Automated
Classification of Sedimentary Units in Drill Cores and Outcrops using
high-resolution Hyperspectral Imaging and Neural Networks
Ragona, Daniel Eduardo1,
Bernard Minster2, Tom Rockwell3 (1) University of
Califorinia, San Diego, La Jolla, CA (2) University of California, San Diego,
La Jolla, CA (3) San Diego State University, San Diego, CA
We present a new methodology for
automatic mapping of sedimentary stratigraphy of drill cores or outcrops using
short wave infrared (SWIR) hyperspectral images and multilayer perceptron
(MLP). Ground-based hyperspectral imaging provides an effective method to study
and store stratigraphic and structural data from cores or field exposures.
Neural networks supply a variety of well-established techniques towards pattern
recognition, especially for data examples with high-dimensionality
input-outputs. High-spatial/spectral resolution data from drill cores were
collected using a portable hyperspectral scanner with 245 continuous channels
measured across the 960 to 2404 nm spectral range. The data were corrected and
pre-processed to obtain the reflectance spectra. For the supervised
classification we built a set using hundreds of reflectance spectra of the
sediments of the cores. The examples were grouped into eight classes
corresponding to the most common materials found in the samples. A MLP was
trained to construct the classification models. The best model achieved 98.4 %
classification accuracy. Using this model we generated classification images of
the cores that show an excellent description of the stratigraphic units of the,
even in areas where the human eye was not capable to identify layers. In
conclusion, the results of this work show that reflectance spectra combined
with neural networks can be used to properly discern and classify sediments of
very similar composition and grain size. Quantitative identification of
geological materials can be used as a fast and objective method to describe,
automatically classify and assist in the correlation of samples, drill cores
and outcrops.
AAPG Search and Discover Article #90063©2007 AAPG Annual Convention, Long Beach, California