First-break picking technology based on level set method and U-Net
-
摘要: 针对使用U-Net网络拾取地震初至存在的拾取结果不连续等问题，提出了一种融合水平集方法的U-Net网络结构模型。首先基于U-Net网络提取地震图像特征图，逐像素计算损失，然后应用水平集方法对特征图计算水平集损失，最后将两者的加权和作为最终的损失函数。改进后的网络模型采用水平集方法保证了地震道之间初至的连续性，同时保留了U-Net网络“端到端”精细化分类的能力；采用Focal Loss损失函数缓解地震数据样本类别不均衡问题；在训练过程采用混合模拟数据和实际数据的方法克服训练集数据不足问题。经正演模拟和实际地震资料测试结果表明，该网络模型不仅精确度较高，而且较好地适应在低信噪比环境下地震信号的初至拾取。Abstract: In order to solve the discontinuous problems in the seismic first-break breaking using the U-Net network, a variantone based on level set method is proposed. Firstly, the feature map of seismic image is extracted based on U-Net network, and the loss is calculated pixel by pixel. Then the level set loss is calculated by using the level set method. Finally, the weighted sum of the two is taken as the final loss function. The improved network uses the level set method to ensure the continuity of first breaks between the adjacent seismic channels, and retains its ability of "end-to-end" fine classification. The Focal Loss function is used to alleviate the imbalance of sample categories of the training set. The mixed training set combined with the synthetic data and the actual field data is used to overcome the training set insufficiency problem. The results tested by the theoretical data and actual seismic data show that the improved U-Net network can improve the picking accuracy and robustness in low signal-to-noise ratio seismic data.
-
Key words:
- first break picking /
- level set /
- semantic segmentation /
- U-Net
-
图 7 不同信噪比下模拟数据拾取结果
Figure 7. First break picking results of simulated data with different SNR noise
表 1 超参数优化参数
Table 1. Hyperparametric values
超参数 方法/值 激活函数 ReLU、Randomized Leaky ReLU、ELU 优化器 Adamx、Adamw、Adadelta 学习率 0.01~0.001 
下载: 导出CSV
表 2 实际数据拾取结果统计
Table 2. The statistical result picked using actual data
模型 拾取准确率/% 拾取误差方差 U-Net 71.8 98.1 改进后U-Net 96.8 17.2
下载: 导出CSV
-
[1] 刘晗, 张建中. 微震信号自动检测的STA/LTA算法及其改进分析[J]. 地球物理学进展, 2014, 29(4): 1708-1714. https://www.cnki.com.cn/Article/CJFDTOTAL-DQWJ201404029.htmLiu Han, Zhang Jianzhong. STA/LTA algorithm analysis and improvement of microseismic signal automatic detection[J]. Progress in Geophysics, 2014, 29(4): 1708-1714. https://www.cnki.com.cn/Article/CJFDTOTAL-DQWJ201404029.htm [2] He D K, Chen C. A new method based on the surface consistence for picking the first arrival time of the seismic data[C]//2012 IEEE 11th International Conference on Signal Processing. October 21-25, 2012, Beijing, China. IEEE, 2012: 2292-2294. [3] 郑晶, 曹子原, 姜天琪, 等. 基于深度信念神经网络的微震波到时拾取方法[J]. 矿业科学学报, 2018, 3(6): 521-526. http://kykxxb.cumtb.edu.cn/article/id/180Zheng Jing, Cao Ziyuan, Jiang Tianqi, et al. Deep belief neural network-based arrival picking for microseismic data[J]. Journal of Mining Science and Technology, 2018, 3(6): 521-526. http://kykxxb.cumtb.edu.cn/article/id/180 [4] 朱盼盼, 李帅朋, 张立强, 等. 基于多任务学习的高分辨率遥感影像建筑提取[J]. 地球信息科学学报, 2021, 23(3): 514-523. https://www.cnki.com.cn/Article/CJFDTOTAL-DQXX202103015.htmZhu Panpan, Li Shuaipeng, Zhang Liqiang, et al. Multitask learning-based building extraction from high-resolution remote sensing images[J]. Journal of Geo-Information Science, 2021, 23(3): 514-523. https://www.cnki.com.cn/Article/CJFDTOTAL-DQXX202103015.htm [5] Papadomanolaki M, Vakalopoulou M, Karantzalos K. A deep multi-task learning framework coupling semantic segmentation and fully convolutional LSTM networks for urban change detection[J]. IEEE Transactions on Geoscience and Remote Sensing, 2021, 59(9): 7651-7668. doi: 10.1109/TGRS.2021.3055584 [6] Ma Y Y, Cao S Y, Rector J W, et al. Automated arrival-time picking using a pixel-level network[J]. GEOPHYSICS, 2020, 85(5): 415-423. doi: 10.1190/geo2019-0792.1 [7] Shelhamer E, Long J, Darrell T. Fully convolutional networks for semantic segmentation[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 39(4): 640-651. doi: 10.1109/TPAMI.2016.2572683 [8] Badrinarayanan V, Kendall A, Cipolla R. SegNet: A deep convolutional encoder-decoder architecture for image segmentation[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 39(12): 2481-2495. doi: 10.1109/TPAMI.2016.2644615 [9] Ronneberger O, Fischer P, Brox T. U-Net: convolutional networks for biomedical image segmentation[C]//Medical Image Computing and Computer-Assisted Intervention-MICCAI 2015, 2015: 234-241. [10] 刘佳楠, 武杰. 