HAYFA AL-SUHAIMI


PRINCE MUHAMMAD UNIVERSITY,
HAYFA.ALSUHAIMI@GMAIL.COM

SALEH AL-DOSSARY


SAUDI ARAMCO,
SALEH.DOSSARY.6@ARAMCO.COM

Abstract

The resolution of seismic imagery greatly impacts a geoscientist’s ability to accurately identify and interpret subsurface features. Here, a novel and automated workflow is proposed to generate high-resolution imagery of subsurface geological features, such as faults, fractures and channels. The workflow consists of two components: 1) noise removal, and 2) geo-body edge detection and enhancement. The technology exploits the Phase Congruency Filtering (PCF) technique to produce a higher-resolution subsurface image.

The two-component workflow is applied in three broad steps. Firstly, a median filter is applied to the input seismic data to reduce random noise, while preserving discontinuities (edges). Secondly, the edges are detected and enhanced using a Generalized Hilbert Transform (GHT), which helps emphasize the geological feature boundaries. Finally, a PCF algorithm is used to further sharpen the geological features by locating them in-phase frequency components.

The proposed workflow was successfully tested on field data to improve fault pattern recognition. Test results indicated a significant improvement in the ability to detect faults by enhancing subsurface discontinuities in the seismic image. In addition, both the efficiency and quality of fault interpretation was enhanced by reducing ambiguities, such as noise and bias, in the interpretation process.

Introduction

One of the most important challenges of the petroleum industry is acquiring accurate information regarding the location and shape of underground geological structures, such as faults, fractures and channels, which give rise to discontinuities (edges) in the seismic data.

The phase congruency filter is a technology that exploits the in-phase characteristics of the edges in the image in the Fourier domain to detect edges and corners in images (Shafiq et al., 2017). Hence, one can find edges in the image by locating the in-phase frequency components of the image. Multiple workflows exist in the industry today to solve this challenge by using phase congruency technology. Some workflows automatically extract faults using Ant Tracking attributes to enhance results (Silva et al., 2005), others use the double Hough transform to extract faults (Jacquemin and Mallet, 2005). Also there’s the workflow of estimating Local Fault Extraction (LFE) volumes and skeletonizing them (Cohen et al., 2006).

Brian Russell’s workflow passes the seismic data through a de-noising algorithm followed by phase congruency (Russell et al., 2009). Having the data first de-noised while preserving as many faults as possible, followed by applying phase congruency to the de-noised data helps enhance the potential faults and eliminate small lineaments.

This paper proposes an automated novel workflow to generate a high-resolution highlight of subsurface geological features, such as faults, fractures and channels, by using 3D Phase Congruency technology, to solve the same problem.

Theory

The phase congruency function is calculated based on the phase congruency measurements, where 0 corresponds to no congruency, and 1 corresponds to maximum phase congruency (Eltanany and Safy, 2020). Phase congruency will precisely respond to seismic reflections (both strong and weak ones), which helps to locate borders and boundaries efficiently (Shafiq et al., 2017).

The proposed workflow consists of three major steps (Figure 1). First of all, we preprocess the seismic data using a median filter to de-noise the data while preserving features (Al-Dossary, 2014). The median filter works by replacing the central sample in a 3D seismic window by the median value of all samples falling within the analysis window. This 3D filter is typically of odd dimension (e.g., 3 × 3 or 5 × 5). The median filter is preferred over the mean filter because of its capability to reduce random noise while keeping the structures (edges) intact, and thus it is an edge-preserving filter. We leveraged 3D adaptive median filter (Al-Dossary, 2014). The adaptive median filter kernel size is dynamically determined: a large kernel size for noisy areas and small size for low-noise areas, to preserve as much fault data as possible. We then compute coherence attribute of the seismic data to enhance edges in the seismic data. Coherence seismic attribute provides a quantitative measure of the changes in waveform across a discontinuity, a measure of change in reflector dip magnitude and azimuth across a discontinuity or a measure of change in reflectivity amplitude as energy across a discontinuity. Such discontinuity measures may highlight possible faults, fractures or channels. Luo et al. (2003) have presented amplitude gradient method based on the generalized Hilbert transform (GHT) that can detect abrupt and gradual amplitude changes associated with fault and channels. We applied GHT to the smoothed seismic data to detect the faults in the seismic data.

