基于深度卷积神经网络（DCNN）的γ能谱分析新技术研究

γ能谱分析是辐射测量中的关键基础技术。然而已有的γ能谱分析技术均有明显缺陷，在测量对象放射性较弱、辐射本底较强、探测器能量分辨率较差等情况下难以给出准确分析结果。近期，深度学习技术的快速发展及其在图像识别、自然语言处理等领域的成功应用经验为突破现有γ能谱分析理论框架、提高复杂条件下能谱分析精度提供了可能性。本项目拟基于当前深度学习中最重要的DCNN模型，利用其强大的特征提取性能和拓扑不变特性开展γ能谱分析新技术研究，重点解决数据预处理算法、专用DCNN设计、DCNN数值寻优与超参数调优算法、性能验证策略等方面的科学难题。本项目的研究成果将为实际应用中弥补现有技术缺陷，实现更准确γ射线测量奠定基础。

γ spectrum analysis is the key basic technology in radiation measurement. However, all the existing γ spectrum analysis technologies have obvious defects. It is difficult for them to give accurate analysis results in the case of weak radioactivity, strong radiation background and poor energy resolution of the detector. Recently, the rapid development of deep learning technology and its successful application experience in image recognition, natural language processing and other fields provide the possibility to break through existing theoretical framework and improve the accuracy of γspectrum analysis under complex conditions. Based on the most important DCNN model in current deep learning field, this project plans to carry out the novel γ spectrum analysis technology research by using its powerful feature extraction performance and topology invariant characteristics. This project will focuses on solving the scientific problems in data preprocessing algorithm, special DCNN design, DCNN numerical optimization and super parameter optimization algorithm, performance verification strategy, etc. The research results of this project will lay the foundation for the practical application to make up the existing technical defects and achieve more accurate γ ray measurement.