Received date: 2024-03-23
Online published: 2025-03-13
Copyright
Deep learning applied to seismic First Break (FB) picking has been developed for many years, and numerous researchers have used Image Semantic Segmentation Networks (ISSNs) for multi-channel FB picking. The existing seismic FB picking method based on ISSNs usually adopts two label calibration and FB determination methods, one is to divide the seismic signal into pre-FB and post-FB, and pick up the FB through mask segmentation; the other is to divide the seismic signal into FB and non-FB, and pick up the FB by extracting the highest confidence point of each trace. The former suffers from FB false pickup with localized regional continuity due to the mask edge blurring problem; The latter, because of the large proportion of positive and negative samples, tends to make the network hard fitting and cannot be applied to data with complex FB waveforms and large size. Based on this, a dual-channel mask interaction seismic FBs picking method is proposed, which ensures the network's FBs feature recognition ability by banded FBs range mask, and enhances the network's FBs accurate picking ability by linear preferred FBs mask, which effectively avoids the shortcomings of the existing methods. Theoretical experiments show that the method has good noise resistance and can be generalized to higher noise level data by training in low noise level data. When the method is applied to the field data, it achieves higher FB picking accuracy than the existing methods, and the number of traces with picking error of 0 ms is as high as 75.9%, which is 7.1%, 26.8%, and 15.6% higher than that of STUNet, SegNet, and Res-Unet, respectively, and greatly improves the efficiency of high-quality seismic FB picking. Meanwhile, the approach adopts a lightweight network model with high inference efficiency and easy engineering deployment, which has practical application value.
Xu ZHANG , Jian ZHANG , CuiCui SHI , Fei DENG , ZhengCai FAN . Seismic first break picking using dual-channel mask interaction[J]. Progress in Geophysics, 2025 , 40(1) : 131 -142 . DOI: 10.6038/pg2025HH0475
图5 理论数据测试集各噪声水平初至拾取对比图(a)理论地震数据;(b)0%噪声水平拾取结果;(c)5%噪声水平拾取结果;(d)15%噪声水平拾取结果;(e)30%噪声水平拾取结果;(f)50%噪声水平拾取结果;其中红点表示本文标签标定方法结果,蓝点表示初至前后标签标定方法结果. Fig 5 Comparison of first break picking at each noise level for the theoretical data test set (a) Theoretical seismic data; (b) 0% noise level pickup results; (c) 5% noise level pickup results; (d) 15% noise level pickup results; (e) 30% noise level pickup results; (f) 50% noise level pickup results; where the red dots indicate the results of the labeling calibration method in this paper, and the blue dots indicate the results of the labeling calibration method before and after the first break. |
表1 理论数据测试集各噪声水平数值结果对比Table 1 Comparison of numerical results for each noise level in the theoretical data test set |
| 噪声水平 | 双通道掩码标注方法 | 初至前后标注方法 | |||||||
|---|---|---|---|---|---|---|---|---|---|
| 平均样点误差/px | 平均时间误差/ms | 零误差占比/% | 推理时间处理个数/s | 平均样点误差/px | 平均时间误差/ms | 零误差占比/% | 推理时间处理个数/s | ||
| 注:px表示像素点(pixel). | |||||||||
| 0% | 0.41 | 0.82 | 88.90 | 24.3 | 0.44 | 0.88 | 81.89 | 24.9 | |
| 5% | 0.67 | 1.34 | 66.81 | 0.66 | 1.32 | 60.01 | |||
| 15% | 0.92 | 1.84 | 54.79 | 0.96 | 1.92 | 49.94 | |||
| 30% | 1.42 | 2.84 | 48.75 | 1.39 | 2.78 | 41.83 | |||
| 50% | 2.01 | 4.02 | 36.25 | 2.07 | 4.14 | 29.91 | |||
表2 各方法不同误差区间占比Table 2 Percentage of different error ranges for each method |
| 方法 | 误差占比 | |||
|---|---|---|---|---|
| 0 px占比/% | ≤5 px占比/% | ≤10 px占比/% | >15 px占比/% | |
| Res-Unet | 60.3 | 97.5 | 99.4 | 0.25 |
| Seg-Net | 49.1 | 95.8 | 99.2 | 0.16 |
| STUNet | 68.8 | 97.3 | 99.0 | 0.77 |
| Ours | 75.9 | 97.7 | 99.1 | 0.47 |
表3 各方法数值结果对比Table 3 Comparison of numerical results for each method |
| 方法 | 平均样点误差/px | 平均时间误差/ms | 计算复杂度/GMac | 推理时间处理个数/s | 训练时长/min |
|---|---|---|---|---|---|
| 注:px表示像素点(pixel), 粗加表示最优数值. | |||||
| Res-Unet | 0.89 | 3.56 | 125.04 | 14.7 | 108.2 |
| Seg-Net | 1.13 | 4.52 | 75.05 | 20.1 | 136.6 |
| STUNet | 0.81 | 3.24 | 24.60 | 22.9 | 69.2 |
| Ours | 0.74 | 2.96 | 2.97 | 46.6 | 56.4 |
感谢审稿专家提出的修改意见和编辑部的大力支持!
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
Zhu L G, Kang H, Lv M Z, et al. 2022. Lyken17/pytorch-OpCounter. https://github.com/Lyken17/pytorch-OpCounter.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
/
| 〈 |
|
〉 |