Low-Information Radiation Imaging using Rotating Scatter Mask Systems and Neural Network Algorithms

Abstract

Developing fast, portable, and accurate radiation imagers remains an objective for many nuclear safety and security applications. While recent studies have demonstrated the directional capabilities of the single-detector rotating scatter mask (RSM) system for discrete, dual-particle environments, there has been little progress towards adapting it as a true imaging device. In this study, two algorithms were developed and tested using an RSM mask design previously optimized for directional detection and simulated 137Cs signals from a variety of source distributions. The first, maximum-likelihood expectation-maximization(ML-EM), was shown to generate noisy images, with relatively low accuracy (145% average relative error) and signal-to-noise ratio (0.27) for most source distributions simulated. The second, a novel regenerative neural network (ReGeNN), performed exceptionally well, with significantly higher accuracy (33% average relative error) over all source types compared to ML-EM and drastically improved signal-to-noise ratio (0.85) in the reconstructed images. This method was experimentally validated using an additively-manufactured mask. Measuring two point and one ring 22Na source distributions, a modified ReGeNN was able to successfully train on simulated noisy signals and accurately predict the relative size and direction of the three sources. Training ReGeNN further revealed potential errors caused from overfitting, suggesting future improvement in ReGeNN architecture and training base is needed to obtain accurate activity profiles. To support future design optimizations, a ray tracing algorithm was also developed as an alternative to more rigorous Monte Carlo RSM simulations. This ray tracing code was shown to significantly improve computational efficiency, at a slight cost to the simulated signal accuracy for more complex designs

Open PDF

Document Details

Document Type
Technical Report
Publication Date
Oct 01, 2020
Accession Number
AD1115152

Entities

People

  • Robert J. Olesen

Organizations

  • Air Force Institute of Technology

Tags

Communities of Interest

  • Energy and Power Technologies

DTIC Thesaurus Topics

  • Accuracy
  • Air Force
  • Artificial Intelligence Software
  • Data Science
  • Databases
  • Detection
  • Detectors
  • Diagnostic Imaging
  • Gamma Rays
  • Gaussian Distributions
  • Information Processing
  • Information Science
  • Machine Learning
  • Neural Networks
  • Predictive Modeling
  • United States
  • United States Government

Fields of Study

  • Physics

Readers

  • Image Processing and Computer Vision.
  • Neural Network Machine Learning.

Technology Areas

  • AI & ML
  • AI & ML - Bayesian Inference
  • AI & ML - Neural Networks