The application of hyperspectral imagery to the mineralogical mapping of host-rock alteration halos has been developed successfully using the example of the samples from the area around Cigar Lake uranium deposit. The approach allowed the identification and differentiation of a wide variety of minerals belonging to most alteration phases already known for this area. The main limitations of the hyperspectral methods are: • the difficulty in spectrally discriminating the different di-octahedral K-phyllosilicates (e.g. muscovite, illite), as well the different types of chlorites (e.g. tri-octahedral chlorite, di-trioctahedral sudoite), • the significant uncertainty on absolute mineral abundances due to the impossibility of quantifying quartz and fresh feldspar contents. The alteration mineral proportions are commonly overestimated due to spectrally-inactive minerals like quartz and unaltered feldspar (Fig. 10). Two reasons can explain this phenomenon: • the investigation depth of this technique is deeper (i.e. 30 to 100 μm) than the thin section thickness (30 μm); • the transparency of the silicates in the SWIR range reveals alteration products that are present behind (below) the spectrally- inactive minerals and thus are not visible optically. This project on mineral alteration mapping on drill cores using a HySpex SWIR-320m hyperspectral camera has highlighted the application of hyperspectral imaging on core samples coming from an unconformity-type uranium deposit environment in Canada. The mineral maps obtained provide information about the sample structure, such as foliation, bedding, and veining that can be present in the core samples. Petrographic observations have validated the hyperspectral mineralogical mapping. The pseudo-modal compositions that were obtained are directly related to the surface spectral proportions of these minerals. However, some overestimation of the alteration mineral proportions due to the technique has to be kept in mind. The proposed mineralogical semi-quantification methodology may be used in other geological contexts and should improve the under- standing of the mineralogical alteration distribution around hydrothermal metal deposits. DISCUSSION AND CONCLUSIONS Qz Ilt Sud + Ilt Ms Chl Sud Sud 30 µm 0 µm 100 µm 1 2 1 2 Quartz Chloritized biotite Argillized feldspar Thin section Investigation depth: Illite Chlorite + Mica or Sudoite 1 2 Hyperspectral imagery A B 2 mm 1 2 A B Figure 10 - Schematic sketch illustrating the local overestimation of clay mineral pro- portions in spectral-class maps. The A-B section is represented in depth by the scheme at the bottom. Sud: sudoite; Ilt: illite; Chl: chloritized biotite; Ms: muscovite; Qz: quartz. 2 mm Qz Ilt Sud + Ilt Chl Ilt Ilt Ilt Ilt Ilt Sud Sud Chl Qz Qz Qz Qz Qz Qz Qz Qz Qz Ms 1 2 A B Muscovite Chlorite Illite Chlorite + Mica or Sudoite Mica + Chlorite Unclassified Legend Color composite images quickly highlight the minerals that characterize the samples without significant amounts of data processing (e.g. Fig. 7). However, minerals presenting a similar spectral behavior can dis- play the same colors in one 3-band color composite image, so it is then necessary to use a series of color composites to differentiate all of these minerals. RESULTS AND VALIDATION Validation of the mineral identification results for alteration minerals obtained using hyperspectral imagery was carried out: i) through optical microscopic petrographic analysis on thin sections, and ii) by visual inter- pretation of images digitized with a VHX 2000 numerical microscope (Keyence) and processed using the Keyence image processing software. The classical mineral identification was done using cross polarized transmitted light. These interpretative mineral maps were superimposed on the hyperspectral maps (Fig. 9). The bivariate comparison diagrams show linear best-fit lines and correlations obtained between hyper- spectral imaging mineral proportions and comparable data from the image processing of petrographic thin sections (Fig. 9). The mineral maps (or spectral-class images, see Fig. 8): i) allow fast identification of the main spectral- ly-active mineral species, especially those resulting from alteration events; and ii) highlight the main petro- graphic textures such as foliation, bedding, veins, and the geometry of pervasive alteration. 