The three-dimensional radiation field is an important database reflecting the radioactivity distribution in a nuclear facility. It is of great significance to accurately and quickly grasp the radiation dose field distribution to implement radiation protection. Presently, majority of radiation field reconstruction algorithms concentrate on two-dimensional reconstruction and can only measure on a regular grid. With the progress of artificial intelligence technology, neural networks have great potential in radiation field reconstruction. In this work, an improved Genetic Algorithm Optimized Backpropagation (GA-BP) neural network was proposed, which can efficiently reconstruct the radiation dose rate at any given position within the three-dimensional space, even under the condition of a low sampling rate. The proposed method achieves a remarkable speed, capable of reconstructing nearly 500 spots in 0.01s. Two Monte Carlo simulations corresponding to the shielded and unshielded cases verified the effectiveness of the proposed method. The method was further tested on datasets with equally spaced and randomly distributed data points. In both simulation scenarios, the proposed method demonstrated the ability to reconstruct the three-dimensional dose rate field using less than 6% of the data for the simulation cases with a low error level of 3% (unshielded) to 8% (shielded). In the real experimental validation, the error is at 15%, and the point error is less than 30% in most areas.
Read full abstract