System and method for identifying mineral deposits utilizing airborne-acquired hyperspectral data

Abstract

The invention relates to a method for detecting a mineral deposit based on hyperspectral data, comprising : (a) acquiring hyperspectral data G by imaging the given area in distinct light bands; (b) forming a virtual cube having x-y- z dimensions, in which the x-y dimensions define the given area, and the z dimension defines reflectance from the area in each said distinct light bands included in the hyperspectral data G; (c) performing a 1st spectral similarity algorithm calculation on all pixels Gi of hyperspectral data G relative to a given spectrum R of a sought mineral, and receiving a separate score Li for each pixel Gi in data G; (d) selecting a group T of pixels having a score Li indicating remoteness from R, defined as having a value above threshold TH (D), LOWER, and averaging the spectra of these pixels at each wavelength band to receive a single spectrum J that is spectrally remote from a material of interest; (e) ratioing the spectrum of each pixel Gi, by individually dividing it by J, thereby forming a ratioed cube K; (f) performing a 2nd spectral similarity algorithm calculation on all the ratioed pixels Ki of K relative to said given spectrum R, and receiving a score Si for each pixel in K; and selecting pixel s having a score Si showing closeness to R, defined as having a value below a threshold TH (T), UPPER.

Description

[0001] System and Method for Identifying Mineral Deposits Utilizing Airborne-Acquired Hyperspectral Data

[0002] Field of the Invention

[0003] The field of the invention generally relates to processing systems and techniques for identifying mineral deposits with the aid of satellite or airborne-acquired hyperspectral images.

[0004] Background of the Invention

[0005] Identifying mineral deposits on Earth or remote celestial bodies by image processing of satellite or airborne-acquired hyperspectral images is well-known and widely used.

[0006] Hyperspectral imaging involves collecting and processing information in a plurality of narrow spectral bands over a continuous spectral range, producing the spectra of all pixels in the scene. A spectral analysis for each pixel in a scene is performed to identify materials, find objects, or detect processes. Three techniques for acquiring hyperspectral images are typically used. The push broom scanner (and the related whisk broom scanner - spatial scanning) reads images over time. A band sequential scanner (spectral scanning) acquires images of an area at different wavelengths. The snapshot hyperspectral imager uses a staring array to instantly generate an image of the scene relating to a broad spectrum. No matter which of the three hyperspectral imaging techniques is used, the result is a recorded spectrum image with fine wavelength resolution, covering a wide range of wavelengths that can be spectrally analyzed, for example, to determine the possible existence of minerals within the imaged geographic area.

[0007] Several spectral similarity-based algorithms are currently used to analyze hyperspectral images, particularly to determine the existence of specific materials within the imaged area. These similarity-based algorithms typically compare the respective spectra of pixels in a given hyperspectral image and specific reference "signatures" (spectra), where each signature characterizes a respective mineral of interest. When applied to a specific pixel and a target mineral spectrum, the spectral similarity algorithm outputs a score relating to the similarity between the pixel's spectrum and the signature of the target mineral sought.

[0008] For example, the Spectral Angle Mapper (SAM) is a spectral similarity algorithm that calculates the angle between spectral vectors in multi-dimensional space. The number of spectral bands sets the number of dimensions of the spectral vectors; a small angle between the vectors represents a close spectral match. Another typical spectral similarity algorithm is the Spectral Information Divergence (SID), also known as the Kullback-Leibler information measure, directed divergence, or cross-entropy. In some cases, the SID and SAM algorithms are combined to form the SID-SAM spectral similarity algorithm. SID-SAM combines both measures and improves spectral discriminability. In the case of the SID-SAM, a score of zero reflects spectral identity, and the score increases with decreasing spectral similarity in the image pixels. Other spectral similarity algorithms may apply other types of scores to define the similarity or dissimilarity of the two compared spectra. In some cases, the spectral similarity algorithms are applied by utilizing MATLAB®'s Image Processing Toolbox - Hyper-spectral Imaging Library.

[0009] Planetary hyperspectral imaging, which acquires hyperspectral images from satellite or airborne vehicles, is used to detect minerals on Earth or in other celestial bodies. The spectral features of the material of interest are often relatively minor contributors to the acquired spectrum, which represents the combined properties of the imaged surface (e.g., the mineralogy, vegetation cover). Various types of "background" noises must be dealt with by the spectral similarity algorithm to increase accuracy. The most significant background noise typically existing within the image pixels is the mineralogical noise, resulting from the mixture of materials within the imaged pixel area, in addition to the material of interest. The Earth's atmosphere is another noise factor, although to a lesser degree.

