基于图像分割的盐丘识别算法研究

doi:10.19596/j.cnki.1001-246x.8588

摘要/Abstract

摘要：

提出一种基于图像分割的盐丘识别方法实现自动化、高精度地识别盐丘。该方法在原始U-Net网络基础上进行迁移学习, 加载预训练模型的SENet用作编码器的主干网络, 对地震图像中的盐丘特征进行增强, 突出图像中的重要特征, 抑制不重要的特征。针对盐丘分割任务特点, 引入Lovasz-Softmax损失函数进行标准化实验, 提升对盐丘边界的分割效果。在TGS盐体识别挑战赛提供数据集上的实验结果表明: 该方法最终在测试集上取得了97.5%的准确率和87.26%的交并比, 与UNet、USKNet相比交并比分别提升了12.59个百分点和1.6个百分点。

关键词: 盐丘识别, U-Net, SENet, 图像分割, Lovasz-Softmax损失

Abstract:

Salt dome identification is of great significance for oil and gas exploration. Many important resources are located near the salt dome, but manual identification is time-consuming, labor-intensive and subjective. To solve this problem, this paper proposes a salt dome identification method based on image segmentation to realize automatic and high-precision identification of salt dome. The method is based on the original U-Net network for migration learning. The SENet loaded with the pre training model is used as the backbone network of the encoder. The salt dome features in the seismic image are enhanced, the important features in the image are highlighted, and the unimportant features are suppressed. In addition, according to the characteristics of salt dome segmentation task, Lovasz Softmax loss function is introduced for standardization experiment to improve the segmentation effect of salt dome boundary. The experimental results on the data set provided by the TGS salt dome identification challenge show that the method has finally achieved 97.5% Accuracy and 87.26% IoU on the test set. Compared with UNet and USKNet, the IoU has increased by 12.59 percentage points and 1.6 percentage points, respectively, reflecting the effectiveness and universality of this method.

Key words: salt dome identification, U-Net, SENet, image segmentation, Lovasz-Softmax loss

娄莉, 张丰侠, 韩柏迅. 基于图像分割的盐丘识别算法研究[J]. 计算物理, 2023, 40(4): 511-518.

Li LOU, Fengxia ZHANG, Boxun HAN. Research on Salt Dome Identification Algorithm Based on Image Segmentation[J]. Chinese Journal of Computational Physics, 2023, 40(4): 511-518.

图/表 11

图1 Squeeze-and-Excitation (SE)模块[19]

Fig.1 Squeeze-and-Excitation (SE)module[19]

图2 SE-Unet网络结构图[18]

Fig.2 SE-Unet network structure[18]

图3 数据集样例图

Fig.3 Dataset samples

图4 数据增强

Fig.4 Data augmentation

图5 训练损失变化

Fig.5 Training loss change

图6 训练准确率变化

Fig.6 Training accuracy change

图7 地震图像分割结果

Fig.7 Segmentation results of seismic image

表1 不同主干网络的实验结果对比

Table 1 Comparison of experimental results of different backbone networks

主干网络	Accuracy/%	IoU/%
ResNet50	96.8	84.05
ResNet101	96.8	85.13
ResNet152	96.9	85.38
SE-ResNet152	97.1	86.12
SENet154	97.5	87.26

表2 不同损失函数的实验结果对比

Table 2 Comparison of experimental results of different loss functions

	Accuracy/%	IoU/%
Cross-Entropy	97.1	85.14
Lovasz-Softmax	97.5	87.26

表3 不同模型的实验结果对比

Table 3 Comparison of experimental results of different models

	Accuracy/%	IoU/%
U-Net	93.9	74.67
基于U-Net方法^[22]	94.0	79.24
PSPNet^[23]	93.2	75.27
USKNet^[24]	96.1	85.66
SeNet154-FPN^[25]	97.5	87.48
SE-Unet(本文方法)	97.5	87.26

表4 SeNet154-FPN与本文算法的参数量对比

Table 4 Comparison of parameters between SeNet154-FPN and the algorithm in this paper

