Statistical modelling of neural networks in γ-spectrometry

doi:10.1016/S0168-9002(96)80068-4

Nuclear Instruments and Methods in Physics Research Section A: Accelerators, Spectrometers, Detectors and Associated Equipment

Volume 369, Issues 2–3, 1 February 1996, Pages 642-647

https://doi.org/10.1016/S0168-9002(96)80068-4 Get rights and content

Abstract

Layered Neural Networks are a class of models based on neural computation and have been applied to the measurement of uranium enrichment. The usual methods consider a limited number of X- and γ-ray peaks, and require calibrated instrumentation for each sample. Since the source-detector ensemble geometry conditions critically differ between such measurements, the spectral region of interest is normally reduced to improve the accuracy of such conventional methods by focusing on the K_αX region where the three elementary components are present. Such measurements lead to the desired ratio. Experimental data have been used to study the performance of neural networks involving a Maximum Likelihood Method. The encoding of the data by a Neural Network approach is a promising method for the measurement of uranium ²³⁵U and ²³⁸U in infinitely thick samples.

References (12)

T.D. Reilly et al.
Report LA-4065-MS
(1970)
L.A. Kull et al.
Atomic Energy Rev.
(1976)
J. Morel et al.
(1978)
P. Matussek
Report KfK-3752
(1985)
R.J.S. Harry et al.
(1980)
R. Hagenauer
Nucl. Mater. Manag. Proc.
(1991)

There are more references available in the full text version of this article.

Cited by (34)

