DSB Seminar: Salt classification using deep learning, Anders U. Waldeland

Traditional methods for salt classication consist of choosing a set of attributes that are sensitive to the characteristics of salt bodies and training a classication algorithm to discriminate between salt and other geological structures. Convolutional neural networks have the advantage of combining attribute extrac- tion and classication in one network. This allows both the attributes and classication to be trainable for the given application. In this work we show how this technique can be applied to salt classication in seismic datasets. The results shows that training a classier on one labelled inline slice is sufcient to classify other slices in the same dataset.