Simulated Thick, Fully-Depleted CCD Exposures Analyzed with Deep Learning Techniques

TL;DR

This paper demonstrates the application of deep learning, specifically panoptic segmentation, to simulated thick, fully-depleted CCD images for identifying and measuring radioactive energy depositions, with promising results in spectrum reproduction.

Contribution

It introduces a novel application of deep learning to analyze simulated CCD data for nuclear physics and dark matter detection, focusing on isotope identification and energy measurement.

Findings

Accurately reproduces the beta spectrum from simulated CCD images.

Shows potential for identifying radioisotope depositions with deep learning.

Highlights challenges in distinguishing gamma rays, electrons, and cosmic muons.

Abstract

Thick, Charge Coupled Devices (CCDs) have recently been explored for applied physics, such as nuclear explosion monitoring, and dark matter detection purposes. When run in fully-depleted mode, these devices are sensitive detectors for energy depositions by a variety of primary particles. In this study we are interested in applying the Deep Learning (DL) technique known as panoptic segmentation to simulated CCD images to identify, attribute and measure energy depositions from radioisotopes of interest. We simulate CCD exposures of a chosen radioxenon isotope, $^{135}$Xe, and overlay a simulated cosmic muon background appropriate for a surface-lab. We show that with this DL technique we can reproduce the beta spectrum to good accuracy, while suffering expected confusion with same-topology gammas and conversion electrons and identifying cosmic muons less than optimally.