2 mm 2 mm Qz Ilt Sud + Ilt Ms Chl Sud Sud 2 mm 2 mm Qz Ilt Sud + Ilt Chl Ilt Ilt Ilt Ilt Ilt Sud Sud Chl Qz Qz Qz Qz Qz Qz Qz Ilt Sample: no. 1 (pegmatite) 1 2 3 4 Ms Qz Qz 2 mm 2 mm 2 mm Ilt Ilt Ilt Ilt Ilt Ilt Ilt Ilt Ilt Ilt Ilt Ilt Ilt Sud Sample: no. 8 (pegmatite) Sud 2 mm Unaltered feldspar Unaltered feldspar 2 mm 2 mm 2 mm Sample: no. 6b (gneiss) 2 mm Ilt Ms Drv Unaltered feldspar Cb Sud Cb Ilt Ms Drv Unaltered feldspar 2 mm 2 mm Ilt + Chl + Qz Ilt + Chl (< 500 µm) Chl ( > 500 µm) + Ilt Chl ( > 500 µm) + Ilt Chl Qz + Ilt Qz + Ilt Qz + Ilt Qz + Ilt 2 mm Sample: no. 9b (pegmatite) 2 mm Ilt + Chl + Qz Ilt + Chl (< 500 µm) Chl ( > 500 µm) + Ilt Chl ( > 500 µm) + Ilt Chl Qz + Ilt Qz + Ilt Qz + Ilt Qz + Ilt Cb Cb Figure 9 - Comparison between thin section images and spectral-class maps for the pegmatite samples no. 1 (A), no. 8 (B), and no. 13 (C), and for a gneiss sample (no. 6, D). 1) Thin section images are in transmitted, polarized, and analyzed light; 2) Mineral maps from thin section interpretation; 3) Spectral-class maps, and 4) Layering of the thin section mineral maps over the spectral-class maps. Ilt: illite; Sud: sudoite; Chl: chloritized biotite; Ms: muscovite; Cb: carbonate; Drv: dravite; Qz: quartz. The diagrams show the comparison of the mineralogical proportions obtained by thin section analysis (X-axis) and by hyperspectral image classification (Y-axis) showing correlations between the methods. 1 2 3 1 2 3 4 4 1 2 3 4 Chlorite Illite + Muscovite Sudoite Kaolins, Dravite, Carbonates y = 0,9256x R² = 0,721 0 10 20 30 40 50 60 0 10 20 30 40 50 60 70 Mineral proportions (relative %) derived from hyperspectral image analysis Mineral proportions (relative %) derived from thin section analysis Chlorite Illite + Muscovite Sudoite Kaolins, Dravite, Carbonates y = 0,6402x R² = 0,114 0 10 20 30 40 50 60 70 0 10 20 30 40 50 60 Mineral proportions (relative %) derived from thin section analysis Mineral proportions (relative %) derived from hyperspectral image analysis Chlorite, Dravite, Kaolins Illite + Muscovite Sudoite Carbonates y = 0,9686x R² = 0,982 0 10 20 30 40 50 60 70 0 10 20 30 40 50 60 70 Mineral proportions (relative %) derived from thin section analysis Mineral proportions (relative %) derived from hyperspectral image analysis Chlorite Illite + Muscovite Sudoite Dravite, Kaolins Carbonates y = 0,9394x R² = 0,629 0 5 10 15 20 25 30 35 40 45 0 10 20 30 40 50 Mineral proportions (relative %) derived from thin section analysis Mineral proportions (relative %) derived from hyperspectral image analysis Dolomite Magnesite Unclassified Rhodocrosite Calcite Dickite + Illite Dickite Kaolinite Illite (basement) Paragonite Phengite Muscovite Illite (sandstone) Mica + Chlorite Chlorite + Mica or Sudoite Chlorite Masked pixels Dravite 5 cm 5 cm 21 1 2 3 4 5 6 7 8 9 10 11 12 13 14 20 2348 2197 2270 (nm) 5 cm 5 cm Figure 7 - HySpex SWIR-320m color composites defined with bands sensitive to the spectral properties of illite (red indi- cates 2348 nm; green indicates 2270 nm; blue indicates 2197 nm) with normalized reflectance [5]. Pegmatites: samples no. 1, 2, 3, 8, 9 and 13; gneisses: samples no. 4, 5, 6, 7, 10, 11, 12 and 14; sandstones: samples no. 20 and 21. Dickite Illite basement Illite sandstone Dolomite Figure 8 - HySpex SWIR-320m spectral-class image of the Waterbury/Cigar Lake project samples. Pegmatites: samples no. 1, 2, 3, 8, 9 and 13; gneisses: samples no. 4, 5, 6, 7, 10, 11, 12 and 14; sandstones: samples no. 20 and 21. See color key for image color-class assignment. Note that unclassified (black) pixels correspond to quartz, feldspar, and sulphide