Conference PaperPDF Available

Core photos lithological interpretation using neural networks

Authors:

Abstract

With a rapid increase in digital data, oil and gas companies are constantly looking for an automatic and accurate method to extract the geological information from the core. However, methods for automatic core analysis are very limited 2. At the moment, this is still done by an expert-geologist, which takes a lot of time and effort. Recently, Chatterje (2012) applied the multi-class support vector machine (SVM) algorithm to classify rocks 1. Hasanov et al. 2016 used color distribution analysis on core photos 3. Still, no comprehensive study was performed on real geological samples with complex structural and textural features. In this study, we propose a non-destructive method for rock-type classification of core images. An algorithm for core images processing and rock-type classification is based on convolutional neural network analysis. We trained our network using 10 cm samples of core images, which uniformly sized to 128x128 px images. The training samples consisted of four lithotypes (laminated sandstone, limestone, sandstone, shale). We also added to the analysis two non-conditional core types (non-cutted and crushed core). The computer successfully distinguished 4 classes with different false rates. The best results achieved for limestones and shales (recall=83 and 91%, on 54 and 138 test images), the worst are for laminated sandstones and sandstones (recall are 53 and 73% on 15 test images each). Other two groups discern with poor scores (crushed core recall 7%, non-cutted-32% on 30 and 57 images). In most cases these two groups got right class as a real rock type and wrong one as image predicted type (compared to true label of image). As it seen from experimental results, the rates highly depend on each class number and quality of images. The performance of low-quality images with a small number of them in class gives the poor precision. Our results show that the convolutional neural network detects the right class of lithology with low boundary recall 85% (average, in grayscale images), even on small amount of data. Additionally, this algorithm allows studying core photos in much faster and accurate way. To improve further the accuracy of the method, more core-photos of different quality, texture characteristic and geological settings should be analyzed.
ISC2018, Québec City
CORE PHOTOS LITHOLOGICAL INTERPRETATION USING NEURAL
NETWORKS
E.E.Baraboshkin1*, A.V.Ivchenko2**, L.S.Ismailova1, D.M.Orlov1, E.Yu.Baraboshkin3, D.A.Koroteev1
1Petroleum Engineering department, Skolkovo Institute of Science and Technology, Nobelya Ulitsa 3, Moscow,
Russia.
2 Radio Engineering and Cybernetics department, Moscow Institute of Physics and Technology, 9 Institutskiy per.,
Dolgoprudny, Moscow Region, Russia.
3 Regional Geology and Earth History department, Lomonosov Moscow State University, Leninskie Gory 1, Moscow,
Russia.
*e-mail: evgenii.baraboshkin@skoltech.ru
** e-mail: ivchenko.a.v@phystech.edu
With a rapid increase in digital data, oil and gas companies are constantly looking for an
automatic and accurate method to extract the geological information from the core. However,
methods for automatic core analysis are very limited2. At the moment, this is still done by an expert-
geologist, which takes a lot of time and effort. Recently, Chatterje (2012) applied the multi-class
support vector machine (SVM) algorithm to classify rocks1. Hasanov et al. 2016 used color
distribution analysis on core photos3. Still, no comprehensive study was performed on real
geological samples with complex structural and textural features.
In this study, we propose a non-destructive method for rock-type classification of core
images. An algorithm for core images processing and rock-type classification is based on
convolutional neural network analysis. We trained our network using 10 cm samples of core
images, which uniformly sized to 128x128 px images. The training samples consisted of four
lithotypes (laminated sandstone, limestone, sandstone, shale). We also added to the analysis two
non-conditional core types (non-cutted and crushed core). The computer successfully distinguished
4 classes with different false rates. The best results achieved for limestones and shales (recall=83
and 91%, on 54 and 138 test images), the worst are for laminated sandstones and sandstones (recall
are 53 and 73% on 15 test images each). Other two groups discern with poor scores (crushed core
recall 7%, non-cutted 32% on 30 and 57 images). In most cases these two groups got right class
as a real rock type and wrong one as image predicted type (compared to true label of image). As it
seen from experimental results, the rates highly depend on each class number and quality of images.
