Luminescence thermometry driven by a support vector machine: a strategy toward precise thermal sensing.

Luminescence thermometry is a promising non-contact temperature measurement technique, but improving the precision and reliability of this method remains a challenge. Herein, we propose a thermal sensing strategy based on a machine learning. By using GdGaO: Er-Yb as the sensing medium, a support vector machine (SVM) is preliminarily adopted to establish the relationship between temperature and upconversion emission spectra, and the sensing properties are discussed through the comparison with luminescence intensity ratio (LIR) and multiple linear regression (MLR) methods. Within a wide operating temperature range (303-853 K), the maximum and the mean measurement errors actualized by the SVM are just about 0.38 and 0.12 K, respectively, much better than the other two methods (3.75 and 1.37 K for LIR and 1.82 and 0.43 K for MLR). Besides, the luminescence thermometry driven by the SVM presents a high robustness, although the spectral profiles are distorted by the interferences within the testing environment, where, however, LIR and MLR approaches become ineffective. Results demonstrate that the SVM would be a powerful tool to be applied on the luminescence thermometry for achieving a high sensing performance.