Fusing Gaussian Processes and Dynamic Time Warping for Improved Natural Gamma Signal Classification
- PDF / 642,720 Bytes
- 24 Pages / 439.37 x 666.142 pts Page_size
- 86 Downloads / 197 Views
Fusing Gaussian Processes and Dynamic Time Warping for Improved Natural Gamma Signal Classification Katherine L. Silversides1 Derek Wyman2
· Arman Melkumyan1 ·
Received: 19 January 2015 / Accepted: 18 May 2015 / Published online: 19 June 2015 © International Association for Mathematical Geosciences 2015
Abstract In the Hamersley Ranges of Western Australia, highly consistent marker shales are used to identify the stratigraphic location of the banded iron formation (BIF) or BIF-hosted ore. These marker shales produce distinctive signatures in the natural gamma downhole logs, which are currently processed by slow manual interpretation. The location of the marker shales is important when distinguishing lithology for geological and ore body mapping. Previously, two methods of automatically identifying these shales in the natural gamma logs have been proposed, using either Gaussian processes (GP) or dynamic time warping (DTW). There are advantages and disadvantages with each method and a combined result is desirable. However, the outputs from each technique are very different and cannot be directly combined. This paper proposes a method of standardizing the results to the same scale and combining them using a weighted average. The method is tested using two natural gamma signatures from a deposit that contains the Marra Mamba sequence of the Hamersley province. The results showed that the combined method outperforms both individual methods. The GP has accuracies of 87.0 to 92.0 %, increasing to 92.9 to 95.7 % when the method is certain. The DTW has accuracies of 90.0 to 90.6 %, increasing to 94.0 to 95.4 % when the method is certain. In comparison, the combined method has accuracies of 90.7 to 93.6 %, increasing to 94.4 to 95.9 % when this method is certain. The combined method balances the strengths of the individual methods, identifying a larger number of positive signatures than the GP, while retaining a higher classification accuracy for these signatures than the DTW.
B
Katherine L. Silversides [email protected]
1
Australian Centre for Field Robotics, University of Sydney, Rose Street Building (J04), Sydney, NSW 2006, Australia
2
School of Geosciences, University of Sydney, Sydney, NSW, Australia
123
188
Math Geosci (2016) 48:187–210
Keywords Banded iron formation · Natural gamma · Geophysics · Mine automation · Signal processing
1 Introduction The iron ore deposits located in the Hamersley Group in the Pilbara region of Western Australia are stratiform deposits that contain banded iron formation (BIF) interlayered with shale bands. The Hamersley Group was deposited between 2.6 and 2.4 Ga and consists mainly of chemical sedimentary rocks with some ash-fall deposits and intrusions (Lascelles 2006). The Marra Mamba and Brockman Iron Formations each contain one of these BIF and shale sequences, and in localized areas the BIF has been mineralized by supergene enrichment to form high-grade martite goethite iron ore (Thorne et al. 2008). The shale bands in the BIF are highly laterally consis
Data Loading...
