Tomographic velocity images by Radial Basis Function Artificial Neural Network
Abstract
The seismic tomography can be used to constrain estimates of the Earth's velocity structure. This kind of problem is usually known to be non-linear, high-dimensional, with a complex search space which may be riddled with many local minima, and results in irregular objective functions. We investigate here the performance and the application of a radial basis function artificial neural network (RBF-ANN) type, in the tomographic velocity reconstruction. The proposed structure has the advantage of being easily trained by means of a backpropagation algorithm without getting stuck in local minima. An adequate cross-validation test is run to ensure the performance of the network on new data sets. The application of such a network to synthetic data shows that the inverted seismic velocity section was efficient. A comparative reconstruction with tow classical methods was performed using Algebraic Reconstruction Technique (ART) Conjugate Gradient (CG). The results clearly show improvements of the quality of the reconstruction obtained by radial basis function artificial neural network.