Corrosion detection from IR thermal images in signed cumulative distribution transform domain

Abstract

This study introduces a novel approach to detect corrosive defects on metal substrates using infrared (IR) thermal images. Among numerous non-destructive techniques, infrared thermography (IRT) is notable for its effectiveness in identifying invisible surface and subsurface corrosion in materials. Existing methods for corrosion detection from IRT images lack the connection to the physical processes governing the emission of heat by defective areas. This paper proposes a transport-based mathematical model to describe the difference in heat-flow characteristics between the non-corroded and corroded material regions. A novel detection technique is then devised, utilizing the signed cumulative distribution transform (SCDT) and a subspace classifier to classify 1D thermal signals derived from IRT image sequences. Experiments demonstrate that the proposed approach is capable of detecting corrosive regions on metal substrates with high accuracy while being data efficient with respect to a number of machine learning-based detection methods.