A fault identification method for deep coalbed methane reservoirs based on an improved TransUNet

Authors

• GAO Congshun, School of Earth and Environment, Anhui University of Science and Technology, Huainan 232001, ChinaFollow
• WU Haibo, School of Earth and Environment, Anhui University of Science and Technology, Huainan 232001, ChinaFollow
• LIU Lei, School of Artificial Intelligence, Anhui University of Science and Technology, Huainan 232001, China
• WANG Zijian, School of Artificial Intelligence, Anhui University of Science and Technology, Huainan 232001, China
• HUANG Xuanyu, School of Artificial Intelligence, Anhui University of Science and Technology, Huainan 232001, China
• JIA Lilan, School of Earth and Environment, Anhui University of Science and Technology, Huainan 232001, China
• XU Aiguo, School of Earth and Environment, Anhui University of Science and Technology, Huainan 232001, China

Abstract

Objective Small-scale faults in deep coalbed methane (CBM) reservoirs significantly affect CBM enrichment and the deployment of development wells. However, deep seismic data tend to suffer from low signal-to-noise ratios (SNRs) and poor continuity of reflection events, while conventional methods are likely to miss small-scale faults. To address these limitations, this study proposes a fault identification method based on an improved TransUNet. Methods The proposed method is developed based on the R50-ViT-B_16 TransUNet architecture. A coordinate attention (CA) module is introduced before the skip connections enter the decoder to enhance the feature representation of the locations and directional continuity of faults. Meanwhile, a joint loss function combining Focal Loss and Dice Loss is adopted to mitigate the class imbalance caused by the low proportion of fault pixels. Results and Conclusions Comparative and ablation experiments on a synthetic seismic dataset indicate that the joint loss function improved the recall of fault pixels, while the CA module enhanced the feature representation of the spatial locations and directions of faults. Owing to their combined effects, the improved CA-TransUNet model achieved Dice coefficient, Intersection over Union (IoU), Precision, and Recall values of 0.7687, 0.6379, 0.7927, and 0.7431, respectively. Compared with the baseline TransUNet, the CA-TransUNet model yielded higher Dice, IoU, and Recall values, especially Recall, but slightly lower Precision. The CA-TransUNet model was further applied to measured seismic data from the Linxing Block in the Ordos Basin. The identified faults were primarily distributed near the target horizon, showing linear or inclined band-shaped patterns and corresponding to bending and dislocated seismic reflection events. After fault morphological constraints were introduced, point-like, lumpy, and boundary anomalies in the predictions were eliminated. The developed method based on the improved TransUNet can enhance the identification accuracy of small-scale faults in deep CBM reservoirs, thereby providing technical support for CBM exploration and development.

Keywords

deep coalbed methane (CBM), fault identification, deep learning, TransUNet, coordinate attention, joint loss function

DOI

10.12363/issn.1001-1986.26.06.0355

Recommended Citation

GAO Congshun, WU Haibo, LIU Lei, et al. (2026) "A fault identification method for deep coalbed methane reservoirs based on an improved TransUNet," Coal Geology & Exploration: Vol. 54: Iss. 9, Article 13.
DOI: 10.12363/issn.1001-1986.26.06.0355
Available at: https://cge.researchcommons.org/journal/vol54/iss9/13

Reference

[1] 王飞,程礼军,刘俊峰,等. 叠后地震属性识别页岩气储层裂缝研究及应用[J]. 煤田地质与勘探,2015,43(5):113−116. WANG Fei,CHENG Lijun,LIU Junfeng,et al. Research and application of post–stack seismic attributes in recognizing shale gas reservoir fracture[J]. Coal Geology & Exploration,2015,43(5):113−116.

[2] BAHORICH M,FARMER S. 3–D seismic discontinuity for faults and stratigraphic features: The coherence cube[J]. The Leading Edge,1995,14(10):1053−1058.

[3] MARFURT K J,KIRLIN R L,FARMER S L,et al. 3–D seismic attributes using a semblance–based coherency algorithm[J]. Geophysics,1998,63(4):1150−1165.

[4] WU Xinming,HALE D. Automatically interpreting all faults,unconformities,and horizons from 3D seismic images[J]. Interpretation,2016,4(2):T227−T237.

[5] GUO Bowen,LIU Lu,LUO Yi. Automatic seismic fault detection with convolutional neural network[C]//International Geophysical Conference. Beijing: Society of Exploration Geophysicists and Chinese Petroleum Society,2018: 1786−1789.

[6] GAO Kai,HUANG Lianjie,ZHENG Yingcai,et al. Automatic fault detection on seismic images using a multiscale attention convolutional neural network[J]. Geophysics,2022,87(1):N13−N29.

[7] 陈青青. 基于语义分割的地震图像断层检测方法研究[D]. 长春: 吉林大学, 2021. CHEN Qingqing. Research on the fault detection method of seismic image based on semantic segmentation[D]. Changchun: Jilin University, 2021.

[8] 芦凤明,孟瑞刚,张军华,等. UNet++和迁移学习相结合的复杂断裂识别方法研究[J]. 地球物理学进展,2022,37(3):1100−1111. LU Fengming,MENG Ruigang,ZHANG Junhua,et al. Research of complex fault recognition method based on UNet++ network and transfer learning technique[J]. Progress in Geophysics,2022,37(3):1100−1111.

[9] 张利霞,高俊涛,马强,等. 基于改进UNet++的地震断层识别方法研究[J]. 计算机技术与发展,2023,33(8):199−205. ZHANG Lixia,GAO Juntao,MA Qiang,et al. Research on seismic fault identification methods based on improved UNet++[J]. Computer Technology and Development,2023,33(8):199−205.

