Introduction

The invention consists in an algorithm able to perform both identification and quantification of the isotopic composition of gamma spectra produced by device with low/medium energy resolution. The use of convolutional neural networks with a directed acyclic graph structure allows to perform both tasks automatically and precisely with short training times.

Technical features

A method for the radioisotope recognition of a gamma spectrum and the quantification of its radioisotopic composition. This task is largely affected by errors in the fields where the energy resolution of the employed gamma detectors is low (industrial, medical, environmental monitoring, security and safety). The invention consists in an algorithm base don convolutional neural networks (CNN) with a directed acyclic graph structure (DAG). The capability of extracting the relevant features (CNNs), combined with the possibility of performing multiple tasks (DAGs), allows to obtain an automatic and precise identification and quantification. After the training, the only required input is the raw spectrum produced by the detector, without human intervention nor intermediate pre-processing.

Possible Applications

• CBRNE protection (Chemical, Biological, Radiological, Nuclear, and high yield Explosives);
• Countermeasures for illicit trafficking of nuclear and radioactive materials;
• Radiation monitoring of sensitive locations;
• Environmental radioactivity monitoring.

Advantages

• Applicable to gamma spectra with low energy resolution;
• Simultaneous classification and quantification of the isotopic composition of a raw gamma spectrum;
• Fast training;
• High accuracy;
• Both simulated and/or experimental datasets can be used;
• High customisability.