Szczegóły publikacji
Opis bibliograficzny
Deep neural networks for separation of overlapping voltammetric signals / Filip CIEPIELA, Szymon WÓJCIK, Małgorzata JAKUBOWSKA // Expert Systems with Applications ; ISSN 0957-4174. — 2025 — vol. 276 art. no. 126985, s. 1-12. — Bibliogr. s. 11-12, Abstr. — Publikacja dostępna online od: 2025-03-04
Autorzy (3)
Słowa kluczowe
Dane bibliometryczne
| ID BaDAP | 158863 |
|---|---|
| Data dodania do BaDAP | 2025-04-29 |
| Tekst źródłowy | URL |
| DOI | 10.1016/j.eswa.2025.126985 |
| Rok publikacji | 2025 |
| Typ publikacji | artykuł w czasopiśmie |
| Otwarty dostęp |  |
| Czasopismo/seria | Expert Systems with Applications |
Abstract
The paper proposes a novel approach to the separation of overlapping peaks employing a deep learning strategy, which correctly reproduces the heights and positions of the components. The models were trained using data obtained from the simulation and based on curves registered in voltammetric experiments. The research was carried out using a loss function and model architecture dedicated to this issue. In various variants where the distance between the two overlapping peaks ranged from 14 to 120 points (in voltammetry, i.e., 28 to 240 mV) and the ratio of peak heights ranged from 1:1 to 1:10, the agreement between the actual signal parameters and the prediction result was at the level of R 0.9995 for the experimental test set. The peak positions were perfectly reproduced throughout the experiments, and the RMSE for the peak heights was less than 0.18 (i.e., 1.8% of the maximum peak amplitude in the dataset). In an analytical experiment designed to verify the model, the positions of the caffeic and ferulic acid peaks were shifted by 16 points. The model successfully reproduced these shifted peaks, and the calibration line parameters remained consistent with those obtained for individual analytes. Moreover, the calibration correlation coefficient for both analytes was greater than 0.999, for both the mixture and pure solutions. In this way, it was shown that the model fulfills the task of separating overlapping peaks in typical chemical analysis of multicomponent systems. The approach can be further developed to handle different shapes and more overlapping signals.
