Support Vector Machine-Based Tagged Neutron Method for Explosives Detection

• Published in: Arabian journal for science and engineering (2011) Vol. 49; no. 7; pp. 9895 - 9908
• Main Authors: Li, Guang-Hao, Jia, Shao-Lei, Lu, Zhao-Hu, Jing, Shi-Wei
• Format: Journal Article
• Language: English
• Published: Berlin/Heidelberg Springer Berlin Heidelberg 01.07.2024
• Subjects: Engineering; Humanities and Social Sciences; multidisciplinary; Research Article-Physics; Science; Support vector machines; Explosives; ray; Grid search method; Tagged neutron method; Characteristic
• ISSN: 2193-567X; 2191-4281
• DOI: 10.1007/s13369-023-08695-8

Tagged neutron detection system combined with support vector machine (SVM) is proposed for the detection of explosives hidden inside walls. The detection system was based on an ING-27 neutron generator as neutron source, two lutetium yttrium silicate (LYSO) detectors as γ -detectors, and one silicon detector as α-detector. The difference in γ -ray counts within the time window for different samples was combined with the peak area ratios of the elemental peaks in the γ -energy spectra and used as input vectors for an SVM. A Gaussian kernel function was used as a kernel function and a grid search method as optimization of the penalty factor c and hyperparameter g of the SVM in this experiment. Fivefold cross-validation was used to evaluate the models developed. The correctness of the support vector machine was found to be 100%, 98.3%, and 95% for the target sample detection in that order, and the fivefold cross-validation accuracy was 100%, 97.5%, and 93.3%, respectively. The detection results showed that the established model effectively avoided the overfitting problem and could accurately detect the explosives hidden in the wall.