제목 미상

요약

Zhenghao Liu, Yuhuai Cheng, Xiaofei Jiang — Journal of Instrumentation · Abstract Accurate localization of radioactive sources in large-scale unknown environments is an important problem in nuclear emergency response and radiation protection. Conventional localization methods under wide-area sparse-measurement conditions are often sensitive to initial values and may suffer from unstable convergence during global search. In addition, purely data-driven deep-learning methods usually rely on large-scale offline training datasets. To address these issues, this paper proposes a two-stage physics-informed neural network framework, termed TS-PINN, for radioactive source localization in large-scale environments. The proposed method decomposes the localization task into two stages: global coarse search and local fine refinement. In Stage I, a PINN is trained over the entire search domain to obtain a coarse estimate of the source position. In Stage II, a local subdomai