WO2024059941A1

WO2024059941A1 - System and method for automating quantitative mineralogy using optical microscopy

Info

Publication number: WO2024059941A1
Application number: PCT/CA2023/051248
Authority: WO
Inventors: Benjamin DE CASTRO; Mostafa BENZAAZOUA; Soumali ROYCHOWDHURY
Original assignee: Clemex Technologies Inc.
Priority date: 2022-09-21
Filing date: 2023-09-20
Publication date: 2024-03-28

Abstract

A system and method for automatically analyzing mineral particles in a raw solid sample are disclosed in the present application. The system comprises: a device for applying a fast curing resin coating to a ground/polished sample obtained from the raw sample to prepare a material ready for analysis; an optical microscope; a camera acquiring digital images of the material through the optical microscope; and a computing device that includes a deep learning module trained on optical images for detecting particles. The computing device analyzes the digital images using to identify the mineral particles in the raw sample using the acquired digital images of the material. The hardening of the fast curing resin results in a bubbled matrix which, when observed under the optical microscope, enables distinction between mineral particles and bubbled matrix in the digital images.

Description

System and Method for Automating Quantitative Mineralogy using Optical Microscopy

TECHNICAL FIELD

[0001] This patent application relates to quantitative mineralogy. This patent application also relates to a system and a method for automating quantitative mineralogy using optical microscopy.

BACKGROUND

[0002] Identification of properties of rocks and minerals by studying their optical properties is known as optical mineralogy. Optical mineralogy involves measuring optical properties of minerals and rocks. Some of these properties of rocks and minerals are macroscopic and can be seen in hand specimens. However, the use of a microscope is generally necessary. Most commonly, rock and mineral samples are prepared as thin sections or grain mounts for study in the laboratory with a specialized optical microscope. Image analysis is then carried out to identify and characterize the properties of rocks and minerals, inferred from the optical properties.

[0003] Many papers have contributed to the development of mineralogical optical image analysis. These include: Donskoi, E., & Poliakov, A. (2020). Advances in Optical Image Analysis Textural Segmentation in Ironmaking. Applied Sciences, 10(18), 6242. Retrieved from httDs://www.mdDi.com/2076-3417/10/18/6242. Lopez, Alfredo & Catalina, Juan Carlos & Alarcon, David & Grunwald, Ursula & Romero, Paulo & Bolibar, Ricardo (2020) Automated ore microscopy based on multispectral measurements of specular reflectance, I - A comparative study of some supervised classification techniques, Minerals Engineering, 146. 10.1016/j.mineng.2019.106136; Castroviejo, R., Berrezueta, E., Lastra, R. (2002). Microscopic digital image analysis of gold ores: a critical test of methodology, comparing reflected light and electron microscopy. Minerals and Metallurgical Processing, 19(2), 102- 109. Pirard, E. (2004). Multispectral imaging of ore minerals in optical microscopy. Mineralogical Magazine, 68(02), 323-333. doi: 10.1180/0026461046820189; Bouzahzah, H., Califice, A., Benzaazoua, M., Mermillod-Blondin, R., Pirard, E. (2008). Modal analysis of mineral blends using optical image analysis versus X-ray diffraction. Paper presented at the Proceedings of International Congress for Applied Mineralogy ICAM08, Brisbane, Australia. AusIMM; and Chopard, Marion, P., Royer, J. -J., Taza, R., Bouzahzah, H., Benzaazoua, M. (2019). Automated sulphide quantification by multispectral optical microscopy (Vol. 131).

[0004] In addition, software systems that are capable of performing mineralogical characterization under optical microscopes have been developed. Examples of such software applications include: MultiSpec®, Defininiens Developer®, Mineral/Recognition® and AMCO®.

[0005] Although, the basic tenets of quantitative mineralogy using optical microscopy are known, improvements in one or more of efficiency, degree of automation, accuracy and cost are desired.

SUMMARY OF THE DISCLOSURE

[0006] In accordance with a part of the present disclosure, there is provided a method for identifying mineral particles, the method comprising: obtaining a ground sample of a raw solid mineral sample, the raw solid mineral sample containing one or more mineral particles; preparing a material adapted for analysis by optical microscopy by applying a resin coating to the ground sample; acquiring a first digital image of the material through an optical microscope using a first camera operating in the visible range of the electromagnetic spectrum; acquiring a second digital image of the material using a second camera operating at least partly outside the visible range the electromagnetic spectrum; analyzing the first digital image using a deep learning module trained on optical images; identifying pixels in the first digital image corresponding to the one or more mineral particles in the material; and identifying a mineral content associated with the one or more mineral particles in the material based on a reflectance profile of the one or more mineral particles as determined from the second image.

[0007] In accordance with another part of the present disclosure, there is provided a system for automatically analyzing one or more mineral particles in a material, the system comprising: an image acquisition unit comprising: an optical microscope; a first camera for acquiring a first digital image of the material through the optical microscope, the material comprising a resin applied to a ground sample obtained from a raw solid sample; a second camera for acquiring a second digital image of the material; a computing device comprising a processor and memory, the memory for storing instructions for a deep learning module trained on a plurality of optical images for detecting the one or more mineral particles, the processor for executing the instructions to: receive the first digital image and the second digital image as input; analyze the first digital image using the deep learning module to identify pixels in the first digital image corresponding to the one or more mineral particles in the material; and identify mineral content associated with the one or more mineral particles based on a reflectance profile of the one or more mineral particles determined from analyzing the second image.

[0008] In accordance with yet another part of the present disclosure, there is provided a method for preparing a polished section for optical mineralogy, the method comprising the steps of: providing liquid acrylic in a container; providing acrylic powder to the container; mixing the acrylic powder and acrylic powder into a resin mixture; introducing the resin mixture into a polished section mold for forming the polished section; and adding powder sample into the resin mixture in the mold, wherein the resin mixture comprises methacrylate polymer, dibenzoyl peroxide, and methyl 2-methylprop-2-enoate.