	Params/M	IoU/%
SeNet154-FPN	893.8	87.48
SE-Unet(本文方法)	461.5	87.26

参考文献 25

1	马梓程, 帅爽, 谢菲, 等. 嵌于扎格罗斯山脉中的"宝石"——盐丘[J]. 中国资源综合利用, 2020, 38 (6): 86-87+123.
2	徐慧. 什么是地下盐丘?[J]. 资源环境与工程, 2016, 30 (4): 666- 667.
3	LECUN Y , BENGIO Y , HINTON G . Deep learning[J]. Nature, 2015, 521 (7553): 436- 444. DOI
4	ZHANG Qingfu , YAO Jun , HUANG Zhaoqin , et al. A multiscale deep learning model for fractured porous media[J]. Chinese Journal of Computational Physics, 2019, 36 (6): 665- 672.
5	DONG Peng, LIAO Xinwei. An Efficient approach for automatic well-testing interpretation based on surrogate model and deep reinforcement learning[J/OL]. Chinese Journal of Computational Physics, 2023, 40(1): 67-80.
6	XIAO Cong , ZHANG Shicheng , MA Xinfang , et al. Model-reduced autoregressive neural network for parameter inversion[J]. Chinese Journal of Computational Physics, 2022, 39 (5): 564- 578.
7	SHI Y , WU X , FOMEL S . Interactively tracking seismic geobodies with a deep-learning flood-filling network[J]. Geophysics, 2021, 86 (1): A1- A5.
8	ZHANG H , CHEN T , LIU Y , et al. Automatic seismic facies interpretation using supervised deep learningFacies interpretation by deep learning[J]. Geophysics, 2021, 86 (1): IM15- IM33. DOI
9	丰超, 潘建国, 李闯, 等. 基于深度神经网络的断层高分辨率识别方法[J/OL]. 地球科学, 48(8): 3044-3052.
10	DI H, WANG Z, ALREGIB G. Real-time seismic-image interpretation via deconvolutional neural network[C]//2018 SEG International Exposition and Annual Meeting, OnePetro, 2018: 2051-2055.
11	DI H, WANG Z, ALREGIB G. Deep convolutional neural networks for seismic salt-body delineation[C]//AAPG Annual Convention and Exhibition, 2018: 90323.
12	SHI Y, WU X, FOMEL S. Automatic salt-body classification using a deep convolutional neural network[M]//SEG Technical Program Expanded Abstracts 2018, Society of Exploration Geophysicists, 2018: 1971-1975.
13	WALDELAND A U, SOLBERG A. Salt classification using deep learning[C]//79th eage conference and exhibition 2017, European Association of Geoscientists & Engineers, 2017: 1-5.
14	WU X , LIANG L , SHI Y , et al. FaultSeg3D: Using synthetic data sets to train an end-to-end convolutional neural network for 3D seismic fault segmentation[J]. Geophysics, 2019, 84 (3): IM35- IM45. DOI
15	ZENG Y, JIANG K, CHEN J. Automatic seismic salt interpretation with deep convolutional neural networks[C]//Proceedings of the 20193rd International Conference on Information System and Data Mining, 2019: 16-20.
16	LI H , HU Q , MAO Y , et al. Deep learning-based model for automatic salt rock segmentation[J]. Rock Mechanics and Rock Engineering, 2022, 55 (6): 3735- 3747. DOI
17	张玉玺, 刘洋, 张浩然, 等. 基于深度学习的多属性盐丘自动识别方法[J]. 石油地球物理勘探, 2020, 55 (3): 475-483+467.
18	RONNEBERGER O, FISCHER P, BROX T. U-net: Convolutional networks for biomedical image segmentation[C]//International Conference on Medical Image Computing and Computer-assisted Intervention. Springer, Cham, 2015: 234-241.
19	HU J, SHEN L, SUN G. Squeeze-and-excitation networks[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2018: 7132-7141.
20	BERMAN M, TRIKI A R, BLASCHKO M B. The lovász-softmax loss: A tractable surrogate for the optimization of the intersection-over-union measure in neural networks[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2018: 4413-4421.
21	HE K, ZHANG X, REN S, et al. Deep residual learning for image recognition[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2016: 770-778.
22	卢新瑞, 黄捍东, 李帅, 等. 基于U-Net的盐体识别方法[J]. 计算物理, 2020, 37 (3): 327- 334.
23	ZHAO H, SHI J, QI X, et al. Pyramid scene parsing network[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2017: 2881-2890.
24	程国建, 刘宁, 万晓龙, 等. 结合SKNet与U-Net的盐体识别方法[J]. 油气地质与采收率, 2022, 29 (1): 62- 68.
25	LIU B, JING H, LI J, et al. Image segmentation of salt deposits using deep convolutional neural network[C]//2019 IEEE International Conference on Systems, Man and Cybernetics. IEEE, 2019: 3304-3309.