An Efcient Approach for Semantic Segmentation of Salt Domes in Seismic Images Using Improved UNET Architecture

Many areas of Earth’s surface with large accumulations of gas and oil even have huge deposits of salt under the surface. Exploring such deposits helps many countries to increase the storage capacity of their Petroleum reserves and explore new ones. But fnding such deposits is a herculean task. Expert seismic imaging requires human interpretation of salt bodies. But this leads to very biased and highly variable translations. So the idea behind this paper is to build an approach that accurately and automatically identifes if the seismic image contains any region of salt deposit or not. If a surface is found to have salt deposits, then it may contain the accumulations of oil or gas and even the salt domes or caverns can be used as a storage site for already available petroleum or oil. Since semantic segmentation classifes every pixel in the given image to its class label, this can be used to segment the salt deposits from the provided seismic images. In this paper, we introduce a variation of UNet, a popular segmentation model, for seismic
image segmentation. We have added a batch normalization
layer following every convolution layer as a deeper network
helps extract better features which turned out to be true