[0009] Other technical advantages may become readily apparent to one of ordinary skill in the art after review of the following figures and description.

BRIEF DESCRIPTIONS OF THE DRAWINGS

[0010] For a better understanding of the embodiments described herein and to show more clearly how the embodiments may be carried into effect, reference will now be made, by way of example only, to the accompanying drawings in which:

[0011] FIG. 1 is a schematic block diagram representation of an system for automatically analyzing mineral particles in a raw solid sample, the system comprising an image acquisition unit and a computing device, the computing device comprising a trained deep learning module and a particle detector;

[0012] FIG. 2 a schematic block diagram of several hardware and data forming part of the computing device of FIG. 1;

[0013] FIG. 3 is a flowchart of a method of automatically analyzing mineral particles in a raw solid sample, using the system of FIG. 1; [0014] FIG. 4 a schematic block diagram illustrating the training and use of the deep learning module forming part of the system of FIG. 1;

[0015] FIG. 5 is a graph of a reflectance profile maintained in a reflectance database used by the system of FIG. 1, for a select number of mineral particles, for use in classification of image pixels;

[0016] FIG. 6 is a flowchart summarizing the steps in a method for preparing a polished section;

[0017] FIG. 7 is a schematic illustration of the method in FIG. 6 at various stages;

[0018] FIG. 8 is an example of a mask image produced by the deep learning module of

FIG. 1;

[0019] FIG. 9 is an example of the annotated image depicted in FIG. 4;

[0020] FIG. 10 is simplified block diagram of a U-net architecture for 32x32 pixels in the lowest resolution; and

[0021] FIG. 11 is a high-level architectural block diagram of an encoder and a decoder in the deep learning module of FIG. 1.

[0022] Unless otherwise specifically noted, articles depicted in the drawings are not necessarily drawn to scale.

DETAILED DESCRIPTION

[0023] Directional terms such as "top," "bottom," "upwards," "downwards," "vertically," and "laterally" are used in the following description for the purpose of providing relative reference only, and are not intended to suggest any limitations on how any article is to be positioned during use, or to be mounted in an assembly or relative to an environment. The use of the word "a" or "an" when used herein in conjunction with the term "comprising" may mean "one," but it is also consistent with the meaning of "one or more," "at least one" and "one or more than one." Any element expressed in the singular form also encompasses its plural form. Any element expressed in the plural form also encompasses its singular form. The term "plurality" as used herein means more than one, for example, two or more, three or more, four or more, and the like.

[0024] In this disclosure, the term "about" means within 5% of the stated value.

[0025] In this disclosure, the terms "comprising", "having", "including", and "containing", and grammatical variations thereof, are inclusive or open-ended and do not exclude additional, un-recited elements and/or method steps. The term "consisting essentially of" when used herein in connection with a composition, use or method, denotes that additional elements, method steps or both additional elements and method steps may be present, but that these additions do not materially affect the manner in which the recited composition, method, or use functions. The term "consisting of" when used herein in connection with a composition, use, or method, excludes the presence of additional elements and/or method steps.

[0026] As used herein, the term "bubble matrix" refers to a mineral sample contained within a resin. During sample preparation, the ground mineral sample (e.g. powder form) is mixed with, and distributed within, the resin to form a matrix. During this mixing process, grains of mineral sample are often trapped in minute air pockets referred to as "bubbles". The resulting product of the mineral sample and resin mixture is a "bubble matrix".

[0027] As used herein, the term "raw solid mineral sample" may include solid samples such as drill cores or other forms of solid specimens, that may be further ground, pulverized or crushed and/or further processed to obtain a "ground sample", "ground/polished mineral sample", section, coarse particles, or mineral sample in "powder form" suitable for dispersion in resin.

[0028] For clarity, and where considered appropriate, reference numerals may be repeated among the Figures to indicate corresponding or analogous elements. In addition, numerous specific details are set forth in order to provide a thorough understanding of the embodiment or embodiments described herein. However, it will be understood by those of ordinary skill in the art that the embodiments described herein may be practiced without these specific details. In other instances, well-known methods, procedures and components have not been described in detail so as not to obscure the embodiments described herein. It should be understood at the outset that, although embodiments are illustrated in the figures and described below, the principles of the present disclosure may be implemented using any number of techniques, whether currently known or not. The present disclosure should in no way be limited to the implementations and techniques illustrated in the drawings and described in this disclosure.

[0029] Various terms used throughout the present description may be read and understood as follows, unless the context indicates otherwise: "or" as used throughout is inclusive, as though written "and/or"; singular articles and pronouns as used throughout include their plural forms, and vice versa; similarly, gendered pronouns include their counterpart pronouns so that pronouns should not be understood as limiting anything described herein to use, implementation, performance, etc. by a single gender; "exemplary" should be understood as "illustrative" or as a non-limiting example, and not necessarily as "preferred" over other embodiments. Further definitions for terms may be set out herein; these may apply to prior and subsequent instances of those terms, as will be understood from a reading of the present description. It will also be noted that the use of the term "a" or "an" will be understood to denote "at least one" in all instances unless explicitly stated otherwise or unless it would be understood to be obvious that it must mean "one".

[0030] Modifications, additions, or omissions may be made to the systems, apparatuses, and methods described herein without departing from the scope of the disclosure. For example, the components of the systems and apparatuses may be integrated or separated. Moreover, the operations of the systems and apparatuses disclosed herein may be performed by more, fewer, or other components and the methods described may include more, fewer, or other steps. Additionally, steps may be performed in any suitable order. As used in this document, "each" refers to each member of a set or each member of a subset of a set.

