An End-to-End Infrared and Visual Fusion Detection Framework for Near-Surface Defect in Moving Mode

Traditional infrared thermal testing (IRT) is susceptible to surface conditions such as variation of surface emissivity, which may lead to false detection and has high requirements for the flatness and cleanliness of the sample. Traditional visual testing (VT) could capture surface texture information with high resolution but lacks the ability to detect sub-surface defects. This paper proposes a multispectral fusion detection framework in moving mode combines the advantages of IRT and VT for real-time detection and classification of both surface defects and sub-surface defects. The designed multispectral fusion detection system can adjust the trajectory and speed for different materials and shapes of specimens and collect time-synchronized infrared and visible image pairs for further processing. The proposed algorithmic framework enables real-time end-to-end fusion detection by fast coarse registration and late fusion. Comparative experiments have been conducted on four standard and natural samples of different materials and shapes to validate the reliability and efficiency of the proposed framework.