The performance of low-quality images with a small number of them in class gives the poor
precision.
Our results show that the convolutional neural network detects the right class of lithology
with low boundary recall 85% (average, in grayscale images), even on small amount of data.
Additionally, this algorithm allows studying core photos in much faster and accurate way. To
improve further the accuracy of the method, more core-photos of different quality, texture
characteristic and geological settings should be analyzed.
References
1 S. Chatterjee. Applied Intelligence. 39, 2012.
2 L. Lepistö, I. Kunttu, J. Autio, A. Visa. WSCG SHORT PAPERS proceedings, WSCG'2003, 2003.
3 I.I. Hasanov, I.A. Ponomarev, A.V. Postnikov, N.A. Osintseva. Geomodel-2016 (in Russian), 2016.
Acknowledgements
This work was supported by RFBR (grants 16-05-00207a, 13-05-00745a).
... Recent works ( Baraboshkin et al., 2018;Ivchenko et al., 2018) have shown that artificial neural networks can extract information from images and easily determine lithotypes. A development of this method is presented below. ...
... In previous work ( Baraboshkin et al., 2018;Ivchenko et al., 2018) we showed how the identification of different rock properties can be automated using a convolutional neural network (CNN) ( LeCun et al., 1989). We implemented new machine vision algorithms as the number of classes grew. ...
... This lets us understand how the system interprets the images and why it makes mistakes. The first network that was tested ( Baraboshkin et al., 2018;Ivchenko et al., 2018 ...
Preprint
Full-text available
The description of rocks is one of the most time-consuming tasks in the everyday work of a geologist, especially when very accurate description is required. We here present a method that reduces the time needed for accurate description of rocks, enabling the geologist to work more efficiently. We describe the application of methods based on color distribution analysis and feature extraction. Then we focus on a new approach, used by us, which is based on convolutional neural networks. We used several well-known neural network architectures (AlexNet, VGG, GoogLeNet, ResNet) and made a comparison of their performance. The precision of the algorithms is up to 95% on the validation set with GoogLeNet architecture. The best of the proposed algorithms can describe 50 m of full-size core in one minute.
... Secondary physical and chemical analyzes, such as hardness, acid tests and other nondestructive geochemical techniques, can also be performed to confirm the mineralogy of the lithofacies (Croudace and Rothwell, 2015;McPhee et al., 2015;Amao et al., 2016). Although physical analysis cannot be analyzed in machine learning, machine learning is an excellent tool for replicating visual pattern analysis and performing more robust and unbiased classification (Thomas et al., 2011;Baraboshkin et al., 2018;Martin et al., 2021). Visual-based machine learning, specifically deep neural networks, has been widely applied in the past 5 years to perform lithological classification (de Lima et al., 2019;Baraboshkin et al., 2020;Koeshidayatullah et al., 2020). ...
Article
Full-text available
Lithofacies classification is a fundamental step to perform depositional and reservoir characterizations in the subsurface. However, such a classification is often hindered by limited data availability and biased and time-consuming analysis. Recent work has demonstrated the potential of image-based supervised deep learning analysis, specifically convolutional neural networks (CNN), to optimize lithofacies classification and interpretation using core images. While most works have used transfer learning to overcome limited datasets and simultaneously yield a high-accuracy prediction. This method raises some serious concerns regarding how the CNN model learns and makes a prediction as the model was originally trained with entirely different datasets. Here, we proposed an alternative approach by adopting a vision transformer model, known as FaciesViT, to mitigate this issue and provide improved lithofacies prediction. We also experimented with various CNN architectures as the baseline models and two different datasets to compare and evaluate the performance of our proposed model. The experimental results show that the proposed models significantly outperform the established CNN architecture models for both datasets and in all cases, achieving an f1 score and weighted average in all tested metrics of 95%. For the first time, this study highlights the application of the Vision Transformer model to a geological dataset. Our findings show that the FaciesViT model has several advantages over conventional CNN models, including (i) no hyperparameter fine-tuning and exhaustive data augmentation required to match the accuracy of CNN models; (ii) it can work with limited datasets; and (iii) it can better generalize the classification to a new, unseen dataset. Our study shows that the application of the Vision transformer could further optimize image recognition and classification in the geosciences and mitigate some of the issues related to the generalizability and the explainability of deep learning models. Furthermore, the implementation of our proposed FaciesViT model has been shown to improve the overall performance and reproducibility of image-based core lithofacies classification which is significant for subsurface reservoir characterization in different basins worldwide.