[0031] Any module, unit, component, server, computer, terminal, engine or device exemplified herein that executes instructions may include or otherwise have access to computer readable media such as storage media, computer storage media, or data storage devices (removable and/or non-removable) such as, for example, magnetic disks, optical disks, or tape. Computer storage media may include volatile and non-volatile, removable and non-removable media implemented in any method or technology for storage of information, such as computer readable instructions, data structures, program modules, or other data. Examples of computer storage media include RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital versatile disks (DVD) or other optical storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other medium which can be used to store the desired information and which can be accessed by an application, module, or both. Any such computer storage media may be part of the device or accessible or connectable thereto. Further, unless the context clearly indicates otherwise, any processor or controller set out herein may be implemented as a singular processor or as a plurality of processors. The plurality of processors may be arrayed or distributed, and any processing function referred to herein may be carried out by one or by a plurality of processors, even though a single processor may be exemplified. Any method, application or module herein described may be implemented using computer readable/executable instructions that may be stored or otherwise held by such computer readable media and executed by the one or more processors.

[0032] Disclosed herein are a system and a method that utilize automated mineralogy to obtain a segmented digital image, wherein pixels of the segmented digital image are assigned to a mineralogical phase. Obtaining a segmented digital image involves the use of a classification procedure that utilizes a large group of algorithms including combined thresholding operations such as k-means clustering, Mahalanobis distance, and linear discriminant analysis.

[0033] An objective behind this classification procedure is to identify each mineral by its reflectance profile as understood through multispectral analysis. The reflectance profile is then compared to standard reflectance profiles compiled in a mineralogical data library to determine the identity of the mineral. In some embodiments, multispectral analysis involves use of optical spectral filters.

[0034] Among the problems with conventional optical microscopy is the fact that optical microscopy, in reflected light mode, is usually limited to the detection of opaque minerals such as sulphides, oxides, alloys and the like. With conventional optical microscopy, transparent phases such as silicates and carbonates, are not easily detected. This leads to a bias in the texture quantification of the mineral particles. In addition, certain opaque phases may become difficult to distinguish from one another due to the similarity of their optical properties. [0035] Existing approaches to automated optical mineralogy include optical multispectral analysis that can be burdensome to automate. Such classification algorithms require a complete multispectral analysis involving the use of about twenty (20) spectral filters. For example, a study by Poliakov, A., & Donskoi, E. (2014), Automated relief-based discrimination of non-opaque minerals in optical image analysis. Minerals Engineering, 55, 111-124 from http://www.sciencedirect.com/science/article/pii/S0892687513002872. doi:https://doi.org/10.1016/j.mineng.2013.09.014 discloses what may currently be among the most advanced methods for detection of transparent gangue minerals. However, the procedure is not suitable for the detection of all transparent phases. Phases that do not show reliefs in optical microscopy in reflected light require extensive digital corrections.

[0036] The approach of Grunwald, U., Catalina, J. C., Lopez, A., Bolibar, R. (2019), A reliable method for the automated distinction of quartz gangue and epoxy resin with reflected light microscopy and its application to digital image analysis, requires specific preparation of mineral samples in polished thin sections. This is a particularly long, tedious, and increasingly expensive preparation technique. This approach is inconsistent with the obvious desire of the mining industry to obtain ready-for-ana lysis materials quickly and inexpensively.

[0037] Described herein is a system for automatically analyzing mineral particles in a raw solid mineral sample. Referring to FIG. 1 and according to an embodiment of such system, there is provided a system 10 comprising a sample preparation unit 12, an image acquisition unit 30, and a computing device 20.

[0038] The sample preparation unit 12 comprises a grinding/polishing device 18 for grinding/polishing drill core 14 to obtain the ground/polished mineral sample 16, which may be in the form of a section, coarse particles or powder. The sample preparation unit 12 further comprises a resin-coating device 19 for applying a fast curing resin coating to a ground/polished sample 16 obtained from the raw solid sample to prepare a material 21 ready for analysis. The sample preparation (i.e., polishing section) may be a manual process that is known in the art. However, in this embodiment, a novel type of resin (acrylic) is used for the mineral sample preparation along with fast curing to present a specific matrix (bubble matrix). This novel approach allows for detection of all particles, including opaque and transparent particles, via deep-learning analysis. [0039] Image acquisition unit 30 comprises an automated optical microscopy unit 32, which itself comprises an optical microscope 34 in optical communication with a camera 36 and a hyperspectral camera 37. In this embodiment, camera 36 is a charge-coupled device (CCD) camera operating in the RGB domain whereas camera 37 is a hyperspectral camera. The cameras 36, 37 acquire images of the material 21 through the optical microscope 34. In this embodiment, camera 36 is a classic RGB digital camera, which uses a CCD sensor to capture images in the visible spectrum; in other embodiments, camera 36 may use a CMOS (complementary metal oxide semiconductor) image sensor. Hyperspectral camera 37 also uses a CCD or CMOS image sensor but captures a much wider spectral range beyond the visible band, and may contain filters.

[0040] Hyperspectral camera 37 may have no specific spatial resolution. In the depicted embodiment, image acquisition is accomplished by a linear translation stage. In one specific embodiment, the specifications of the camera 37 are as follows: Spectral range = 330nm - 800nm; Spectral Channels = 255; Spectral Bandwidth = 1.84nm; spectral Resolution = FWHM 2.8nm; Spatial Pixels: 1500; and maximum frame rate = 142fps. Camera 36 may be in the form of the Clemex St RGB (CCD) camera employing a Sony ICX814 CCD color image sensor having a resolution of 3376 x 2704 = 9.1 Megapixels.

[0041] Computing device 20 includes a processor and memory, the memory storing processor executable instructions in the form software and data structures. The software in computing device 20 includes a deep learning module 22 trained, or configured to be trained, on a plurality of optical images for detecting particles. Software executing on the computing device 20 also includes a particle detector module 24, a numerical analyser module 24, a mineralogical classification module 28 and a mineralogical quantification module 29.