Finally, a 3D Phase Congruency filter is applied to enhance potential faults according to the continuity and direction of the faults, which will allow us to eliminate small lineaments and enhance the main faults.

The following sections explain the algorithm used in each step one by one.

Figure 1. Proposed workflow chart of fault skeletonization

Phase Congruency Filter

The phase congruency formula used in the proposed workflow is as follows (Shafiq et al., 2017),

where o is number of orientations, n is number of instances, A(i) is the Fourier components amplitude, Φ is the Fourier components phase, To is noise influence estimation, is a small threshold to avoid division by zero, W(i) is the weighting function at an orientation, and the operator “[⫞] ” is for applying soft threshold.

To apply phase congruency, the data needs to be converted from space-time domain to Fourier domain. It is also necessary to apply a band-pass filter to smoothen the image due to its high sensitivity to noise, since phase congruency values range only between 0 to 1 (Jiang et al., 2016). The Log Gabor function is chosen for this purpose (Russell et al., 2009),

where r is the zero-frequency radius, ls is the scale value, σ is the log Gaussian function width, and is low pass butterworth filter (Russell et al., 2009).

Afterwards, the radial Log Gabor filters are multiplied by the angular filters, then applied to the input image. The angular filters are created over M orientation angles. The default value of the number of angular filters is 6, which means angles range from 0o in increment of 30o to 150o.

After the filters have been applied successfully, the image is inversed back to the space-time domain using an Inverse Fast Fourier Transform. Some modifications are then done to the data including noise threshold and weighting, and special domains of each scale are then added up to have images corresponding to each orientation. Analysis of the resulting image is performed to get the ending results of maximum (M) and minimum (m) moments with the use of their respective equations below, having maximum moments correspond to edges and minimum moments correspond to corners (Russell et al., 2009).

The following applies to each variable:

To test our proposed workflow and compare it to current industry workflows, we processed the synthetic seismic data shown in Figure 2a; Figure 2b shows the same input slice with lines over the faults detected in the image. The two outputs are shown side-by-side in Figures 2c and 2d. Comparison of the results shows that the proposed workflow has more well-defined and exact edges, with minimal loss of features. In contrast, the current industry workflow fails to remove data that are not edges, and delineation is of lower overall quality.

Figure 2. Synthetic example process of phase congruency. (a) Input synthetic seismic data. (b) Input synthetic seismic data with lined faults. (c) Phase congruency application alone. (d) Sobel operator followed by phase congruency application.

2-Dimensional Seismic Data

To demonstrate the effectiveness of our workflow, we applied it to a 2-dimensional seismic data time-slice to determine detailed mapping of the faults. First, we applied median filtering to the time slice to enhance structural continuity and minimize random noise, while keeping the faults intact. We then computed coherence from the median filter output, resulting in Figure 3a. Then the Phase Congruency filter was applied to scale each sample between 0 and 1, adding up the edges (Figure 3b) and corners (Figure 3c) as outputs. The resulting image is shown in Figure 3d, where values of 1 correspond to maximum phase congruency and values of 0 correspond to no congruency.

Figure 3. Phase congruency results. (a) Noise Reduction and Coherence Attribute image used as input. (b) Edge detection through calculating the maximum moments. (c) Corner detection through calculating the minimum moments. (d) Adding results (b) and (c) to get phase congruency final output.

For comparison, Figure 4 shows the result of a current industry workflow (Figure 4b) with the same data slice (Figure 4a). Comparing our proposed workflow output alongside the current workflow output, noticeable differences can be observed. Our proposed workflow provides a more precise output and enhanced feature details, which appear to be lost in the current workflow’s output. The output of the latter also appears to have increased noise. The proposed workflow output provides a better overall quality, and more accurately defined skeletonized faults, as shown in Figures 5a and 5b.

Figure 4. Current workflow input and output. (a) Noise Reduction and Coherence Attribute image used as input. (b) Output of going through current industry workflow; noise reduction, edge detection, radon transform, and skeletonization.

Figure 5. Current workflow vs. proposed workflow results. (a) Current workflow; noise reduction, edge detection, radon transform, skeletonization. (b) Proposed workflow; noise reduction, edge detection, phase congruency.

Conclusions

This paper proposes a new workflow to enhance fault recognition by sequential application of a Median filter, followed by Coherence filter and final Phase Congruency filter.

