Integrating two spectral imaging systems in an automated mineralogy application

Abstract

A system for the automated analysis and sorting of mineral samples has been developed to assist in the concentration of heavy mineral samples in the diamond exploration process. These samples consist of irregularly shaped mineral grains ranging from 0.3 to 1mm in diameter, and the application requires the treatment of this material at the rate of tens of grams per hour. Specific mineral grains of interest are identified and extracted from a mineralogically highly diverse background. Material is treated in batches, with trays of mono-layered material presented to various imaging systems. The identification of target grains is achieved by means of spectral imaging in two wavelength bands (Visible, and Long Wave Infrared). Target grains are then extracted by means of a SCARA robot, in a pickand-place arrangement. The successful development required the solution of various challenges in the image processing and materials handling domains. A previous paper has concentrated on the physical handling of the small, irregularly shaped grains. In this paper we concentrate on the imaging challenges, specifically associated with the registration between separate spectral images to a high spatial accuracy. The pick-and-place system required that the pneumatic nozzle is directed to a target position with an accuracy of 100 micron, in terms of both the particle plane and the height above the particle bed. We describe the registration and image distortion correction methodologies employed, in order to achieve this accuracy