[0042] In this embodiment, deep learning module 22 is trained on a plurality of optical images for detecting particles having associated annotations identifying specific minerals associated with groups of pixels that constitute a mineral composition.

[0043] FIG. 2 depicts a simplified block diagram of various elements of computer system 20. As shown, computer system 20 has a number of physical and logical components including a processor 44, which may be the form of a central processing unit ("CPU"), as well as random access memory ("RAM") 48, an input/output ("I/O") interface 52, a network interface 56, and non-volatile storage 60. A high-speed interface circuit 64 enables processor 44 to communicate with the other components. Processor 44 executes processor executable instructions in the form of at least an operating system, and a one or more applications including the software modules depicted in FIG. 1.

[0044] RAM 48 provides relatively responsive volatile storage to processor 44. I/O interface 52 allows input to be received from one or more peripheral devices, such as a keyboard, a mouse, etc., and outputs information to output devices, such as a display and/or speakers. Network interface 56 permits wired or wireless communication with other computing devices over computer networks such as the Internet. Non-volatile storage 60 stores the operating system and programs, including computer-executable instructions for implementing the software implemented portions of modules 22, 23, 26, 28 and 29 and associated code, data structures and objects, depicted in FIG. 1. During operation of computer system 20, the operating system, the programs and the data may be retrieved from non-volatile storage 60 and placed in RAM 48 to facilitate execution.

[0045] In the depicted embodiment, deep learning module 22 has been trained on a training data set of over 1000 optical images from RGB images at different magnifications including: 25x, 50x, lOOx, 200x where the bubbled matrix resin and particle with various shape and grain-size occur.

[0046] An example of a deep learning algorithm employed in one embodiment is used to detect the mineral particles is a modified version of the vanilla UNet that was first used in biomedical image segmentation described in Ronneberger, 0., Fischer, P., & Brox, T. (2015) U-net: Convolutional networks for biomedical image segmentation, paper presented at the International Conference on Medical image computing and computer-assisted intervention. Vanilla UNet architecture includes a contracting path to capture context and a symmetric expanding path that enables precise localization. Unet is a fully convolution network with several feature channels in the convolution layers (contracting path) and the up-convolutional layers (expanding path), it does not have any fully connected layers and hence the network weights of a pre-trained network with a particular input size can be used to hyper tune the network with different input sizes. FIG. 10 depicts a simplified block diagram, of an example of a U-net architecture for 32x32 pixels in the lowest resolution, where each blue colored box corresponds to a multi-channel feature map; the number of channels is denoted on atop of the box, the x-y-size is provided at the lower left edge of the box; white boxes represent copied feature maps; and arrows denote the different operations.

[0047] The modified version of vanilla UNet as used in this embodiment has different layer sizes, different type of filters (convolutional layers), difference in the initialization of the filters etc. The input image sizes, and the output mask sizes have also been customized. The vanilla Unet in Ronneberger et al. (2015) was implemented using Caffe and C++ whereas the modified version used in this embodiment uses TensorFlow and Keras with underlying Python. The training strategy of the modified vanilla UNet heavily relies on the strong use of data augmentation, that is, augmentation of the annotated training images, to use the available annotated images more efficiently and to become invariant to small changes in lightning conditions, resulting shadows and color changes along with several translation and rotational changes. The network is trained end-to- end with a small set of images and consequently it is seen to perform a very precise semantic segmentation of the mineral particles from the background resins.

[0048] Another example of a deep learning algorithm that can be employed in the embodiments described is discussed in : Chen, Liang-Chieh, Yukun Zhu, George Papandreou, Florian Schroff, and Hartwig Adam. "Encoder-decoder with atrous separable convolution for semantic image segmentation." In Proceedings of the European conference on computer vision (ECCV), pp. 801-818. 2018, the contents of which are incorporated by reference herein. The aforementioned paper describes an encoder-decoder structure, which employs DeepLabv3 as an encoder module and provides a simple decoder module. A high-level architecture of the encoder and decoder in deep learning module 22 is depicted in FIG. 11. DeepLabv3 is a semantic segmentation architecture that employs Atrous (dilated) convolution with upsampled filters to extract dense feature maps and to capture long range context. Specifically, to encode multi-scale information, the architecture's cascaded module gradually doubles the atrous rates while an atrous spatial pyramid pooling module augmented with image-level features, probes the features with filters at multiple sampling rates and effective field-of-views. The structure, allows arbitrarily control of resolution of extracted encoder features by atrous convolution to trade-off precision and runtime; and adapts the Xception architecture for the segmentation task and applies depthwise separable convolution to both the spatial pyramid pooling module and the decoder module. The Xception architecture is a deep convolutional neural network architecture that involves depthwise separable convolutions.

[0049] Using system 10, an automatic analysis of mineral particles in a raw solid sample can be carried out. FIG. 3 depicts a flowchart for a method 100 of automatically analyzing mineral particles in a raw solid sample via system 10, in one embodiment.

[0050] Method 100 commences with step 102 which involves grinding/polishing the raw solid sample to obtain a ground/polished sample.

[0051] Step 104 involves preparing a material that is ready for analysis, by applying a fast curing resin coating to the ground/polished sample.

[0052] Step 106 involves acquiring digital images of the material using optical microscopy. The digital images may include RGB images obtained with a classic RGB camera such as camera 36, containing pixel values in the visible range of the electromagnetic spectrum; as well as hyperspectral and multispectral images obtained using the hyperspectral camera 37 that additionally contain pixel components outside the visible range. Multispectral imagery generally involves a few (e.g. 3 to 10) relatively wide bands which may be in visible range (red, green, blue; 400 to 800nm) and near-infrared range of the electromagnetic spectrum. On the other hand, hyperspectral imaging utilizes much narrower bands (around 1 nm) but may include hundreds or thousands of bands.