基于全卷积神经网络的地震初至波拾取[J]. 信息技术与网络安全, 2018, 37(11): 58-63. https://www.cnki.com.cn/Article/CJFDTOTAL-WXJY201811014.htmLiu Jianan, Wu Jie. Seismic first arrival picking based on fully convolutional networks[J]. Information Technology and Network Security, 2018, 37(11): 58-63. https://www.cnki.com.cn/Article/CJFDTOTAL-WXJY201811014.htm [11] 丁建群, 何永清, 李伟波, 等. 基于U-Net网络交错网格改进的地震数据初至自动拾取方法[C]//SPG/SEG南京2020年国际地球物理会议论文集(中文). 石油地球物理勘探, 2020: 955-958. [12] 杨双瑜, 郭小龙, 盛冠群. 基于U-Net和DenseNet的微地震初至拾取[C]//SPG/SEG南京2020年国际地球物理会议论文集(中文). 石油地球物理勘探, 2020: 1162-1165. [13] 陈德武, 杨午阳, 魏新建, 等. 一种基于改进的U-Net网络的初至自动拾取研究[J]. 地球物理学进展, 2021, 36(4): 1493-1503. https://www.cnki.com.cn/Article/CJFDTOTAL-DQWJ202104017.htmChen Dewu, Yang Wuyang, Wei Xinjian, et al. Research on first-break automatic picking based on an improved U-Net network[J]. Progress in Geophysics, 2021, 36(4): 1493-1503. https://www.cnki.com.cn/Article/CJFDTOTAL-DQWJ202104017.htm [14] David Cova, 刘洋, 丁成震, 等. 人工智能和视速度约束的地震波初至拾取方法[J]. 石油地球物理勘探, 2021, 56(3): 419-435, 409. https://www.cnki.com.cn/Article/CJFDTOTAL-SYDQ202103001.htmDavid Cova, Liu Yang, Ding Chengzhen, et al. First break picking method based on artificial intelligence and apparent velocity constraint[J]. Oil Geophysical Prospecting, 2021, 56(3): 419-435, 409. https://www.cnki.com.cn/Article/CJFDTOTAL-SYDQ202103001.htm [15] Osher S, Sethian J A. Fronts propagating with curvature-dependent speed: Algorithms based on Hamilton-Jacobi formulations[J]. Journal of Computational Physics, 1988, 79(1): 12-49. doi: 10.1016/0021-9991(88)90002-2 [16] Chan T F, Vese L A. Active contours without edges[J]. IEEE Transactions on Image Processing, 2001, 10(2): 266-277. doi: 10.1109/83.902291 [17] Kim Y, Kim S, Kim T, et al. CNN-based semantic segmentation using level set loss[C]//2019 IEEE Winter Conference on Applications of Computer Vision. January 7-11, 2019, Waikoloa, HI, USA. IEEE, 2019: 1752-1760. [18] 张宇豪, 徐磊, 白一清. 基于融合U-Net和水平集的肝脏CT图像分割[J]. 电子测量技术, 2021, 44(9): 116-121. https://www.cnki.com.cn/Article/CJFDTOTAL-DZCL202109019.htmZhang Yuhao, Xu Lei, Bai Yiqing. Liver CT image segmentation based on fusion of U-Net and level set[J]. Electronic Measurement Technology, 2021, 44(9): 116-121. https://www.cnki.com.cn/Article/CJFDTOTAL-DZCL202109019.htm [19] Oktay O, Schlemper J, Folgoc L L, et al. Attention U-net: learning where to look for the pancreas[EB/OL]. arXiv: 1804.03999, 2018. https://arxiv.org/abs/1804.03999. [20] Lin T Y, Goyal P, Girshick R, et al. Focal loss for dense object detection[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2020, 42(2): 318-327. doi: 10.1109/TPAMI.2018.2858826 [21] 刘强. 基于时频变换的低信噪比地震信号初至拾取[J]. 地球物理学进展, 2020, 35(3): 995-999. https://www.cnki.com.cn/Article/CJFDTOTAL-DQWJ202003024.htmLiu Qiang. First brick pick of low SNR seismic signal based on time-frequency transformation[J]. Progress in Geophysics, 2020, 35(3): 995-999. https://www.cnki.com.cn/Article/CJFDTOTAL-DQWJ202003024.htm [22] 董烈乾, 王岩, 霍禹, 等. 基于中值滤波和seislet阈值法联合压制混叠噪声方法[J]. 物探化探计算技术, 2019, 41(6): 705-711. https://www.cnki.com.cn/Article/CJFDTOTAL-WTHT201906001.htmDong Lieqian, Wang Yan, Huo Yu, et al. Blending noise suppression based on Median filter and seislet domain threshold denoising approach[J]. Computing Techniques for Geophysical and Geochemical Exploration, 2019, 41(6): 705-711. https://www.cnki.com.cn/Article/CJFDTOTAL-WTHT201906001.htm [23] Louboutin M, Lange M, Luporini F, et al. Devito (v3.1.0): an embedded domain-specific language for finite differences and geophysical exploration[J]. Geoscientific Model Development, 2019, 12(3): 1165-1187. -
点击查看大图
图(9) / 表(2)
计量
- 文章访问数: 439
- HTML全文浏览量: 139
- PDF下载量: 39
- 被引次数: 0