| Convolutional Versus Unsupervised Vector Quantizer Neural Networks for 3D Automated Gas Chimney Detection, West Offshore, Nile Delta, Egypt |
| Paper ID : 1041-ISCH |
| Authors |
|
Amir Ismail *1, Amer Shehata2, Ahmed B. Ahmed3, Takeshi Tsuji4, Mohamed Ahmed5 1Department of Geology, Faculty of Science, Helwan University, Egypt 2Associate Professor 3Department of Systems Innovation, Graduate School of Engineering, The University of Tokyo 4Department of Systems Innovation, Graduate School of Engineering, The University of Tokyo, Tokyo 113-8656, Japan 5Department of Physical and Environmental Sciences, Texas A&M University – Corpus Christi |
| Abstract |
| Seismic attributes refer to a numerical evaluation of a specific seismic property, forming a crucial component in the interpretation of seismic reflections. It quantifies both the waveshape (i.e., amplitude) and morphological aspects observed within seismic data. The initial stage in the chimney detection procedure involves utilizing several seismic attributes to enhance the chimney features in seismic volume. A convolutional neural network (CNN) is a division of machine learning that has been used in the automatic prediction of seismic features. CNN comprises four main elements: convolutional (trainable) filters, activation functions, down-sampling, and fully connected layers. Supervised interpreted seismic sections are used for the training process along with real-time data augmentation. We employ other sections for data validation. Furthermore, we use the entire 3D seismic cube to test the trained CNN model. Three statistical measures were used to assess the model performance for the studied attribute models. Preliminary findings indicate that there are strong correlation coefficients for the chaos, amplitude, and RMS attributes, which exhibit the highest levels of accuracy and loss function. These statistical results indicate that chaos, amplitude, RMS, and variance attributes exhibit superior performance in the prediction process versus Unsupervised the vector quantizer neural network (UVQ-NN). Utilizing seismic attributes as CNN inputs proves effective in detecting gas chimneys versus the UVQ-NN. The observations presented show the prediction procedure relies on CNN algorithms to provide a precise flowchart showing visual evidence to improve gas chimneys and associated gas seepage detection, reducing exploration and development risks. |
| Keywords |
| Offshore Nile Delta; gas chimneys; seismic attributes; hazard evaluation, convolutional neural network (CNN) |
| Status: Abstract Accepted (Oral Presentation) |