BackgroundThe high-dimensionality and inherent noise in gamma spectrum data often lead to low identification accuracy and poor computational efficiency in conventional nuclide identification methods. Therefore, a new method that can provide high accuracy and computational efficiency in portable devices is needed.PurposeThis study aims to develop an efficient method for nuclide identification by combining neighborhood rough set (NRS) theory and K-nearest neighbor (KNN) classifier.MethodsThis method involved three-stage process. Firstly, principal component analysis (PCA) was applied to gamma spectrum data for dimensionality reduction and reduction of linear redundancy. Subsequently, a heuristic search strategy based on NRS was employed to optimize the feature subset by eliminating redundant attributes. Lastly, the KNN classifier was used for efficient nuclide identification in the reduced low dimensional feature space. In the meantime, the identification method was implemented on an STM32F407ZGT6 platform, and experimental verification was performed with a LaBr₃(Ce) detection system and a 1 024-channel multichannel analyzer.ResultsThe experiment used a dataset of 224 gamma spectra, comprising 12 single-nuclide and 2 mixed-nuclide categories was collected in the experiment, and verification results show that an average identification accuracy of 98.5% was achieved by proposed method on the test set with a neighborhood radius of δ=0.2. The processing time for a single identification is within 140 ms.ConclusionsExperimental results confirm the superior generalization performance and high computational efficiency of the proposed method across various scenarios, including mixed-nuclide conditions, hence providing a reliable and efficient technical pathway for efficient nuclide identification in portable radiation detection devices.
Chen et al. (Sun,) studied this question.