The precise analysis of throat structures in rock cast thin section images is crucial for understanding reservoir microstructure and transport properties.However, existing throat recognition methods suffer from low accuracy and misidentification when processing these images.To address these limitations, this paper proposes GAB-ETransUNet, a deep learning algorithm for throat recognition built on the TransUNet architecture.Built upon such architecture, the algorithm upgrades the Transformer layer to an ETransformer structure to better focus on throat features, and second, it introduces a Group Aggregation Bridging(GAB) module to promote multi-scale feature fusion and reduce misidentification.Experimental results demonstrate that GAB-ETransUNet significantly outperforms the original TransUNet, achieving a CR of 83.63% (an increase of 5.99%) while reducing UR and OR by 4.22% and 3.82%, respectively; these improvements are accomplished with a 30.1% reduction in model parameters, thereby effectively enhancing both the accuracy and reliability of throat recognition.
从技术本质来看,喉道识别的核心任务是准确区分孔隙区域中的连通狭窄部分,这一过程依赖于图像分割技术对微观结构的颜色特征提取和边界识别能力。在图像语义分割深度学习的研究中,Ronneberger等[7]提出了UNet网络,通过跳跃连接巧妙地将语义信息与空间信息整合在一起。这种跳跃连接为上采样过程直接提供了低级纹理信息,对高级语义特征的构建至关重要,让反卷积层能获取基本的高分辨率特征[8]。Chen等[9]提出的TransUNet通过将CNN的局部特征提取能力与Transformer的全局上下文建模优势相结合,在编码阶段利用CNN捕获图像细节特征,在深层网络引入Transformer建立长距离依赖关系,提升了医学图像的分割精度。Liu等[10]提出了CM-UNet,采用CNN编码器提取局部特征,并创新地使用Mamba架构的解码器聚合全局上下文,通过引入通道-空间注意力门控和多尺度融合模块,显著提升了遥感图像分割的效率和精度。Kirillov 等[11]提出了Segment Anything Model (SAM),通过引入提示式分割任务,构建了一个通用的分割基础模型。该模型在超过十亿掩码的监督下进行训练,实现了对任意提示的快速响应,有效解决了传统方法在交互效率和跨领域适应性方面的局限。
FakhimiA, Alavi GharahbaghE.Discrete element analysis of the effect of pore size and pore distribution on the mechanical behavior of rock[J].Int J Rock Mech Min Sci, 2011, 48(1): 77-85.
[2]
GongY, LiuK.Pore throat size distribution and oiliness of tight sands-A case study of the Southern Songliao Basin, China[J].J Petrol Sci Eng, 2020, 184: 106508.
[3]
Professional Standardization Technical Committee for Oil and Gas Field Development. Determination of Rock Pore Structure Characteristics by Image Analysis Method: SY/T 6103-2004 [S]. Beijing: National Development and Reform Commission, 2004: 1.
TavanaeiA, SalehiS.Pore, throat, and grain detection for rock sem images using digitalwatershed image segmentation algorithm[J].J Por Media, 2015, 18(5): 507-518.
[6]
TuB Y, XiongS H, TengQ, et al.Throat segmentation algorithm for rock cast thin section images[J].Modern Computer, 2018, 630(30): 22-27.
CaiY H.Research on algorithms for pore and grain size analysis in cast thin section images [D].Chengdu:Sichuan University, 2021.
[9]
蔡宇恒.铸体薄片图像孔隙及粒度分析相关算法研究[D].成都:四川大学, 2021.
[10]
RonnebergerO, FischerP, BroxT.U-Net: Convolutional networks for biomedical image segmentation [M]//Medical Image Computing and Computer-assisted Intervention-MICCAI 2015, 2015: 234-241.
[11]
ZeilerM D, KrishnanD, TaylorG W, et al.Deconvolutional networks[C]//2010 IEEE Computer Society Conference on Computer Vision and Pattern Recognition, 2010: 2528-2535.
[12]
ChenJ, LuY, YuQ,et al.TransUNet: Transformers make strong encoders for medical image segmentation[PP/OL].[2025-07-22].
[13]
LiuM, DanJ, LuZ, et al.CM-UNet: Hybrid CNN-Mamba UNet for remote sensing image semantic segmentation[PP/OL].[2025-08-19].
[14]
KirillovA, MintunE, RaviN, et al.Segment anything[C]//2023 IEEE/CVF International Conference on Computer Vision (ICCV), 2023: 3992-4003.
[15]
DengL.Research on image analysis and processing of microscopic macro-model displacement [D].Chengdu:Sichuan University, 2022.
[16]
邓亮.微观大模型驱替图像分析与处理研究[D].成都:四川大学, 2022.
[17]
YueF Y, HeX H, TengQ Z, et al.A throat segmentation algorithm for cast thin section images based on improved UNet++ [J].New Generation of Information Technology, 2023, 6(23): 15-22.
ChenJ, MeiJ, LiX, et al.TransUNet: Rethinking the U-Net architecture design for medical image segmentation through the lens of transformers[J].Med Image Anal, 2024, 97: 103280.
[20]
MollerM.Efficient training of feed-forward neural networks[M]//Neural Network Analysis, Architectures and Applications.Boca Raton: CRC Press, 2024: 136-173.
[21]
ZhangQ, YangY B.Rest: An efficient transformer for visual recognition[J].Advances in Neural Information Processing Systems, 2021, 34: 15475-15485.
[22]
RuanJ, XieM, GaoJ, et al.Ege-unet: An efficient group enhanced unet for skin lesion segmentation [C] //International Conference on Medical Image Computing and Computer-assisted Intervention, 2023: 481-490.
[23]
HowardA G, ZhuM, ChenB, et al.Mobilenets: Efficient convolutional neural networks for mobile vision applications[PP/OL].[2025-08-05].
[24]
SuJ, LiuZ, ZhangJ, et al.DV-Net: Accurate liver vessel segmentation via dense connection model with D-BCE loss function[J].Knowl Based Syst, 2021, 232: 107471.
[25]
RuanJ, LiJ, XiangS.Vm-unet: Vision mamba unet for medical image segmentation[PP/OL].[2025-08-12].
[26]
ZhouZ, SiddiqueeM M R, TajbakhshN, et al.UNet++: Redesigning skip connections to exploit multiscale features in image segmentation[J].IEEE Trans Med Imaging, 2020, 39(6): 1856-1867.
[27]
WangH, CaoP, WangJ,et al.UCTransNet: Rethinking the skip connections in U-Net from a channel-wise perspective with Transformer[J].Proceedings of the AAAI Conference on Artificial Intelligence, 2022, 36(3):2441-2449.
[28]
WangH, CaoP, YangJ, et al.Narrowing the semantic gaps in U-Net with learnable skip connections: The case of medical image segmentation[J].Neural Netw, 2024, 178: 106546.
[29]
ZhuL, WangX, KeZ, et al.BiFormer: Vision transformer with bi-level routing attention[C]//2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2023: 10323-10333.
[30]
OuyangD, HeS, ZhangG, et al.Efficient multi-scale attention module with cross-spatial learning [C]//ICASSP 2023 - 2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2023: 1-5.
[31]
SelvarajuR R, CogswellM, DasA, et al.Grad-CAM: Visual explanations from deep networks via gradient-based localization[C]//2017 IEEE International Conference on Computer Vision (ICCV), 2017: 618-626.