| Téléchargement | - Voir le manuscrit accepté : Overfitting one-dimensional convolutional neural networks for Raman spectra identification (PDF, 5.1 Mio)
|
|---|
| DOI | Trouver le DOI : https://doi.org/10.1016/j.saa.2022.120961 |
|---|
| Auteur | Rechercher : Hamed Mozaffari, M.; Rechercher : Tay, Li-Lin1 |
|---|
| Affiliation | - Conseil national de recherches du Canada. Centre de recherche en métrologie
|
|---|
| Format | Texte, Article |
|---|
| Sujet | CNN; deep learning; handheld spectrometer; spectral matching; AI in chemometrics |
|---|
| Résumé | Dedicated handheld spectrometers have been adopted by first responders and law enforcement agencies for in situ identification of unknown substances. Real-time spectral matching process is a pixel-by-pixel comparing of the unknown spectra with reference data. In fact, the success rate of this process using a miniaturized portable Raman spectrometer relies mainly on the variety of reference data carried on the memory. This is a hurdle in miniaturizing and affordability of the current handheld spectrometers due to limited memory and computational power. In this study, we aim to mitigate this issue by utilizing the power of one-dimensional Convolutional Neural Networks (1DCNN) trained on millions of Raman spectra augmented from standard available reference databases. Specifically, an intentionally overfitted 1DCNN model can be substituted with the reference database of handheld spectrometers to alleviate the memory size and increase the identification process speed and accuracy. Our experimental results revealed that 1DCNN could identify one pure unknown Raman instance from thousands of classes with a high accuracy. |
|---|
| Date de publication | 2022-01-29 |
|---|
| Maison d’édition | Elsevier |
|---|
| Dans | |
|---|
| Langue | anglais |
|---|
| Publications évaluées par des pairs | Oui |
|---|
| Exporter la notice | Exporter en format RIS |
|---|
| Signaler une correction | Signaler une correction (s'ouvre dans un nouvel onglet) |
|---|
| Identificateur de l’enregistrement | 040c9591-60fb-43d1-8d6c-9590ce9c4cf1 |
|---|
| Enregistrement créé | 2022-02-24 |
|---|
| Enregistrement modifié | 2022-03-22 |
|---|
