Development of OpenMC based AI Model for Gamma Spectral Decomposition for Low Activity Estimation in Environmental Samples

Bulk gamma-ray spectrometry of environmental soil samples using sodium iodide detectors is limited by severe photopeak overlap, density-dependent self-absorption and the lack of certified volumetric calibration standards for trace fallout. In this study, a computational framework couples OpenMC particle transport modeling of a cylindrical NaI(Tl) crystal in sealed polymer Marinelli container geometry with an AI-based spectrum decomposition (AI-SD) architecture. Twenty-three environmental soil samples characterized by high-purity germanium (HPGe) spectrometry served as an experimental reference for validating simulated response functions of principal radionuclides (\(\:{}^{226}\text{Ra},{}^{232}\text{Th},{}^{40}\text{K}\) and \(\:{}^{137}\text{Cs}\)). A dataset of 100,000 synthetic spectra spanning variable bulk soil densities was generated to train a heterogeneous soft-voting ensemble combining 1D-CNN peak-shaping filters, residual baseline-preservation pathways and an attention-gated U-Net deconvolution network. Dynamic attention gating selectively isolates the 661.7 keV \(\:{}^{137}\text{Cs\:}\)photopeak from adjacent 609.3 keV \(\:{}^{214}\text{Bi}\:\)down-scattering without requiring empirical background subtraction. Across an independent synthetic test set of 15,000 spectra, the ensemble achieved 100% identification accuracy with mean absolute percentage errors below 4.3% for primordial decay series and 96.5% accuracy (0.81 Bq/kg mean absolute error) for trace \(\:{}^{137}\text{Cs}\). When benchmarked against 21 intact physical soil samples, the framework demonstrated strong agreement with laboratory HPGe measurements (\(\:{R}^{2}>0.98\)), completely eliminating the uranium-series divergence typical of classical matrix unfolding. Executing in under 10 ms per spectrum on standard central processing units, the methodology provides a practical calibration equivalent for autonomous environmental soil radiometry.

Authors

Institutions

Publication Details

Journal
Research Square
Published
2026-11-18
DOI
https://doi.org/10.21203/rs.3.rs-11647722/v1
Primary Topic
Radiation Detection and Scintillator Technologies
Type
preprint

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Development of OpenMC based AI Model for Gamma Spectral Decomposition for Low Activity Estimation in Environmental Samples

Awais Zahur, Danish Gulzar, Rizwan Ahmed, Muhammad Sohail
Research Square
Radiation Detection and Scintillator Technologies
preprint

Development of OpenMC based AI Model for Gamma Spectral Decomposition for Low Activity Estimation in Environmental Samples

Awais Zahur, Danish Gulzar, Rizwan Ahmed, Muhammad Sohail
preprint en

Abstract

Bulk gamma-ray spectrometry of environmental soil samples using sodium iodide detectors is limited by severe photopeak overlap, density-dependent self-absorption and the lack of certified volumetric calibration standards for trace fallout. In this study, a computational framework couples OpenMC particle transport modeling of a cylindrical NaI(Tl) crystal in sealed polymer Marinelli container geometry with an AI-based spectrum decomposition (AI-SD) architecture. Twenty-three environmental soil samples characterized by high-purity germanium (HPGe) spectrometry served as an experimental reference for validating simulated response functions of principal radionuclides (\(\:{}^{226}\text{Ra},{}^{232}\text{Th},{}^{40}\text{K}\) and \(\:{}^{137}\text{Cs}\)). A dataset of 100,000 synthetic spectra spanning variable bulk soil densities was generated to train a heterogeneous soft-voting ensemble combining 1D-CNN peak-shaping filters, residual baseline-preservation pathways and an attention-gated U-Net deconvolution network. Dynamic attention gating selectively isolates the 661.7 keV \(\:{}^{137}\text{Cs\:}\)photopeak from adjacent 609.3 keV \(\:{}^{214}\text{Bi}\:\)down-scattering without requiring empirical background subtraction. Across an independent synthetic test set of 15,000 spectra, the ensemble achieved 100% identification accuracy with mean absolute percentage errors below 4.3% for primordial decay series and 96.5% accuracy (0.81 Bq/kg mean absolute error) for trace \(\:{}^{137}\text{Cs}\). When benchmarked against 21 intact physical soil samples, the framework demonstrated strong agreement with laboratory HPGe measurements (\(\:{R}^{2}>0.98\)), completely eliminating the uranium-series divergence typical of classical matrix unfolding. Executing in under 10 ms per spectrum on standard central processing units, the methodology provides a practical calibration equivalent for autonomous environmental soil radiometry.

Research Square
Pakistan Institute of Engineering and Applied Sciences (PK)
Pakistan Institute of Engineering and Applied Sciences
Radiation Detection and Scintillator Technologies
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon