TECHNICAL FIELD

[0001] The present invention generally relates to mining of geological structures.

BACKGROUND

[0002] The reference to any prior art in this specification is not, and should not be taken as an acknowledgement or any form of suggestion that the prior art forms part of the common general knowledge.

[0003] A mobile mining spectral scanner can be used to perform hyperspectral scanning of a mine including scanning of mine faces, muck piles, cores, stockpiles and other geological structures.

[0004] It is desirable to mine geological structures where there is a higher abundance of minerals.

[0005] The preferred embodiment provides a useful analytical tool for such a purpose.

SUMMARY OF THE INVENTION

[0006] According to one aspect of the present invention, there is provided a geological analysis method for determining abundance results for one or more valuable materials in a geological structure, the method involving: receiving scan data relating to a geological structure; processing the data to determine a continuum; and generating an abundance result of the valuable materials using the continuum.

[0007] Advantageously, the abundance result can be used for mining applications whereby locations of the geological structure having higher abundance are more desirable to mine. [0008] Preferably, the method is automated.

[0009] The method may involve capturing the scan data using a hyperspectral imaging device. The method may involve forming a hyperspectral data cube. The hyperspectral data cube may include reflectance spectra over a spatial period.

[00010] The continuum may be determined using a heuristic, and preferably a greedy heuristic. The determined continuum may be optimized based upon minimal curvature and/or minimal offset from reflectance spectra. The continuum may be removed from reflectance spectra. The continuum may be determined by solving a least squares problem. The continuum may be determined by iteratively solving least squares problems. The continuum may be determined by splitting a solution vector into a passive and an active set, and iteratively updating the passive set with a highest residual.

[00011] The continuum may be determined by using pre-calculated QR decomposition for at least one least squares problem.

[00012] The abundance result may include a mineral percentage at a location of the geological structure.

[00013] According to another aspect of the present invention, there is provided geological analysis system for determining abundance results for one or more valuable materials in a geological structure, the system including: a scanner for capturing scan data relating to a geological structure; a processor for processing the data to determine a continuum; and a generator for generating an abundance result of the valuable materials using the continuum.

[00014] The scanner may include a hyperspectral camera. The scanner may be mobile.

[00015] Any of the features described herein can be combined in any combination with any one or more of the other features described herein within the scope of the invention. BRIEF DESCRIPTION OF THE DRAWINGS

[00016] Preferred features, embodiments and variations of the invention may be discerned from the following Detailed Description which provides sufficient information for those skilled in the art to perform the invention. The Detailed Description is not to be regarded as limiting the scope of the preceding Summary of the Invention in any way. The Detailed Description will make reference to a number of drawings as follows:

[00017] Figure 1 is a schematic diagram of a geological analysis system in accordance with an embodiment of the present invention;

[00018] Figure 2 is a higher level flowchart of a geological analysis method performed using the system of Figure 1 ;

[00019] Figure 3 is a lower level flowchart of the illumination pre-processing step of the higher level flowchart of Figure 2; and

[00020] Figure 4 shows outputs of the illumination pre-processing step of Figure 3 at different iterations.

DETAILED DESCRIPTION OF PREFERRED EMBODIMENTS

[00021] According to an embodiment of the present invention, there is provided a geological analysis system 100 or tool as shown in Figure 1. The system 100 includes a mobile hyperspectral scanner 102, which is an imaging device for capturing scan data 103 relating to a mine 104. The mine 104 may include mine faces, muck piles, core, stockpiles and other like geological structures, including minerals or other valuable materials.

[00022] The system 100 also includes a base processor 106 for processing of the hyperspectral scan data 103 to determine a continuum. The processor 106 (i.e. generator) also generates an abundance result of the minerals using the determined continuum.

[00023] An electronic display 108 is provided for displaying the abundance result to a geologist 110 who can then determine the best place to mine. The base processor 106 is typically in wireless communication with the remote scanner 102, and is also networked to the cloud for enabling the geologist 110 to utilise stored scan data 103 and software via a web-based portal.

[00024] Overview

[00025] Figure 2 is a flowchart of a geological analysis method 200 performed using the system 100. The automated method 200 is used to determine abundance results for one or more minerals in the mine 104.

[00026] At step 202, the scan data 103, from the scanner 102 and relating to the mine 104, is received by the processor 106. The mobile scanner 102 drives around the mine 104 capturing the data at various locations of interest and a hyperspectral data cube is formed. The hyperspectral data cube includes reflectance spectra over a spatial period.

[00027] At step 204, the method 200 involves illumination pre-processing the data 103 to determine a continuum. Illumination pre-processing of the hyperspectral data (reflectance spectra) produces an illumination spectrum using either known methods, or the method detailed below.

