InterPore2025

Structure Tensor-Based Identification of Laminated Rocks in μ-CT Digital Rock Images

20 May 2025, 10:05

1h 30m

Poster Presentation (MS10) Advances in imaging porous media: techniques, software and case studies Poster

Speaker

Felipe Bevilaqua Foldes Guimarães (Federal University of Rio de Janeiro)

Description

The identification of lamination patterns in rock samples is important for understanding petrophysical properties behavior in heterogeneous rocks as laminated structures can influence fluid flow patterns, affecting both routine core analysis (RCAL) and special core analysis (SCAL) measurements [1, 2]. Usually, the identification of these structures rely on subjective human interpretation, which can lead to inconsistencies and inefficiencies in large-scale analyses.

Meanwhile, Digital Rock imaging has emerged as a promising approach for automating predictions of petrophysical properties and leveraging rock physics knowledge [3]. However, one aspect often overlooked in Digital Rock analysis is the automation of rock structure characterization, particularly the identification of laminated rock samples. Several aspects can complicate the development of automated methods for identifying these structures from rock images, including varying levels of image noise, diverse lamination types with distinct roughness, frequencies, discontinuities, and orientations characteristics, as well as distinct types of rock heterogeneities and lithologies that could further difficult this process.

In this work, we propose a method for identifying laminated samples from $\mu$-CT images of rock plugs using the structure tensor method. The structure tensor is a mathematical tool that can be used to determine local orientations within images [4, 5]. In our approach, we use these local orientations calculated from the images to compare the distribution of these orientations with a uniform distribution using the Wasserstein distance. Our underlying assumption here is that we expect laminated samples to show a more peaked distribution, indicating a predominant orientation in the image, while non-laminated samples would show a more random distribution that would be closer to a uniform distribution. To determine if a sample is laminated, we adjusted a threshold over the calculated Wasserstein distances based on lamination annotations from three human evaluators. For the lamination orientation, we use the mode of the orientations histogram. Furthermore, we classify each sample into vertical, horizontal, and inclined laminations by comparing the calculated orientation with pre-defined ranges of orientation for each of this categories. A visual representation of our methodology can be seen in the Figure 1.

The proposed methodology is able to distinguish laminated patterns in rock samples and determine their orientation, a crucial aspect for interpreting the impact of these structures on petrophysical properties. Our method has been validated against annotations from three human evaluators across more than 4000 rock $\mu$-CT images, indicating its effectiveness across diverse rock types and lithologies, primarily from carbonate fields in the Brazilian pre-salt formations.