[0010] A common technique for eliminating atmospheric spectral disturbances and for highlighting the spectral features of the material of interest is to divide the spectra obtained from an image portion relating to an area of particular interest by a collection of "background" pixels, which seemingly possess no mineral information. This technique is called "ratioing". The ratioing emphasizes the material of interest's spectral features and increases its detectability. While this technique and other currently used techniques allow the identification of material deposits to some degree, they are still limited in their accuracy and success rate, particularly in areas with a low concentration of the material. Furthermore, the choice of the "background" spectrum, which serves as the denominator to which the spectrum of interest is ratioed, is manual. Although best-practice recommendations exist for the choice of a "background" spectrum, there are no clear criteria for a choice that maximizes the sensitivity and specificity of detection.

[0011] • Viviano, C. E. , et al. (2014), Revised CRISM spectral parameters and summary products based on the currently detected mineral diversity on Mars, J. Geophys. Res. Planets, 119, 1403-1431, teaches a manual-based radioing method for analyzing hyperspectral data;

[0012] • Wray, J. et al. (2010), Identification of the Ca-sulfate bassanite in Mawrth Vailis, Mars, Icarus, 209(2), 416— 421, also teaches a manual-based radioing method for analyzing hyperspectral data;

[0013] • Clark, R. N., et al. (2003), Imaging spectroscopy: Earth and planetary remote sensing with the USGS Tetracorder and expert systems, J. Geophys. Res., 108, 5131, discloses a USGS tetracorder system that identifies and maps materials by analyzing their spectral features using a combination of algorithms and expert system rules within imaging spectroscopy data; and

[0014] • Schodlok, M.C. et al.. Implications of new hyperspectral satellites for raw materials exploration. Miner Econ 35, 495-502 (2022) discloses a Spectral Feature Fitting (SFF) approach, which is also a manual method; and

[0015] • Pearlshtien et al., PRISMA,sensor evaluation: a case study of mineral mapping performance over Makhtesh Ramon, Israel, International Journal of Remote Sensing, Volume

[0016] 42, 2021, Issue 10, Pages 0882-5914. It is an object of the invention to provide a method that, based on image processing of aircraft or satellite-acquired hyperspectral images, increases the success rate in identifying areas with a material of interest, for example, mineral deposits.

[0017] Another object of the invention is to increase the sensitivity of detecting materials from hyperspectral images so that these materials can be detected at lower abundances than when utilizing existing techniques.

[0018] Another object of the invention is to provide a method for automatically selecting optimal "background" pixels to serve as the denominator against which spectra of interest are ratioed.

[0019] Another object of the invention is to generalize the system and method of the invention, thereby detecting various materials.

[0020] Still, another object of the invention is to provide a fully automatic method that does not require manual or visual intervention.

[0021] Another object of the invention is to adapt the method and system of the invention to the detection of minerals in remote celestial bodies.

[0022] Other objects and advantages of the invention become apparent as the description proceeds. Summary of the Invention

[0023] The invention relates to a method for detecting a mineral deposit based on hyperspectral data, comprising: (a) acquiring hyperspectral data G by imaging the given area in distinct light bands; (b) forming a virtual cube having x-y-z dimensions, in which the x-y dimensions define the given area, and the z dimension defines reflectance from the area in each said distinct light bands included in the hyperspectral data G; (c) performing a 1stspectral similarity algorithm calculation on all pixels Gi of hyperspectral data G relative to a given spectrum R of a sought mineral, and receiving a separate score Li for each pixel Gi in data G; (d) selecting a group T of pixels having a score Li indicating remoteness from R, defined as having a value above threshold TH(D>,LOWER, and averaging the spectra of these pixels at each wavelength band to receive a single spectrum J that is spectrally remote from a material of interest; (e) ratioing the spectrum of each pixel Gi, by individually dividing it by J, thereby forming a ratioed cube K; (f) performing a 2ndspectral similarity algorithm calculation on all the ratioed pixels Ki of K relative to said given spectrum R, and receiving a score Si for each pixel in K; and selecting pixels having a score Si showing closeness to R, defined as having a value below a threshold TH(T),UPPER.

[0024] In an embodiment of the invention, the group T comprises clusters of contiguous pixels, all having a score TH(D),LowER<Li, where TH(D>,LOWER defines a score threshold relatively remote from a match with the spectrum of R.

[0025] In an embodiment of the invention, the group T comprises clusters of contiguous pixels, all having a score TH(D>,LOWER <Li<TH(D),UPPER, where TH(D>,LOWER defines a score threshold relatively remote from match with the spectrum of R, and TH (D),LOWER<TH (D),UPPER.

[0026] In an embodiment of the invention, each cluster includes at least n contiguous pixels.

[0027] In an embodiment of the invention, each of the 1stand 2ndspectrum similarity algorithms are SID-SAM.

[0028] In an embodiment of the invention, the selection of n depends on the signal to noise ratio of the spectral data.

[0029] In an embodiment of the invention, for a lower signal to noise ratio, a higher n is selected.

[0030] In an embodiment of the invention, the method is used for mapping of mineral deposits.

