Digitization of Raster Logs A Deep Learning Approach M Quamer Nasima b Narendra Patwardhana Tannistha Maitiaand Tarry Singha

2025-04-24 0 0 1.66MB 18 页 10玖币

侵权投诉

Digitization of Raster Logs: A Deep Learning

Approach

M Quamer Nasima b, Narendra Patwardhana, Tannistha Maitiaand Tarry Singha

aDeepkapha.ai, Amsterdam, Netherlands ∗

bIndian Institute of Technology, Kharagpur, India

Abstract

Raster well-log images are digital representations of well-logs data generated over

the years. Raster digital well logs represent bitmaps of the log image in a rectangular

array of black (zeros) and white dots (ones) called pixels. Experts study the raster

logs manually or with software applications that still require a tremendous amount of

manual input. Besides the loss of thousands of person-hours, this process is erroneous

and tedious. To digitize these raster logs, one must buy a costly digitizer that is not

only manual and time-consuming but also a hidden technical debt since enterprises

stand to lose more money in additional servicing and consulting charges. We propose

a deep neural network architecture called VeerNet to semantically segment the raster

images from the background grid and classify and digitize the well-log curves. Raster

logs have a substantially greater resolution than images traditionally consumed by

image segmentation pipelines. Since the input has a low signal-to-resolution ratio, we

require rapid downsampling to alleviate unnecessary computation. We thus employ a

modiﬁed UNet-inspired architecture that balances retaining key signals and reducing

result dimensionality. We use attention augmented read-process-write architecture.

This architecture eﬃciently classiﬁes and digitizes the curves with an overall F1 score

of 35% and IoU of 30%. When compared to the actual las values for Gamma-ray and

derived value of Gamma-ray from VeerNet, a high Pearson coeﬃcient score of 0.62 was

achieved.

Keywords: raster log, digitization, transformer, deep learning, well-log curves

1 Introduction

Well-logging is the process of taking measurements of various rock properties along the

length of the well down into the ground by drilling tools. The digital log responses are

functions of lithology, porosity, ﬂuid content, and textural variation of formation. The well-

logging parameters are used to derive lithofacies groups and facies-by-facies descriptions of

rock properties. Before the advent of digital logging instruments, well-logging data were

∗Corresponding author email: quamer.nasim@deepkapha.com

arXiv:2210.05597v1 [physics.geo-ph] 11 Oct 2022

drawn on the parameter graph in curve format. Well-logging parameter graphs have many

disadvantages: large size, ample memory space, and interference like gridlines. Therefore,

it is necessary to convert well-logging parameter graphs into X-Y coordinates, where X

represents parameter values and Y represents depth values. Raster logs are scanned copies

of paper logs saved as image ﬁles.

Well log data saved as depth-calibrated raster images provide an economic alternative

to digital formats for preserving this valuable information into the future (Cisco, 1996).

Although often discarded after vectorization, raster imaged well logs may be the key to a

global computer-readable format for legacy hardcopy data. This legacy data is stored on

multiple media and contains information for various applications in addition to resource

exploration and development, such as environmental protection, water management, global

change studies, and primary and applied research.

Experts such as geologists/reservoir engineers revisit and study these raster logs manu-

ally or with software applications requiring a tremendous amount of manual input. Besides

the loss of thousands of person-hours, the existing process is erroneous and tedious. To

digitize these raster logs and eﬃciently use them in conventional as well as unconventional

analysis, one needs to buy an expensive digitizer which is a manual and time-consuming task

but also there is a hidden technical debt since enterprises stand to lose more money in ad-

ditional servicing and to consult charges. The commercially used logging curve digitization

software, Neuralog, is based on SCTR (Scanning, Compressing, Tracing, and Rectifying).

However, due to the interference of the background grid, this software frequently pauses dur-

ing curve tracking.Several unsupervised computer vision methods have been implemented

to digitize the log data embedded in the binary image.

Well-log digitization could be performed through two kinds of approaches: Pixel-based

methods and non pixel-based methods. Pixel-based methods include the thinning process

and the Global Curve Vectorization (GCV) method (Zheng et al., 2005; Hilaire and Tombre,

2001; Nagasamy and Langrana, 1990). The thinning method reduces the width of a line

to only one pixel, leaving only the skeleton that can characterize its features. The main

disadvantage of the thinning process is that it has a high time complexity, loses line width

information, and is prone to deformation and wrong branches in the intersection area. GCV

method is suitable for line processing but poor for point line processing (Yuan and Yang,

2019).