[00028] At step 206, the processor 106 generates an abundance result of the minerals using the continuum. The processor 106 continues processing the data 103, using machine learning with the illumination continuum as input, to produce abundance percentages useful for mining applications. The processor 106 takes the output of the illumination pre-processing step 204 (in this case the optimal fit of the continuum, removed from reflectance spectra) and generates a result in the form of abundances of minerals.

[00029] At step 208, the abundance result is displayed on the display 108 or can be printed for the geologist 110. The abundance result includes a mineral percentage at a location of the mine 104. Advantageously, the abundance result can be used for mining applications whereby locations of the geological structure having higher abundance are more desirable to mine.

[00030] Illumination pre-processing [00031] A known step in the processing of hyperspectral data analysis is to estimate and remove the illumination continuum from the reflectance spectra. The continuum line connects the local maximas of the reflectance spectra and is frequently estimated using the method developed by Clark & Roush (1984) and summarised by Clark (1999) and improved upon by Clark et al. (2003).

[00032] There are other various methods published in the scientific literature, including Mutanga et al. (2004).

[00033] Instead of using the foregoing Convex Hull method, a different method approach is developed which is more complex, aligns with underlying physics and gives better results. A model of the continuum is built and optimized having: (1) minimal curvature, (2) minimal offset from the captured reflectance spectra. In an ideal world this leads to a low pass filter to fit the reflectance spectra using off-the-shelf quadratic program algorithms, this however is computationally slow.

[00034] Part A - Greedy Heuristic to Improve speed

[00035] A first approach to getting reasonable runtimes is using a Greedy Heuristic (see Appendix I below). The solution vector is split into a passive and an active set (which is a common approach in Quadratic Programming algorithms). The passive set is the set of solution vector elements that are precisely zero and the active set includes the parts that are non-zero. The method then involves iterating multiple times based on a heuristic.

[00036] Figure 3 shows the illumination pre-processing step 204. In the least squares formulation, setting an element of the solution vector to zero is equivalent to cutting a row out of the A matrix.

[00037] The greedy heuristic steps are as follows:

204.1. Initialise the passive set to the empty set (i.e. the whole solution vector is active)

204.2. Solve the least squares problem via a QR factorization 204.3. Find the residual with the highest positive value

204.4. Add this residual to the passive set

204.5. Solve the least squares problem on this smaller problem

204.6. Iterate this procedure until there are no more positive residuals (with a tolerance).

[00038] Steps 204.1 and 204.6 involve initialising the working points set to be empty (i.e. the whole reflectance spectra is contained in the active set).

[00039] Steps 204.2 form the problem which is solved using a least squares via a QR factorization. This can then use standard methods, such as case Bjorck (1996), to calculate the solution. A known library, QRUpdate library (20220), can be used in implementing steps 204.2 and 204.5.

[00040] Steps 204.3 and 204.4 utilise the “greedy heuristic” approach by iteratively updating the passive set with the highest residual to provide speed optimisation.

[00041] Part B - Iterative updates to a pre-computed AR Set

[00042] Step 204.5 exploits the fact that, in the least squares matrix used to solve step 204.2, setting an element of the solution vector to zero is equivalent to removing a column out of the active matrix. Since the matrix A and the result vector b do not change within iterations (see Appendix I below), the method uses the pre-calculated QR decomposition that is needed for each least squares iteration.

[00043] These updates are performed directly to the pre-calculated QR factorisation which was original output via step 204.2, but is taken as an input to Bjdrck’s (1996) corrected semi-normal equation (CSNE) approach on each subsequent iteration. This greatly speeds up the solution time as it avoids the need for a full matrix solve at every iteration (as the rate limiting factorisation step is avoided).

[00044] Figure 4 shows outputs of the illumination pre-processing step of Figure 3 at different iterations. [00045] Since the matrix A and the result vector b do not change within iterations, the method can pre calculate the QR decomposition that is needed for each least squares iteration. As the passive set is updated, this is equivalent to removing rows from the A matrix. These updates are performed directly to the QR factorisation via Bjdrck’s corrected semi-normal equation (CSNE) approach (REF). This greatly speeds up the solution time as it avoids the need for a full matrix solve at every iteration (as the rate limiting factorisation step is avoided).

[00046] Von Neumann boundary conditions can be used for continuity of derivate at the ends to the signal. The regularization parameter can be tuned based on repeated scans of the same sample. Runtime was observed down to approximately 30 seconds on a single thread for a Chip scan. It was implemented to make use of multithreading and saturated at about 4 threads at approximately 2 times quicker. This physics informed normalisation procedure for the samples enables results of this decomposition to be fed into a subsequent machine learning (ML) processing pipeline.