[0031] Brief Description of the Drawings

[0032] In the drawings:

[0033] Fig. la shows an example of hyperspectral imaging (in this case, imaging of Mars);

[0034] Fig. lb shows an example of a 2D reflectance image in a single band (i.e., single wavelength);

[0035] Fig. 1c shows an entire (multi-pixel) region viewed as a 3D cube, summarizing the accumulated response in all 544 bands; Fig. Id shows a combined spectrum of a single pixel indicated in Fig. 1c (based on reflectance in the plurality of individual bands); Fig. le is similar to Fig. 1c, showing within the cube a reflectance from a region (dimensions x-y), over a plurality of bands (dimension z);

[0036] Fig. If shows the three matrices, each relating to one of the three RGB channels, respectively, from among the channels included in the cube of Fig. le;

[0037] Fig. 1g symbolizes the RGB combination of Fig. If;

[0038] Fig. Ih illustrates an example of how three separate R+G+B images, each image representing a region's reflectance at one specific band (as shown), are combined into a single composite image that enables a visual evaluation;

[0039] Fig. li schematically demonstrates how, based on an RGB combination of three selected bands (where R=2.4pm, B=1.9pm, and G=2.1pm), various materials (or material families) can be observed within a composite image;

[0040] Fig. Ij shows an example of a laboratory olivine spectrum and a spectrum obtained by CRISM (Compact Reconnaissance Imaging Spectrometer for Mars) and assigned to olivine, given their spectral similarity;

[0041] Fig. Ik provides examples illustrating how the selection of background pixels affects the desired spectrum for verification;

[0042] Fig. 11 shows an example of the results of a ratioing procedure;

[0043] Fig. Im summarizes the typical prior art process for detecting mineral deposits in a hyperspectral image;

[0044] Fig. 2 is a flow diagram illustrating a process according to the present invention for detecting minerals based on hyperspectral data;

[0045] Fig. 3 shows gypsum (and bentonite and kaolinite) deposits detected by Pearlshtien et al. (2021) within Ramon Crater; Fig. 4a shows the findings of gypsum by the procedure of the invention at the Ramon Crater;

[0046] Fig. 4b shows vein fillings in contact with overlying carbonates as observed in a visit within location A (at Ramon Crater) determined by the procedure of the invention; Fig. 4c shows massive vein fillings farther from overlying carbonate units, as observed at a visit at location A within the Ramon Crater;

[0047] Fig. 4d shows weathered gypsum pebbles found at location A away from the veins, on desert pavement stabilized to variable degree on soil terraces;

[0048] Fig. 4e shows Gypsum vein fillings near the marl-carbonate contact as found in location B within the Ramon Crater;

[0049] Fig. 5a shows a location where Calvin et al. (2009) reported gypsum founding in Mars' North Polar Region;

[0050] Fig. 5b shows the results by the method of the invention, where high concentrations of gypsum were found throughout the entire area;

[0051] Figs. 5c and 5d compare findings by Wray et al. (2010), who applied prior art techniques, to those obtained by the present invention's techniques; and

[0052] Fig. 6 is a table summarising prior art findings in locations of previous detections on Mars.

[0053] Detailed Description of Preferred Embodiments

[0054] Fig. la shows an example of hyperspectral imaging (in this case, imaging of Mars). The pixel's spatial (x-y) resolution is 18m per pixel. In this example, the entire image region is hyperspectrally imaged by 544 bands in the 362nm-3920nm wavelength range, such that each band spans 6.55nm. An example of a 2D reflectance image 112 in a single band (i.e., single wavelength) is shown in Fig. lb. The entire (multi-pixel) region may be viewed as a 3D cube 102 summarizing the accumulated response (hyperspectral data) in all the 544 bands, as shown in Fig. 1c, where the x and y dimensions indicate spatial directions, and the z dimension represents the reflectance in the various 544 bands, respectively. Practically, cube 102 includes (in this specific case) 544 "slices" like slice 112 shown in Fig. lb, each relating to one particular wavelength band. Fig. Id shows a combined spectrum 103 of a single pixel 107 indicated in Fig. 1c (based on reflectance in the plurality of individual bands). Fig. le is similar to Fig. 1c, showing within cube 102 the reflectance from a region (dimensions x-y), over the plurality of bands 114 (dimension z).

[0055] Hyperspectral data are difficult to visualize in map view. This difficulty is typically overcome by degrading the full spectrum of each pixel into a Red-Green-Blue (RGB) false-color composite. In each pixel, each of the three RGB channels typically represents a spectral feature at a particular wavelength (e.g., the reflectance at 400nm) or a simple algebraic combination of spectral features (e.g., the ratio of the reflectance at 400nm to that at 450nm). The RGB color composite thus represents only a few bands of the hyperspectral data, but it can easily be visualized in map view. The resulting image is sometimes referred to as a "browse product" that allows rapid qualitative assessment of the spectral variability of the surface.

