波形
计算机科学
稳健性(进化)
特征提取
残余物
噪音(视频)
信号处理
实时计算
语音识别
数据挖掘
模式识别(心理学)
人工智能
算法
电信
生物化学
雷达
化学
图像(数学)
基因
作者
Xuan Jiang,Tong Shen,Mingjie Tan,Jianlin Wang
标识
DOI:10.1109/raiic59453.2023.10280928
摘要
Seismic P-wave first arrival picking is a fundamental and vital component of seismic monitoring, with its accuracy playing a critical role in subsequent data processing. In order to address the problems of low signal-to-noise ratio data pickup accuracy and poor robustness of the traditional seismic wave first arrivals, this paper proposes an improved U-Net3+ seismic P-wave first arrivals picking method. Compared with the traditional method, the model does not require manual setting of thresholds and pre-processing operations such as filtering of data, and only requires input of normalized three-component seismic waveform data to intelligently identify seismic P-wave first arrivals. To minimize information loss during seismic waveform extraction and enhance the utilization of effective feature information at the P-wave first arrival time, the residual unit is incorporated solely in the encoder and decoder part of U-Net3+. The model exhibits a lightweight structure and excels in feature extraction, enabling effective identification of low signal-to-noise ratio data. The results demonstrate hit rates of 81.4%, 86.5%, and 87.5% for data with signal-to-noise ratios below 5dB, 10dB, and 20dB, respectively. The average pickup error is 0.054s, outperforming mainstream deep learning methods like U-Net and U-Net3+. These findings present a novel approach for automated P-wave first arrival picking.
科研通智能强力驱动
Strongly Powered by AbleSci AI