[0053] Image acquisition may be accomplished via linear pixel acquisition. The stage of the microscope 34 is synchronised by this linear pixel acquisition to provide a hyperspectral cube of data of each particle previously detected by the deep-learning module 22 in sample 16. Hyperspectral camera 37 can be used to obtain a monochromatic image according to the wavelengths acquired. System 10, using hyperspectral camera 37 can generate these specific images, which may be called multispectral images, and perform imaging treatment to identify the minerals.

[0054] At step 108, analysis of the digital images obtained at step 106, is carried out using a deep learning algorithm in module 22. As noted above, deep learning module 22 is trained on a large set of optical images for detecting particles. Particles in the raw solid sample, including opaques (e.g. sulfides, oxides) and transparent minerals (e.g., silicates, carbonates) are thus readily detected just by using the acquired digital image of the material 21.

[0055] As will be appreciated by persons of skill in the art, conventional approaches were not able to reliably detect transparent minerals such as silicates and carbonates by reflected light optical microscopy alone. There have been many prior attempts to detect minerals by a combination of transmitted light and reflected light optical microscopy. These include: Berry, R., Walters, S., & McMahon, C. (2008) Automated mineral identification by optical microscopy. Australasian Institute of Mining and Metallurgy Publication Series, 91-94; Hunt, J., Berry, R., & Bradshaw, D. (2011) Characterising chalcopyrite liberation and flotation potential: Examples from an IOCG deposit. Minerals Engineering, 24(12), 1271-1276. doi: 10.1016/j.mineng.2011.04.016). However, these attempts required a specific preparation of raw solid samples involving polishing thin section, which is not efficient for the mining industry as it involves a time consuming and labour intensive task. Predictably, these prior art approaches were not suitable for application in mining which often requires quick production of ready-to-ana lysis samples.

[0056] Other approaches proposed the use of polished thin sections combined with fluorescent resin, have allowed to obtain by grey thresholding processing two distinct spectral signatures between fluorescent resin and transparent phases in order to discriminate them. See for example, Grunwald, U., Catalina, J. C., Lopez, A., & Bolibar, R. (2019) A reliable method for the automated distinction of quartz gangue and epoxy resin with reflected light microscopy and its application to digital image analysis. However, this study included a specific preparation of raw solid sample: polishing thin section.

[0057] Many other authors have been interested in the discrimination of transparent minerals for iron ore characterization. Publications that used a deep-learning approach to detect particles include: Maitre, J., Bouchard, K., Bedard, L. P. (2019). Mineral grains recognition using computer vision and machine learning. Computers & Geosciences, 130, 84-93. doi: https://doi.Org/10.1016/j.cageo.2019.05.009; Okada, N., Maekawa, Y., Owada, N., Haga, K., Shibayama, A., Kawamura, Y. (2020). Automated Identification of Mineral Types and Grain Size Using Hyperspectral Imaging and Deep Learning for Mineral Processing. Minerals, 10(9), 809; and Koh, E. J., Amini, E., McLachlan, G. J., & Beaton, N. (2021). Utilising convolutional neural networks to perform fast automated modal mineralogy analysis for thin-section optical microscopy. Minerals Engineering, 173, 107230.

[0058] Very few significant studies in the literature have hitherto used deep learning for automated mineralogical quantification in light microscopy. These include the works of Maitre et al, (2019), Okada et al. (2020) and Koh et al., (2021). The former two studies are not based directly on optical analyzes and the samples do not involve special preparation i.e., the sample was directly analyzed as sand or trapped into resin with no special preparation steps. The roughness effects that particles present may thus disrupt the quality of images taken and unfortunately skew the training of the algorithm. Moreover, these approaches are dependent on the method of sample preparation, as well as the origin mineralogical composition of the sample.

[0059] The study of Koh et al., (2021) proposed what may currently be the most advanced procedure for the detection of mineral particles known in the field. The study has proposed two algorithms to locate and classify minerals grains (segmentation) including transparent gangue particles and opaque particles (pyrite). The first is referred to as Mask Region-based Convolutional Neural Networks (Mask R-CNN) while the second is known as Segmenting Objects by Locations v2 (SOLO v2). These two algorithms were trained on one hundred and six (106) training images provided by an optical microscope acquired on polished thin sections samples. The SOLO v2 algorithm is reported to perform the best detection of particles with a segmentation speed of 12, 000 particles/minute. Nevertheless, the algorithms were performed from thin sections optical images. As with a previous study, Grunwald et al. 2019, the procedure includes specific preparation steps, which are not suitable for application in the mining industry.

[0060] Such studies use a specific algorithm of border-based discrimination of transparent minerals. The aim is to identify transparent minerals borders by performing a dynamic thresholding based on the optical relief that presents transparent particles such as quartz. Many other corrections are performed to improve this detection and is described in the papers noted above. Optical relief is linked to the difference of refractive index between the particle and the resin used to mount the polished section. Consequently, this optical propriety depends on the mineralogical specie of the particle. (Ref: Delbem, I. D., Galery, R., Brandao, P. R. G., & Peres, A. E. C. (2015). Semi-automated iron ore characterisation based on optical microscope analysis: Quartz/resin classification. Minerals Engineering, 82, 2-13. doi: 10.1016/j.mineng.2015.07.021 ; Poliakov, A., & Donskoi, E. (2014). Automated relief-based discrimination of non-opaque minerals in optical image analysis. Minerals Engineering, 55, 111-124. Retrieved from http://www.sciencedirect.com/science/article/pii/S0892687513002872. doi: https://doi.Org/10.1016/j.mineng.2013.09.014).

[0061] Embodiments of the system and method described herein utilize deep learning combined with a special preparation of the polished section using acrylic resin to detect particles including transparent and opaque particles.

