Barlow Twins for semi-supervised learning in NIR spectroscopy

Mark Wohlers; Andrew Mcglone; Eibe Frank; Geoffrey Holmes

doi:10.1016/j.chemolab.2026.105664

Home
Search
Barlow Twins for semi-supervised learning in NIR spectroscopy

https://doi.org/10.1016/j.chemolab.2026.105664

Barlow Twins for semi-supervised learning in NIR spectroscopy

May 1, 2026
Chemometrics and Intelligent Laboratory Systems
Mark Wohlers +3 more

Listen

Abstract
Literature Map
References
Similar Papers

Abstract

Near-infrared (NIR) spectroscopy is a widely used technology in the horticulture industry for non-destructive fruit grading. Partial Least Squares (PLS) regression is the dominant method for producing fruit quality predictions from measured spectra. Alternative deep learning methods have shown promise, but often require large amounts of labelled data to train. This study proposes a semi-supervised method based on Barlow Twins to include unlabelled data in the training process. We adopt the Barlow Twins method by using repeated measurements on the same fruit from different devices as different “views” to encode into the same latent space and combine the encoder network with a regression head for prediction. Our approach demonstrates improved performance over PLS with up to 17% lower RMSE, especially when the labelled data is limited. The Barlow loss function also improves calibration transfer results. • Novel application of Barlow Twins contrastive learning to NIR spectroscopy using repeated measurements from different devices. • Semi-supervised learning allows for unlabelled spectra to assist in training. • Up to 50% reduction in RMSE in calibration transfer tasks compared to training on the MSE loss only. • The best performance gains were observed at small labelled training sizes. • The Barlow Twins loss was not detrimental at large training set sizes.

Similar Papers

Conference Article
10.1109/icnc.2007.458

Laplacian Regularized Least Squares Regression and its Dynamic Parameter Optimization for Near Infrared Spectroscopy Modeling

Jan 01, 2007
Hui-Hua Yang + 5 more +5

Partial least square (PLS) is the most commonly used algorithm for near infrared (NIR) modeling. NIR modeling features that it's cheap, easy and fast to measure the NIR spectroscopy while expensive, difficult and time-consuming to measure the reference value for this spectroscopy. PLS often faces the challenge of that limited samples are available in training set to build a predicative model. To tackle this problem, a novel NIR modeling method - Laplacian regularized least squares regression (LapRLSR) and its dynamically adaptive parameters optimization method was presented. Based on the semi-supervised learning framework, LapRLSR can take the advantage of many unlabeled spectra to promote the prediction performance of the model though there are only few labeled samples. The proposed LapRLSR modeling algorithm was applied to the online monitoring of the concentration of salvia acid B in the column separation procedure of TCM manufacturing, and the results demonstrated that its prediction capability outperformed PLS and regularized least square regression method.

Conference Article
10.23919/ccc63176.2024.10661910

Semi-Supervised Learning Based Calibration Modeling of Near-Infrared Spectroscopy for Monitoring 1,3-Propanediol Fermentation Process Components

Jul 28, 2024
Xi Chen + 4 more +4

Concerning the real-time monitoring of important component contents during a fermentation process of 1,3-propanediol, a near-infrared (NIR) spectral calibration modeling method is proposed based on a semi-supervised learning strategy. To solve the problems of high dimensionality and serious information overlap of NIR spectra, a binary mayfly optimization algorithm (BMA) is proposed to select the characteristic wavenumbers of the measured spectra for model calibration. To address the issue of insufficient labeled sample dataset, a robust semi-supervised probabilistic principal component regression (RSSPPCR) modeling method is presented based on the expectation maximization (EM) algorithm, which could fully utilize all labeled and unlabeled spectral data. The student t-distribution is introduced into the modeling to improve the model tolerance of outliers, so as to ensure the reliability of simultaneous detection of multiple components during the fermentation process. The experimental results show that the proposed model can effectively predict the important contents of glycerol (substrate), 1,3-propanediol (product), and biomass during a fermentation process of 1,3-propanediol.