Hyperpixels: pixel filter arrays of multivariate optical elements for optimized spectral imaging

Résumé fourni par la source

Spectral imaging systems are critical for revealing new information about the structure and composition of diverse samples, but traditional approaches that use generic 'bandpass' spectral filters are sub-optimal in challenging scenarios with low data signal-to-noise ratio (SNR).To address this, we introduce the concept of hyperpixels: a novel, compact, application-specific spectral filter array approach compatible with integration atop CMOS image sensors.Each element of a hyperpixel filter array is engineered to selectively transmit specific spectral components optimized for the target application, analogous to arrays of multivariate optical elements designed for analyte sensing.Hyperpixels achieve spectral tailoring through precise height engineering of multiple sub-pixel Fabry-Perot resonators covering each pixel area.Building on our earlier work exploring the feasibility of this approach [1], we present the first experimental validation of this concept.First, we develop a design framework based on matched filter theory and demonstrate its effectiveness by creating a set of 4 hyperpixels optimized to discriminate among 4 distinct spectral reflectance targets.Fabricated 2×2 arrays of hyperpixels are compared to optimal bandpass filter arrays through both spectral and imaging characterization.Our results demonstrate that hyperpixels outperform optimal bandpass filters in separating spectral components, achieving a 2.4× improvement in unmixing matrix condition number (𝑝=0.031) based on measured spectra, and a 3.47× reduction (𝑝=0.020) in condition number during imaging experiments.Importantly, simulations confirm that this advantage is robust even in the presence of significant fabrication errors.These findings demonstrate the superior spectral discrimination capability of hyperpixels over traditional approaches.With straightforward customization, scalable fabrication, and compatibility with CMOS sensors, hyperpixels offer a highly versatile solution for real-time imaging.Potential use cases include micro-endoscopy, capsule endoscopy, industrial inspection, and machine vision, where compact form factors and ability to handle low raw SNR are critical.Future improvements in design and fabrication will further enhance performance, enabling new possibilities for application-specific spectral imaging.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.