... The automated methods for core description have been developed for a long time. Different approaches have been used to meet the desired result: color distribution analysis and computer vision [1][2][3][4][5], machine learning [6][7][8], or deep learning [9][10][11][12][13] application. Recent developments in computer vision and machine learning allowed the creation of a system for automated core image extraction and description [14]. ...
Conference Paper
Full-text available
The automated methods for core description have been developed for a long time. Different approaches have been used to meet the desired result: color distribution analysis and computer vision [1-5], machine learning [6-8], or deep learning [9-13] application. Recent developments in computer vision and machine learning allowed the creation of a system for automated core image extraction and description [14]. The system works with daylight core box images. Algorithms based on computer vision and deep learning have been applied to extract core column images from a core box with semantic segmentation [15] and automatically depth-reference them. These algorithms also helped to classify rocks into different rock and structure types [16]. The system works as a cloud solution. It was applied to various oil and mining industries' tasks-the system allows changes to the data produced by segmentation and classification algorithms. The speed of classification is about 10m\s. In this study, we overview the system application for sedimentological description. The quality of the images can be low (the minimum requirement is around 100dpi, which is about 390px for a 10 cm core column width). After a classification algorithm was developed for 14 different types of rocks (3 rock types: sandstone, siltstone, shale\argillite, and five structures: bioturbated, rippled, laminated, massive, and cross-bedded) [17], it was uploaded to the system, and an expert applied it for a description of 10 different wells. An expert sedimentologist checked each well after classification and made some changes. The automated classification results have been reviewed and statistically analyzed compared to the expert's results. The classification accuracy on new data demonstrated a good accuracy\speed ratio. With a 7-times well analysis speed increase, the minimum accuracy gained is 70%. Most lithologies have been detected correctly, but the structure was sometimes mixed. Also, some lithologies have not been included in the model (e.g., coal). The number of samples in the training set was small (less than 40 10x10 cm images), so the expert added these types to the system. The difference in NTG value measurement on the expert and algorithm classification results with a 1m sliding window is only 5%. The automated core description application results are valuable for the industry in terms of their speed and NTG calculation. Also, it makes the life of an expert much easier. It allows him to concentrate on the results instead of the description. The overall description speed based on daylight core box images was 7-times increased. Another benefit of such a system application is automated depth referencing for the core images-the expert can interactively see the depth. The applied classification models help increase classification speed, improve the overall description quality, and be used as the first mandatory stage of initial preparation before the trip to the core storage or as a stand-alone stage for a quick core description. Even an untrained person can learn to classify different rock types while examining the results produced by an algorithm.
... В России эти исследования также проводятся многими специалистами [6,7]. Группа исследователей из Сколтеха и Digital Petroleum работала над решением этой проблемы [8][9][10][11]. В результате совместных исследований и консультаций со специалистами ЗАО «МиМГО» удалось разработать программный продукт, который позволяет решать конкретные практические задачи, важные при проведении седиментологических работ. Данный продукт позволяет с достаточной точностью и высокой скоростью классифицировать фото керна на различные типы пород и текстур, а также определять другие характеристики, в зависимости от задачи [12]. ...
Conference Paper
Current technologies allow to automate and ease many tasks. The work presents an automated full-bore core description system application and integration into workflow example. A machine learning model for sedimentological core description developed by mutual efforts. The model describes 14 lithotypes. The speed of core description increased by 7 times after model integration in the automated core description system. An expert made changes to the generated description in 20 % of cases which sufficiently reduced his workload for core description
... Группа исследователей из Сколтеха и Digital Petroleum на протяжении длительного времени также занималась проблемами в области автоматизации описания керна [E. E. Baraboshkin et al., , 2018E. Baraboshkin et al., 2019. ...
