| 不同结构深度神经网络的时间域航空电磁数据成像性能分析 |
| |
| 引用本文： | 李金峰,刘云鹤.不同结构深度神经网络的时间域航空电磁数据成像性能分析[J].世界地质,2020,39(1):159-166. |
| |
| 作者姓名： | 李金峰 刘云鹤 |
| |
| 作者单位： | 吉林大学地球探测科学与技术学院,长春,130026 |
| |
| 基金项目： | 中央级公益性科研院所基本科研业务费专项经费项目(JYYWF20180103)。 |
| |
| 摘 要： | 时间域航空电磁系统采样密集,数据量大,所以在该领域较为实用的数据处理方法主要为一维反演和电阻率成像法。笔者从成像问题出发,建立了庞大的数据模型训练集,研究并分析了不同结构的神经网络的成像精度。通过对比分析测试结果,获得了在一定条件下适用于航空电磁成像的最优网络模型结构,包含其神经元个数和层数等信息。本文采用早停法训练神经网络,压制数据中噪声对成像结果的影响。
|
| 关 键 词： | 深度神经网络 成像 地球物理 电磁数据 |
Performance analysis of different structure deep neural networks in time-domain airborne EM data imaging |
| |
| LI Jin-feng,LIU Yun-he.Performance analysis of different structure deep neural networks in time-domain airborne EM data imaging[J].World Geology,2020,39(1):159-166. |
| |
| Authors: | LI Jin-feng LIU Yun-he |
| |
| Institution: | (College of Geo-exploration Science and Technology,Jilin University,Changchun 130026,China) |
| |
| Abstract: | The time-domain airborne EM system is with intensive sampling and can generate large amount of data,so the practical data processing methods are mainly one-dimensional inversion and resistivity imaging.Taking imaging problem as the research object,the authors establish a huge data-model training set and analyze the imaging accuracy of neural networks with different structures.By comparing the test results,the optimal network model structure under certain conditions has been obtained,which contains the information on number of neurons and layers.The early stopping method is used to train the neural network to suppress the influence of noise on the imaging results. |
| |
| Keywords: | deep neural network imaging geophysics EM data |
| 本文献已被 CNKI 维普 万方数据 等数据库收录！ |