minerals, and also to small phyllosilicate grains that display a noisy spectral signature. Legend (images 3 and 4) Muscovite Illite Unclassified Mica + Chlorite Chlorite Dolomite Dickite Chlorite + Mica or Sudoite 21 1 2 3 4 5 6 7 8 9 10 11 12 13 14 20 Minerals can be identified in each pixel of the entire image due to a spectral resolution high enough to resolve the relatively narrow OH absorption features. Minerals can be highlighted using color composites or by classifying each pixel of the hyperspectral data cube. Two types of methods exist: i) unsupervised, which are automatic, based on iterative processes with no operator intervention, and ii) supervised, which are manual, with operator intervention. Most of unsupervised methods do not demonstrate proper mathematical sophistication; supervised methods have generally been used on hyperspectral data, such as the Spectral Angle Mapper method (SAM) [19] and the Mixture-Tuned Matched Filtering method (MTMF) [20]. Conse- quently, an algorithm was developed and implemented in the IDL language to classify minerals based on their spectral signature in two steps (see flowchart in Fig. 4 and Fig. 5). The first step (Fig. 4, box 1) looks for the maximum and minimum absorp- tion feature position in a selection of characteristic ranges determined for each mineral type listed in Table 2. The second step (Fig. 4, box 2) uses the combination of minimum and max- imum found in step one to classify each pixel in a class number. Some of the minerals (e.g. carbonates, tourmaline, and kaolins) have sufficiently contrasting signatures to be classified using only the position of their maxi- mum absorption features (Fig. 4, box 2a). Other minerals having common absorption bands (e.g. micas and kaolins at 2207 nm) require additional conditions to differentiate them (Table 3 and Fig. 4, box 2b). Since illite and muscovite are similar in chemistry, their differentiation requires comparisons in multiple ranges of the spectrum. This is done using other parameters, such as: i) depth of the absorption feature linked to the water (around 1900 nm, Eq. 1), ii) Index mica value de- fined by Eq. 2, and iii) spectral slope characteristics of some mineral mixtures. An example is presented in Fig. 6 to illustrate the parameters defined by the equations above. The "chlorite index" (same expression as the "mica index" Eq. 2, with different spectral ranges) is used to classify the various chlorites, with- out species discrimination. The “pseudo-modal” composition is the spectral proportion of identifiable minerals present in each sample and is defined by Eq. 3. min 2080 1808 ref ref ref Depth O H × + × = 2 1000 2 − + × × = 1 ref ref ref Index mica min 2 min max 2 1000 ( ) 100 × = ∑ p mineral mineral N Pixel % Proportion with: ref1808the reflectance value at 1808 nm ref2080the reflectance value at 2080 nm refmin the minimum reflectance value in the range 1808-2080 nm Eq. 1 Eq. 2 with: refmin the first minimum reflectance value in the range 2139-2294 nm refmaxthe maximum reflectance value in the range 2207-2338 nm ref2minthe second minimum reflectance value in the range 2294-2377 nm with: Proportion the proportion of a mineral (class) in the chosen sample, in % the sum of the pixels belonging to the mineral Eq. 3 Illite Muscovite Illite 1808 nm 2080 nm 2139 nm 2207 nm 2294 nm 2338 nm 1808 nm 1 2 3 4 ref a) b) max ref 2min ref min Class Carbonates Kaolins Tourmaline (dravite) Micas Chlorites Reflectance minimum (research interval 1) 2106 - 2402 2153 - 2192 2324 - 2397 2139 - 2294 2187 - 2275 Reflectance maximum (research interval 1) 2331 - 2417 2178 - 2207 - 2207 - 2338 2251 - 2343 Reflectance minimum (research interval 2) - 2192 - 2241 - 2294 - 2377 2275 - 2387 Reflectance maximum (research interval 2) - 2207 - 2260 - - - Table 2 - Spectral ranges (in nm) defined to select maximum and minimum absorption feature positions to classify minerals. Mineral Membership conditions for the mineral class Minimum reflectance