[0062] Embodiments of the resin are supplied by Struers, a company with headquarters in Denmark that provides a complete range of materialographic solutions, products, and services. A specific composition that may be used is VERSOCIT-2 POWDER™ provided by Struers.

[0063] An example preparation method for the polished section is described below with reference to FIG. 6 which is a flowchart summarizing the preparation method.

[0064] As shown in flowchart 600, initially at step 602, liquid acrylic is poured into a container.

[0065] At step 604 acrylic powder is added to the container. The acrylic liquid and acrylic powder may be obtained from an acrylic kit supplied by Struers. In one specific embodiment, the ratio of volumes of the acrylic liquid to the acrylic powder may be 1:2 (e.g., 5ml_ of liquid for 10 ml_ of powder).

[0066] At step 606, the liquid acrylic and acrylic powder are mixed for a predetermined amount of time (e.g. for 30 seconds) to obtain an acrylic resin mixture.

[0067] At step 608, the acrylic resin mixture obtained at step 606, is poured into a polished section mold, and then a ground form of the raw solid sample is added as powder.

[0068] At step 610, the mixture in the polished section mold is thoroughly blended to obtain an appropriate or even optimal dispersion of particles of the sample throughout the resin in the mold. Step 610 may take approximately 2 minutes 30 and seconds by which time the process of hardening may start. After a completion of hardening process, which may take up to 10 minutes, the polished section may be removed from the mold and is ready to be polished.

[0069] In some embodiments, the resin mixture may comprise 60% to under 100% for methacrylate polymer; less than 1% for dibenzoyl peroxide; and less than 1% for methyl 2-methylprop-2-enoate.

[0070] FIG. 7 is a schematic illustration of the steps in in FIG. 6 at various stages. As shown, at Stage A, liquid acrylic is poured into a container 702 followed by addition of an acrylic powder which is poured into the same container.

[0071] At stage B: the liquid and powder are mixed (e.g. for 30 seconds) to obtain an acrylic resin mixture. At stage C, the acrylic resin mixture obtained at Stage B is poured into a mold, to prepare the polished section and then the sample added (as powder).

[0072] At stage D, the preparation in the polished section mold in order to obtain an appropriate (or even optimal) dispersion of particle sample through the mold. As noted above it may take approximately 2 minutes 30 and seconds by which time the process of hardening lasting up to 10 minutes may start.

[0073] Hardening of the fast curing resin results in a bubbled matrix that, when viewed under optical microscopy, enables distinction between of the mineral particles and those of the bubbled matrix.

[0074] Accordingly, pixels of the mineral particles in the digital images can be readily distinguished from pixels of the bubbled matrix by deep learning module 22, and annotated.

[0075] In operation, the computing device 20 receives the digital images from the image acquisition unit 30 as input, and analyzes the digital images using the deep learning module 22 to detect the mineral particles. The mineral particles are georeferenced in particle detector module 24. Thus identification of mineral particles is in the raw solid sample is accomplished by using the digital images of the material 21. [0076] Polishing and grinding operations in system 10, together with the resin coating, convert the raw mineral powder sample 16 into a ready-for-ana lysis material 21. Optical analysis via reflected light optical microscopy can then be performed on said ready-for- analysis material 21. The optical properties of the surface of material 21 can be georeferenced using deep learning module 22 to automatically annotate and describe mineral content contained in material 21, and by inference the mineral content of the powder sample 16. The entire process can thus be automated and provide quick analysis without the use of chemical analysis and other labor intensive or expensive intermediate processes.

[0077] Hardening of the fast curing resin results in a bubbled matrix that when observed under the optical microscope 34, enables distinction between mineral particles and bubbled matrix in the resulting digital images obtained by cameras 36, 37.

[0078] The software in computing device 20 enables, through analysis of optical hyperspectral images, a digital segmentation based on a more efficient algorithm of mineral classification in module 28.

[0079] The segmentation uses a classification method, which is suitable for a spectral analysis of the particles detected by deep learning module 22. The classification procedure includes comparing the reflectance behavior of each pixel of image of the particle with a database of minerals, which may be a proprietary reflectance database of minerals.

[0080] In an embodiment, the reflectance database is a propriety database specifically set up for mineral identification as part of the present invention. The classification method uses algorithms such as Mahalanobis distance and linear discriminate analysis to attribute a pixel to a mineral specie to obtain a segmented image i.e., mineralogical mapping of the particles.

[0081] Other embodiments of the method described herein may include the step of building a new reflectance database by measuring the reflectance of chosen pixels on samples whose mineral compositions are already known to the user. [0082] Mineralogical identification of using at least some embodiments disclosed is thus no longer limited to a few wavelengths selected using spectral filters, but the much wider spectral range of a hyperspectral camera 37 utilized in the image acquisition unit 30.

[0083] In the visible range, optical images 40 used for the deep-learning module 22 are provided by the RGB camera 36. Utilization of optical hyperspectral image analysis makes system 10 faster than tools currently available in the field of automated mineralogy under optical microscopy which perform multispectral analysis.

[0084] Specific mineral compositions in a sample may be more efficiently identified using spectral mineralogical identification at different spectral ranges. For a given mineral composition of a sample, a particular spectral range may be advantageous because its reflectance characteristics in that range, compared to another ranges, for the purposes of mineralogical identification.

[0085] The spectral mineralogical identification is dependant of the wavelengths used to acquire the multispectral images. If the minerals in the samples have spectral reflectance very close to each other at the filters (wavelengths) available in the automated optical microscope 24, then system 10 may not be able to discriminate among them for effective mineralogical identification. In embodiments of the present invention, a system such as system 10 provides enough filters to cover a large spectral range so that it can perform an effective mineralogical identification regardless of the mineralogical composition.

