Detecting the Bull’s-Eye Effect in Seismic Inversion Low-Frequency Models Using the Optimized YOLOv7 Model  

利用优化的YOLOv7模型自动检测地震反演低频模型中的‘牛眼’效应

作　　者：Jun Li Jia-bing Meng Pan Li 李军;孟佳兵;李攀(防灾科技学院,河北三河065201;河北省高校智慧应急应用技术研发中心,河北三河065201)

机构地区：[1]Institute of Disaster Prevention,Sanhe 065201,China [2]Hebei Province University Smart Emergency Application Technology Research and Development Center,Sanhe 065201,China

出　　处：《Applied Geophysics》2024年第4期766-776,880,881,共13页应用地球物理（英文版）

摘　　要：To detect bull’s-eye anomalies in low-frequency seismic inversion models,the study proposed an advanced method using an optimized you only look once version 7(YOLOv7)model.This model is enhanced by integrating advanced modules,including the bidirectional feature pyramid network(BiFPN),weighted intersection-over-union(wise-IoU),efficient channel attention(ECA),and atrous spatial pyramid pooling(ASPP).BiFPN facilitates robust feature extraction by enabling bidirectional information fl ow across network scales,which enhances the ability of the model to capture complex patterns in seismic inversion models.Wise-IoU improves the precision and fineness of reservoir feature localization through its weighted approach to IoU.Meanwhile,ECA optimizes interactions between channels,which promotes eff ective information exchange and enhances the overall response of the model to subtle inversion details.Lastly,the ASPP module strategically addresses spatial dependencies at multiple scales,which further enhances the ability of the model to identify complex reservoir structures.By synergistically integrating these advanced modules,the proposed model not only demonstrates superior performance in detecting bull’s-eye anomalies but also marks a pioneering step in utilizing cutting-edge deep learning technologies to enhance the accuracy and reliability of seismic reservoir prediction in oil and gas exploration.The results meet scientific literature standards and provide new perspectives on methodology,which makes significant contributions to ongoing eff orts to refine accurate and efficient prediction models for oil and gas exploration.为了检测地震反演低频模型中的“牛眼”异常,提出了一种利用优化的YOLOv7模型的前沿方法。该模型通过集成高级模块,包括双向特征金字塔网络(BiFPN)、加权交并比(Wise-IoU)、高效通道注意力(ECA)和空洞空间金字塔池化(ASPP)进行了增强。BiFPN促进了强大的特征提取,通过在不同网络规模之间实现双向信息流动,增强了模型捕获地震反演模型中复杂模式的能力。Wise-IoU通过其对交并比加权的方法,提升了储层特征定位的细腻度和准确性。ECA优化了通道间的相互作用,促进了有效的信息交换,并改善了模型对微妙反演细节的总体响应。此外,ASPP模块在多个尺度上战略性地处理空间依赖性,进一步提升了模型识别复杂储层结构的能力。通过协同集成这些高级模块,提出的模型不仅在检测“牛眼”异常方面展现了优越性能,也标志着利用最先进的深度学习技术来增强油气勘探中地震储层预测的准确性和可靠性方面的开创性步骤。结果不仅符合科学文献标准,还为方法论提供了新的视角,为精炼更准确、更高效的油气勘探预测模型的持续努力做出了重大贡献。

关 键 词：bull’s-eye YOLO bidirectional feature pyramid network weighted intersection-over-union atrous spatial pyramid pooling 

分 类 号：P631.4[天文地球—地质矿产勘探]