position in the range (nm) Maximum reflectance in the range (nm) Other necessary conditions Magnesite 2300 - 2310 2331 - 2402 - Dolomite 2310 - 2330 2331 - 2402 - Calcite 2330 - 2335 - Rhodocrosite 2335 - 2336 2331 - 2402 - Kaolinite 2160 - 2170 2207 - 2260 - Dickite 2170 - 2180 - Mixed Illite-Dickite - 2207 - 2338 positive slope between 1379 and 1398 nm during mica research Paragonite 2190 - 2195 - Illite (basement) 2192 - 2210 Depth H2O (1900 nm) > 1100 Illite (sandstone) 2192 - 2210 Index Mica > 750 Muscovite 2206 - 2210 Depth H2O (1900 nm) < 1200 Index Mica < 400 Phengite 2217 - 2222 - Micas + Chlorites - negative slope between wavelength ref min and wavelength ref max during mica research Chlorites < 2260 Index Chlorite > 16 Chlorite + Micas or Sudoite - negative slope between 2187 nm and wavelength ref min during chlorite research Tourmaline (Dravite) 2367 - 2369 - - Table 3 - Spectral ranges (in nm) defined to classify minerals based on additional spec- tral features. Figure 4 - Processing flowchart of the algorithm developed and implemented in IDL to classify minerals from a hyperspectral image. 1 2 2a 2b Tourmaline Kaolins Carbonates Dravite Kaolinite Dickite Magnesite Calcite Dolomite Rhodocrosite Micas Chlorites (Fe-chlorite) (Int-chlorite) (Mg-chlorite) Paragonite Illite (basement) Muscovite Phengite Minimum of reflectance position (in wavelength) research and shoulder research (maximum of reflectance) Chlorite + mica or sudoïte Mica + chlorite Illite + dickite Constraints : - 1900 nm absorption depth (H2O) - Mica index - Slope calculation Constraints : - Slope calculation - Chlorite index - Second reflectance minimum position Illite (sandstone) Chlorites Classified image bn b1 Input data (hyperspectral cube) Figure 6 - Spectral ranges and bands used to calculate the water absorption depth near 1900 nm and the micas index. a) Spectral ranges used for the parameter re- search. 1: minimum of reflectance research window for the H2O feature; 2: re- search window for the first minimum of reflectance for Al-OH mica feature (refmin); 3: research window for the maximum of reflectance for the Al-OH mica feature (refmax); 4: research window for the second minimum of reflectance for the Al-OH mica feature (ref2min). b) representation of the refmin, refmax, and ref2min parameters to differentiate illite from muscovite. min max 2min 2min 2 min max * MINERALOGICAL MAPPING METHODOLOGY Figure 5 - HySpex SWIR-320m spectra of minerals characterizing Waterbury/Cigar Lake project samples: a) illite, b) muscovite and phengite, c) chlorite, d) dickite, e) dolomite, and f) dravite. Continuum removal has been applied to enhance the dif- ferences in shape between spectra. The short vertical lines indicate wavelength location of important absorption features. Illite 1 Illite 2a Illite 2b Illite 2c Muscovite 1 Muscovite 2 Phengite OH, H2O H2O OH, H2O 1910 2207 2348 2217 2197 2353 2207 2197 2343 2202 2348 Mg-OH Al-OH Mg-OH Al-OH H2O 1413 1910 1413 a) b) - - Dolomite Dravite 1934 2324 1408 1442 1934 2207 2241 2368 H2O (Mg,Ca)CO3 Mg-OH Al-OH H2O OH e) f) - - Mg-Fe chlorite Mg chlorite Dickite 1403 2256 2343 2251 2333 1384 1413 2178 2207 Mg-OH Fe-OH OH, H2O Al-OH OH c) d) - - 1808 2080 min min max 2min “Pseudo-modal” composition because it takes into account only the minerals having a spectral signa- ture in the SWIR range (i.e. excludes mainly quartz, unaltered feldspars, and sulfides). In summary, class membership needs three conditions: i) absorption feature positions have to be defined in a nanometer-wide range, ii) a maximum is necessary to identify the spectral shape, and iii) for some minerals (e.g. micas, clays and chlorites) additional constraints are necessary. ∑ mineral Pixel mineral Proportion HySpex SWIR-320m Motorized platform Spectralon® (reference) Computer (data acquisition) Halogen lamp Figure 3 - a) HySpex SWIR-320m camera: the device is mounted on the scanning system while samples are placed on a motorized platform (the red arrow