Salt Body Segmentation in Seismic Images using Mask R-CNN

Salt body exploration has always been a major area of study in Geophysics and the associated area of seismic interpretation, owing to its commercial significance in the oil and petroleum industry. In recent years, with the advancement in the area of machine learning; deep learning models are used for the efficient detection and segmentation of salt body from seismic images. In this paper, a deep neural model with instance segmentation capability (Mask R-CNN) has been proposed for the detection and segmentation of salt body from a set of seismic images. The model accepts a seismic image as input, performs the required classification of the pixels and outputs an image with annotated marks, clearly showing the segmented salt regions. The model has been trained and tested using the Seismic image dataset provided by Kaggle. The results obtained have been compared with the already known standard CNN model in this area and the experimental results shows that the proposed model is better.