Evaluating Machine Learning Algorithms in CCDs for Environmental Tritium Detection

Detecting tritium, a radioactive isotope of hydrogen, is a key area of research for nuclear science and security. As a by-product of nuclear fission, tritium is key signature for detecting nuclear activities. Tritium emits low energy beta rays that stop after traveling only a short distance. Through the use of charge-coupled devices (CCDs), we are studying environmental tritium detection. With CCDs we can collect faint beta ray signals emitted by tritium, and produce high resolution information-rich images of particle tracks. To establish an efficient model to separate faint tritium signals from background signals, we are studying a variety of machine learning algorithms, such as boosted decision trees (BDTs) to train on simulated and experimental data. We present analysis results from experimental and simulated data and discuss future tritium detection using CCDs.