The Application of Neural Networks to Imaging and Signal Processing in Astronomy and Medicine

The growing use of computer technology throughout astronomy and medicine, as well as other fields of scientific research, has led to greatly increased rates of data acquisition. The advent of the computer also gave birth to the new research field of Artificial Intelligence (AI). AI techniques may allow the construction of intelligent filtering systems, to reduce the amount of acquired data which must be stored, and allow the vast quantities of scientific data already stored on computer to be analysed without the need for human intervention. The resurgence of neural networks within AI research is driven in pert by their ability to solve image and signal recognition problems, two areas where AI's rule based systems have performed poorly. Neural networks also have the ability to learn by example and generalise, allowing them to be trained on a representative set of examples for a given problem.

Neural nets have been applied to the problems of star/galaxy classification of telescope plates. Classification rates in excess of 90% for both stars and galaxies have been obtained to a limiting B-magnitude of 20. Event selection on the INTEGRAL gamma-ray satellite, using both neural networks and rule generation systems, has shown that signal (photons which passed through the mask) and background (photons which did not) events cannot be separated using a statistical classifier.. Finally the use of neural networks for electrical impedance tomographic image characterisation is considered. The network is shown to be able to characterise simulated images but this property does not extend to real images. Further clinical validation is required in order to improve the image data available for training the network

University of Southampton
Astronomy Group

Written by Ade Miller,
last updated on 26/07/1994.