Machine-learning-assisted temperature measurement of localized heated air using femtosecond filament plasma...

Non-contact temperature measurement of localized hot air is useful for thermal-flow and plasma diagnostics. We recorded femtosecond-filament plasma spectra at seven temperatures from 155 to 432°C and analyzed baseline-corrected N₂/N₂⁺ spectral intervals normalized to the 357.7 nm N₂ band. Single-peak fitting yielded an RMSE of 18.0°C on a random hold-out set, whereas Gaussian process regression (GPR) achieved an RMSE of 3.3°C, MAE of 2.7°C, and R² of 0.9987. These results demonstrate accurate, target-free temperature estimation of localized heated air under laboratory conditions.