基于Resnet神经网络的地层核素能谱快速识别方法研究

Method Research of Resnet Neural Network-based Fast Identification for Stratigraphic Nuclide Energy Spectrum

铀裂变瞬发中子测井技术在实际操作中易受地层环境影响,致使铀矿的解释含量误差较大,故亟需发展地层环境核素快速识别方法,实现环境校正。文章针对传统的伽马能谱核素识别方法预处理繁琐、时间复杂度高、识别效率低、易受众多不可控因素影响等问题,提出了一种基于Resnet神经网络模型的伽马能谱核素识别方法。在网络的构建上,根据伽马能谱数据的特点,选取小卷积核多层卷积的结构,保护输入能谱特征,减少参数数量,提高网络的识别精准度与训练效率。并通过蒙特卡罗(MC)方法获得核素能谱仿真数据,将一维能谱信号转化为二维能谱灰度图像作为神经网络的训练样本,进行对比分析实验。实验结果表明该方法可精准地识别单源、双源及三源核素,从而更好地保护能谱数据的原始特征,在保证识别精度的同时提高识别方法的运算效率。

Uranium fission transient neutron logging technology in practice is sensitive to the influence of the stratigraphic environment,resulting in a large error in the interpretation of uranium content,so there is an urgent need to develop a rapid identification method of stratigraphic environmental nuclides to achieve environmental correction.The paper proposed a gamma energy spectrum nuclide identification method based on the Resnet residual neural network model to solve the problems of tedious pre-processing,high time complexity,low identification efficiency and susceptibility to many uncontrollable factors in the traditional gamma energy spectrum nuclide identification method.In the construction of the network,according to the characteristics of gamma energy spectrum data,the structure of multi-layer convolution with small convolution kernels was selected to protect the input energy spectrum features and reduce the number of parameters to improve the recognition accuracy and training efficiency of the network.And the simulated data of nuclide energy spectrum was obtained by Monte Carlo(MC)method,and the one-dimensional energy spectrum signal was transformed into two-dimensional energy spectrum grayscale image as the training sample of the neural network for comparative analysis experiments.The experimental results show that the method can accurately identify single-source,dual-source and triple-source nuclides,which can better protect the original features of the energy spectrum data and improve the computational efficiency of the identification method while ensuring the identification accuracy.