A genetic algorithm scheme for pairing meteorite finds

Abstract

A genetic algorithm is employed to perform the pairing of meteorite fragments based on various characteristics measured from thin sections using an image analysis program and from analyses routinely carried out during classification. The genetic algorithm searches for best group pairings by generating a population of trial pairs, linking them together to form groups, and evolving the population, so that only pairs that are members of likely pairing groups survive to the next generation of the population. In this way, meaningful pairing groups will emerge from the population, as long as characteristics from within real pairing groups have variance sufficiently small compared to the variance between groups. What constitutes 'sufficient small' is discussed and investigated by testing the genetic algorithm method on artificial data, which shows that, in principle, the mehtod can achieve a 100% success rate. The method is then tested on real data whose pairing groups are definitely known. This is achieved by gathering data from the image processing of several scenes of the same meteorite thin section, treating each scene as a separate fragment. Using thin sections from the Reg el Acfer meteorite population, we find that the genetic algorithm identifies almost all of the main pairing groups, with about half the groups being found in their entirety; the pairwise success rate being 76%. Although this methodology requires some refinement before it could be applied to a population of meteorite fragments, these preliminary results are encouraging. The potential benefit of an automated approach lies in the tremendous savings in time and effort, allowing meaningful and reproducible pairings to be made from data sets that are prohibitively large for a human being.