LSNet: efficient nuclide spectrum recognition network

To address the interference from non-radioactive components in the environment and achieve efficient identification of mixed nuclide spectra, we propose a novel isotope identification method, LSNet. Experimental results demonstrate that, with comparable parameter counts and computational complexity, LSNet outperforms models such as the Gate-Recurrent-Unit (GRU) across various evaluation metrics. Additionally, LSNet shows significantly higher accuracy in identifying anomalous nuclide spectra compared to other models. Finally, we demonstrate that LSNet effectively identifies mixed nuclides.