Machine-Learning Image Recognition Enhances Rock Classification
You have access to this full article to experience the outstanding content available to SPE members and JPT subscribers.
Automated image-processing algorithms can improve the quality and speed in classifying the morphology of heterogeneous carbonate rock. Several commercial products have produced petrophysical properties from 2D images and, to a lesser extent, from 3D images. Images are mainly microcomputed tomography (µCT), optical images of thin sections, or magnetic resonance images (MRI). However, most successful work is from homogeneous and clastic rocks. In the complete paper, the authors have demonstrated a machine-learning-assisted image-recognition (MLIR) approach to determine the porosity and lithology of heterogeneous carbonate rock by analyzing 3D images from µCT and MRI.
The authors’ literature review has revealed the pressing need to perform 3D image processing instead of 2D. Achieving this goal requires an interdisciplinary approach. This study deployed new analysis and verification approaches, including 3D micromodels (3DMM) with various micropore sizes and uses 3DMM as an image-processing calibration reference. Additionally, a new image-resolution enhancement for quality segmentation is developed. Porosity was determined mainly using two methods. The first is a standalone image processing, in which image-information extraction was successful. The second is MLIR. The difference between image processing and image analysis is important. If image processing enables extraction of meaningful data, then image analysis is the ability to interpret this data through numerical analysis.
3D µCT Images. Not every image comes with the needed resolution for optimal image analysis. The authors acquired 3D µCT images for dry (meaning the pore contains air only) carbonate rock. One of the 2D slices is displayed in Fig. 1. The figure demonstrates that micropores in carbonate rock appear in more than one gray value, although they should appear as black. The cause of this is low resolution, when the pore size is smaller than the pixel resolution used to acquired the image. The pore must be black in color (as air fills pores), and the pixel must hold a value of 0. To measure porosity accurately, images must be corrected to a 0 value for pores. The challenge is how to achieve this correction without mistakenly converting true nonvoid to void. In the figure, the top image shows a 38-mm-diameter sample acquired with 40-µm resolution (each pixel is 40×40 µm). The image inside the red circle is the zone of interest for identifying pores (black) and solids (light). To solve the resolution effect, a convolutional Gaussian kernel was created for improving the resolution of 3D images.
3D MRI Images. For the same carbonate rock, 3D MRI images were acquired. Pores of this rock sample were flooded with crude oil. 3D MRI images were acquired with three different sampling rates, meaning different slice thicknesses along the z-axis, along with the core-sample cylindrical shape. This process is detailed in the complete paper.
To solve for the resolution effects, an experiment was performed involving building 3DMM. Several types of micromodels were created to enable quantifying the blurriness of the image and correcting for it. To increase resolution and reduce blurriness, an image-correction model had to be built and, at the same time, efficiency had to be ensured. Therefore, a convolutional Gaussian image-processing filter was the first choice. After several trials, a novel optimized convoluted Gaussian image-processing algorithm was developed. The authors call the new algorithm an image-resolution-optimized Gaussian algorithm (IROGA).
Methodology of MLIR
The study followed the following steps. The machine-learning step is detailed in the next section.
- Machine learning
- Image processing. This is defined as tasks a human expert performs on the image (2D, 3D, 4D, or 5D) that produce a new set of information or a new version of an image that provides more insight to a human expert.
- Combination of machine learning and image processing. The combination scenario was termed MLIR, defined as the ability of a machine to perform tasks such as training sets and image interpretation on new images in a faster, more-accurate manner than a human expert can.
Porosity for the rock core plug was measured using helium gas, considered to be the ultimate reference of porosity value. The same core plug was also imaged with µCT and MRI.
Porosity From Experimental Measurements. Core-plug porosity was measured with three methods. The first method used the weight difference of rock under dry and wet conditions. The second used nuclear magnetic resonance on the same rock saturated with fluid (crude oil). The third method used helium gas on the same sample after cleaning and drying.
Porosity From Machine Learning. For this portion of the study, the authors performed 3D µCT image recognition using a random-forest machine-learning algorithm. The primary machine-learning-performed tasks adhered to the following process:
1. A human expert, a petrophysicist in the case of this study, provided a data set, identifying two types of minerals (limestone and pyrite), two physical statuses (pore and solid), and four image intensities. The data set contains:
- An independent-parameter data column, featuring image-identified properties with four shadings: black, dark gray intensity, light gray intensity, and white
- A dependent data column containing the desired classes of limestone, pyrite, and pore; solids here include limestone and pyrite
2. The human expert chooses a machine-learning algorithm; in this study, the random-forest algorithm was selected because of its superior capability in performing classification.
