Seismic Facies Segmentation

Project about Semantic Segmentation in Seismic Images (Facies).

Objective

Implement deep neural networks for Semantic Segmentation for Facies Clasification.

Dataset

The dataset used was the Netherlands F3 block, which is a fully-annotated 3D geological model open-sourced by Alaudah et al. It has six classes, where each one represents a facies with the exception of one that is the union of two facies because it was difficult to define the limits between them. The three-dimensional block has a dimension of 600x900x255. In order to get a model that generalizes correctly, ranges were defined to split the data in one block for training and two testing blocks.

Implementation

Docker available for this proyect:

docker pull smitharauco/tensorflow_1.13:latest

First, .txt files are generated for loading the sections after performing the split in training, validation and test set. For this use the notebook generate_and_load_sections.ipynb.

For training the model, you can see an example in Baseline_UNet_Train.ipynb.

Results

The following table shows the two best results, which managed to outperform the results of the paper that presented the data. For more details, see Towards a Benchmark for Sedimentary Facies Classification: Applied to the Netherlands F3 Block.

Model	Pixel Accuracy	Mean Class Accuracy	Frequency-Weighted Intersection over Union
Alaudah et al.	0.905	0.817	0.832
BiAtrousUNetConvLSTM	0.942	0.848	0.894
Atrous UNet	0.943	0.871	0.895

References

@InProceedings{campos2020f3,
    author="Campos Trinidad, Maykol J. and Arauco Canchumuni, Smith W. and Cavalcanti Pacheco, Marco Aurelio",
    title="Towards a Benchmark for Sedimentary Facies Classification: Applied to the Netherlands F3 Block",
    booktitle="Information Management and Big Data",
    year="2021",
    publisher="Springer International Publishing",
    address="Cham",
    pages="211--222",
    isbn="978-3-030-76228-5"
}

https://link.springer.com/chapter/10.1007/978-3-030-76228-5_15

@InProceedings{campos2021convlstmf3,
author="Campos Trinidad, Maykol J.
and Arauco Canchumuni, Smith W.
and Queiroz Feitosa, Raul
and Cavalcanti Pacheco, Marco Aurelio",
title="Seismic Facies Segmentation Using Atrous Convolutional-LSTM Network",
year="2021",
booktitle="Proceedings of the XLII Ibero-Latin-American Congress on Computational Methods in Engineering and III Pan-American Congress on Computational Mechanics, ABMEC-IACM",
}

https://cilamce.com.br/anais/arearestrita/apresentacoes/252/10005.pdf