Enhancing TGS Salt Identification with U-NET and Graph Neural Networks

Seismic imaging's ability to accurately demarcate salt bodies is vital for several oil and gas applications, including hydrocarbon exploration and reservoir assessment. Algorithms that attempt to recognize salt bodies in seismic data automatically can be tested on the TGS Salt Identification Challenge dataset. This research presents a new method for improving the accuracy of salt detection that combines U-Net with Graph Neural Networks (GNNs). This approach uses GNNs' relational reasoning capabilities in conjunction with U-Net's hierarchical feature representation capabilities to extract global and local contextual information from seismic imagery. The model successfully represents the intricate structural relationships in seismic data by enhancing the U-Net architecture with graph convolutional layers. Tested on the TGS Salt Identification Challenge dataset, the strategy outperforms state-of-the-art approaches. According to the experiments, the suggested U-Net with GNNs successfully identifies salt bodies in seismic pictures. This might lead to improvements in subsurface imaging and exploration for oil and gas