Developing a Conditional Variational Autoencoder to Guide Spectral Data Augmentation for Calibration Modeling

To deal with the typically insufficiently labeled samples involved in practical spectroscopy measurements, a conditional variational autoencoder (CVAE) is proposed to guide the spectral data augmentation calibration modeling method for in situ measurement. First, the CVAE is designed to generate the...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:IEEE transactions on instrumentation and measurement Jg. 71; S. 1 - 8
Hauptverfasser: Mu, Guoqing, Chen, Junghui
Format: Journal Article
Sprache:Englisch
Veröffentlicht: New York IEEE 2022
The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
Schlagworte:
Calibration
Calibration model building
conditional variational autoencoder (CVAE)
Data augmentation
Data models
Fermentation
In situ measurement
Numerical models
Regression models
semi-supervised
Semi-supervised learning
Semisupervised learning
Spectra
Spectroscopy
Spectrum analysis
Task analysis
Training
virtual sample
ISSN:0018-9456, 1557-9662
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:To deal with the typically insufficiently labeled samples involved in practical spectroscopy measurements, a conditional variational autoencoder (CVAE) is proposed to guide the spectral data augmentation calibration modeling method for in situ measurement. First, the CVAE is designed to generate the virtual spectra such that the augmentation training set is employed to develop the calibration model. To use the generated unlabeled samples for modeling with online measurement purposes, a semi-supervised ladder network (S2-LN)-based regression learning model is developed. The proposed method incorporates all generated virtual unlabeled samples with real labeled samples. An important advantage of this approach is that it ensures that the generated virtual spectra and the real labeled spectra are the same distribution, which in turn ensures the effectiveness of semi-supervised learning. A numerical simulation example and an experimental example of the glucose fermentation process illustrate the effectiveness of the approach.
Bibliographie:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:0018-9456
1557-9662
DOI:10.1109/TIM.2022.3142060