Advancing AI-driven spectral material detection under varying illumination conditions

Imaging and optical sensing technologies are rapidly becoming indispensable tools across diverse indoor and outdoor applications – ranging from industrial material sorting and quality control to precision agriculture (crop variation monitoring) to infrastructure inspection (corrosion or defect detection), threat detection in defense applications (UXOs, IEDs, adversarial drones) and autonomous systems (object identification and road surface monitoring). Among these, spectral image data offers significantly richer physical information than conventional RGB vision, enabling fine-grained material detection and surface characterization.

A fundamental bottleneck in deployment is interpreting measured spectral data under varying illumination conditions. In real-world environments (whether subject to shifting indoor ambient lighting or changing outdoor solar angles, cloud cover, and scene geometries) captured spectral signatures fluctuate significantly. Furthermore, wide dynamic ranges in light intensity require adaptive sensing strategies (e.g., auto-exposure, multi-exposure interleaving) that complicate analysis.

The objective of this PhD is to design AI-driven methodologies for robust, interpretable material detection across indoor and outdoor scenes under varying light conditions. The researcher will develop innovative AI models that integrate environmental context (such as dynamic lighting and background spectra) and quantify uncertainty ranges to improve detection confidence and reliability. This includes exploring hybrid approaches combining physics-informed representations with calibrated spectrometers.

The candidate will work directly with state-of-the-art IMEC spectral sensor technologies, gain hands-on experience with modern vision systems, and build expertise at the intersection of AI, computer vision, and physical optics.