Machine learning-assisted luminescence thermometry using Mn5 + -doped near-infrared phosphor with improved accuracy and precision

Abstract

This study provides a thorough investigation of machine learning-assisted luminescent thermometry using a Mn5+-doped Ca6Ba(PO4)4O phosphor. A novel, slightly modified Principal Component Analysis (PCA), where data normalization was observation-based rather than feature-based, was used to analyze near-infrared emission spectra collected over a temperature range of 293–373 K. This method showed significantly improved thermometric performance compared to traditional single-parameter and multiparametric approaches. Based on statistical analysis of cross-validation experimental data, the PCA-based method achieved exceptional average temperature resolution (δT = 0.135 K) and accuracy (ΔT = 0.077 K) across the entire temperature range, with even better performance in the physiological temperature range (δTphy = 0.074 K, ΔTphy = 0.032 K). This method utilizes full spectral data through dimensionality reduction, offering insights into the most thermometrically significant spectral regions while keeping the computation simple with basic mathematical operations. Compared to traditional thermometry techniques, which involve calculating emission band intensity ratios, finding spectral positions, and fitting emission decays, PCA-assisted thermometry greatly simplifies and speeds up the computational process, while also enhancing the accuracy and precision of temperature measurement.