Coordinate transformation by radial basis function neural network

Abstract

The Turkish National Geodetic Network (TNGN) datum (ED50) was changed to the Turkish National Fundamental GPS Network (TNFGN) datum (WGS84) in 2001 in parallel with the increasing use of GPS technology. Due to this reference frame change it became necessary to transform the existing coordinate information between ED50 and WGS84. The two-dimensional (2D) affine transformation is widely used for coordinate transformation. The objective of this study is proposing a radial basis function neural network (RBFNN) that has been more widely applied in function approximation as an alternative coordinate transformation method. 2D affine transformation (Affine) method and RBFNN are evaluated over a study area, in terms of the root mean square error (RMSE). The results showed that RBFNN transformed the plane coordinates (Y, X) of the check points with a better accuracy (± 0.011 m, ± 0.013 m, respectively) than Affine method and pointed out that RBFNN can be used for coordinate transformation. Key words: Coordinate transformation, artificial neural network, radial basis function, affine transformation.