Automated lithological classification from drill hole geochemistry

[摘要] Summary-- Classification methods which do not take spatial information into account are typically noisy when plotted in geographic space.-- In order to reduce noise each geochemical variable is assigned to spatially contiguous domains of similar composition (1D drill hole domains).-- For multivariate data the domains from different variables are merged.-- The domain compositions are compared by means of a similarity matrix calculated using symmetrical Kullback-Liebler divergence.-- Spectral clustering is applied to the similarity matrix to obtain clusters of similar domains; domains are labelled with a rock type.

[发布机构] CSIRO

[学科分类] 地球科学(综合)