Gaussian process regression algorithm for radiation source localization based on scintillation fiber optic sensor networks

This paper introduces a novel method that combines sensor networks and machine learning to achieve real-time monitoring and accurate localization of radiation sources. By deploying multiple sensors to create a monitoring network and developing a Gaussian process regression model from machine learning to reconstruct radiation field data from sensor measurements, the position of the radiation source is identified based on the peak coordinates of the reconstructed radiation field. The effectiveness of this approach is validated through simulation experiments involving localization with a single radiation source, two sources, and obstructions. Using LYPS:Ce crystal material as radiation sensors, an all-optical fiber multi-point sensor array was assembled. Radiation-induced luminescence tests were performed on a 60 Co radiation source, and radiation field reconstruction was achieved using the Gaussian process regression model. The results show that within a 5 m × 5 m radiation environment, the average positioning error with 25 sensors is 0.14m when using the Gaussian process regression algorithm. Combining sensor networks with Gaussian process regression algorithms enables real-time, remote dose monitoring and precise positioning of radiation sources.