Authors:
Carolina L. S. Cipriano
1
;
Domingos A. D. Junior
1
;
Petterson S. Diniz
1
;
Luiz F. Marin
2
;
Anselmo C. Paiva
1
;
João O. B. Diniz
1
;
3
and
Aristófanes C. Silva
1
Affiliations:
1
Applied Computer Group NCA-UFMA, Federal University of Maranhao (UFMA), Sao Luís, Brazil
;
2
Tecgraf Institute, Pontifical Catholic University of Rio de Janeiro (PUC-Rio), Rio de Janeiro, Brazil
;
3
Fábrica de Inovaç ão, Instituto Federal do Maranhão, Grajaú, Brazil
Keyword(s):
Hydrocarbons, Seismic Images, MLP-Mixer, U-Net, DenseNet, ResNet, Machine Learning.
Abstract:
The seismic data acquired through the seismic reflection method is important for hydrocarbon prospecting. As an example of hydrocarbon, we have natural gas, one of the leading and most used energy sources in the current scenario. The techniques for analyzing these data are challenging for specialists. Due to the noisy nature of data acquisition, it is subject to errors and divergences between the specialists. The growth of deep learning has brought great highlights to tasks of segmentation, classification, and detection of objects in images from different areas. Consequently, the use of machine learning in seismic data has also grown. Therefore, this work proposes an automatic detection and delimitation of the natural gas region in seismic images (2D) using MLP-Mixer and U-Net. The proposed method obtained competitive results with an accuracy of 99.6% (inline) and 99.55% (crossline); specificity of 99.79% (inline) and 99.73% (crossline).