[0086] Some conventional systems use a predetermined number of filters. For example, the AMCO® system uses 20 wavelengths for image acquisition. However, such a conventional approach entails a long multispectral image acquisition of the sample. In contrast, embodiments of the present invention utilize hyperspectral camera 37 that quickly and directly acquires spectral data on the particles georeferenced and detected by the module 24, resulting in a faster process.

[0087] As noted above, it is desirable to have a classification procedure is to identify each mineral by its reflectance profile as understood through multispectral analysis. The reflectance profile is then compared to standard reflectance profiles compiled in a mineralogical data library, such as those depicted in FIG. 5, to determine the identity of the mineral. In some embodiments, multispectral analysis involves use of optical spectral filters.

[0088] The automation of the detection of mineral particles, both transparent and opaque, is possible by the combination of fast curing resin coatings used for polished section preparation in resin coating device 19 and the deep learning algorithms in deep learning module 22.

[0089] The resin coating which in one embodiment, may comprise methacrylate polymer, dibenzoyl peroxide, and methyl 2-methylprop-2-enoate, allows fast preparation of polished sections, typically in less than three (3) hours including ten (10) minutes for hardening the resin and polishing for coarse and powder mineral material. This novel fast preparation is suitable for use in the mining industry and allows mining companies to obtain a quickly and inexpensively, ready-for-ana lysis materials 21.

[0090] The hardening resin used is made possible by a fast polymerization. The composition of one embodiment of the resin, permits fast polymerisation. The fast curing is typical of acrylic resin. This provides a notable bubbled matrix under optical microscopy and makes it possible to easily distinguish mineral particles and from the matrix. The disclosed detection is utilized to develop deep learning algorithms in module 22.

[0091] In embodiments of the present invention, several deep learning architectures have been trained on a bank of optical (RGB) images of annotated mineral particles into the bubbled resin. FIG. 4 schematically illustrates the process.

[0092] Training images 70 are used to train the untrained deep learning module 22', which is then tested with test images 72. In the depicted embodiment, more than 700 training images are used to train the deep learning module 22'. The trained deep learning module 22 is then used to classify acquired images 40 received from the image acquisition unit 30 to produce annotated image 82 which may be further processed to contain mineral content description 84. FIG. 9 depicts an example of an annotated image.

[0093] The deep learning algorithms of are applied to new optical images such as image 40 of the ready-for-analysis material 21 containing the bubbled resin, and make it possible to obtain a new mask image. A mask image is a numerical image produced by the deep learning module 22 after detection of the particle performed by the deep-learning algorithm where only the particle delimitation is available for further image processing. An example of a mask image is shown in FIG. 8.

[0094] Digital processing of the new mask image allows detection of all mineral particles within the material 21. The mineral particles may be opaque or transparent. In one embodiment, deep learning module 22 is used only for particle detection without identifying the minerals making up the detected particles.

[0095] Reflected light optical microscopy is a more accessible technique for automated quantitative mineralogical analysis. The sample preparation unit 12 allows for the preparation a mineral sample in different forms: sections, coarse particles, and powder. The process involves a ten (10) minute fast hardening resin coating in the form of a polished section. Combined with polishing, a polished section mineral sample is ready for optical analysis in less than two (2) hours. The sample preparation unit 12 also enables polishing preparation technique directly a raw solid sample such as a piece of drill core 14 which may be over 100 cm in length. The selection of pieces of cores for additional analyzes is very common in mining exploration. The embodiments supplement these data with analysis of mineralogical quantification.

[0096] Microscope 34 may be a high-performance optical microscope, such as optical microscope models sold by Clemex Technologies Inc. In some embodiments, a reverse or inverted optical microscope model such Leica® DMi8 microscope platform available from Leica Microsystems. The microscope 34 may be made up of different parts: objectives, automated stage, various light sources (halogen and ultraviolet), spectral filters, classic RGB camera and hyperspectral camera. Associated with a control software, the image acquisition unit 30 allows the acquisition of multispectral, hyperspectral, and polarized light optical images as well as automatic, rapid, and complete digitization of the various study supports. For example, up to six units of ready-for-analysis material 21 may be processed at a time.

[0097] The software in system 10, and in particular in computing device 20, ultimately enables fully automated mineralogical quantification with hyperspectral optical analyzers utilizing deep learning algorithms in deep learning module 22. [0098] Conveniently, the system 10 allows detection of all the particles, both opaque and transparent, of a mineral sample in material 21. This is enabled by the application of a deep learning algorithm trained on a bank of annotated optical reference images of mineral particles.

[0099] Additionally, the software includes mineralogical classification algorithms based on multispectral / hyperspectral data analysis. The algorithm allows comparing the spectral thresholding behaviour analysis to a reflectance database to assign for each pixel of a hyperspectral/multispectral image 38 a mineralogical category (mineral species). Combined with a processing of optical images 40 including various digital correction operations, a raw solid sample can be fully analyzed for segmentation and mineralogical quantification including modal composition, granulo-mineralogy, degree of liberation / exposure and mineralogical association.

Advantages

[0100] Automated quantitative mineralogy is becoming increasingly attractive in the mining industry. The process however, has only developed around scanning electron microscopy associated with chemical microanalysis. Optical microscopy in reflected light is a more accessible alternative. Unlike electron microscopy, light microscopy uses the optical properties of minerals to identify them.

[0101] Advantageously, embodiments disclosed utilize an optical microscope. The optical microscope, often referred to as "reflected light optical microscope," is a type of microscope that uses visible light and a system of lenses to magnify images of small samples. Optical microscopes are the oldest design of microscope and basic optical microscopes are relatively simple and inexpensive, although there are many complex design variations that can be used to improve resolution and sample contrast.