3. The computer uses the random-forest algorithm to learn from the data set and build the prediction model with all its governing equations.
4. The computer inputs the new data set, the 3D μCT image, and uses the learned prediction model with its governing equations to predict output classes.
Running 3DMM Experiments for Constructing IROGA
The authors developed a 3DMM technique with various micropore sizes to use as measurement control. The process of experimental design and imaging for the bundle of capillary tubes used in the study is detailed in the complete paper.
To perform the task of running the 3DMM experiments, experimental screening using three different methods was designed and conducted (two machine-learning methods and one novel image-processing algorithm). These methods, detailed in the complete paper, included the following:
- Machine-learning Gaussian random forest
- Novel IROGA image-processing algorithm
- Machine-learning difference Gaussian random forest (MLDGRF)
IROGA Prediction and Validation for Carbonate Rock
Prediction and validation of IROGA on a carbonate core plug using MRI images was performed. For validation of IROGA prediction, the reference helium porosity was used; the resulting accuracy was 91.8%.
The experiment, modeling, and use of IROGA is explained in the following steps:
1. 3DMM built with 600-μm-inner-diameter tubes
2. 3DMM flooded with crude oil collected from a lower Cretaceous offshore formation
3. 3D stacked MRI images acquired for 3DMM
4. Iterative image-resolution-enhancement algorithm depending on a convoluted Gaussian matrix was structured
5. IROGA model validated with geometrical reference calculation of the 3DMM
6. 3D stacked MRI images acquired of a fluid-saturated carbonate rock core plug
7. IROGA applied to the core-plug MRI image to measure porosity and locate fluid distribution
8. IROGA porosity results validated with the reference helium porosity
3D µCT Carbonate Rock Machine-Learning Image-Recognition Porosity
Because the success of MLIR in achieving high accuracy when measuring 3DMM porosity has been demonstrated, the work was extended to 3D µCT. Using the MLDGRF algorithm to measure 3D µCT porosity, the authors compared MLDGRF results with three porosity measurements. The accuracy of MLDGRF reached 94.37%. The approximately 1.7-porosity-unit difference in value between MLDGRF and the reference helium porosity is attributed to the difference of Gaussian function. Further tuning of the function parameters is recommended to increase accuracy.
- Image recognition using the MLDGRF algorithm is superior compared with machine-learning image recognition using the Gaussian random-forest algorithm. Experimental validation is recommended to quantify the algorithm accuracy.
- The MLDGRF algorithm works well for different types of images (e.g., µCT and MRI). If the domain expert labels the desired features correctly and chooses the best algorithm, machine-learning image recognition can save years of tedious work.
- Improving the resolution of 3D MRI needs adequate image-processing algorithms. Enhancement can lead to over- or undercorrecting. The authors developed an image-enhancement algorithm for two-class segmentation.
- The work on porosity determination using machine learning has helped achieve a better understanding of rock heterogeneity and has provided insight into analyzing and digitally classifying rock.
Machine-Learning Image Recognition Enhances Rock Classification
01 October 2020
Artificial Intelligence in Operation Monitoring Discovers Patterns Within Drilling Reports
The complete paper provides an approach using machine-learning and sequence-mining algorithms for predicting and classifying the next operation based on textual descriptions.
Machine-Learning Approach Determines Spatial Variation in Shale Decline Curves
The complete paper describes an automated machine-learning approach to determine the spatial variation in decline type curves for shale gas production, based on existing data of production, completion, and geological parameters.
Wireline Sand-Detection Tool Locates, Quantifies Downhole Sand Production
Field examples presented in the complete paper describe principles of data acquisition with a sand-detection tool when run in combination with a production logging string and results of logging in slightly deviated wells completed with sand screens.
Don't miss out on the latest technology delivered to your email weekly. Sign up for the JPT newsletter. If you are not logged in, you will receive a confirmation email that you will need to click on to confirm you want to receive the newsletter.
19 October 2020
26 October 2020