Active wavelength selection for mixture analysis with tunable infrared detectors

Huang, Jin; Gutierrez‐Osuna, Ricardo

doi:10.1016/j.snb.2014.10.094

Cited by 2 publications

(2 citation statements)

References 43 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…For this reason, we designed a mixture construction policy so that the chosen problems would be neither trivial nor unsolvable. Namely, we randomly selected a large number of 50-component 8 mixtures and calculated their classification rate with a set noise level 9 . We then selected five mixtures that could be correctly classified of the times.…”

Section: Results On Synthetic Datamentioning

confidence: 99%

See 1 more Smart Citation

Active wavelength selection for mixture identification with tunable mid-infrared detectors

Huang

Gutierrez‐Osuna

2016

Analytica Chimica Acta

Self Cite

View full text Add to dashboard Cite

This article presents a wavelength selection framework for mixture identification problems. In contrast with multivariate calibration, where the mixture constituents are known and the goal is to estimate their concentration, in mixture identification the goal is to determine which of a large number of chemicals is present. Due to the combinatorial nature of this problem, traditional wavelength selection algorithms are unsuitable because the optimal set of wavelengths is mixture dependent. To address this issue, our framework interleaves wavelength selection with the sensing process, such that each subsequent wavelength is determined on-the-fly based on previous measurements. To avoid early convergence, our approach starts with an exploratory criterion that samples the spectrum broadly, then switches to an exploitative criterion that selects increasingly more relevant wavelengths as the solution approaches the true constituents of the mixture. We compare this "active" wavelength selection algorithm against a state-of-the-art passive algorithm (successive projection algorithm), both experimentally using a tunable spectrometer and in simulation using a large spectral library of chemicals. Our results show that our active method can converge to the true solution more frequently and with fewer measurements than the passive algorithm. The active method also leads to more compact solutions with fewer false positives.

show abstract

Section: Results On Synthetic Datamentioning

confidence: 99%

“…Our work builds on a previous algorithm for active wavelength selection [9] based on multi-modal solvers. In that early work, a multi-modal solver was used to generate multiple candidate spectra that fit the measurements well, and the wavelength with maximum variance across the candidate spectra was chosen as the next measurement.…”

mentioning

confidence: 99%