shows the translational movement of the platform during data ac- quisition). b) Core sample set on the motorized platform. SAMPLING 16 barren drill core sections (10 to 30 cm long), taken from drill holes within Waterbury/Cigar project area and from 1 to 8 km away from the deposit, were selected. They represent different: i) lithologies from the basement (pelitic gneiss and pegmatites) and the sandstone, and ii) alteration types and degrees (from retrograde to hydrothermal alteration, and from weak to strong hydrothermal alteration degree). HYPERSPECTRAL DATA ACQUISITION Gneiss, pegmatite, and sandstone samples were scanned using a HySpex SWIR-320m hyperspectral camera (Table 1 and Fig. 3) in order to detect the presence and abundance of key minerals like clays and tourmaline. a) b) Spectralon® Core sample Support Motorized platform Support for the Spectralon® 4 cm (approximate scale) Manufacturer Norsk Elektro Optikk (NEO) Dimensions (L × W × H; cm) * 36 × 14 × 15.2 Weight (kg) 7 Optical device Lens (30 cm x 100 cm x 300 cm) Field Of View (FOV; degrees) 14 Type of detector SWIR/HgCdTe Spatial pixels (line) 320 Spectral range (nm) 1300-2500 Re-sampling interval (nm) User-defined Integration time (ms) User-defined Number of bands 239 Pixel size (mm × mm) 0.5 × 0.5 Table 1 - HySpex SWIR-320m technical specifications. *camera support not taken into account SAMPLES AND TOOLS GEOLOGICAL SETTING: CIGAR LAKE DEPOSIT The Cigar Lake deposit is located in the eastern part of the Athabasca Basin (Fig. 1), at the un- conformity between the Manitou Falls Formation (Athabasca Group) and the Archean-Pro- terozoic basement complex (Wollaston Group). The entire orebody forms a continuous lens approximately 2 km long by 25 to 100 m width, lying at a depth of approximately 450 m below the surface. The high-grade core of the deposit is approximately 600 m in length and 100 m in width. ALTERATION The hydrothermal host-rock alteration affects both the Athabasca sandstones and the base- ment rocks, in particular close to reactivated fault damage zones (Fig. 2). In the sandstones, the alteration forms haloes that are geometrically arranged around the orebody, extending up to 300 m from the main mineralization. In the basement, two hydrothermal alteration zones are evident. The first zone, located near the unconformity, is composed of illite and chlorite (mainly sudoite which replaces retrograde chlorites), replacing the original rock-forming silicate minerals (biotite, garnet, amphibole, feld- spar), and local dravite (alkali-deficient dravitic tourmaline: magnesiofoitite sub-group). This zone is also frequently depleted in graphite and sulphides. In the second zone the rock is clay-altered, but the original textures and quartz are preserved, as well as graphite and sulphi- des, and illite predominates over sudoite. MINERALIZATION The main orebody forms a lens lying at and above the unconformity within the Athabasca Group sandstones; however, a minor part of the uranium mineralization occurs in the sub-Athabasca basement rocks (Fig. 2). 20 m 20 m Faults Perched mineralization Illite 1Mc + illite 2M + dickite Quartz zone Illite 1Mc + dickite Hematitic zone Main mineralization Sudoite + hydromuscovite 1Mt Sudoite + hydromuscovite 1Mt S N UNCONFORMITY MANITOU FALLS Illitized zone Wollaston Group: Augen textured metapelitic gneiss Fine-grained metapelitic gneiss Calc-magnesian gneiss OVERBURDEN WOLLASTON GROUP Figure 2 - Schematic cross-section showing the geology, ge- ometry, and the alteration haloes of the Cigar Lake deposit (modified from 15 to 18). Figure 1 - Simplified geological map showing the location of the Water- bury/Cigar project property and the Cigar Lake deposit within the Athabasca Basin. Other uranium deposits are highlighted by red dots. Projection: UTM Zone 13, datum NAD 1983. INTRODUCTION Hyperspectral imaging is a technique that combines both spectral and spatial imaging methods. It is based on the representation of objects in