[0056] In one example, the browse product is configured based on the known spectral signature of the specific mineral sought. For example, suppose the signature of the sought mineral is characterized by having three absorbance points at the frequencies fi, f2, and fs. In that case, red, green, and blue colors may be assigned to these three frequencies, respectively. In some cases, the reflectance at one frequency is artificially amplified relative to the others within the browse product to emphasize the absorbance in this specific frequency.

[0057] The term mineral as used herein is meant to also include naturally occurring minerals from which metals and other elements can be extracted (e.g., phosphorus), namely ores. The metals usually exist in the form of oxides, sulfides, sulfates, silicates, and may also be present in the ores in the elemental form.

[0058] To better visualize data of interest from cube 102, selected data (for example, data resulting from selected bands) from cube 102 are introduced into 3 RGB matrices 107, respectively. Fig. If shows the three matrices, each relating to one of the three RGB channels, respectively. Each cell 107 within a matrix 104 represents a reflection from a single pixel, respectively. In some cases, the data within the three RGB matrices 104 results from a selected manipulation of cube 102. The combination of the three RGB matrices' 104 data, as symbolically shown in Fig. 1g, results in a composite image, which can be evaluated visually. More specifically, a rapid qualitative assessment of the spectral variability of the region's surface becomes possible.

[0059] Fig. Ih illustrates an example of how three separate R+G+B images 112a-112c, respectively, each image representing a region's reflectance at one specific band (as shown), are combined into a single composite image 113 that enables a visual evaluation. Fig. li schematically demonstrates how, based on an RGB combination of three selected bands (where R=2.4pm, B=1.9pm, and G=2.1pm), various materials (or material families) can be observed within a composite image.

[0060] Fig. Ij shows an example of a laboratory olivine spectrum and a spectrum obtained by CRISM (Compact Reconnaissance Imaging Spectrometer for Mars) and assigned to olivine, given their spectral similarity.

[0061] While the browse product briefly and visually highlights areas of interest, the prior art uses various spectral similarity algorithms to pixel-wise confirm that specific minerals indeed exist in those areas of interest. Those spectral similarity algorithms determine the similarity between a given (known) signature of a mineral of interest and the reflected spectrum (as obtained from cube 102) at each pixel of interest. SID and SAM are examples of spectral similarity (vector similarity in general) algorithms that provide a similarity score between the two (given and examined) spectrum signatures. The SID-SAM combines the results determined separately using the SID and SAM techniques. The combined SID-SAM score is a linear score (0<score) estimating the possibility of detecting the mineral in question in an area relating to a given image pixel. The spectral similarity algorithm (for example, the SID-SAM) may be applied to specific pixels within the image of cube 102 or to the entire image by repeatedly applying the technique to additional pixels.

[0062] To demonstrate the invention's applicability, the inventors used the combined SID-SAM algorithm; however, any other spectral similarity algorithm may be used, according to the present invention. In SAM: The angle a between a given observed spectrum t and a given reference spectrum r, both of a length of C (i.e., they both span the same spectral range and resolution = total C i.e., N of elements), is calculated as:

[0063] Where t is the spectrum of the inspected pixel, and Y is a reference spectrum relating to a specific mineral, as obtained from some database.

[0064] In discretized form, the above equation becomes:

[0065] The SAM score is then:

[0066] SID score is defined as: where p and q are the distribution values of test and reference spectra, respectively, and i represents the channel index:

[0067] Finally, the SID-SAM combined score (a linear value) is:

[0068] The SID-SAM score is a number S, where 0≤S (other spectral similarity algorithms may use different ranges to indicate similarity or remoteness). A lower S score means a higher match with the reference mineral signature, meaning a higher probability that the mineral exists at the respective pixel location. A score of zero means a full match, while a high number indicates no match.

[0069] The analysis (such as by SID, SAM, or SID-SAM) of given hyperspectral data obtained from a region of interest typically begins, as an initial stage, by ratioing the spectrum obtained relative to a "background" (denominator) spectrum and by doing so, highlighting (enhancing) relevant spectral features and reducing disturbances such as (a) spectral noise resulting from a mixture of materials at the pixel location; (b) the effects of dust cover; and (c) residual atmospheric effects. This ratioing step requires selecting a denominator spectrum while the numerator represents the region's target spectrum. An accurate denominator is critical to increasing the sensitivity of detection of the mineral of interest and minimizing false mineral detections at the regions of interest (ROIs). However, an appropriate finding and selection of spectrally neutral "background" pixels (i.e., spectra of pixels that do not include the material in question) for the denominator is challenging. Fig. Ik provides examples illustrating how the selection of background pixels affects the desired spectrum for verification. Example 170 shows a case where an appropriate selection of background pixels is made, namely, the selection of spectrally neutral background pixels that presumably do not contain specific mineral data (or at least not the mineral of interest). The spectra of the selected background pixels were averaged and used as the denominator during ratioing. The resulting ratioed spectrum emphasizes the spectral features of materials present at the surface. Example 172 shows a case where an inappropriate selection of background pixels was made (namely, pixels containing mineral spectral features - see the overlapping absorbance points); the averaging of their spectra results in a ratioed target spectrum with spurious spectral features.

