Neural Network Interpretation Scheme For A Frequency-Domain Electromagnetic Ellipticity Surveys

A system of neural networks is used for the interpretation of electromagnetic ellipticity data from a frequencydomain<br>electromagnetic system developed at the Laboratory for Advanced Subsurface Imaging. The in-field data<br>interpretation consists of networks operating in mapping mode (1D) and object mode (3D, not yet implemented).<br>The purpose of the interpretation system is to provide a fast, approximate yet reasonable interpretation in the field<br>during a survey. Our approach to the neural network processing is to divide the interpretation into many parts<br>using several small networks. Three network paradigms were used for training, a backpropagation algorithm,<br>radial basis functions, and modular neural networks. We have tested the networks for a frequency range of 1kHz<br>to 1MHz with synthetic data and from a field survey in Wyoming.