Today, we start a series of posts about Machine Learning Workflows supported in OpendTect. Each workflow is presented in the form of a short video. We will show that you do not have to be a data science expert to use Machine Learning solutions in day-to-day seismic interpretation work. Some videos show workflows from our original Neural Networks plugin, which is now an integral part of the Machine Learning plugin. Other videos describe new workflows for deep learning algorithms such as CNNs and Unets. Amongst others we will show that labels can be created with a paintbrush (image-to-point applications) and by drawing polygons (image-to-image applications).
We kick-off the series with an old-time favorite: Quick UVQ Waveform Segmentation. This very simple but extremely useful workflow visualizes seismic patterns along a mapped horizon. The user selects a time-window along a mapped horizon to be segmented into a user-defined number of segments. The network automatically finds the cluster centers in a training phase. It then follows up by applying the trained network to all trace segments along the horizon resulting in two output grids: a segmentation grid and a match grid expressing the confidence in the segmentation result on a scale of 0 to 1.