Airborne gamma spectrum smoothing method utilizing a generalized support vector regression machine

Abstract

The suppression of statistical fluctuations is crucial for the accurate qualitative and quantitative analysis of gamma-ray energy spectra. The nonlinear nature of these spectra in complex measurement environments presents a significant challenge to traditional smoothing methods, which are often constrained by their reliance on pre-defined models. To address this, support vector regression (SVR), an efficient supervised learning algorithm, is well-suited for managing nonlinear datasets. This paper proposes a generalized support vector regression (GSVR) model for gamma-ray spectrum smoothing, based on the principle of structural risk minimization. The performance of the proposed model was verified through a comparative study with traditional methods: multipoint moving average smoothing (MMAS), wavelet threshold denoising (WTDM), and noise-adjusted singular value decomposition (NASVD). Model performance was evaluated using a suite of metrics, including Smoothing Goodness (SG), Root Mean Square Error (RMSE), Energy Spectrum Distortion (ESD), and Signal-to-Noise Ratio (SNR). The comparison reveals that the proposed GSVR model demonstrates significant improvements. It achieves superior smoothing performance and better preservation of spectral peak shapes compared to all traditional methods evaluated. These results confirm the efficacy of the proposed model, offering an effective solution for smoothing gamma-ray energy spectra.