Feature extraction of gamma spectra based on a CNN-GRU fusion classification model

The development of nuclide recognition methods based on marine gamma spectra is crucial for enhancing the performance of radiological monitoring. Marine gamma ray spectra often exhibit significant fluctuations and noise, complicating the measurement and identification of nuclides. Existing methods include denoising, background subtraction, and feature extraction. During these processes, information from low-count channels may be lost. This study proposes a CNN-GRU fusion classification model. To ensure the model's effectiveness and generalization capability, the training dataset was constructed based on both marine field experiments and laboratory standard source experiments, as well as Monte Carlo simulations. The study compares a 2D-CNN model, a 1D-GRU model, and the CNN-GRU fusion classification model. Model performance was evaluated using metrics such as F1 score, recall, precision, accuracy, and loss curves. The results demonstrate that the CNN-GRU fusion classification model not only performs well in identifying drift peaks and weak peaks but also extracts more detailed information from complex spectra, showing significant improvements in both accuracy and algorithm efficiency.