Speaker
Description
This project entails the development of a machine learning-based gamma detection
system with tracking and prediction capabilities. It involves integrating gamma spectroscopy via an artificial intelligence model, including a custom
neural network trained on spectra from various isotopes and a compact
detector. The model would be integrated into a compact, low-cost, micro-controller
enabled system with additional sensors to provide the hardware. This includes, but is
not limited to, a thermal sensor, GPS sensor, multi-spectral camera, etc. and would
be used to run and manage the model. The overall system with the neural network and multi-spectral system would be designed to provide a useful and inexpen-
sive complement to current radiation safety methods and additionally serve as an early
warning system (e.g. in the case of radiation leaks). Additionally, the treatment of noise in the obtained spectra, both environmental and
systematic, will be investigated in this research via neural networks and Kalman filters.
This is crucial for monitoring out in the field, where low-level monitoring often suffers
from background interference. The value is increased by the packaging of thermal and
GPS sensors, together with a standard camera and infrared camera for mobile object detection,
complementing the gamma-ray spectra identification.
Thermal sensors combined with an infrared camera can allow the detection of heat
signatures, which is useful for finding the source of emissions. Through the use of recursive
algorithms, an interactive multiple-model estimator could offer prediction capabilities through modelling trajectories of
gamma radiation within the environment. Eventually, a self-correcting system with identification and tracking can be used to
provide early warning or other useful information regarding radiation in the environ-
ment. Although a stationary system with stationary sources will be tested, one could
easily apply this to a dynamic mobile system (e.g. drones and surveillance cameras)
with the addition of motors or compact vehicles for transportation. This system could
be compared to existing radiation safety methods and warning systems, as well as ma-
chine learning benchmarks for spectra prediction. For example, one could test how
well it identifies radiation leaks, the number of false positives, and the accuracy of the
tracking system.
It is also believed that the supervised training process of the
network can include examples of noise, to assist in obtaining as clean a spectrum as
possible, and to guide the network to the true reaction data.
Ideally, the network will learn to identify the isotopes by peaks and possibly backscatter
patterns. This will prevent the inclusion of unanticipated sources confusing the network
when making predictions.
The consideration of radiation monitoring environments, such as nuclear waste disposal
sites and nuclear power plants (e.g. Koeberg), yields insights to the value of this
research. A low-cost, efficient radiation monitoring device could assist in radiation
protection cases, capable of detecting gamma emissions and hotspots in the surrounding
environment. There is also the use in the field, for remote monitoring and built-in GPS for local-
ization. Moreover, a mobile system can be used to investigate stationary and moving
5
sources in the field (e.g. geological vaults), providing information regarding the ener-
gies and intensities of various gamma sources. There is also the knowledge gained by investigating further development of micro-
controller-enabled systems with deep learning-based object recognition and tracking
software. There is a direct comparison to be made with existing surveillance and monitoring technology.
Level for award;(Hons, MSc, PhD, N/A)?
PhD
| Apply to be considered for a student ; award (Yes / No)? | Yes |
|---|
