Neutron penumbral image reconstruction with a convolution neural network using fast Fourier transform

Area of Science:

• Nuclear Fusion Science
• Plasma Physics
• Diagnostic Techniques

Background:

• Hot spot asymmetry in Inertial Confinement Fusion (ICF) impacts implosion performance.
• Neutron penumbral imaging is crucial for diagnosing hot spot shape in ICF.

Purpose of the Study:

• To develop an advanced algorithm for reconstructing 2D neutron emission images from penumbral detector images.
• To improve the accuracy and robustness of hot spot imaging in ICF diagnostics.

Main Methods:

• Development of a 16-layer neural network, the Fast Fourier Transform Neural Network (FFTNN), incorporating FFT, convolution, and fully connected layers.
• Generation of training datasets using a phenomenological hot spot model due to experimental data limitations.
• Comparison of FFTNN performance against traditional Wiener filtering and Lucy-Richardson algorithms.

Main Results:

• The FFTNN demonstrates superior reconstruction performance compared to conventional methods, particularly in high-noise scenarios.
• Evaluation metrics such as mean squared error and structure similar index measure confirm FFTNN's enhanced accuracy.
• The neural network effectively reconstructs 2D neutron emission images from penumbral data.

Conclusions:

• The developed FFTNN offers a significant advancement in neutron imaging reconstruction for ICF.
• This approach enhances the integration of neutron imaging diagnosis into ICF research.
• FFTNN provides a more accurate and reliable method for analyzing hot spot characteristics.