An artificial neural network is used for the classification of minerals. Op
tical data using thin sections is acquired using the rotating polarizing mi
croscope stage, which extracts a basic set of seven primary images during e
ach sampling. A selected set of parameters based on hue, saturation, intens
ity and texture measurements are extracted from the segmented minerals with
in each data set. Parameters such as pleochroism, plane light hue, and grad
ient homogeneity were a few that proved to yield class-discriminating prope
rties. Texture parameters are shown to have the ability to classify colourl
ess minerals. The neural network is trained on manually classified mineral
samples. The most successful artificial network to date is a three-layer fe
ed forward network using generalized delta error correction. The network us
es 27 distinct input parameters to classify 10 different minerals. Testing
the network on previously unseen mineral samples yielded successful results
as high as 93%. (C) 2001 Elsevier Science Ltd. All rights reserved.