Radiation detectors and machine learning

Developing a machine learning approach to discriminating between different types of radiation in cutting edge scintillator detectors

Project status



Scintillator detectors capture signals from multiple types of incident radiation and it's necessary to discriminate between these different types with both high accuracy and high computational efficiency. Data-driven approaches can learn to accurately discriminate between complex noisy signals and uncover the true underlying distribution of the data, and numerous frameworks have been developed to generate high performance models. Developing a successful method would produce a better signal-to-noise ratio than existing methods while keeping the latency costs low enabling more efficient neutron scintillator detectors for facility science, which has application to a wide variety of physical, biological, and material sciences.

Header image: Science and Technology Facilities Council

Explaining the science

Scintillator detectors capture incident radiation to produce an excitation in the form of emitted light which is then converted to an electronic signal which can be digitised and recorded. Neutron scintillator detectors detect both incident neutron and gamma particles. The gamma particles are generally considered noise which one wishes to ignore, while neutrons provide the signal from which one can infer structural properties of matter. The discrimination of the neutron pulse shape from the gamma pulse shape is non-trivial and is further complicated by limited information at lower energies.

Project aims

The aims of this project are to explore different computational approaches to solving the Pulse Shape Discrimination (PSD) problem for neuron-gamma scintillator detectors with application to ISIS Neutron and Muon Source. The desired outcome is to produce a technique which provides a better signal-to-noise ratio without creating additional latency overheads. 

An additional outcome will be to create a close collaboration between the Detector Technology group and the Scientific Machine Learning group based at Rutherford Appleton Laboratory to enable better cross facility engagement. 


Accurate and efficient pulse shape discrimination allows us to achieve a better signal-to-noise ratio in the detection of neutron events. A more accurate and efficient detector ultimately leads to a reduced count time, facilitating an increase in efficiency for experiments carried out a neutron sources such as ISIS Neutron and Muon Source.