The Median filter is first used to enhance the signal-to-noise ratio while preserving major faults, and the Coherence filter is then applied to enhance edges. Finally, a Phase Congruency filter is applied to further enhance fault planes for improved automated picking.

Our test results indicate that the proposed workflow is able to output precise fault recognition regardless of fault width. Additionally, it is apparent that although the Phase Congruency filter is relatively robust when used for major fault recognition, weaker faults may be overlooked. This workflow may be extended for future research to detect fault surfaces automatically.

References

Al-Dossary, S, 2014, Random noise cancellation in seismic data using a 3-D adaptive median filter, 84th Annual International. Meeting Society of Exploration Geophysicists, Expanded Abstracts, 2512-2516

Bahorich, M., and S. Farmer, 1995, The coherence cube: The Leading Edge, 14, 1053–1058

Cohen, Israel et al. “Detection And Extraction Of Fault Surfaces In 3D Seismic Data”. SEG Library, 2006, https://library.seg.org/doi/full/10.1190/1.2215357.

Eltanany, Abdelhameed, and Mohammed Safy. “SAR Images Co-Registration Based On Phase Congruency Algorithm”. Ieeexplore.Ieee.Org, 2020, https://ieeexplore.ieee.org/document/9096867.

George, Ginu et al. “A Survey On Various Median Filtering Techniques For Removal Of Impulse Noise From Digital Image”. Ieeexplore.Ieee.Org, 2018, https://ieeexplore.ieee.org/document/8544273.

Gersztenkorn, A., and Marfurt, K. J., 1996a, Coherency computations with eigenstructure: 58th Internat. Mtg., Eur. Assn. Geoscientist and Engineers, Extended Abstracts, x031.

Han, Ke et al. “Fingerprint Image Enhancement Method Based On Adaptive Median Filter”. Ieeexplore.Ieee.Org, 2018, https://ieeexplore.ieee.org/document/8633498.

Jacquemin, Pierre, and Jean-Laurent Mallet. “Faults Extraction From Seismic Cube Using A Double Hough Transform”. Ring-Team.Org, 2005, https://www.ring-team.org/research-publications/ring-meeting-papers?view=pub&id=5167.

Jiang, Xia et al. “Automated Identifying Granule Features On The Solar Photosphere Using Phase Congruency Technique”. Ieeexplore.Ieee.Org, 2016, https://ieeexplore.ieee.org/document/7528532.

Luo,Y., S.Al-Dossary, M. Marhoon, and M. Alfaraj, 2003, Generalized Hilbert transform and its application in Geophysics: The Leading Edge, 22, 198–202

Luo, Y., W. G. Higgs, and W. S. Kowalik, 1996, Edge detection and stratigraphic analysis using 3-D seismic data: 66th Annual International Meeting, SEG, Expanded Abstracts, 324–327.

Russell, Brian H. et al. “Applying the Phase Congruency Algorithm to Seismic Data Slices – A Carbonate Case Study.” First Break 28 (2009): n. pag.

Shafiq, Muhammad et al. “Salt Dome Detection Within Migrated Seismic Volumes Using Phase Congruency”. SEG Library, 2017, https://library.seg.org/doi/epdfplus/10.1190/segam2017-17443246.1.

Silva, Celina et al. “Automatic Fault Extraction Using Ant Tracking Algorithm In The Marlim South Field, Campos Basin”. SEG Library, 2005, https://doi.org/10.1190/1.2148294.

About the Authors

Hayfa Alsuhaimi holds a Bachelor’s degree in Computer Science from Prince Mohammad bin Fahd University. Hayfa’s interests shifted towards machine learning which led her to further develop herself in this rapidly evolving field. Hayfa pursued her internship at Aramco in the Geophysical Applications Division under the Exploration Application Services Department, where she explored the different seismic processing techniques, gaining valuable hands-on experience.

Saleh A. Al-Dossary began his work at Saudi Aramco in the Dhahran Geophysical Research Group, contributing to the edge-preserving and smoothing developments. He now works in the Exploration Application Services Department, developing new seismic processing, attributes and machine learning and quantum computing algorithms. Saleh holds BS in Computer Science and minor in Geophysics from NMT; MS from Stanford University; PhD in Geophysics from University of Houston.

Saleh holds fourteen patents, and is an applicant for five additional patents in seismic edge preserving and detection technology. He published multiple papers with SEG and EAGE, and authored two books on seismic attributes and noise reduction.