Automatic velocity analysis method based on a joint model
Journal of Seismic Exploration · 18.06.2026 · 35 (4) · 025010141
DOI: 10.36922/jse025010141 ↗Kurz gefasst
Fachartikel in der Zeitschrift Journal of Seismic Exploration, online erschienen am 18.06.2026. Verfasst von Sixin Zhu und 3 weiteren Autor:innen.
Abstract (Originalsprache)
Accurate seismic velocity analysis is essential for normal moveout correction, stacking, migration imaging, and subsequent seismic interpretation. However, manual velocity picking is time-consuming and depends strongly on interpreter experience, while existing automatic methods may miss weak and divergent energy clusters in deep velocity spectra with low signal-to-noise ratios. To address these limitations, this study proposes a joint detection and prediction framework for automatic velocity spectrum picking. In the detection stage, the You Only Look Once detector is improved by incorporating a Squeeze-and-Excitation attention mechanism and an EfficientNetV2 backbone, which enhances the recognition of weak and multi-scale energy clusters. In the prediction stage, a Grey-Wolf-Optimization-enhanced bidirectional long short-term memory network is used to learn the ordered evolution of energy-cluster coordinates along the time axis and to suppress anomalous picks. The proposed method was evaluated using field seismic velocity spectra from an oilfield in eastern China. To reduce the risk of spatial information leakage, the training, validation, and test data were selected from spatially separated areas under similar acquisition conditions. The method was compared with a traditional similarity-weighted calculation method and a convolutional-neural-network-based velocity-picking baseline. In addition to common detection indicators, shallow and deep root-mean-square velocity errors were used to assess the geophysical relevance of the picked velocity trends. The results show that the proposed framework improves weak-energy recognition, reduces missed and false picks, and generates velocity-picking curves that are more consistent with expert manual interpretation. Ablation experiments further demonstrate that the attention mechanism, feature extraction backbone, bidirectional sequence prediction model, and parameter optimization strategy each contribute to the overall performance. These findings suggest that the proposed framework provides a practical and robust approach for automatic seismic velocity analysis.
Zitieren
APA
Zhu, S., Hu, Y., Zhao, X., & Cai, S. (2026). Automatic velocity analysis method based on a joint model. Journal of Seismic Exploration, 35(4), 025010141. https://doi.org/10.36922/jse025010141
Vancouver
Zhu S, Hu Y, Zhao X, Cai S. Automatic velocity analysis method based on a joint model. Journal of Seismic Exploration. 2026;35(4):025010141. doi:10.36922/jse025010141
BibTeX
@article{Zhu2026,
author = {Zhu, Sixin and Hu, Yijie and Zhao, Xu and Cai, Shuo},
title = {Automatic velocity analysis method based on a joint model},
journal = {Journal of Seismic Exploration},
year = {2026},
volume = {35},
number = {4},
pages = {025010141},
doi = {10.36922/jse025010141}
}
EUVIMED