Feature Extraction Enhancement Using Near-Infrared Light for Multiview Metallic Surface Image Fusion

Abstract

AbstractProblem statement: Difficult to approach wearable parts are omnipresent: they are found in turbines, engines, pipes and even the human body. Smart hard-to-reach surface inspection systems based on videoendoscopy, imaging synchronization and adaptive backlight are designed. One of the main problems of endoscopic inspection is space confinement that causes field of view shortage. To solve it, multichannel approach and multiview image fusion can be used, providing significant field of view expansion. It is known that metal reflectance increases along with light source wavelength, making surface details more contrasting. Consequently, image fusion, particularly feature detection and extraction, can be enhanced by using light sources of longer wavelength range. Purpose of research: Comparing visible and near-infrared light sources efficiency for multiview metallic surface image fusion as a part of intelligent hard-to-reach surface inspection system. Results: The study involves unrefined steel plate and corrosion-damaged steam turbine blade as test objects. The efficiency of near-infrared 940 nm LEDs is compared to seven light colors composed with RGB LEDs. Results show that near-infrared light source proves to be the most efficient for unrefined metallic surface image fusion. Small quantity of features found using near-infrared light (37% less than average) is compensated by the highest feature significance (108% more than average). However, near-infrared light results for steam turbine blade corrosion-damaged surface are mediocre (8% better than average). Azure light source leads with overall efficiency 57% better than average for corrosion-damaged steam turbine blade surface. Practical significance: results obtained during the study define the best spectral ranges to improve image fusion quality for metallic surface inspection with intelligent multichannel videoendoscopic inspection system.KeywordsVideoendoscopyIntelligent inspection systemHard-to-reach surface examinationFeature matchingImage fusionMultichannel systemPanoramic imagingImage stitching