Combination of thermal fundamentals and Deep Learning for infrastructure inspections from thermographic images. Preliminary results

Abstract

The application of Deep Learning (DL) models using the measurements acquired by Non-Destructive Testing (NTD) tools as input data stands as a versatile solution for highly automated analysis. However, DL models using thermal images as input data are quite scarce when it comes to analysing defects in medium- and large-scale bodies. Therefore, this paper proposes the application of a thermal criterion and a DL model, Mask R-CNN, in thermal images acquired from different infrastructures with thermal bridges and moisture. The thermal criterion is first applied to the input data, showing its utility to improve DL models performance