[00047] In conclusion, the foregoing method is used for preprocessing the Hyperspectral data that is subsequently used for regressing XRF chemical species percentages from hyperspectral images that is performed. The crux problem in turning diffuse reflectance spectra into absorption spectra for mineral species classification is the determination of an unknown “continuum”. This is the remaining unknown after calibration for the known illumination spectra of a scan. In Hyperspectral literature this is typically estimated via the fitting of a convex hull to the reflectance spectra. There is no solid theoretical foundation for this assumption of convexity or linearity in the continuum. As a counterpoint to this modelling assumption a basic application Kubelka Munk theory of scattering from infinetesimal layers suggests a model that is the reciprocal of a low order polynomial for the continuum [REF]. More principled ways of estimating the continuum via variational methods are provided herein.

[00048] References

Bjorck, A. (1996). Numerical methods for least squares problems. SIAM.

Clark, R.N., Roush, T.L. (1984). Reflectance spectroscopy: Quantitative analysis techniques for remote sensing applications.: Journal of Geophysical Research: Solid Earth 89, 6329-6340. Clark, R. N. (1999). Spectroscopy of rocks and minerals and principles of spectro-scopy. In Manual of Remote Sensing, edited by A. N. Rencz, John Wiley, New York.

Clark, R. N., King, T. V. V. and Gorelick, N. S. (1987): Automatic continuum analysis of reflectance spectra. Proceedings of the Third Airborne Imaging Spectrometer Data Analysis Workshop, 30. 138-142.

Friedlander, M., Le Provost, M., Saba, E. (2022). QRUpdate Julia library (v1.0.00 [Computer software]. GitHub. httpsV/github.com/mpf/QRupdate.jl

Mutanga, O. and Skidmore, A. K. (2004). Hyperspectral band depth analysis for a better estimation of grass biomass (Cenchrus ciliaris) measured under controlled laboratory conditions International Journal of applied Earth Observation and Geoinformation, 5, 87-96.

Clark, R. N., G. A. Swayze, K. E. Livo, R. F. Kokaly, S. J. Sutley, J. B. Dalton, R. R. McDougal, and C. A. Gent, Imaging spectroscopy. (2003) Earth and planetary remote sensing with the USGS Tetracorder and expert systems, J. Geophys. Res., 108(E12), 5131 , doi:10.1029/2002JE001847.

[00049] A person skilled in the art will appreciate that many embodiments and variations can be made without departing from the ambit of the present invention.

[00050] In compliance with the statute, the invention has been described in language more or less specific to structural or methodical features. It is to be understood that the invention is not limited to specific features shown or described since the means herein described comprises preferred forms of putting the invention into effect.

[00051] Reference throughout this specification to 'one embodiment’ or 'an embodiment’ means that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one embodiment of the present invention. Thus, the appearance of the phrases 'in one embodiment’ or 'in an embodiment’ in various places throughout this specification are not necessarily all referring to the same embodiment. Furthermore, the particular features, structures, or characteristics may be combined in any suitable manner in one or more combinations. APPENDIX I

Reflectance Spectra Decomposition

A fundamental assumption of this method is that the illumination corrected reflectance spectra that we sample, p(λ), are the product of a wavelength dependent absorption spectrum, K(λ), and a continuum C(λ), i.e. p(λ) = K(λ)C(λ) (1)

The question is then framed as follows,

“how do we estimate the continuum K(λ) for a given reflectance spectra p(λ), without knowing the absorption spectra K(λ)?”.

This is essentially a blind source separation problem where we can use known attributes of the absorption and continuum spectra to pull these two signals

apart. The two crucial known pieces of information that allow for such a de- composition are

1. The continuum is strictly greater than the reflectance spectrum, i.e. C(λ) ≥ p(λ) ∀λ

2. The continuum is a smoother function than the absorption spectrum, which is typically composed of narrow absorption bands (where we as- sume that hold the information about material composition).

To make progress towards this decomposition we can make the following mod- elling assumptions

We model the continuum as the sum of the reflectance spectra plus an offset, i.e.

The ideal continuum will have minimal curvature m n C (3) |

The ideal continuum will have minimal offset to the refelctance spectra

Using the above desiderata and assuming the 12 norm in (3) and (4) as a linear objective yields the following optimisation problem where the regularisation parameter vi(A) has been introduced to control the smoothness of the estimated continuum.

Without the constraint that ο(λ) ≥ 0, equation (5) is simply a second order low pass filter on the reflectance spectrum and may be solved by a variety of simple methods on a discrete wavelength basis. The positivity constraint changes the nature of the problem however, it can be shown that this is instead a Quadratic Program via formulation of the dual form (REF), i.e.

(6) where Q = A'A, and c = — A’y for the equivalent least squares problem

Note that in our case on a discrete wavelength basis where x is the discrete basis o and Ais the finite difference operator.

Naive solution of a Quadratic Program requires exponential runtime, with quadratic complexity for some algorithms for positive definite Q (note we are only guaranteed semi-positive definite for our formulation of the problem). Both