Conference Paper
Full-text available
The automatization is a modern trend in various field of geology. In this work we present a system which were constructed based on convolutional neural network (CNN) for automated core description. The system was successfully applied to production data. The application of the system speeds up the core description process in 7x. A sedimetnologist spent 40 minutes to describe 60 meters of core in a scale of 1:10cm instead of 5 hours. The results are stored in digital format which removes all paperwork. The system helps to describe most of required lithologic types (rock type and its structure). In case of missed rare lithotype – user can add it to the system. A pipeline to prepare and train the CNN model described.
Conference Paper
In strategising development of hydrocarbon reservoirs, substantial uncertainty in recovery potential is often attributed to subsurface heterogeneity. Challenged reservoir characterisation is proposed to be directly due to the inability of correlating spatial scales: core analyses to well logging data. This study’s central goal is to propose a ‘Multiscale link’ by challenging empirical correlations of multiphase displacement and ‘upscaling’ processes of reservoir characterisation by exploiting Artificial Intelligence and ‘Digital Rock Technology’, aiming at minimising geological risk. By exploiting 40 years of a North Sea field's appraisal and production and formulating an AI-compatible ‘multiscale’ data set, petrophysical correlations have integrated a further innovative concept: borehole image processing to characterise geological features and oil potential. In binding the ‘Multiscale’, fundamental multiphase dynamics at pore-scale have been critically associated to most affine reservoir modelling ‘deep learning’ frameworks, leading to ideating an AI workflow linking field-scale rates, well logs and core analyses to the continuously-reconstructed pore network, whilst extracting invaluable multiphase dependencies. The preliminary results implementing selected Machine Learning algorithms, coupled with advanced digital technologies in reservoir simulation, have been showcased in proposing a solution to the ‘Multiscale link’ in reservoir characterisation, providing the groundworks for its programming realisation. Importantly, it was concluded that the layers of complexity within learning algorithms, which constrained its execution within this project, undoubtedly require multidisciplinary approach. By conceiving a physically and coding-robust workflow for advanced reservoir characterisation and modelling permitting ‘multiscale’ representative multiphase simulations, identification of optimal EOR becomes attainable. This leading edge represents potential to minimise geological risk, thus de-risking reservoir management (in turn FDP) of mature and live fields; but also expected to set a starting point for further developments of Artificial Intelligence in the oil and gas industry.
Article
Segmentation and analysis of individual pores and grains of mudrocks from scanning electron microscope images is non-trivial because of imaging artifacts, variation in pixel grayscale values across images, and overlaps in grayscale values among different physical features such as silt grains, clay grains and pores, which make identifications difficult. Moreover, because grains and pores often have overlapping grayscale values, direct application of threshold-based segmentation techniques is not sufficient. Recent advances in the field of computer vision have made it easier and faster to segment images and identify multiple occurrences of such features in an image, provided that ground-truth data for training the algorithm are available. Here we propose a deep learning SEM image segmentation model, MudrockNet based on Google's DeepLab-v3+ architecture implemented with the TensorFlow library. The ground-truth data were obtained from an image-processing workflow applied to scanning electron microscope images of uncemented muds from the Kumano Basin offshore Japan at depths <1.1 km. The trained deep learning model obtained a pixel-accuracy > 90%, and predictions for the test data obtained a mean intersection over union (IoU) of 0.6663 for silt grains, 0.7797 for clay grains and 0.6751 for pores. We also compared our model with the random forest classifier using trainable Weka segmentation in ImageJ, and it was observed that MudrockNet gave better predictions for silt grains, clay grains and pores in most cases. The size, concentration, and spatial arrangement of the silt and clay grains can affect the petrophysical properties of a mudrock, and an automated method to accurately identify the different grains and pores in mudrocks can help improve reservoir and seal characterization for petroleum exploration and anthropogenic waste sequestration.
Applied Intelligence
  • S Chatterjee
S. Chatterjee. Applied Intelligence. 39, 2012.