Dataset for Three-Dimensional Radiation Field Reconstruction with Sparse Sampling via Physics-Informed Neural Network

This dataset contains the supporting data for the paper titled “Three-dimensional radiation field reconstruction with sparse sampling via physics-informed neural network.” The dataset was generated using the Geant4 Monte Carlo simulation toolkit, with the simulation scenarios designed by reference to real radiation working environments. A Cs-137 point source was used to establish three typical three-dimensional gamma radiation field scenarios, including a Single-source radiation field without shielding, a Dual-source radiation field without shielding, and a Single-source radiation field with shielding. For each radiation field, sparse sampling datasets were constructed by collecting dose-rate data at randomly distributed spatial points, while the test datasets were constructed from complete spatial dose-rate data covering the whole study domain. These datasets were used for model training, reconstruction, and quantitative evaluation.This dataset provides detailed dose-rate data for sparse-sampling three-dimensional radiation field reconstruction under different source configurations and shielding conditions. It supports the development and validation of reconstruction models, especially physics-informed learning methods, and may serve as a reference for comparative studies, algorithm benchmarking, and related research in three-dimensional radiation field reconstruction, radioactive source localization, and radiation-aware path planning.