基于U-net网络的频散曲线自动拾取方法研究
Research on Automatic Picking Method of Dispersion Curve Based on U-net Network
卜凯旭 1姚振岸 1任望 1李红星 1王向腾 1毕升博 1陈振昊1
作者信息
- 1. 东华理工大学江西省防震减灾与工程地质灾害探测工程研究中心,江西南昌 330013
- 折叠
摘要
频散曲线拾取是面波勘探的关键环节,旨在通过频散曲线反演出地下横波速度结构.然而目前频散曲线拾取工作主要依靠人工拾取,耗时耗力.为此,本文通过将频散曲线拾取问题看成是图像分割问题,引入U-net网络,发展出一种频散曲线的自动拾取方法.该方法使用频散能量图并使其作为数据集,使用人工手动拾取的频散曲线作为标签集;通过卷积神经网络经由上采样、下采样和跳层链接等步骤学习图片特征,实现频散曲线的自动拾取.模型测试结果验证了利用U-net网络提取频散曲线的准确性.最后本文将训练好的网络模型应用于冰岛南部Ölfusa河岸的Arnarbaeli周边试验场地的实际数据频散曲线提取,并将提取结果与手动拾取的频散曲线进行对比.结果表明,利用U-net网络提取频散曲线预测速度快,预测512 × 512 × 3大小的图片耗时为96 ms,预测准确度高.
Abstract
The picking up of dispersion curves is a crucial step in surface wave exploration,aiming to invert the velocity structure of underground shear waves through dispersion curves.However,currently the work of picking up dispersion curves mainly relies on manual picking,which is time-consuming and labor-intensive.Therefore,this article regards the problem of picking up dispersion curves as an image segmentation problem,introduces U-net networks,and develops an automatic method for picking up dispersion curves.This method uses a dispersion energy map as the dataset and manually picked dispersion curves as the la-bel set.By using convolutional neural networks to learn image features through upsampling,downsampling,and skip layer linking,automatic picking of dispersion curves is achieved.The model test results verified the accuracy of using U-net network to extract dispersion curves.Finally,this article applies the trained network model to the actual data dispersion curve extraction of the Arnabæli surrounding experimental site on the banks of the Ölfusá River in southern Iceland,and compares the extraction results with manually picked disper-sion curves.The results show that using U-net network to extract dispersion curves has a fast prediction speed,a prediction time of 96ms for images of size 512 × 512 ×3,and high prediction accuracy.
关键词
瑞雷波勘探/频散曲线拾取/深度学习/卷积神经网络/U-net网络/人工智能Key words
Rayleigh wave exploration/dispersion curve picking/deep learning/convolution-al neural network/U-net network/artificial intelligence引用本文复制引用
基金项目
国家自然科学基金青年项目(42004113)
江西省自然科学基金项目(20212BAB211003)
江西省自然科学基金项目(20232BAB213079)
东华理工大学江西省大学生创新训练项目(S202310405006)
出版年
2024
NETL
NSTL
万方数据