[0102] Embodiments of system 10 and method 100 thus permit considerable cost reduction, and provide a more accessible alternative for mining and other industries which aim to carry out fast mineralogical quantification analysis for their operations. A good example is geometallurgical characterization of mining projects. [0103] Although specific advantages have been enumerated above, various embodiments may include some, none, or all of the enumerated advantages.

[0104] Persons skilled in the art will appreciate that there are yet more alternative implementations and modifications possible, and that the above examples are only illustrations of one or more implementations. The scope of the claims should not be limited by the example embodiments set forth herein, but should be given the broadest interpretation consistent with the description as a whole.

Claims

WHAT IS CLAIMED IS:

1. A method for identifying mineral particles, the method comprising:

(a) obtaining a ground sample of a raw solid mineral sample, the raw solid mineral sample containing one or more mineral particles;

(b) preparing a material adapted for analysis by optical microscopy by applying a resin coating to the ground sample of the raw solid mineral sample;

(c) acquiring a first digital image of the material through an optical microscope using a first camera operating in the visible range of the electromagnetic spectrum;

(d) acquiring a second digital image of the material using a second camera operating at least partly outside the visible range the electromagnetic spectrum;

(e) analyzing the first digital image using a deep learning module trained on optical images;

(f) identifying pixels in the first digital image corresponding to the one or more mineral particles in the material; and

(g) identifying a mineral content associated with the one or more mineral particles in the material based on a reflectance profile of the one or more mineral particles as determined from the second image.

2. The method as claimed in claim 1, wherein the ground sample is obtained by grinding the raw solid mineral sample.

3. The method as claimed in claim 1, wherein the step of identifying the mineral content associated with the one or more mineral particles in the material comprises a full spectral range analysis of the second digital image.

4. The method as claimed in claim 1, wherein the one or more mineral particles comprise transparent mineral particles. The method as claimed in claim 1, wherein the one or more mineral particles comprise opaque mineral particles. The method as claimed in claim 1, wherein the resin comprises any one or all of methacrylate polymer, dibenzoyl peroxide, and methyl 2-methylprop-2-enoate. The method as claimed in claim 1 wherein the material adapted for analysis by optical microscopy is prepared by applying the resin coating to the ground sample and hardening of the resin by fast polymerization. The method as claimed in claim 1, wherein the first camera is an RGB camera and the second camera is a hyperspectral camera. The method as claimed in claim 1, wherein the raw solid mineral sample comprises a piece of drill core. The method as claimed in claim 1, wherein the material adapted for analysis by optical microscopy is prepared by applying the resin coating to the ground sample and polishing the resin coating. A system for automatically analyzing one or more mineral particles in a material, the system comprising :

(a) an image acquisition unit comprising:

(i) an optical microscope;

(ii) a first camera for acquiring a first digital image of the material through the optical microscope, the material comprising a resin applied to a ground sample obtained from a raw solid sample;

(iii) a second camera for acquiring a second digital image of the material;

(b) a computing device comprising a processor and memory, the memory for storing instructions for a deep learning module trained on a plurality of optical images for detecting the one or more mineral particles, the processor for executing the instructions to: (i) receive the first digital image and the second digital image as input;

(ii) analyze the first digital image using the deep learning module to identify pixels in the first digital image corresponding to the one or more mineral particles in the material; and

(iii) identify mineral content associated with the one or more mineral particles based on a reflectance profile of the one or more mineral particles determined from analyzing the second image. The system as claimed in claim 11, wherein the resin is a fast curing resin. The system as claimed in claim 11, wherein the resin comprises resin comprises methacrylate polymer, dibenzoyl peroxide, and methyl 2-methylprop-2- enoate. The system as claimed in claim 12, wherein hardening of the fast curing resin results in a bubbled matrix that when observed under the optical microscope, enables distinction between mineral particles and the bubbled matrix in the digital images. The system as claimed in claim 11, wherein the deep learning module comprises an encoder and a decoder, the encoder comprising a semantic segmentation architecture that employs dilated convolution with upsampled filters to extract dense feature maps, and the decoder recovers object boundaries. The system as claimed in claim 11, further comprising a sample preparation unit comprising: a device for applying the fast curing resin to the ground sample to prepare the material. The system as claimed in claim 16, wherein the sample preparation unit further comprises: a grinding/polishing device for grinding the raw solid sample to obtain the ground sample. The system as claimed in claim 11, wherein the first camera is an RGB camera and the second camera is a hyperspectral camera. The system as claimed in claim 18, wherein the RGB camera comprises one of a CCD sensor and a CMOS sensor. A method for preparing a polished section for optical mineralogy, the method comprising the steps of:

(a) providing liquid acrylic in a container;

(b) providing acrylic powder to the container;

(c) mixing the acrylic powder and acrylic powder into a resin mixture;

(d) introducing the resin mixture into a polished section mold for forming the polished section; and

(e) adding powder sample into the resin mixture in the mold, wherein the resin mixture comprises any one or all of methacrylate polymer, dibenzoyl peroxide, and methyl 2-methylprop-2-enoate. The method as claimed in claim 20, wherein the ratio of volumes of the acrylic liquid to acrylic powder is 1:2. The method as claimed in claim 21, wherein the volume of acrylic liquid is 5ml_ and the volume of acrylic powder is lOmL. The method as claimed in claim 20, wherein the powder sample added into the resin mixture in the polished section mold is thoroughly blended to increase dispersion of the ground sample throughout the resin in the mold. The method as claimed in claim 23, further comprising removing the resin from the mold and polishing the resin to form the polished section. The method as claimed in claim 20, wherein the resin comprises methacrylate polymer, dibenzoyl peroxide, and methyl 2-methylprop-2-enoate in proportions of: (i) about 60% to under 100%, for methacrylate polymer; (ii) less than 1% for dibenzoyl peroxide; and (iii) less than 1% for methyl 2-methylprop-2-enoate.