Non-pixel-based methods mainly fall into two categories: contour-based and adjacency

graph-based. The contour-based approach (Hori and Tanigawa, 1993) is to extract the

contour of the image ﬁrst and then ﬁnd the matched contour pairs. The adjacency graph

method ﬁrst applies run-length encoding to graphs, then analyses the segments and gen-

erates various adjacency graph structures, such as line adjacency graph (LAG) and block

adjacency graph (BAG) (Pavlidis, 2012). (Li et al., 2003) used SCTR (Scanning, Compress-

ing, Tracing, and Rectifying) approach by employing the LAG data structure. Yang (2009)

improved the SCTR method and put forward the PCTR (Preprocessing, Compressing,

Tracing, and Rectifying) method. Yuan and Yang (2019) proposed an algorithm for erasing

grid lines and reconstructing strokes in Chinese handwriting based on BAG. However, such

methods are diﬃcult to deal with the complex situation in well-logging parameter graphs,

especially the analysis of nodes. Yuan and Yang (2019) used morphological image process-

ing and pixel statistics method to eliminate gridlines, isolating the curves and the gridlines.

Then, the remaining grid lines and noise points are cleared according to the characteristics

of the small size of their connected components. However, all these existing methods need

manual intervention, which is not the desired option, especially when the paper logs have

a huge size >10 MB. In the present study, we propose a novel transformer-based deep

learning model named VeerNet which employs self-attention mechanisms to identify indi-

vidual curves from a single track. The straightforward design of transformers allows the

processing of various modalities (e.g., image, video, text, and speech) using similar process-

ing blocks. Transformers demonstrate high scalability to large-scale deep neural networks

and large datasets. These strengths have led to exciting progress on several vision tasks

using Transformer networks Khan et al. (2021).

We train our model on synthetic data as well as on real raster logs. In the multi-

segmentation model of digitizing a track with three curves, VeerNet performs with a pre-

cision of 0.94, 0.48 and 0.39 respectively. Whereas the VeerNet trained on real data has a

precision of 0.6 for three curves.

2 Work Flow in commercial software vs our proposed

solution

Fig.1 illustrates the overall workﬂow of the existing system for digitizing well-log graphs.

This workﬂow is from the recognized logging curve digitization software - Neuralog Due to

interference with the background grid, this software frequently pauses during curve tracking.

The software interface is based on a ribbon structure and includes a few modules. The steps

include (a) raster calibration and (b) digitization. During the raster calibration, the user

must provide a rectangular region that captures the header and scale areas. Also, a set

of points is deﬁned to capture the depth of the log tracking. User input is also required

to determine the left and right axis values and the type of scale, whether logarithmic or

linear. For the digitization ribbon, one needs to identify the track’s top and bottom, deﬁne

the image track with width, and add depth points and grids. Finally, the user must pick

points in the curve, and the software’s auto-trace functionality will trace the rest of the

curve. Our proposed solution (Fig. 2) operates on minimal manual intervention. The user

is not required to provide left or right input points; they don’t need to specify the width of

the track. VeerNet architecture does not have any metadata requirements related to depth

grids. Therefore, the algorithm can eﬃciently diﬀerentiate between grids and curves. The

user only needs to provide a cropped raster or paper log section.

3 Dataset: Synthetic and Real log curves

3.1 Synthetic Dataset

Reconstruction of well log curves through machine learning techniques like the random

forest, SVM (Akinnikawe et al., 2018), and deep neural network-based prediction models

Kim et al. (2020) is used to generate synthetic logs. These synthetic curves are generated for

target ﬁelds and require domain dependence. Here we aim to develop a generalized model

without any speciﬁc targeted oil ﬁeld. We hypothesized that the curves in a particular track

would have random overlapping and wrapping factors. The synthetic curves are generated

based on mean and standard deviation, adding random noise. Our dataset consists of

two well-log curves per image. The synthetic curves were developed based on parameters

文档加载中……请稍候！
如果长时间未打开，您也可以点击刷新试试。

下载文档到电脑，查找使用更方便

10 玖币 0人已下载

立即下载

摘要：

DigitizationofRasterLogs:ADeepLearningApproachMQuamerNasimab,NarendraPatwardhana,TannisthaMaitiaandTarrySinghaaDeepkapha.ai,Amsterdam,Netherlands*bIndianInstituteofTechnology,Kharagpur,IndiaAbstractRasterwell-logimagesaredigitalrepresentationsofwell-logsdatageneratedovertheyears.Rasterdigitalwelllog...

展开>> 收起<<

Digitization of Raster Logs A Deep Learning Approach M Quamer Nasima b Narendra Patwardhana Tannistha Maitiaand Tarry Singha.pdf

共18页,预览4页

还剩页未读，继续阅读

声明：本站为文档C2C交易模式，即用户上传的文档直接被用户下载，本站只是中间服务平台，本站所有文档下载所得的收益归上传人(含作者)所有。玖贝云文库仅提供信息存储空间，仅对用户上传内容的表现方式做保护处理，对上载内容本身不做任何修改或编辑。若文档所含内容侵犯了您的版权或隐私，请立即通知玖贝云文库，我们立即给予删除！

Digitization of Raster Logs A Deep Learning Approach M Quamer Nasima b Narendra Patwardhana Tannistha Maitiaand Tarry Singha

相关推荐

开通VIP享超值会员特权

作者详情

相关内容

热门标签

举报选择: