基于深度学习的地震叠加速度谱自动拾取算法研究  

Research on deep learning-based automatic pickup algorithm for seismic stacked velocity spectra

作　　者：朱四新 孟凡可 姜彤 ZHU SiXin;MENG FanKe;JIANG Tong(North China University of Water Resources and Electric Power College of Geosciences and Engineering,Zhengzhou 450053,China)

机构地区：[1]华北水利水电大学地球科学与工程学院,郑州450053

出　　处：《地球物理学报》2024年第3期1223-1236,共14页Chinese Journal of Geophysics

基　　金：河南省科技攻关“多神经网络断层速度自动拾取方法研究”(222102210310)资助。

摘　　要：为解决人工拾取地震叠加速度谱时耗时长、效率低等问题,本文提出了一种基于深度学习的地震速度谱自动拾取算法模型VSAP(Velocity Spectrum Accurate Pickup).该算法运用卷积神经网络Faster R-CNN模型构建的多分类任务拾取目标能量团,然后将初步拾取后的能量团坐标输入循环神经网络LSTM(Long-Short Term Memory)模型来进行目标能量团拾取时坐标的取舍和微调,最后输出模型分析和调整过的速度谱自动拾取图像.并通过实际的地震数据集拾取结果验证了该算法模型在叠加速度谱复杂信息的干扰中自动、准确拾取速度谱中能量团的能力,同时验证了该模型的准确性以及鲁棒性.经过改进,该算法模型有效地提高了速度谱拾取的效率和拾取精度.In order to address the issues of time-consuming and inefficient manual pickup of seismic stacked velocity spectra, this paper proposes a deep learning-based automatic pickup algorithm model called VSAP (Velocity Spectrum Accurate Pickup). The algorithm utilizes a convolutional neural network Faster R-CNN model to construct a multi-classification task for picking target energy clusters. Subsequently, the coordinates of the preliminary picked energy clusters are fed into a recurrent neural network LSTM (Long-Short Term Memory) model to refine and adjust the pickup coordinates of the target energy clusters. Finally, the model outputs the analyzed and adjusted automatically picked velocity spectrum images. The proposed algorithm model is validated through the pickup results obtained from real seismic datasets, demonstrating its ability to automatically and accurately pick energy clusters from velocity spectra, even in the presence of complex information interference in stacked velocity spectra. The accuracy and robustness of the model are also verified. With improvements made, the algorithm model effectively enhances the efficiency and precision of velocity spectrum pickup.

关 键 词：速度谱 自动拾取 卷积神经网络 循环神经网络 

分 类 号：P631[天文地球—地质矿产勘探]