The gamma-ray spectrum contains a signature of radionuclides, and its analysis is essential to evaluate quantitative radioactivity in laboratory and on site for a wide range of applications in fields of radiation detection, nuclear technology, nuclear emergencies, etc. Conventional gamma-ray spectrum analysis focuses on the spectrum peaks. The peak structure forms the photoelectric peak due to the photoelectric effect, in which the gamma ray loses all its energy in the detector. The performance of conventional gamma-ray spectrum analysis methods for low-resolution measurements is often poor.

With the rapid development of artificial intelligence, gamma-ray spectrum analysis based on machine learning has mostly been investigated using artificial neural networks (ANNs). Although ANNs have been applied to low-resolution gamma-ray spectra, the resulting methods show limitations regarding the ANN architecture and training set construction, thus being unsuitable for large-scale applications.

Architecture of proposed ResNet for gamma-ray spectrum analysis

Considering the limitations of existing methods for gamma-ray spectrum analysis based on ANNs, CIRP established a method for constructing a comprehensive training set and a modified ResNet (residual nets) architecture to perform low-resolution gamma-ray spectrum analysis. The constructed training set reproduces the shape feature diversity of spectra acquired from real detection scenarios by using various parameter settings in Monte Carlo simulations. The proposed ResNet constitutes the largest and deepest architecture ever applied to gamma-ray spectrum analysis, containing 51 layers and more than 10⁷parameters. Test and comparison results based on simulated spectra show that the average precision of the proposed ResNet is substantially higher than that of a conventional CNN network and an FC network. Additionally, the proposed ResNet provides robust weak ray identification, false peak discrimination, overlapping peak discrimination, and sparsity preservation in the prediction results.

This study demonstrates the feasibility of applying deep learning to gamma-ray spectrum analysis and introduces an approach to achieve general, accurate, sensitive, and reliable gamma-ray spectrum analysis.

Contact: official@cirp.org.cn