[0070] Fig. 11 shows an example of the ratioing procedure. Signal A shows the background spectrum of a pixel (typically acquired by the prior art from the average of spectra from spectrally neutral pixels within the image). Signal A is used as the denominator during the ratioing step. Signal B is the target pixel spectrum acquired from a region of interest and used as the numerator. Signal C is the ratioed signal obtained by dividing signal B (target spectrum) by signal A (background spectrum). Finally, signal D is a lab reference spectrum (signature) of a specific mineral sought at the region of interest.

[0071] Various databases such as CRISM (Compact Reconnaissance Imaging Spectrometer for Mars), EnMAP, or PRISMA contain precalculated ratioed spectra that are typical to specific minerals. For example, hydrated minerals typically display absorption at ~1.9 / im. Prior art techniques manually and intuitively select the pixels of interest (i.e., pixels for manual evaluation of the presence / absence of the sought mineral) based on their color (assuming the color reflects relevant spectral parameters) within the Browse Product. The spectra of the selected pixels of interest are then ratioed relative to a "background" denominator spectrum to eliminate the various types of noises. The background pixels for the denominator were also manually and intuitively selected from the browse products, based on their "darkness" - presumably they reflect bland pixels' spectra (bland indicates a lack of strong features or characteristics and therefore uninteresting).

[0072] Fig. 1m summarizes the typical prior art process 130 for detecting mineral deposits in a hyperspectral image. In step 132, the hyperspectral data A are acquired from a region of interest. The hyperspectral data A are typically represented by a plurality of channel matrices (for example, 544 channels when CRISM is used to acquire the hyperspectral data; any other number of channels is applicable. In step 134, the hyperspectral data A are reduced to an RGB browse product B, in the manner as discussed above. In step 136, a group (one or more) of background (spectrally neutral) pixels are manually and intuitively selected and merged to form a single background spectrum C. In step 138, the raw spectra of pixels of interest the raw spectra. In step 140, a spectral similarity procedure is performed on each of the ratioed spectra of interest di,

[0073]

[0074] The manual-intuitive selection of "dark" (background, hopefully spectrally neutral) pixels to determine the ratioing denominator is common to all known prior art techniques. The inventors have found that this manual-intuitive selection of pixels for the denominator is not optimal. The invention provides an automatic method for selecting background pixels for the ratioing denominator, which, based on experiments, provides significantly improved results in detecting mineral deposits compared to the prior art technique of Fig. Im.

[0075] The inventors found that the ratioing procedure performed by the prior art procedure 130 is not optimal and can be significantly improved. Specifically, the inventors found that the manual-intuitive selection of the spectrally neutral pixels from the browse product - whose distinct spectra are then averaged to form the denominator of the ratioing stage (138) is not optimal. The invention provides an automated procedure for selecting pixels that are verified to be spectrally distant from the material of interest compared to those intuitively chosen by the prior art. As is elaborated below, the verification is based on remoteness of pixel signatures from the reference signature, as determined by a spectral similarity algorithm. Fig. 2 is a flow diagram illustrating a process 230 according to the present invention for detecting minerals based on hyperspectral data. In step 232, the hyperspectral data are acquired from a region of interest forming a hyperspectral data cube G (similar to cube 102 of Fig. 1c). The hyperspectral cube is in fact virtual, as according to the present invention there is no necessity to visualize its data. In step 234, a 1stspectral similarity procedure relative to a reference signature R is conducted on all the pixels of G, resulting in first similarity scores Li for each pixel (i indicates the pixel's index). Reference signature R is the signature of the sought mineral, as obtained from a public library, or otherwise. In step 236, a group T of pixels with a score Li indicating their remoteness of spectrum from spectrum R is selected. For example, if a SID-SAM algorithm is used (in which a score Li=0 indicates a full match and remoteness is indicated by scores Li greater than 0), the selected pixels are those having a score TH(D>,LOWER<L1 (where TH(D>,LOWER is a predefined threshold indicating sufficient remoteness from a full match). The spectra of all the selected pixels in T are then averaged, resulting in a single spectrum J representing background pixels that are spectrally remote from the material of interest. Any type of algebraic averaging may be applied in this stage. In step 238, all the pixel spectra within virtual cube G (hyperspectral data) are ratioed by dividing each pixel spectrum Gi (spectrum Gi represents the spectrum of pixel I within hyperspectral data G) by the spectrally remote signature J obtained in step 236 (where in each such division, the respective Gi is the numerator and J is the denominator). The result is a ratioed virtual cube K in which the spectra of all the cube's pixels are ratioed to emphasize their features. Then, in step 240 a 2ndspectral similarity algorithm is calculated separately for each ratioed pixel spectrum Ki of cube K relative to the signature R (of the sought mineral), resulting in a second (final) set of similarity scores S1, S2, S3-Sn. The scores Si, S2, S3-Sn are used as an indication of the likelihood of finding the mineral relating to R within the respective area pixel in K. For example, in an algorithm in which a full match is represented by a score of 0, any score S<TH(T),UPPER (where TH(T>,UPPER is a predefined threshold indicating sufficient similarity to a full match) indicates a high likelihood of finding the material R in the physical region related to pixel Ki.

