Overcoming crosstalk in luminescence images of mineral grains

Luminescence imaging systems are becoming available for use in luminescence dating, and could potentially allow the dating of sediment and rock at a microscopic scale. For this to be achieved, analytical methods must be developed for turning the data-rich images into reproducible luminescence signals. At present, luminescence signals are collected from images after identifying Regions of Interest (ROIs) - a group of pixels mapped to a luminescent region or grain; the sum of the net ROI signal provides the measure of luminescence for each grain. However, the design of luminescence imaging systems requires a trade-off between signal focus and signal intensity. To maximise signal intensity, commercial systems use a lens combination which also induces optical aberrations, affecting the focus of the image. The variable focus of the image, combined with sample movement between measurements, means that the ROI signals may suffer from reproducibility problems and that signal crosstalk is a significant problem. Instead, the images should be paramaterised so that the inherent signal from each grain can be decontaminated from nuisance factors. We describe a data reduction method which uses a Bayesian hierarchical model to resolve the signal from each grain, with input from an incrementally expanding ROI. When tested with an artificial mixed population of grains, the method is better at recovering the known doses than the standard ROI approach, and has significant potential if combined with optimised measurement systems and pre-processing software.