Quantitative textural assessment for sample selection guided by drill core hyperspectral imagery

Introduction

Mining operation success is reliant on making adequate decisions throughout the mining value chain, from mineral exploration to metal refinement and waste treatment. Current technological advances (Schodlok et al., 2016) facilitate an increasingly data-driven decision-making process (Suazo, Kracht, and Alruiz, 2010), potentially improving productivity and environmental outcomes (Cracknell et al., 2018; Merrill et al., 2017). Despite the extensive rock property data collected as part of a typical mining operation, metallurgical and environmental test data is relatively scarce. This is largely due to the high costs associated with sample collection, preparation, transportation, and laboratory testing. Moreover, metallurgical sample selection protocols are often selected based on grade and spatial distribution, resulting in units that are critical for generating informative and reliable metallurgical and environmental models being underrepresented or completely missed (Parbhakar-Fox and Lottermoser, 2015).

In this study we propose a method for quantitatively assessing textural differences derived from hyperspectral drill core mineral maps. Statistical analyses of mineral maps was used to define discrete spatial patterns, which may have implications for mineral processing (Lamberg et al., 2013).

Methodology

Mineral co-occurrence probability fields

Based on the grey level co-occurrence matrix (Eichkitz, Amtmann, and Schreilechner, 2013) algorithm, a feature extraction method was developed to derive mineral co-occurrence probability fields (MCOPF), which computes the angle-distance dependent probability that a pixel is one of the target minerals with respect to a reference pixel in a given mineral map (Fig. 1). MCOPF are calculated for every pair of minerals, accounting for not only textural relations within minerals themselves but also with other minerals.

The input parameters for the calculation of the MCOPF are as follows:

1. Maximum distance: corresponds to the farthest pixel pairs considered, e.g., the drill core diameter

2. Radial step: the number of pixels between the reference pixel and its pair on each step until reaching the maximum distance

3. Angular step: increment of the angle on each step, which starts from 0 until reaching π radians (180°)

Processing time increases with decreasing radial and angular step or increasing maximum distance, resulting in higher resolution MCOPF.

Minimum rotational difference

The quantitative textural difference between two sections of drill core was calculated using a developed method minimum rotational difference (MRD; Fig. 2). This method proceeds by finding the minimum value obtained for the sum of the absolute difference between the MCOPF of two sections, while rotating one with respect to the other from 0 to π radians, that way the quantitative difference measurement becomes independent from the orientation of both: the drill hole perforation and the rock structures. This allows for the result to exclusively represent textural differences, separate from other spatial phenomena.

Results

Hyperspectral imagery of 700 m of drill core was sectioned into 30-cm pieces, each representing a sample candidate and MCOPF were calculated. The minimum rotational difference was derived for all samplecandidates, allowing to distinguish individual units.

The textural difference matrix in Figure 2 combines 100 sections (100 x 100) of drill core. Samples that are unique with respect to the common textures observed are represented with high values, whereas more common textures are displayed with low values. This visualisation provides geological information that focuses attention on drill core sections with unique properties.

Discussion and Conclusions

The proposed methodology successfully establishes a quantitative textural comparison among drill core sample candidates, enabling the future development of automated sampling protocols. Metallurgical and environmental knowledge may be incorporated through phenomenological understanding of the effect of specific minerals and textural patterns into downstream processes, allowing to correlate test data with textural and compositional data from hyperspectral or other imaging characterization techniques. We hope this sampling protocol improves data representation and relevance across the entire mining value chain from feasibility studies and plant design, through process control and optimization, to mine closure and waste management.