EAGE London: Semi-universal Geo-Crack Detection by Machine Learning

EAGE Local Chapter London

WEBINAR: Semi-universal Geo-Crack Detection by Machine Learning

<p>The talk will happen online - link to the webinar will be provided via e-mails to registered attendees: first e-mail will be sent <strong>two days before the event</strong> and the second one just <strong>2 hours before the event</strong>. Do not register too late!</p><h3></h3><h3>Agenda (UK time)</h3><p><strong>18:30-18:40</strong> Introduction and Announcements</p><p><strong>18:40-19:20</strong> Semi-universal Geo-Crack Detection by Machine Learning</p><p><strong>19:20-19:40</strong> Q&A </p><p><strong>19:40-21:00</strong> Informal discussion and relaxed networking </p><p>                                  (cameras + microphones + drinks ON)</p><h3></h3><h3>Speaker</h3><h3>Professor Gerard T. SchusterKing Abdullah University of Science and Technology</h3><p><strong>Gerard Schuster</strong> is a Professor of Geophysics at King Abdullah University of Science and Technology (KAUST) and an adjunct Professor of Geophysics at University of Utah, where he was a professor from 1985 to 2009. He was the founder and director of the Utah Tomography and Modeling / Migration consortium from 1987 to 2009 and has extensive experience in developing innovative migration and inversion methods for both exploration and earthquake seismology. Jerry, as his friends call him, helped pioneer seismic interferometry and its practical applications in applied geophysics, through his active research program and extensive publications, including his books "Seismic Interferometry" in  2009, "Seismic Inversion" in 2017 and "Machine Learning in Geosciences" to be published in 2021. He received a number of teaching and research awards while at University of Utah. He was editor of GEOPHYSICS from 2004-2005 and was awarded SEG’s Virgil Kauffman gold medal in 2010 for his work in seismic interferometry.</p><p></p><h3>Abstract</h3><p>Cracks are an outcrop’s  stress fingerprint etched by a region’s tectonic forces and weathering patterns. To deduce the stress history, the cracks of outcrops must be detected and catalogued over a wide area, and then interpreted by the structural geologist. Tens of thousands of high-resolution photographs can now be conveniently obtained by inexpensive drone surveys, so that cracks to a resolution of centimeters can be recorded. Here, we will use the term crack to indicate any discontinuity on the surface of the rock face,  whether it is a bedding surface, a shearing surface, a fracture or a joint where the rock  face separates due to cooling. </p><p>The problem with detecting cracks in petabytes of digital images is that it demands a highly-accurate and fast computerized method for accurate detection in the presence of noise such as shadows, vegetation, sedimentary interfaces and linear discolorations due to weathering. In the presence of such noise, deterministic edge detectors will typically fail. To overcome this problem we develop a modified U-Net architecture with transfer learning to detect cracks with an accuracy greater than 97.5% in our examples. We train this crack detector on labeled drone photos from a drone survey over large sandstone massifs in Saudi Arabia (see Figure 1a-1b). The bottleneck with the training the sandstone U-net model is that it required more than 100 hours of manually labeling more than 127 sub-photos. Once the U-Net was properly trained we discovered that it can be transformed into a semi-universal geo-crack detector using transfer training.  For example, with a  few hours of crack labeling in photos of volcanic outcrops located in Idaho, the trained sandstone U-Net with transfer training was able to achieve an estimated accuracy of 95% in detecting cracks in volcanic outcrops near a failed Idaho dam (see Figure 1c-1d).  We also found that this detector with transfer training could detect cracks in photos of a Martian floodplain taken by a Mars orbiter (see Figure 1e-1f).   We believe our semi-universal sandstone U-Net model with transfer training can accurately detect cracks in photos of many different geological environments. We also show that a machine learning algorithm can also separate one type of geo-signal from another to give a more interpretable image. </p>

<p>Figure 1. AlUla photographs a) before and b) after crack detection by the sandstone U-net model. Idaho dam photographs c) before and d) after crack detection; Mars orbiter photos e) before and f)  after crack detection.
</p><p></p>

Our {communityGuidelinesLink} describe the sort of content we prohibit on Eventbrite. If you suspect an event may be in violation, you can report it to us so we can investigate.

To report other categories of prohibited or illegal content, submit {link}

EAGE London: Semi-universal Geo-Crack Detection by Machine Learning

More events from EAGE Local Chapter London

Discover more events from EAGE Local Chapter London, from Science & Tech to other experiences you might love.

Still looking for the right event?

Explore all online events to browse and filter by date, category, and more.

EAGE London: Semi-universal Geo-Crack Detection by Machine Learning

More events from EAGE Local Chapter London

Discover more events from EAGE Local Chapter London, from Science & Tech to other experiences you might love.

More Science & Tech events

Browse more Science & Tech events with different dates, prices, and formats to find your next great experience.

Still looking for the right event?

Explore all online events to browse and filter by date, category, and more.