Research on neutron spectrum reconstruction methods based on space neutron spectrometer

The measurement of space neutron energy spectra faces significant uncertainty in spectrum unfolding for neutron detectors due to the lack of prior information. This work proposes a spectrum unfolding method combining a Backpropagation (BP) neural network with iterative or maximum entropy techniques, which can effectively enhance the accuracy of space neutron energy spectrum reconstruction. Traditional neural network-based unfolding methods are often limited by the small size of available datasets, which restricts their accuracy. This work expands the neural network dataset through linear superposition of common neutron energy spectra, achieving better reconstruction results. However, the model showed poor robustness to noise and could not effectively handle noise interference. To address this, the study further improved the training process by considering the impact of noise, enhancing the model’s performance in noisy environments, and ultimately achieving more ideal spectral reconstruction results. Finally, based on the neural network as a prior spectrum, combined with iterative methods and the maximum entropy method, a neutron spectrum deconvolution scheme suitable for space neutron spectrometer was proposed, which can effectively reconstruct the neutron spectrum and provide a feasible solution for practical applications.