Sparse Infrared Spectroscopy for Detection of Volatile Organic Compounds

TL;DR

This paper introduces Sparse Infrared Spectroscopy (SIRS), a data-driven method that reduces hardware complexity in VOC detection by identifying minimal spectral filters, achieving high sensitivity with fewer samples.

Contribution

The paper presents a novel, task-specific spectral collection approach using non-negative matrix factorization to optimize infrared spectroscopy hardware for VOC detection.

Findings

Detects VOCs at 1-10 PPM levels

Uses 20-50 spectral samples instead of 1000+

Robust to noise and mixture variations

Abstract

To reduce the complexity of infrared spectroscopy hardware while maintaining performance, a data informed, task-specific, spectral collection approach termed Sparse Infrared Spectroscopy (SIRS) is developed. Using a numerically based virtual experiment based on a quantitatively accurate infrared database, non-negative matrix factorization is used to identify the spectral pass bands of a minimal number of filters necessary to identify volatile organic compounds (VOC) within either an inert background or mixture of gases. The data-driven approach is found capable of identifying contaminants at the 1-10 part per million level (PPM) with $\mathrm{\sim~20-50}$ spectral samples as opposed to the more than 1,000 typical of a traditional infrared spectrum. Reasonably robust to both noise and the characteristics of the base compound in a mixture, the task-specific spectral sampling points to simplified hardware design that maintains performance.