微弱核辐射信号稀疏重构模型及核素快速识别方法研究

Non-connect detection and recognition of radioactive material are open problems in nuclear arms control examination, the nuclear material smuggling and nuclear nonproliferation, etc. Because of the environmental background noise and Compton scattering events interference have great effect to week nuclear radiation signals detection, and gamma energy spectrum are easily distorted by the effect of shield reflection/attenuation or vehicle movement in complex testing situation. In this project, a novel week nuclear radiation signal sparse reconstruction method and a practical spectral distortion correction method are employed to improve the week signal detection and recognition performance under strong noise background. Through the characteristics analyzing and statistical modeling of typical noise, the noises can be classified into two categories: structured and unstructured. An over-complete dictionary is constructed by using the structured noise signal and the theoretical nuclear radiation signal waveform, which used for representation of the potential true signal in noise background. Then, using the sequential Bayesian analysis method to identify valid nuclear pulses, and applying sparse signal reconstructing method extract the corrupted week nuclear radiation signal. By analyzing the spectral distortion interference factors in the testing situation, build the mathematical description between spectral distortion and interference factors, the spectral distortion interference factors in the testing situation is analyzed, and the mathematical description between spectral distortion and interference factors is established. Based on evidence theory take the advantage of the redundant information and complementary information to solve the energy spectrum measurement uncertainty, which caused by full-energy peak does not meet the Gaussian hypothesis in short time detection. Finally the types of radioactive nuclide are accurately recognized.

本项目针对非接触式放射性材料探测中由环境本底噪声和康普顿散射引起的微弱核辐射信号难以检测的问题，以及复杂检测情境下由屏蔽体反射、衰减或运载工具运动等因素引起的γ能谱畸变问题，以提高强噪声背景下信号检测能力为切入点，探索核辐射信号的稀疏重构和能谱畸变校正方法，使之满足放射性材料微弱信号提取与核素识别的需求。通过对典型噪声的特征分析和统计建模，将噪声分为结构化和非结构化噪声，利用结构化噪声和核辐射信号理论波形建立用于表征隐藏在含噪信号下本真信号的过完备字典，并基于序贯贝叶斯分析进行核辐射特征脉冲甄别，采用稀疏信号重构的方式，提取被噪声污染的微弱核辐射信号；研究检测情境中干扰因素对能谱畸变的影响规律及影响关系的数学描述，去除由确定性因素引起的能谱测量误差；并基于证据理论对冗余信息和互补信息进行融合，解决在短时检测中γ射线全能峰不满足高斯假设而引起的能谱量测不确定性，最终实现核素种类的准确识别。

针对微弱核脉冲时间序列信号提取和核素快速识别问题，通过分析典型探测器噪声和环境本底噪声特征，提出了基于Gabor变换和稀疏表示的脉冲信号提取模型，并采用多吸引子混沌电路检测脉冲发生时刻提高信号提取效率；利用迁移矩阵建模由环境温度变化引起的伽马能谱差异，并基于标准核素能谱矩阵构建稀疏表示字典，之后通过多任务学习实现能谱校正和复杂核素识别；基于视觉和核脉冲信号两个模态的信息，利用视觉传感获取运动物体的位置信息，并通过位置变换改进经典基于序贯贝叶斯核素快速识别算法，提出了用于运动微量核素快速识别的改进算法；采用稀疏表示将伽马能谱在区分性最好的稀疏原子上进行投影，提出强噪声测量情况下的核素快速识别方法。利用项目研究所提出的微弱脉冲信号检测、脉冲提取和核素快速识别方法，设计并实现了一套用于模拟运动放射源实验的直线运动平台系统验证了所提算法的可行性；利用项目成果设计的剂量率测量电路和核素识别装置应用到了核应急探测项目和核退役放射性检测项目中，验证了方法的有效性。