A new background discrimination method using support vector machine (SVM) with gaussian kernel in low-level H liquid scintillation measurement.

In this study, a new background discrimination method using SVM with gaussian kernel in low-level H liquid scintillation measurement is developed. Dimensionality reduction techniques were employed to process the full waveform data of each event following time discrimination. The reduced-dimensional data were then used to construct the classification model. To ensure a fair comparison with traditional method, the advantages of offline processing were employed to systematically traverse all possible parameter configurations within reasonable ranges, enabling the determination of the optimal parameters. In machine learning-based background event discrimination, the high-quality labeled datasets are obtained through long-term measurements of blank samples and short-term measurements of medium-activity H samples. Additionally, a series of low-activity H samples with activity levels of approximately 1 Bq, 3 Bq, and 5 Bq were prepared through quantitative dilution. The results demonstrate that the proposed method achieves a significant improvement in discrimination capability compared to traditional method. It effectively minimizes background levels across the entire energy range while preserving detection efficiency, significantly enhancing the measurement capability for low-activity H samples.