cultural heritage objects exposed outdoors: the contribution of advanced algorithms and dual-branch U-Net

Abstract

AbstractThis study investigates the use of solar loading thermography (SLT) for thermal non-destructive testing (TNDT) and image stabilization of cultural heritage objects, specifically focusing on a century-old ancient book. The irregular contours and deteriorated areas of the book posed significant challenges for feature extraction due to non-uniform temperature variations. To address these challenges, a convolutional neural network (CNN) based dual-branch network of U-Net was used to stabilize the dataset across three degrees of freedom with the ancient book. The stabilization process involved tracking feature lines across each frame of the time-domain datasets, correcting for frame misalignment caused by sample movement during prolonged data acquisition. The effectiveness of this stabilization technique was evaluated by comparing the results of principal component analysis (PCA), fast Fourier transform (FFT), and fast iterative filtering (FIF) algorithms before and after stabilization. Significant improvements were observed, particularly in the clarity and accuracy of defect detection, indicating that this technique provides a robust foundation for further analysis and processing of SLT datasets in cultural heritage preservation. This research demonstrates the potential of combining advanced image processing techniques with SLT to enhance the quality and reliability of NDT in preserving valuable historical artifacts.