Year
2023
File Attachment
Abstract
Portable radiation detection systems can be equipped with contextual sensors to allow free-moving 3D gammaray imaging through scene data fusion (SDF). While developed for free-moving measurements, SDF also has
applications in cases that require multiple static measurements to generate quantitative 3D images, where the scene
information can improve results. The scene information captured by these devices can be leveraged to determine
optimal measurement positions for quantification measurements, such as in safeguards applications, where limited
time is provided to quantify nuclear material present.
In this work the scene information obtained from contextual sensors is used to construct a 3D voxel grid.
A 3D object of interest within this grid is selected by an operator either manually or automatically through an
object detection machine learning framework. A number of potential measurement positions and orientations (poses)
around this object of interest are identified, and the sensitivity to each of the object voxels from each of these
poses is computed. Ray tracing is used to estimate any attenuation term, and the detector response for the Compton
imaging modality is used in computing sensitivity. The combination of positions which results in maximum total
sensitivity with minimum variance in sensitivity across object voxels are identified as the optimal measurement
positions. For large measurement spaces the optimization is computationally intensive, so a genetic algorithm is
used to accelerate the computation of optimal positions.
We first demonstrate the feasibility of the optimization framework with a toy model. We then use a SDF-enabled
gamma-ray imager to find optimal measurement positions for a mock material holdup loop with distributed sources.
We present our approach and results by performing an optimization to produce quantitative Compton images in
these low-count-rate environments and the minimization of associated uncertainties.