NEW HYBRID TECHNIQUE FOR GEOMETRIC CORRECTION OF HIGH RESOLUTION SATELLITE IMAGERY

Abstract

An Artificial neural networks (ANN) MATLAB software was developed with multi-layer perceptron (MLP) technique to derive the geometric correction coefficients. The Artificial neural network training was done using the deduced control points in a way that, image coordinates were used as input and the ground coordinates as output till reaching stabilization state of the neural network parameters. A change in the nature of the distribution of errors has been noted, as a result of the numerical stability of the neural network. A new technique was developed using neural networks to predict the earth coordinates of a set of new regular image points in the same area of the deduced random point’s data set and a new DDSM model. The RFM model was reused by implementing regularized points to reach the final model coefficients between satellite imagery space domain and ground space domain. The new technology improved accuracy by reducing the planimetric error by 39% and the elevation error by 45% of the error recorded when using traditional RFM model.