[10] CHEN Jieneng,LU Yongyi,YU Qihang,et al. TransUNet: Transformers make strong encoders for medical image segmentation[EB/OL]. arXiv: 2102.04306,2021.

[11] 王新,张薇,陈同俊,等. 基于CBAM–TransUNet的地震断层识别方法[J]. 煤炭学报,2025,50(2):1192−1202. WANG Xin,ZHANG Wei,CHEN Tongjun,et al. Seismic fault identification method based on CBAM–TransUNet[J]. Journal of China Coal Society,2025,50(2):1192−1202.

[12] WU Xinming,LIANG Luming,SHI Yunzhi,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.

[13] LIN Lei,ZHONG Zhi,CAI Zhongxian,et al. Automatic geologic fault identification from seismic data using 2.5D channel attention U–Net[J]. Geophysics,2022,87(4):IM111−IM124.

[14] SUN Qifeng,WANG Xin,NI Hongsheng,et al. Fault identification of U–Net based on enhanced feature fusion and attention mechanism[J]. Electronics,2023,12(12):2562.

[15] ZENG Lili,NIU Yixiao,REN Weijian,et al. A method for intelligent identification of faults in seismic using an attention–based ES–UNet network with model re–training learning[J]. Journal of Applied Geophysics,2024,223:105344.

[16] WANG Zhiwei,YOU Jiachun,LIU Wei,et al. Transformer assisted dual U–Net for seismic fault detection[J]. Frontiers in Earth Science,2023,11:1047626.

[17] TANG Zhanxin,WU Bangyu,WU Weihua,et al. Fault detection via 2.5D transformer U–Net with seismic data pre–processing[J]. Remote Sensing,2023,15(4):1039.

[18] ZHOU Tong,MA Yue,SUI Yuhan,et al. Fault transformer: An automatic fault detection algorithm on seismic images using a transformer enhanced neural network[J]. Interpretation,2024,12(3):SE55−SE64.

[19] ZU Shaohuan,ZHAO Penghui,KE Chaofan,et al. ResACEUnet: An improved transformer UNet model for 3D seismic fault detection[J]. Journal of Geophysical Research: Machine Learning and Computation,2024,1(3):e2024JH000232.

[20] LIN Lei,LI Chenglong,KUANG Yanbin,et al. Applications of deep learning–based resolution–enhanced seismic data in fault identification[J]. Geophysical Prospecting,2025,73(2):523−542.

[21] WANG Jing,MA Siteng,LIU Yue,et al. AttentionFaultFormer: An attention–enhanced 3D CNN & transformer model for seismic fault detection[J]. Journal of Applied Geophysics,2025,238:105707.

[22] RONNEBERGER O,FISCHER P,BROX T. U–Net: Convolutional networks for biomedical image segmentation[M]//Medical image computing and computer–assisted intervention: MICCAI 2015. Cham: Springer International Publishing,2015: 234−241.

[23] HOU Qibin,ZHOU Daquan,FENG Jiashi. Coordinate attention for efficient mobile network design[C]//2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Nashville: IEEE,2021: 13708−13717.

[24] LIN T Y,GOYAL P,GIRSHICK R,et al. Focal loss for dense object detection[C]//2017 IEEE International Conference on Computer Vision (ICCV). Venice: IEEE,2017: 2999−3007.

[25] MILLETARI F,NAVAB N,AHMADI S A. V–Net: Fully convolutional neural networks for volumetric medical image segmentation[C]//2016 Fourth International Conference on 3D Vision (3DV). Stanford: IEEE,2016: 565−571.

[26] WU Xinming,SHI Yunzhi,FOMEL S,et al. Convolutional neural networks for fault interpretation in seismic images[C]//2018 SEG International Exposition and Annual Meeting. Anaheim: SEG,2018: SEG–2018–2995341.

[27] LOSHCHILOV I,HUTTER F. SGDR: Stochastic gradient descent with warm restarts[C]//5th International Conference on Learning Representations. Toulon: ICLR,2017: 1−16.

[28] BADRINARAYANAN V,KENDALL A,CIPOLLA R. SegNet: A deep convolutional encoder–decoder architecture for image segmentation[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence,2017,39(12):2481−2495.

[29] 王志壮,吴鹏,孙强,等. 临兴区块深部煤层气井生产特征及影响因素[J]. 煤田地质与勘探,2024,52(8):69−78. WANG Zhizhuang,WU Peng,SUN Qiang,et al. Production characteristics of deep coalbed methane wells in the Linxing Block and associated their influencing factors[J]. Coal Geology & Exploration,2024,52(8):69−78.

[30] 顾娇杨,张兵,郭明强. 临兴区块深部煤层气富集规律与勘探开发前景[J]. 煤炭学报,2016,41(1):72−79. GU Jiaoyang,ZHANG Bing,GUO Mingqiang. Deep coalbed methane enrichment rules and its exploration and development prospect in Linxing Block[J]. Journal of China Coal Society,2016,41(1):72−79.

[31] 高聪顺,吴海波,贾李岚,等. 基于煤田三维地震数据的小断层智能识别技术研究进展[J]. 绿色矿山,2026,4(2):6−15. GAO Congshun,WU Haibo,JIA Lilan,et al. Research progress in small–fault intelligent identification based on 3D seismic data of coal field[J]. Journal of Green Mine,2026,4(2):6−15.