地震反演低频模型中的“牛眼”效应严重影响反演结果的精度和可靠性,尤其在弱对比度、模糊边界和复杂地质背景下难以精准识别。为解决这一问题,在以往研究的基础上,提出了一种轻量级、高鲁棒性的目标检测模型MWLB-YOLOv11。该模型基于YOLOv11,集成了四项关键模块以提升检测性能。通过采用MobileNetV4作为主干网络,模型借助更深的层次结构和参数优化增强了特征提取能力。为了扩大感受野并捕捉低频结构信息,设计了C3k2-WTConv模块,同时避免了过度参数化的问题。在此基础上,引入了大核可分离注意力机制LSKAttention,进一步提升了模型对空间上下文的建模能力,并有效控制了计算开销。此外,BiFPN特征融合结构促进了不同尺度间的信息交互,优化了小目标检测性能。该模型针对地震图像中的“牛眼”类环状异常进行了优化,实验结果表明,MWLB-YOLOv11在精度、效率和鲁棒性上优于传统方法,能够更准确地检测和定位牛眼效应区域。该模型为地震反演分析及其他异常检测任务提供了有效的解决方案,展现了广泛的应用前景。
Abstract
The "Bull's-Eye" effect in low-frequency seismic inversion models severely impacts the accuracy and reliability of inversion results, particularly under conditions of low contrast, blurred boundaries, and complex geological backgrounds, where precise identification becomes challenging. To address this issue, this paper proposes a lightweight and highly robust object detection model, MWLB-YOLOv11. Based on YOLOv11, the model integrates four key modules to enhance detection performance. By adopting MobileNetV4 as the backbone network, the model benefits from a deeper architecture and parameter optimization to improve feature extraction capabilities. To expand the receptive field and capture low-frequency structural information, the C3k2-WTConv module is introduced, while avoiding excessive parameterization. Additionally, a Large Separable Kernel Attention (LSKAttention) mechanism is incorporated to further enhance spatial context modeling while maintaining computational efficiency. The BiFPN feature fusion structure facilitates information exchange across multiple scales, improving performance in small object detection. The model is optimized specifically for detecting ring-shaped "Bull's-Eye" anomalies in seismic images. Experimental results demonstrate that MWLB-YOLOv11 outperforms traditional methods in terms of accuracy, efficiency, and robustness, enabling more precise detection and localization of Bull's-Eye regions. This model offers an effective solution for seismic inversion analysis and other anomaly detection tasks, showcasing broad application potential.
关键词
地震反演 /
牛眼效应 /
YOLOv11 /
MobileNetV4 /
WTConv /
LSKAttention /
BiFPN
Key words
Seismic Inversion /
Bull's-Eye Effect /
YOLOv11 /
MobileNetV4 /
WTConv /
LSKAttention /
BiFPN
{{custom_sec.title}}
{{custom_sec.title}}
{{custom_sec.content}}
参考文献
[1] Wang S,Lin C.The analysis of seismic data structure and oil and gas prediction[J].Applied Geophysics,2004,1(2):75-82.
[2] Sun Q,Zong Z.Building initial model for seismic inversion based on semi‐supervised learning[J].Geophysical Prospecting,2024,72(5):1800-1815.
[3] 孔炜,杨瑞召,彭苏萍.地震多属性分析在煤田拟声波三维数据体预测中的应用[J].中国矿业大学学报,2003(4):105-108.
[4] Nie Z,Feng B,Wang H,et al.Facies-Constrained Kriging Interpolation Method for Parameter Modeling[J].Remote Sensing,2024,17(1):102-102.
[5] Pochet A,Diniz B H P,Lopes H,et al.Seismic Fault Detection Using Convolutional Neural Networks Trained on Synthetic Poststacked Amplitude Maps[J].IEEE Geoscience and Remote Sensing Letters,2019,16(3):352-356.
[6] 方路平,何杭江,周国民.目标检测算法研究综述[J].计算机工程与应用,2018,54(13):11-18+33.
[7] Liu W,Anguelov D,Erhan D,et al.SSD:Single Shot MultiBox Detector[J].CoRR,2015,abs/1512.02325
[8] Redmon J,Divvala K S,Girshick B R,et al.You Only Look Once:Unified,Real-Time Object Detection[J].CoRR,2015,abs/1506.02640.
[9] Bingnan Y,Jiaxin L,Zhaozhao Y,et al.AIE-YOLO:Auxiliary Information Enhanced YOLO for Small Object Detection[J].Sensors,2022,22(21):8221-8221.
[10] 石燕妮,王武魁,吴明晶,等.MAFF-YOLO:面向造林验收的明穴目标检测模型[J].北京林业大学学报,2025,47(4):142-154.
[11] Finder S E,Amoyal R,Treister E,et al.Wavelet Convolutions for Large ReceptiveFields[J].arXiv preprint arXiv:2407.05848,2024.
[12] Wai K L,Lai-Man P,Ur A Y R.Large Separable Kernel Attention:Rethinking the Large Kernel Attention design in CNN[J].Expert Systems With Applications,2024(2):121352. 1-121352.15.
[13] M Tan,R Pang,QV Le.EfficientDet:Scalable and efficient object detection[J].Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition,2020,10778-10787.
基金
中央高校基本科研业务费研究生科技创新基金资助,项目编号：ZY20260323; 廊坊市科技计划自筹经费项目“基于深度学习的井震联合分析在岩性预测中的应用”,项目编号：2024011013
