An RBF Neural Network Approach in Radionuclide Identification of Unknown Sources Utilizing γ-Ray Spectra

Pola-Lydia Lagari, Purdue University


At first, simulated ?γ-ray spectra for a set of 25 radionuclides, have been produced using the "Gamma Detector Response and Analysis Software (GADRAS)". For each of these profiles (counts/kev vs energy), a Gaussian "Radial Basis Function" (RBF) network has been trained to represent it by an analytic closed form expression. Hence a library consisting of 25 RBF-networks, for the corresponding radionuclides, has been built. Secondly, a method for identifying the presence of radionuclides in the spectrum of an unknown source has been developed, assuming that the source contains a mixture of the considered radionuclides only. A linear combination of the library profiles is compared to the actual spectrum, and constrained optimization techniques are applied to minimize the deviation in a least squares sense.




Tsoukalas, Purdue University.

Subject Area

Electrical engineering|Nuclear engineering|Computer science

Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server