Machine Learning technique for isotopic determination of radioisotopes using HPGe γ-ray spectra
2023, Nuclear Instruments and Methods in Physics Research, Section A: Accelerators, Spectrometers, Detectors and Associated Equipment
$γ$ -ray spectroscopy is a quantitative, non-destructive technique that may be utilized for the identification and quantitative isotopic estimation of radionuclides. Traditional methods of isotopic determination have various challenges that contribute to statistical and systematic uncertainties in the estimated isotopics. Furthermore, these methods typically require numerous pre-processing steps, and have only been rigorously tested in laboratory settings with limited shielding. In this work, we examine the application of a number of machine learning based regression algorithms as alternatives to conventional approaches for analyzing $γ$ -ray spectroscopy data in the Emergency Response arena. This approach not only eliminates many steps in the analysis procedure, and therefore offers potential to reduce this source of systematic uncertainty, but is also shown to offer comparable performance to conventional approaches in the Emergency Response Application.
Artificial neural network modeling in environmental radioactivity studies – A review
2022, Science of the Total Environment
Citation Excerpt :
Due to their representativity and possibilities to solve mapping problems with a high number of input and output data, ANNs have been found to be useful for multivariate calibration in nuclear spectrometry. Vigneron et al. (1996) have been exploring their performances for spectrometric data analysis and estimation of parameters in the measurement of uranium enrichment. In this application, the region of interest for determining the total uranium activity is narrowed to the KαX energy range from 83 keV to 103 keV, which contains the main uranium lines and is narrow enough to assume constant efficiency.
The development of nuclear technologies has directed environmental radioactivity research toward continuously improving existing and developing new models for different interpolation, optimization, and classification tasks. Due to their adaptability to new data without knowing the actual modeling function, artificial neural networks (ANNs) are extensively used to resolve the tasks for which the application of traditional statistical methods has not provided an adequate response. This study presents an overview of ANN-based modeling in environmental radioactivity studies, including identifying and quantifying radionuclides, predicting their migration in the environment, mapping their distribution, optimizing measurement methodologies, monitoring processes in nuclear plants, and real-time data analysis. Special attention is paid to highlighting the scope of the different case studies and discussing the techniques used in model development over time. The performances of ANNs are evaluated from the perspective of prediction accuracy, emphasizing the advantages and limitations encountered in their use. The most critical elements in model optimization were identified as network structure, selection of input parameters, the properties of input data set, and applied learning algorithm. The analysis of strategies and methods for improving the performance of ANNs has shown that developing integrated and hybrid artificial intelligent tools could provide a new path in environmental radioactivity modeling toward more reliable outcomes and higher accuracy predictions. The review highlights the potential of neural networks and challenges in their application in environmental radioactivity studies and proposes directions for future research.
Estimation of uranium concentration in ore samples with machine learning methods on HPGe gamma-ray spectra
2022, Nuclear Instruments and Methods in Physics Research, Section A: Accelerators, Spectrometers, Detectors and Associated Equipment
Within the scope of determining the concentration of uranium in ore samples by gamma-ray spectrometry, we tested a series of machine-learning (ML) algorithms with a database including 1288 HPGe gamma spectra measured by Orano Mining. Instead of detecting and identifying peaks, a global interpretation of the spectra is carried out. Two different approaches were used. First, we reduced the size and the dimension of the dataset by selecting 728 spectra acquired with a same experimental setup and by resampling their 8192 channels into 168 energy bands according to the important peaks due to the natural uranium, thorium and potassium activity. Classical ML algorithms have been trained on this reduced dataset and the best uranium concentration predictions show Symmetric Mean Absolute Percentage Error lower than 6%. In a second step, the complete dataset with 1288 gamma spectra including six different measurement setups was used to train deep neural network with a re-sampling of the spectra into 2048 channels. Despite the small dataset, a Convolutional Neural Network (CNN) model provides satisfactory results with mean errors lower than 15% on this broader and more complex dataset in terms of uranium concentrations and experimental setups. These outcomes shows that ML methods can predict uranium concentration with similar uncertainties as classical gamma-ray spectroscopy (10% to 20%), but without requiring an expert knowledge to interpret the spectra.
Spectral unmixing applied to fast identification of γ-emitting radionuclides using NaI(Tl) detectors
2020, Applied Radiation and Isotopes
Citation Excerpt :
Combined to full-spectrum analysis, this property is important to take into account optimally the variability of the observed spectrum in decision-making at low-statistics. Contrary to classifiers such as artificial neural networks (Keller et al., 1995; Vigneron et al., 1996), the multiplicative update rule for spectral unmixing does not need a training phase based on a large dataset. The algorithm was first investigated in the case of the identification of radionuclides usually applied for RPM's testing (57Co, 60Co, 133Ba, 137Cs, 241Am) for the prevention of illegal nuclear material trafficking.
Spectral unmixing was investigated for fast spectroscopic identification in γ-emitter mixtures at low-statistics in the case of measurements performed to prevent illegal nuclear material trafficking or for in situ environmental analysis following a radiological or nuclear accident. For that purpose, a multiplicative update algorithm based on full-spectrum analysis was tested in the case of a 3″x3″ NaI(Tl) detector. Automatic decision-making was addressed using Monte Carlo calculations of decision thresholds and detection limits. The first results obtained with a portable instrument equipped with a 3″x3″ NaI(Tl) detector designed for the control of food samples by non-expert users following a radiological or nuclear accident, are also presented.
A comparison of machine learning methods for automated gamma-ray spectroscopy
2020, Nuclear Instruments and Methods in Physics Research, Section A: Accelerators, Spectrometers, Detectors and Associated Equipment
Citation Excerpt :
There have also been a number of published papers which apply NNs to automated isotope identification. NNs have been applied to peak fitting [3], isotope identification [4,5], and activity estimation [4,6,7]. Previous work applying NNs to spectroscopy have focused on fully-connected architectures.
Pattern recognition algorithms such as artificial neural networks (NNs) and convolution neural networks (CNNs) are prime candidates to perform automated gamma-ray spectroscopy. The way these models train and operate mimic how trained spectroscopists identify spectra. These models have shown promise in identifying gamma-ray spectra with large calibration drift and unknown background radiation fields. In this work, two algorithms for mixtures of radioisotopes based on NN and CNN are presented and evaluated.
Gamma spectral analysis by artificial neural network coupled with Monte Carlo simulations
2020, Nuclear Instruments and Methods in Physics Research, Section A: Accelerators, Spectrometers, Detectors and Associated Equipment
Citation Excerpt :
Common and attractive algorithms are Levenberg–Marquardt (LM), Bayesian regularization (BR), and Scaled conjugate gradient (SCG) [6,8]. ANN has been used in a few NAA studies such as identification [10–14] and quantification [15,16] analysis of chemical elements using ANN, performed for the purposes of explosive detection [11,17,18], drug identification [17], radon contamination [19], uranium detection [20], gamma spectra unfolding [10], uncertainty estimation in gamma spectra [21], analysis of interaction types [22], anomaly detection in gamma spectra [23], and fast neutron spectral analysis [24]. The purpose of this study lies not only in element identification, but also quantification using ANN.
Neutron activation analysis has been widely used for quantitative analysis. It can quantify elements in parts per million or billion. Artificial neural network is an attractive technique to analyze complex gamma spectra obtained from neutron activation. This study offers an improved methodology to analyze neutron activation gamma spectra using an artificial neural network. The methodology was demonstrated by quantifying five trace elements (Br, Na, Zn, K, Au) in common kidney stones. First, Monte Carlo simulations were used to create a large training data set. Then, an artificial neural network was employed for chemical elements identification analysis. For quantitative analysis, a Levenberg–Marquardt algorithm with $5 - 23 - 5$ structure artificial neural network was used. The artificial neural network for analysis of simulated gamma spectra resulted in estimated element concentrations. The differences between true and estimated concentrations are 1.8% for Br, 3.4% for Na, 5.4% for Zn, 2.8% for K, and 1.6% for Au. For real gamma spectral analysis, the largest difference was found to be 28.2% for Zn in a calcium oxalate monohydrate type of kidney stones.

View all citing articles on Scopus

View full text

Statistical modelling of neural networks in γ-spectrometry

Abstract

Report LA-4065-MS

Atomic Energy Rev.

Report KfK-3752

Nucl. Mater. Manag. Proc.