[0076] As shown, in step 236 (which follows the performance of the 1stspectral similarity calculation) pixels having spectra remote from R are selected to serve as background pixels. In contrast, in step 242 (which follows the performance of the 2ndspectral similarity calculation) pixels having spectra close to R are selected, as presumably including the material R (in the respective physical locations).

[0077] The inventors found that (for at least cases where SID-SAM is used) it is preferable to select in step 236 clusters consisting of at least a number n of adjacent (contiguous) pixels. The selection of n depends on the signal to noise ratio of the spectral data. The lower the signal to noise ratio, the higher the preferable value of n, as the averaging of more selected contiguous pixels reduces the effect of the noise. Therefore, in an ideal case (i.e., where the signal to noise ratio is very high), n may be 1, while in more practical cases, n is 2 or larger than 2. Various techniques, conventional or machine-learning based may be used to determine from the environment the level of the noise (and the related signal to

[0078] that the selected background pixels are remote in their score from the much lower score expected from pixels that supposedly include the sought mineral (whose score may be, for example, Li<4.0). The results of the 1stspectral similarity-based calculation are less accurate, as this calculation operates on raw (not ratioed) pixel spectra; however, it has been found that the accuracy is sufficient to find the spectrally remote pixels that are then used in the ratioing step 238. Moreover, steps 234 and 236 automatically determine spectrally remote pixels, in contrast to the manual-intuitive selection performed by the prior art techniques. The method of the invention also eliminates the necessity to create a browse product.

[0079] As will be further discussed below, the inventors have performed two main experiments to prove the concept of the invention, as follows: a.A first series of experiments was performed based on hyperspectral data obtained from locations (scenes) on the surface of Mars by the CRISM instrument on the Mars Reconnaissance Orbiter spacecraft. The existence of the calcium sulfate minerals gypsum (CaSO4x2H2O) and bassanite (CaS04xO.5H2O), as evaluated on the basis of the invention's present process, were compared to the findings of previous works. All of the mineral deposit locations discovered by previous works were also found by the invention's present process. In addition, the present process has found additional areas of mineral deposits, proving a better detection sensitivity. b. A second series of experiments was performed utilizing hyperspectral data obtained by EnMAP relating to the existence of gypsum deposits in Ramon crater in Israel. The experimental results were also compared to the findings of the previous work of Pearlshtien et al. (2021) (Listed above in the "Background of the Invention" chapter). Essentially all of the mineral deposits detected by Pearlshtien et al. (2021) in Ramon crater were also found by the invention's present process. In addition, the present process has found additional areas of gypsum deposits that were not detected by the previous works. The inventors visited said additional locations within the Ramon crater and have confirmed the existence of gypsum deposits at these locations. In some cases, the invention's process detected gypsum at much lower abundances than those present in the locations previously documented by Pearlshtein et al. (2021). These experimental results demonstrate that the invention's process has better sensitivity and accuracy compared to Pearlshtien et al. (2021), who relied on prior art procedures.

[0080] Experiment 1

[0081] The inventors applied the method of the invention to detect gypsum and bassanite deposits on the surface of Mars. The results were compared to previously known findings by other researchers, as follows:

[0082] 1.Calvin, W. M., Roach, L. H., Seelos, F. P., Seelos, K. D., Green, R. 0., Murchie, S. L., & Mustard, J. F. (2009). Compact reconnaissance imaging spectrometer for mars observations of northern martian latitudes in summer. Journal of Geophysical Research: Planets, 114,0-11.

[0083] 2. Weitz, C. M., Bishop, J. L., & Grant, J. A. (2013).

[0084] Gypsum, opal, and fluvial channels within a trough of noctis labyrinthus, mars: Implications for aqueous activity during the late hesperian to amazonian. Planetary and Space Science, 87, 130-145.

[0085] 3.Mangold, N., Roach, L., Milliken, R., Mouelic, S. L.,

[0086] Ansan, V., Bibring, J. P., Masson, P., Mustard, J. F., Murchie, S., & Neukum, G. (2010). A late amazonian alteration layer related to local volcanism on mars. Icarus, 207, 265-276.

[0087] ICARUS.2009.10.015;

[0088] 4.Wray, J. J., Murchie, S. L., Squyres, S. W., Seelos, F.

