Review of unsupervised machine learning methods in active infrared thermography for defect detection and analysis

The presence of internal defects poses a serious challenge to the structural integrity and performance of composite materials such as polymers and cultural heritage. Therefore the application of non-destructive testing (NDT) techniques is essential. Active infrared thermography (AIRT) is becoming increasingly attractive among many NDT techniques due to its low–cost and wide–area coverage advantages. However, thermograms often involve non–uniform backgrounds and measurement noise caused by uneven heating and environmental reflections, necessitating post-processing procedures. Among the research topics in this area, unsupervised machine learning methods have shown promising success in AIRT for defect detection. This paper aims to provide a recent overview of unsupervised machine learning-aided thermography for defect detection. Six perspectives on the role of machine learning in thermogram sequence processing are presented: image denoising, non-uniform background removal, image super-resolution enhancement, feature extraction, image segmentation, and depth prediction. In particular, deep learning methods for thermographic data analysis are reviewed and emphasised. A step-by-step review of treatment options offers a guide for inexperienced readers and investigators entering the field. Additionally, the development of machine learning-based thermography methods for different scenarios is summarised from an application perspective. Finally, an outlook on the prospects and potential of these methods is provided.