Improving the planimetric position accuracy of the absolute GPS receiver using artificial neural networks

Abstract EN

A back propagation neural network (BPN) architecture has been trained with 149 patterns derived from a Navigator GPS observations and a satellite image (acquired by IKONOS imaging system in 2000 and covered the center of Khartoum).

the observed GPS coordinate values were used to generate the input part of the learning patterns.

the corresponding image coordinates of the same points were used to generate the output part.

a performance test sessions was carried out by applying the trained models to the same training patterns and to 25 new test patterns.

the shell was found to be able to process the whole set of the training and test patterns.

the absolute average error in the resultant direction for the test points was reduced from ± 32.478 m (computed from actual GPS observation) to ± 8.387 m (obtained from BPN computations).

an improvement of 75.92% was achieved by applying the BPN architecture against the direct GPS observations.