| Autor: |
B. Conche, Denis Akhiyarov, Mauricio Araya-Polo, A. Gherbi |
| Rok vydání: |
2021 |
| Předmět: |
|
| Zdroj: |
82nd EAGE Annual Conference & Exhibition. |
| DOI: |
10.3997/2214-4609.202113203 |
| Popis: |
Summary Oil slicks from natural seeps and man-made spills of hydrocarbons need to be monitored in real-time to prevent environmental hazards. Remote sensing techniques are suitable for the task. Unfortunately, human intervention prevents real-time monitoring due to the limited area that can be processed by remote sensing specialists in GIS applications at a time. In this study, the focus is on how Machine Learning -in particular Deep Learning (DL)- can help automating the above mentioned task. Thus, a DL architecture is trained with a Total’s proprietary dataset for segmenting oil slick regions from satellite images. Further, the training stage is scaled and enhanced with high-performance computing techniques. These techniques allowed the original task to be solved up to 2.5× faster than the baseline for single GPU and pseudo-linear scalability for the distributed case. The later takes this application closer to real-time use case level. |
| Databáze: |
OpenAIRE |
| Externí odkaz: |
|
