针对人工解释地震资料耗时长、效率低、受主观因素影响较大的不足，提出了一种基于ResUNet和全连接条件随机场（dense conditional random field，Dense CRF）模型的裂缝识别方法。该方法首先使用ResUNet模型提取地震振幅数据体中裂缝的不同分辨率的特征，实现地震裂缝识别；然后利用Dense CRF模型进一步优化识别结果，从而实现地震裂缝的精准识别。将该方法与传统UNet、ResUNet模型在合成地震振幅数据体和F3工区地震数据体进行了实验比较，结果表明运用所提方法识别的裂缝更准确、裂缝尺寸更细、连续性更好。
Aiming at the problems of time-consuming, low efficiency, and high subjective influence in artificial interpretation of seismic data, a crack identification method based on ResUNet and dense conditional random field (Dense CRF) model is proposed. First, the method uses the ResUNet model to extract the features of different resolution levels from the cracks in the seismic amplitude data volume to achieve seismic crack identification, then it uses the Dense CRF model to further optimize the recognition results, so as to achieve accurate recognition of seismic cracks. The proposed method is compared with the traditional UNet and ResUNet methods based on the synthetic seismic amplitude data volume and the seismic amplitude volume data of the F3 work area. Experimental results show that the proposed method performs higher accuracy, finer size and better continuity in crack identification.
[1] Ata E, Michelena R J. Mapping distribution of fractures in a reservoir with P-S converted waves[J]. The Leading Edge, 1995, 14(6): 664-676
[2] Dok R V, Gaiser J E, Jackson A R, et al. 3-D converted-wave processing: wind river basin case history[J]. Seg Technical Program Expanded Abstracts, 1949, 16(1): 1206.
[3] Gaiser J E, Dok R. Borehole calibration of PS-waves for fracture characterization: Pinedale field, Wyoming[J]. Seg Technical Program Expanded Abstracts, 2005, 24(1): 2668.
[4] Grechka V, Kachanov M. Seimic characterization of multiple fracture sets: does orthotropy suffice?[J]. Geophysics, 2006, 71(3): D93-D105
[5] Convers C, Hanitzsch C, Curia D, et al. Elastic parameter estimation for the identification of sweet spots, Vaca Muerta Formation, Neuquén Basin, Argentina[J]. The Leading Edge, 2017, 36(11): 948a1-948a10.
[6] Jaiswal P, Varacchi B, Ebrahimi P, et al. Can seismic velocities predict sweet spots in the Woodford Shale? a case study from McNeff 2–28 Well, Grady County, Oklahoma[J]. Journal of Applied Geophysics, 2014, 104: 26-34.
[7] Aliouane L, Ouadfeul S A. Sweet spots discrimination in shale gas reservoirs using seismic and well-logs data. A case study from the Worth basin in the Barnett shale[J]. Energy Procedia, 2014, 59: 22-27.
[8] Zeng Q C, Deng Y, Hou H X, et al. Quantitative prediction of shale gas sweet spots based on seismic data in Lower Silurian Longmaxi Formation, Weiyuan Area, Sichuan Basin, SW China[J]. Petroleum Exploration and Development, 2018, 45(3): 422-430.
[9] Wu X M, Liang L M, Shi Y Z, et al. FaultSeg3D: using synthetic data sets to train an endto-end convolutional neural network for 3D seismic fault segmentation[J]. Geophysics, 2019: 1-36.
[10] Hu G, Hu Z, Liu J, et al. Seismic fault interpretation using deep learning-based semantic segmentation method[J]. IEEE Geoscience and Remote Sensing Letters, 2020, 99: 1-5.
[11] Zhao T, Vikram J, Atish R, et al. A comparison of classification techniques for seismic facies recognition[J]. Interpretation, 2015, 3(4): SAE29-SAE58.
[12] Huang L, Dong X S, Clee T D. A scalable deep learning platform for identifying geologic features from seismic attributes[J]. The Leading Edge, 36(3), 2017: 249-256.
[13] Xiong W, Ji X, Ma Y, et al. Seismic fault detection with convolutional neural network[J]. Geophysics, 2018, 83(5): 97-103.
[14] Kai Z, Zuo W, Chen Y, et al. Beyond a Gaussian denoiser: residual learning of deep CNN for image denoising[J]. IEEE Transactions on Image Processing, 2016, 26(7): 3142-3155.
[15] Agarap A F. Deep learning using rectified linear units (ReLU)[J/OL]. arXiv preprint arXiv: 1803.08375, 2018[2020-08-30]. https://arxiv.org/abs/1803.08375.
[16] Ronneberger O, Fischer P, Brox T. U-Net: convolutional networks for biomedical image segmentation[J]. Cham: Springer, 2015.
[17] Rayner A. Keras[J]. School Librarian, 2013, 24(8): 259-261.
[18] Finney D J. Probit analysis: a statistical treatment of the sigmoid response curve[M]. Cambridge: Cambridge University Press, 1952.
[19] Desmaison A, Bunel R, Kohli P, et al. Efficient continuous relaxations for dense CRF[C]//European Conference on Computer Vision. Cham: Springer, 2016: 818-833.
[20] Wu X M, Dave H. 3D seismic image processing for faults[J]. Geophysics, 2016, 81(2): IM1-IM11.
[21] Salakhutdinov R, Hinton G E. Replicated softmax: an undirected topic model[C]//Advances in Neural Information Processing Systems 22: 23rd Annual Conference on Neural Information Processing Systems 2009. British Columbia, Canada. 2009: 1607-1614.
[22] Srivastava N, Hinton G, Krizhevsky A, et al. Dropout: a simple way to prevent neural networks from overfitting[J]. The journal of Machine Learning Research, 2014, 15(1): 1929-1958.
[23] Kingma D P, Ba J. Adam: a method for stochastic optimization[J/OL]. arXiv preprint arXiv: 1412.6980, 2014.[2020-08-30] https://arxiv.org/abs/1412.6980.
[24] Zhi L, Wu Q, Yun Z, et al. Adaptive least squares support vector machines filter for hand tremor canceling in microsurgery[J]. International Journal of Machine Learning and Cybernetics, 2011, 2(1): 37-47.
[25] Diebold F X, Mariano R S. Comparing predictive accuracy[J]. Journal of Business & Economic Statistics, 2002, 20(1): 134-144.
