Fractal pore structure exists widely in natural reservoir and dominates its transport property. For that, more and more effort is devoted to investigate the control mechanism on mass transfer in such a complex and multi-scale system. Apparently, effective characterization of the fractal structure is of fundamental importance. Although the newly emerged concept of complexity assembly clarified the complexity types and their assembly mechanism in a fractal system, equivalent extraction of the complexity types is the key for effective characterization. For these, we proposed a deep learning-based method to extract the original and behavioral complexity assembled in bed-packing fractal porous media for simplification and without loss of generality. In detail, the UNeXt network model was trained to obtain the independent connected regions of scaling objects with different scales, the edge detection and clustering analysis algorithms were employed to extract the number-size relationship between two successive scaling objects, and the unique inversion of fractal behavior was realized by taking the number-size model and fractal topography together. Consequently, an equivalent characterization method for fractal complex pore structure was developed based on the concept of complexity assembly. Our investigation provides a theoretical guidance and method reference for the quantitative characterization of fractal porous media that will guarantee the fundamental requirement for the accurate evaluation of the transport properties of natural reservoir.

分形孔隙结构广泛存在于天然储层中，并主导着其输导特性。为此，越来越多的努力致力于研究这种复杂、多尺度系统中的传质控制机制。显然，分形结构的有效表征至关重要。尽管新出现的复杂性组装概念阐明了分形系统中的复杂性类型及其组装机​​制，但复杂性类型的等效提取是有效表征的关键。为此，我们提出了一种基于深度学习的方法来提取床填充分形多孔介质中组装的原始行为复杂性，以进行简化且不失一般性。具体来说，训练UNeXt网络模型以获得不同尺度缩放对象的独立连通区域，利用边缘检测和聚类分析算法提取两个连续缩放对象之间的数量-大小关系，以及分形的独特反演行为是通过将数字大小模型和分形地形结合起来实现的。因此，基于复杂性组装的概念，提出了一种分形复杂孔隙结构的等效表征方法。我们的研究为分形多孔介质的定量表征提供了理论指导和方法参考，保证了准确评价天然储层输导特性的根本要求。