The classification of industrial sand-ores by image recognition methods

The chemical and physical composition of the feldsparquartz sand-ore differ from location to location in a given open pit mine mine and the utilisation of the raw-ore, as well as its resale value will depend on these properties. Thus. a very important aspect of the operation is to determine quickly and accurately the properties of the sand at a given location. Laboratory methods are accurate but take time to execute, while relying on the experience of the mine foreman is often eclectic and may lead to unsatisfactory results. The aim of this paper is to formulate a pattern recognition algorithm to classify, with a very low probability of error, samples of sand-ore of given classes. Such classification should be fast and on-line. so that the sand grabber can use the information automatically.. The algorithm presented operates in two stages. In the first stage of operation. classification has been accurate over 92%, while after the refinement stage, precision has reached on average 96%. With so few objects misclassified, part of the remaining imprecision may be due to uncertainties in the classification ascribed, as it will be shown below. Details are given on how to implement this algorithm on-line, to guide the actual production process.