hun- dreds of narrow and contiguous spectral bands, with a spectral resolution of 10 nm or less [1]. The acquired images correspond to a "hyperspectral cube" [2] endowed with two spatial dimensions (i.e. X and Y) and an electromagnetic spectral dimension (i.e. Z). Hyperspectral imaging systems have been used in mineral exploration to map minerals of economic interest at different scales: from space-borne [3, 4] and airborne scales [4, 5] that cover large areas, to more local scales such as outcrops on the ground [6] or laboratory-scale on collected samples [7]. Unconformity-type uranium deposits are formed at the redox interface between oxidized uranium-bearing fluids and a reducing environment [8, 9]. The mineral- ization is intimately associated with alteration minerals that can be used in mineral exploration as proxies for uranium ore [10, 11, 12]. To characterize such mineralogical alterations in the field, hand-held infrared spectrometers are used, with spectral data being acquired along drill cores at dis- crete sampling intervals (e.g. every 2 to 3 m). To date, only a few studies on core sample material from uranium deposits using hyperspectral imagery have been performed [13, 14], however, this type of study has not been conducted on uncon- formity-type U deposits. The aim of this study is to evaluate the utility of mapping host-rock alterations on sampled drill core material using hyperspectral imaging. This analytical tool will in- crease the data density, as well as data quality, to better identify alteration miner- als due to higher spatial and spectral resolutions. A series of samples were collected from drill holes on the Waterbury/Cigar explo- ration project managed by AREVA Resources Canada Inc. (ARC; Fig. 1). The samples were scanned (hyperspectral imaging), analyzed, and interpreted for mineralogy, with calibration using petrography by optical microscopy. A computer algorithm was developed to classify and discriminate minerals based on both the position and the depth of diagnostic absorption bands. INTRODUCTION Using hyperspectral data to characterize alteration minerals in drill core from Cigar Lake U deposit MAGALI MATHIEU , REGIS ROY , PATRICK LAUNEAU , MICHEL CATHELINEAU , DAVID QUIRT 1,3 1 2 3 1 1) AREVA Resources Canada Inc., P.O. Box 9204, 810 - 45th Street West, Saskatoon, SK S7K 3X5, Canada 2) Laboratoire de Planétologie & Géodynamique/UMR-CNRS 6112, Université de Nantes, BP 92209, F-44322 Nantes Cedex 3, France 3) Université de Lorraine, CNRS, CREGU GeoRessources Laboratory, BP 70239, F-54506 Vandoeuvre-lès-Nancy, France References [1] Goetz et al. (1985), Science 228 (4704), 1147-1153. [2] Vane and Goetz (1988), Remote Sens. Environ. 24, 1-29. [3] Kruse et al. (2003), Geosci. Remote Sens. 41(6), 1388-1400. [4] Kruse et al. (2012), Int. J. Remote Sens. 33 (6). [5] Roy et al. (2009), Geochem. Geophys. Geosyst. 10. [6] Kurz et al. (2013), Int. J. Remote Sens. 34 (5), 1798-1822. [7] Baissa et al. (2011), J. Afr. Earth Sci. 61 (1), 1-9. [8] Hoeve and Quirt (1987), Canada. Bull. Mineral. 110, 157–171. [9] Jefferson et al. (2007), Fifth Decennial International Conference on Mineral Exploration, 741-769. [10] Hoeve and Quirt (1984), Saskatchewan Research Council, SRC Technical Report, 187 pp. [11] Quirt (2010), GeoCanada 2010, Calgary. May 2010, 4 pp. [12] Quirt (2013), The 15th International Clay Conference, July 7-11 2013, Rio de Janeiro, Brazil. [13] Zhang et al. (2013), Uranium Geology 29 (4), 249-255. [14] Sun et al.(2015), AER-Advances in Engineering Research 9, 392-395. [15] Bruneton (1987), Economic Minerals of Saskatchewan, Special publication no. 8, 99-119. [16] Cramer (1986), Canadian Nuclear Society, Winnipeg, Man, 697-702. [17] Fouques et al. (1986), Canadian Institute of Mining and Metallurgy (33) 218-229. [18] Pacquet and Weber (1993), Can. J. Earth Sci. 30, 674-688. [19] Kruse et al. (1993), Remote Sens. Environ. 44, 145-163. [20] Williams and Hunt (2002), Remote Sens. Environ. 82 (2-3), 446-456.