EFFICIENT COMPUTATION AND NEURAL PROCESSING OF ASTROMETRIC IMAGES

In this paper we show that in some peculiar cases, here the generation of astronomical images used for high precision astrometric measurements, an optimised implementation of the DFT algorithm can be more efficient than FFT. The application considered requires generation of large sets of data for the training and test sets needed for neural network estimation and removal of a systematic error called chromaticity. Also, the problem requires a convenient choice of image encoding parameters; in our case, the one-dimensional lowest order moments proved to be an adequate solution. These parameters are then used as inputs to a feed forward neural network, trained by backpropagation, to remove chromaticity.