Automated Mineral Identification by Optical Microscopy

Mineral maps of æintactÆ and large particle textures have been produced routinely from optical microscope images. Polished thin sections of large particles (1 - 3 mm) provide a balance between representative sampling and the preservation of primary rock textures. Optical microscopy is an efficient way to generate mineral maps suitable for geometallurgical applications. The automated optical mineral identification was carried out using a Leica DM6000¬ microscope with a high precision stage suitable for direct tiling of image frames and good registration of multiple image layers. Standard images collected were 10 x 10 frames (20 mm x 16 mm). Four image layers were recorded: ætransmitted light plane polarisedÆ, ætransmitted light crossed polarsÆ, ætransmitted light crossed polars with 1l tint plateÆ and æreflected lightÆ. All image analysis was carried out using Definiens Professional V5¬. The four image bands were read directly in 12 greyscale bands. A segmentation algorithm produced objects based on homogeneity across all bands. These objects were then classified using a complex set of rules (Process Tree) that are based on image brightness in all 12 bands. Opaque and non-opaque minerals were classified in the same sample. Using this protocol up to 12 minerals can be satisfactorily classified in one sample. The opaque minerals were recognised using colour ratio and reflectance intensity in the reflected light image. Non-opaque minerals are more difficult because of the complex variable colours of many minerals and the complexity of interference colours. However, object-oriented classification using three transmitted light image layers and multiple rules is a powerful technique for recognition of these minerals.