Parametric optimization and multi-task learning for pseudo gamma spectroscopy of plastic scintillation detector

Abstract

Plastic scintillation detectors have been widely used in radiation measurements owing to their low cost and large volume, However, they are difficult to conduct energy calibration and not adequate spectroscopic measurements because of poor energy resolution and absence of full energy peaks in their spectrum. In this thesis, I study parametric optimization and multi-task learning to allow plastic scintillation detectors to spectroscopic capability. In parametric optimization, I address an optimization problem to calibrate measured channel to physical energy and to find energy broadening parameter for simulating response function of the detectors. In multi-task learning, I develop a deep learning model to generate full energy peaks and to predict relative activities of gamma-ray sources from plastic gamma spectra.