A radioisotope identification system from gamma emissions using neural networks

Identification of radionuclides based on analysis of their gamma emissions requires as input the gamma spectra of the nuclides to be identified. Due to limitations, aging and variations of operational conditions of the involved electronic equipment, the resultant spectra presents distortions in such a way that the majority of identification systems is only capable to identify the nuclides when calibrated, imposing the need of a periodic manual calibration intervention by an operator. Our intent is to increase the availability and the robustness of this type of system by reducing the necessity of manual recalibration of the system by the operator. This work proposes as solution to the problem of identification under shifting of the spectra, the use of a neural network associated to a transformation of the spectra that permits the correct identification of the nuclide from its shifted spectra. (author)