Salt segmentation beneath earth surface
- 0 Collaborators
Several areas of Earth with large accumulations of oil and gas also have huge deposits of salt below the surface. But unfortunately, knowing where large salt deposits are precisely is very difficult. Professional seismic imaging still requires expert human interpretation of salt bodies. This leads to very subjective, highly variable renderings. More alarmingly, it leads to potentially dangerous situations for oil and gas company drillers. So I am trying to do research and propose a model which can segment them very precisely with some good models baseline like Unet. ...learn more
Project status: Under Development
Intel Technologies
AI DevCloud / Xeon,
Intel Opt ML/DL Framework,
Movidius NCS
Overview / Usage
Imaging salt has been a huge topic in the seismic industry, basically since they imaged salt the first time. The Society of Exploration geophysicist alone has over 10,000 publications with the keyword salt. Salt bodies are important for the hydrocarbon industry, as they usually form nice oil traps. So there's a clear motivation to delineate salt bodies in the subsurface.
One common seismic attribute is called "chaos" or "seismic disorder". So if you talk to cynic geophysicists, you'll hear "that deep learning better outperform the Chaos attribute".
Methodology / Approach
Currently, the difficulties fo segment the salt rocks that the texture is so similar so hard to segment.
So I am trying using deep convolution network which has an architecture like autoencoder with and convolution encoder and deconvolution decoder which can learn feature combine them in the dense features after that deconvolution upsamples and segment the salt from other parts of images.
We are using a UNET like structure used for segmentation in medical imaging to get a kickstart because it connects convolution to deconvolution layers to make like FPN networks and propagate even small detailed features forward in the network to get better results.
Technologies Used
I am using intel dev cloud to train the model with high-end Intel processors and better GPUs to reduce my time to train and try different experimentation. In near future trying to optimize the models with Intel optimized Tensorflow and Intel Movidius Neural Compute Stick to make a product that can be used by others.