Skip to content Skip to sidebar Skip to footer

The majority of our users operate above ground and a few even under water, but very few take eCognition technology underground. Our customer ExxonMobil has done just this. Over the past 5+ years, I have been working together with a team of innovative geologists at Exxon’s headquarters in Houston, Texas to revolutionize the way they process geological thin sections (slices of rock from a geologic boring). Yes, you read correctly, we are using Trimble eCognition technology to analyze microscopic images of rock. But, at the center of it all was an image analysis problem and outcome is a method we call  Quantitative Digital Petrography (QDP).

Together, we have published a paper entitled “Quantitative Digital Petrography: Full Thin Section Quantification of Pore Space and Grains” in the Society of Petroleum Engineers. This initial publication was followed up by a second paper in May 2019 entitled “Quantitative Digital Petrography: A Novel Approach to Reservoir Characterization” that I have again had the honor of co-authoring with Antonio Buono, Kelly Luck and PJ Moore and Shawn Fullmer from ExxonMobile Upstream Research Company.

ExxonMobil approached the Trimble eCognition team nearly 5 years ago with an interesting and challenging image analysis problem: is it possible to automate the analysis of carbonate thin sections?

Traditionally, this type of analysis has been based on a manual method called point counting “which are qualitative to semi-quantitative and vary greatly depending on the petrographer”. Essentially, the expert petrographer casts a net of 300+ sample points across the thin section and characterizes the thin section at each point. The results are then extrapolated to the entire thin section.

There are several problems with this methodology: 1) it does not provide truly wall-to-wall quantitative information on the thin section, 2) it relies heavily on the petrographer’s individual expert interpretation, 3) it is time consuming (between 1-8 hours per thin section). If we are examining only a select few thin sections, this may not sound too bad. But, when we consider 1000s upon 1000s of images coming in on a continual basis the bottlenecks in the traditional methodology are immediately felt.

As the name of our paper implies, we have developed a new method called Quantitative Digital Petrography (QDP) to automate the analysis of carbonate thin sections and resolve the 3 primary issues identified above. QDP uses object-based image analysis techniques thatautomatically isolates the sample on a high-resolution (i.e. <1μm/pixel) scanned thin section, segments the image, and assigns those segments to predefined categories”.

The thin section analysis workflow can be broken down into 4 general steps:

1) the isolation of the region of interest (ROI) within the scanned thin section – here we are automatically discriminating between the actual sample and the surrounding epoxy material. For this purpose, down-sample the image, as full resolution is not required, and apply image filtering to smooth the image. Finally, a contrast split segmentation is used allowing the eCognition software to automatically determine the best parameters for differentiating between the sample and surrounding epoxy material.

2) A detailed multiresolution segmentation is then run on the ROI generating image objects reflecting homogeneity in the RGB color spectrum. This step creates the image objects that will subsequently be classified into classes “such as  pores, cement, grains, and Svugs”.

3) To automatically assign image objects to the various classes, a series of class definitions were created using the unique RGB and hue, saturation, intensity (HSI) characteristics. The goal is to use a class definition that is as accurate as possible and yet still highly transferable allowing the classification to be “applied in batch to a large number of samples simultaneously”. The use of multiple of quantitative parameters in combination with the automated tools “minimizes user subjectivity”

4) The final step is to export the image object information from eCognition – “ the isolation of each class allows further analysis including pore size, pore type abundance, and shape statistics” all of which are calculated and exported for each image object. This information is easily compiled into spreadsheets for “further analysis and integration with existing data, such as CCAL and SCAL”.

The paper examines three QDP examples that we performed on different datasets. Example 1 looks at a Limestone sample where grains, pores, cement and micrite features have been extracted.

The second example examines a fine grained Dolomite sample and the classification isolates dolomite grains, pores, bitumen and anhydrite. “A comparison of point count to QDP shows fairly close alignment in grain size assessment but with a significant increase in data density (270 points via point counting and 1,048,575 points via QDP)”.

The third example we looked at utilized Scanning Electron Microscopy (EDS) data for a Gray Shale sample. In combination with the EDS data we also have a BSE image available.  “This example illustrates the versatility of the QDP approach beyond optical images, where more complex heterogeneous rocks require multiple datasets to accurately characterize mineral assemblages, geometries, relative abundance, and distribution”. Although the general QDP approach can be applied to a sample like this, there were parts of the Rule Set that had to be adapted, for example, the grain classification was not possible on the BSE image alone. For this the EDS element maps were integrated – allowing “us to segment grains that have no grayscale variance”. The unique data fusion capabilities of eCognition allow for the integration of these two different data sets; “this allows for a more realistic integration with the core plug data for samples that need mineralogical mapping”.

The final phase of our eCognition development focused on the creation of a point counting tool – “The versatility of the Trimble eCognition software also allows for scripting a manual point count” tool allowing the user to efficiently maintain the project workflow within a single software package.

The tools we developed allow the user to overlay a traditional point count grid on the sample and record manual measurements that can be compared to the QDP results. 

In conclusion, the QDP technique that we have developed “greatly expands our ability to digitally analyze large numbers of high-resolution thin-section images in a time efficient manner and provides a level of detail in our analysis that has not been previously attainable by any practical method”.

Follow Us on Social Media

For more news on eCognition