Powerful pattern recognition software
The Neural Network plugin supports supervised and unsupervised neural networks to combine multiple attributes into "meta-attributes". The main application of unsupervised networks is a clustering of attributes and/or waveforms for seismic facies analysis. The supervised approach is used for more advanced seismic facies analysis and to create object "probability" cubes such as TheChimneyCube and TheFaultCube. Supervised neural networks are also used in reservoir characterization work flows to predict rock properties from seismic volumes.
"OpendTect provides state-of-the-art pattern recognition tools to complement your eyeball quantitatively."
Supervised neural networks can find highly non-linear relationships in data sets comprising examples of input vectors and desired outputs. In seismic object detection examples are created by the seismic interpreter who picks locations of objects to recognize (e.g. chimneys) as well as counterexamples (points where the object is absent - non-chimneys). At all picked locations multiple attributes are extracted on which the network is trained. Application of the trained network yields a new volume with the "probability" of the target output (chimneys).
Supervised neural networks can also be trained on seismic attributes versus well logs whereby the training set is created by sliding along well tracks. The trained network will then predict rock properties such as Porosity or Vshale e.g. from inverted Acoustic Impedance and/or EEI input volumes.
Unsupervised neural networks reveal structure in the data itself. A fast work flow to reveal seismic patterns along a mapped horizon is the "Quick UVQ" method in OpendTect.
For optimal results the neural network plugin should be combined with the "dip-steering" plugin as this allows attributes to be extracted along the seismic reflectors.