Professorial Lecture | Next Generation Vision Measurement Systems: by Professor Melanie Po-Leen Ooi

Can AI control light to improve what it sees? Discover how frontend machine learning may lead to a physical AI vision-based measurement.   The journey from "what you see" (the physics of light) to "how you see" (the measurement process) and ultimately "how you decide" (data analysis and decision-making) is fraught with complexities. In vision systems, light is fundamentally an operational measurement whereby its meaning is determined by how the system operates, not just the underlying physics or the machine learning model applied afterward.   Currently, state-of-the-art machine learning algorithms function as passive, "backend" tools that analyse images only after they have been captured. This traditional approach forces a reliance on impossibly large training datasets to account for every real-world variable, creating an insurmountable bottleneck for real-world vision measurement systems in industries like manufacturing, smart agriculture, and environmental monitoring.   This Professorial Lecture provides an overview of the vision-based measurement and data analysis pipeline. It explores how light is controlled and captured, how systems are calibrated, and how the resulting data is reported. Most importantly, it questions whether the next generation of vision measurement systems can push machine learning to the "frontend" of the pipeline. By allowing AI to actively control the physical illumination source in a closed-loop, the system can physically tune its incoming images based on the rules it discovers during modelling. Through this approach, the AI can test its own decisions in real-time, for example, by adjusting light wavelengths to verify the presence of a specific material of defect, resulting in systems that may someday learn to see the unknown, and the unseen. #PublicLecture #ProfessorialLecture