Thermography-based material classification using machine learning

Infrared thermography has been widely used today for nondestructive evaluation and testing of materials and other qualitative approaches. However, the field of thermography is much less developed. Most of the existing research uses a relatively simple model, while more realistic models are currently in development. One interesting scenario for thermography is determining the material composition of objects based on their thermal response to excitation, which could lead to applications such as multimodal human-computer interaction, teleoperation and non-contact haptic mapping. This paper presents a system that is capable of classification between a range of different materials in real time, using laser excitation step thermography and a set of machine learning classifiers. Experimental results demonstrate a consistently high accuracy in determining the label of the material, even when the dataset is composed of multiple different sessions of data acquisition.