Bootstrap-Based Multitask Deep Learning Method for Uncertainty Estimation and Correction in TSOM

Through-focus scanning optical microscopy (TSOM) is an optical nondestructive measurement and inspection technique for nanoscale measurement. Uncertainty estimation is fundamental for reliable measurement results. However, this uncertainty estimation has not been addressed in deep learning-based TSOM. To fill this gap, we propose an uncertainty estimation and correction framework based on bootstrap and deep learning. This proposed method evaluates data uncertainty and model uncertainty, where data uncertainty is quantified by bootstrap combined with a pretrained model, and model uncertainty is calibrated to ensure it encompasses reference values by bootstrap combined with variable-output models. The effectiveness of the proposed method is verified through comparison with the Markov chain Monte Carlo (MCMC) and other methods. The results demonstrate its potential as a reliable uncertainty estimation method for TSOM and other measurement technologies.