[0089] P., & Tornabene, L. L. (2009). Diverse aqueous environments on ancient Mars revealed in the southern highlands. Geology, 37(11), 1043-1046.

[0090] 5.Wray, J. J., Squyres, S. W., Roach, L. H., Bishop, J. L., Mustard, J. F., & Noe Dobrea, E. Z. (2010). Identification of the Ca-sulfate bassanite in Mawrth Vallis, Mars. Icarus, 209(2), 416-421.

[0091] 6.Wray, J. J., Squyres, S. W., Roach, L. H., Bishop, J. L.,

[0092] Mustard, J. F., & Noe Dobrea, E. Z. (2010). Identification of the Ca-sulfate bassanite in Mawrth Vallis, Mars. Icarus, 209(2), 416-421.

[0093] 7.Ackiss, S. E., & Wray, J. J. (2014). Occurrences of possible hydrated sulfates in the southern high latitudes of Mars. Icarus, 243, 311-324. https: / / doi.org / 10.1016 / J.ICARUS.2014.08.016

[0094] The table in Fig. 6 summarises prior art findings in locations of previous detections. The "Target ID" is the name provided in the original publication. The "Gypsum score" and "Bassanite score" columns refer to the SID-SAM score for these minerals at the tested location. Note that the locations tested are in many cases approximate, as the information in the original publications is insufficient for the determination of the exact location.

[0095] Again, the inventors utilized the procedure of the invention as in Fig. 2. For the purpose of the invention, the inventors used the following parameters: a. Satellite hyperspectral data: CRISM; b. Spectral similarity algorithm (in both the 1st

[0096] ("denominator") and 2nd("target") runs): SID-SAM; c. TH(T),upper used (following the 2ndrun of the spectral similarity algorithm): 1.7 for gypsum and bassanite d. TH(D),LOWER used (following the 1strun of the spectral similarity algorithm): 10; e. TH(D),UPPER used (following the 1strun of the similarity algorithm): 20; f. Number of contiguous pixels (n) for inclusion in averaging to find spectrally distant ("denominator") spectrum: 10; g. Number of contiguous pixels (N) to be considered a detection: 4.

[0097] In Fig. 5a, Calvin et al. (2009) report gypsum in Mars' North Polar Region, at the small area marked by the tip of arrow A. Fig. 5b shows the findings by the method of the invention, where high concentrations of gypsum were found throughout the entire area, as shown by warm colors indicating spectral similarity with gypsum.

[0098] In another example shown in Fig. 5c, Wray et al. (2010), who applied prior art techniques, report bassanite deposits within the area marked by the dashed rectangle C in a region called Mawrth Vallis on Mars. The inventors have found high concentrations of bassanite deposits not only in this location, but throughout the entire area, as shown in inset D - see also Fig. 5d.

[0099] In still another example, Weitz et al. (2013), who applied prior art techniques, reported the existence gypsum in a region called Noctis Labyrinthus on Mars. When applied by the inventors, the invention's procedure shows that Weitz et al. (2013) were mistaken and that the deposits consist of bassanite, not gypsum. The correct identification of the mineral on the surface at this location became possible due to the high sensitivity of the invention's algorithm.

[0100] In addition to detecting gypsum or bassanite in all previously reported locations, thereby demonstrating the invention algorithm's validity, the algorithm of the present invention also detected new occurrences of gypsum and bassanite that were not detected in previous studies. At times, these new detections were only several hundred meters from the locations of past gypsum and bassanite detections. In other cases, the new detections were in completely different locations (i.e., different satellite "scenes"). The extensive detection of the minerals of interest became possible by the present invention's automatic and relatively rapid algorithm.

[0101] Experiment 2

[0102] The inventors have conducted an experiment to detect gypsum deposits in Ramon Crater, Israel. The findings were compared to the findings of Pearlshtien et al. (2021):

[0103] For the purpose of the experiment, the inventors used the following hyperspectral source: EnMap, which is the latest database used by Pearlshtien et al., though not the database used in their 2021 publication (those authors used other databases during their earlier works).

[0104] The gypsum (and bentonite and kaolinite) deposits detected by Pearlshtien et al. (2021) within Ramon Crater are shown in Fig. 3. Gypsum deposits are shown in magenta and encircled by ellipses. In their experiment, Pearlshtien et al. used mainly SAM but also other approaches like RMSE, ASDS as the spectrum similarity algorithm. They intuitively and manually selected spectrally neutral pixels for the denominator of the ratioing step, and performed essentially the procedure depicted in Fig. Im, as also described above.

