基于BP神经网络的多GM计数器γ辐射定向测量算法研究

针对放射源搜寻作业的角度定向测量需求，设计了由4个GM计数器和十字铅屏蔽结构组成的γ辐射定向测量装置，屏蔽体厚1 cm、径向宽7 cm。利用GEANT4模拟计算了定向测量装置对0.058~3 MeV单能光子的角响应，将其作为训练样本建立了基于BP神经网络的角度反演方法。设计了神经网络训练样本的输入输出参数模型，并比较了3种算法模型的角度分辨精度，其中样本映射重组算法效果最好，角度反演精度为±1.5°；引入计数涨落时，该方法也能实现约±10°以内的角度偏差。利用¹³⁷Cs源进行实测，结果显示角度偏差在±6.25°范围内。

Algorithm Research on Direction Finder for GM Counters γ Radiation Source Based on BP Neural Network

To meet the requirement of angle directional measurement for radiation source searching, a γ ray direction finder was designed, which consists of four GM counters and a cross lead shielding structure. The shielding body is 1 cm in thickness and 7 cm in radial width. The angle responses of the direction finder to 0.058-3 MeV single energy photons were simulated with GEANT4, and then used as training samples of the BP neural network algorithm to establish the angle inversion method. The input and output parameter models of neural network training samples were designed, and the angular resolution accuracies of three algorithms were compared. It is shown that the algorithm based on mapping and restructuring the training samples gives the best angle inversion accuracy of ±1.5°, and within ±10° when introducing counting fluctuation. The tests for the direction finder to ¹³⁷Cs radiation source show that its angular deviation is within ±6.25°.