Fast nuclide identification method based on hybrid dynamic Bayesian network

An efficient and precise nuclide identification method is essential in various contexts. This paper treats the detector output pulse events as a sequence of energy event models corresponding to monoenergetic rays and employs hybrid dynamic Bayesian network modeling, grounded in the energy spectrum response of the detector. The results from Monte Carlo simulations are utilized to assess the belief that each pulse event corresponds to different monoenergetic ray energy events. Furthermore, this paper introduces a probabilistic propagation algorithm that updates the prior probability of the particle transport model and continuously refines the parameters of the Bayesian network model according to the information of each pulse event, thereby enhancing the alignment with radiation detection scenarios. Building upon this foundation, the study employs a sequential test method to further enhance the speed and accuracy of nuclide identification. During implementation, this study constructs a noise model based on authentic background measurement data, simulating radiation detection scenarios for single-nuclide, dual-nuclide, and multi-nuclide cases. The results demonstrate that in single-nuclide scenarios, when the relative intensity ratio between background noise and nuclide radiation reaches 1:7, the identification accuracy exceeds 91.3 %. Under conditions where the relative intensity ratio of background, ⁶⁰Co, and ¹³⁷Cs is 10:10:1, the detection rate for ¹³⁷Cs surpasses 81 %, while the detection rate for ⁶⁰Co remains approximately at 100 %. When the relative intensity ratio of background, ¹³³Ba, ⁶⁰Co, ¹³⁷Cs and ²²Na is set to 8:1:1:1:1, the respective identification rates for ¹³³Ba, ⁶⁰Co, ¹³⁷Cs, and ²²Na reach 99.6 %, 99.1 %, 84.4 %, and 81.3 %, with the false alarm rate for non-target nuclides staying below 0.8 %. These findings validate the feasibility of the proposed method and highlight its significant potential in rapid nuclide identification.