Neural-SPGA: a neural-embedded serial – parallel genetic algorithm for neutron coded-aperture imaging reconstruction

Neutron coded-aperture imaging (NCAI) is a critical diagnostic tool for inertial confinement fusion (ICF), where high-fidelity reconstruction is essential for characterizing the spatial distribution of the fusion hot-spot. Although heuristic methods, such as genetic algorithms, often outperform deterministic methods, their high computational cost hinders their practical use in ICF diagnostics. In this work, we propose a neural-embedded serial – parallel genetic algorithm (Neural-SPGA) for NCAI reconstruction. Built on an adaptive genetic algorithm (AGA) framework, Neural-SPGA incorporates three key components: (i) an elite-guided serial – parallel hybrid reproduction mechanism that dynamically balances global exploration and local exploitation; (ii) a lightweight convolutional neural network initialization module that injects high-quality elite individuals to reduce ineffective early-stage evolution; and (iii) an online surrogate screening module that selects high-potential candidates for true fitness evaluation. Simulation results show that Neural-SPGA consistently outperforms representative deterministic methods as well as the heuristic baseline AGA and its variants in reconstruction quality. Notably, Neural-SPGA achieves a speedup ratio of approximately 8.0 over AGA and remains robust under additive Gaussian noise levels up to 5%. By coupling fast deep learning inference with iterative genetic optimization, Neural-SPGA provides a high-fidelity and computationally efficient reconstruction framework for ICF diagnostics.