CTMNR: A Hybrid CNN-Transformer Approach for Enhanced Multi-Nuclide Classification in Radiation Spectroscopy

Multi-nuclide detection is essential for ensuring nuclear safety, particularly in complex radiation fields where gamma-ray spectra from multiple radionuclides overlap significantly. Traditional spectral analysis methods and existing neural network approaches often struggle with accurate identification under such conditions. To address this challenge, we propose a hybrid deep learning model that integrates convolutional neural networks (CNNs) for local feature extraction with Transformers for long-range dependency modeling. A dataset of about 6 × 105gamma-ray spectra covering twelve radionuclides and their combinations was constructed. Preprocessing strategies, including denoising, data augmentation, and one-dimensional to two-dimensional spectral transformation were applied. Experimental results demonstrate that the proposed model achieves a 95.3% identification accuracy in scenarios involving up to eight mixed radionuclides, outperforming one-dimensional CNNs(1D-CNNs) and two-dimensional CNNs(2D-CNNs) by 43.5% and 18.6% respectively. These findings highlight the effectiveness of combining CNNs and Transformers for complex spectral analysis and provide technical support for applications in nuclear emergency response and environmental radiation monitoring.