[0105] The inventors utilized the procedure of the invention as in Fig. 2. For the purpose of the experiment, the inventors used the following parameters: a. Satellite hyperspectral image of Ramon Crater obtained by EnMap; b. Spectral similarity algorithm (in both the 1stand 2ndruns): SID-SAM; c. TH(T),upper used (following the 2ndrun of the spectral similarity algorithm): 1; d. TH(D),LOWER used (following the 1strun of the spectral similarity algorithm): 10; e. TH(D),UPPER used (following the 1strun of the spectral similarity algorithm): 20; f. Number of contiguous pixels (n) for inclusion in averaging to find spectrally distant ("denominator") spectrum: 10; g. Number of contiguous pixels ( ) to be considered a detection: 1.

[0106] Fig. 4a shows the findings of gypsum by the procedure of the invention at the Ramon Crater. In addition to a strong signal of gypsum in the locations reported also by Pearlshtein et al. (2021) and known to host massive gypsum layers (marked C in Fig. 4a), the inventors found weaker suspected detections at locations A and B (among others), which are absent from the report of Pearlshtien et al. (2021).

[0107] To confirm the results of the hyperspectral analysis, the inventors visited locations A and B marked in Fig. 4a.

[0108] Location A

[0109] In location A, gypsum vein fillings in the Ora Formation marl appear to be the product of pyrite (FeS2) oxidation and carbonate dissolution, which elevated Ca2+and SO42~ concentrations in vein fluids and resulted in gypsum precipitation. The gypsum veins were observed throughout the marl, perhaps with a higher concentration near the contact with the overlying carbonate units. For example, vein fillings in contact with overlying carbonates are shown in Fig. 4b and massive vein fillings farther from the overlying carbonate units are shown in Fig. 4c. Weathered gypsum pebbles are found away from the veins, on desert pavement stabilized to variable degree on soil terraces (Fig. 4d). Brittle gypsum cement is found in the shallow subsurface of the soil terraces (gypcrete). Samples of the veins, pebbles and cements were taken for analysis by X-ray powder diffraction, which verified the mineralogy to be gypsum.

[0110] Several similar stratigraphic sections were investigated within a radius of several kilometers, and all contained the gypsum veins and their weathering products. The inventors reemphasize that Pearlshtien et al. (2021) have not detected gypsum at location A.

[0111] Location B

[0112] In location B, gypsum vein fillings are abundant within the upper marly unit in the Hevion Formation, mostly at the contact with the overlying carbonate units. The gypsum weathers from the rock and is found in the talus of the crater (Makhtesh) walls. Gypsum vein fillings near the marl-carbonate contact are shown in Fig. 4e. Samples of the veins and their weathering products were taken for analysis by X-ray powder diffraction, which verified the mineralogy to be gypsum. We reemphasize that Pearlshtien et al. (2021) have not detected gypsum at location B.

[0113] Experiment 2 summary

[0114] In locations A and B, the algorithm of the present invention outperformed the state of the prior art by detecting low- abundance occurrences of gypsum at the level of vein fillings and pebbles.

Claims

Claims1. A method for detecting a mineral deposit in a given area based on hyperspectral data, comprising: acquiring hyperspectral data G by imaging the given area in distinct light bands; forming a virtual cube having x-y-z dimensions, in which the x-y dimensions define the given area, and the z dimension defines reflectance from the area in each said distinct light bands included in the hyperspectral data G; performing a 1stspectral similarity algorithm calculation on all pixels Gi of hyperspectral data G relative to a given spectrum R of a sought mineral, and receiving a separate score Li for each pixel Gi in G; selecting a group T of pixels having a score Li indicating remoteness from R, defined as having a value above a threshold TH(D>,LOWER, and averaging the spectra of these pixels at each wavelength band to receive a single spectrum J spectrally remote from a material of interest; ratioing the spectrum of each pixel Gi, by individually dividing it by J, thereby forming a ratioed cube K; performing a 2ndspectral similarity algorithm calculation on all the ratioed pixels Ki of cube K relative to said given spectrum R, and receiving a score Si for each pixel in cube K; and selecting pixels with a score of Si showing closeness to spectrum R by having a value below a threshold of TH(T),UPPER, said selected pixels indicate the existence of the mineral relating to spectrum R.

2. The method of claim 1, wherein said group T comprises clusters of contiguous pixels, all having a score TH(D),LowER<Li, where TH<D),LOWER defines a score threshold relatively remote from a match with the spectrum of R.

3. The method of claim 1, wherein said group T comprises clusters of contiguous pixels, all having a score TH(D>,LOWER <Li<TH(D),UPPER, where TH(D>,LOWER defines a score threshold relatively remote from match with the spectrum of R, and where TH(D),LOWER<TH(D),UPPER.

4. The method of claim 3, wherein each cluster of contiguous pixels includes at least n contiguous pixels.

5. The method of claim 1, wherein each of the 1stand 2ndspectral similarity algorithms are SID-SAM.

6. The method of claim 4, wherein the selection of n depends on the signal-to-noise ratio of the spectral data.

7. The method of claim 6, wherein for a lower signal to noise ratio, a higher n is selected.

8. The method of claim 1, used for mapping of mineral deposits.

Images