Application of support vector regression in removing Poisson fluctuation from pulse height gamma-ray spectra

Analysis of pulse height gamma-ray signals is crucial in a variety of applications regarding safeguards and homeland security. Because of the inherent random nature of radiation measurements, the spectra obtained from gamma-ray sources exhibit a high variance that can be modeled as Poisson fluctuation. This variance imposes serious difficulties to spectrum analysis and isotope identification algorithms. To that end, artificial intelligence offers a variety of tools for automated, accurate, and the fast processing of gamma-ray signals. This paper discusses the use of a support vector regression (SVR) based methodology for removing Poisson fluctuation from pulse height radiation spectra. The proposed methodology utilizes an interval based smoothing of the spectrum and subsequently suppresses the variance. Methodology performance is tested on gamma-ray spectra taken with a low-resolution sodium iodide detector having a length of 1024 bins. Furthermore, this SVR technique is benchmarked against the 3-point and 7-point simple moving average methods. The results of this benchmarking demonstrate the effectiveness of the proposed methodology in removing Poisson fluctuation over the other methods tested.