System and method for predicting mineral concentrations

Abstract

The present invention relates to the practical application of machine learning techniques in exploring the Earth's subsurface and uncovering potential locations of mineral deposits. The present invention is directed to a method for predicting mineral concentrations in a land area, the method comprising: acquiring RF data elements representing RF scan returns from the land area in one or more polarizations from a radar mounted on a mobile platform; analyzing the RF data elements to generate a SAR data structure, the SAR data structure including one or more polarization layers each representing one or more polarizations, each polarization layer including a plurality of patch data elements representing a respective plurality of sub-regions of the scanned land area; and applying an ML model to the SAR data structure to predict mineral concentration bin values ​​in at least one sub-region of the scanned land area.

Description

[Technical Field]

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS This application claims the benefit of priority to U.S. Provisional Patent Application No. 63 / 394,186, filed August 1, 2022, entitled "SYSTEM AND METHOD OF PREDICTING MINERAL CONCENTRATION," which is incorporated herein by reference in its entirety as if fully set forth herein.

[0002] The present invention relates generally to geology, and more particularly to the practical application of machine learning and artificial intelligence techniques in exploring beneath the Earth's surface using synthetic aperture radar (SAR) to reveal potential locations of mineral deposits. [Background technology]

[0003] As is known, the problem of exploring the Earth's subsurface to find mineral deposits has always remained a research topic. Today, a huge number of approaches to the exploration of the Earth's subsurface have been developed. These include various types of drilling techniques, geological surveys, geochemical and geophysical exploration, and remote sensing.

[0004] Remote sensing is the process of detecting and monitoring the physical properties of a distant area (typically from a satellite, aircraft, or drone). In geological sciences, remote sensing is used as a complementary data acquisition method to support on-site observations, as it allows mapping of the geological properties of an area without physical contact with the area being explored. "Sensing" is performed through the detection of reflected electromagnetic radiation, which can be induced naturally (e.g., by the sun) or artificially (e.g., by radar mounted on a satellite or aircraft).

[0005] Synthetic Aperture Radar (SAR) is known to be one of the most advanced technologies applied to remote sensing. SAR utilizes the movement of the radar antenna over the target area and the Doppler effect to provide finer spatial resolution than traditional stationary beam scanning radar. The distance the SAR device moves over the target while the target scene is illuminated creates a large synthetic antenna aperture (antenna size), which provides very detailed analysis with a relatively small physical antenna.

[0006] While SAR technology could potentially provide detailed and scalable survey and exploration of mineral deposits at reasonable cost, in practice this has proven difficult. Even with detailed SAR analysis data of scanned land areas, the definition of the dependency between specific SAR data features and the presence or absence of specific mineral deposits, let alone the estimable concentrations of minerals, remains underdeveloped. As a result, the results of such analyses are insufficiently effective and reliable. Summary of the Invention [Problem to be solved by the invention]

[0007] Therefore, there is a need for a system and method for predicting mineral concentrations in a particular land area, which would increase the efficiency and reliability of applying SAR remote sensing techniques for mineral deposit exploration purposes. [Means for solving the problem]

[0008] In order to overcome the shortcomings of the prior art, the following invention is provided.

[0009] In a general aspect, the invention may be directed to a method of predicting mineral concentrations in a land area by at least one processor, the method including: acquiring, from a radar mounted on a mobile platform, radio frequency (RF) data elements representing reflections of an RF scan from the land area in one or more polarizations; analyzing the RF data elements to generate a synthetic aperture radar (SAR) data structure, the SAR data structure including one or more polarization layers each representing one or more polarizations, each polarization layer including a plurality of patch data elements representing a respective plurality of sub-regions of the scanned land area; and applying a machine learning (ML) model to the SAR data structure to predict a range value of mineral concentration in at least one sub-region of the scanned land area.

[0010] In another general aspect, the invention can be directed to a system for predicting mineral concentrations, the system comprising: a non-transitory memory device having modules of instruction code stored thereon; and at least one processor associated with the memory device and configured to execute the modules of instruction code, wherein upon execution of the modules of instruction code, the at least one processor is configured to: acquire, from a radar mounted on a mobile platform, radio frequency (RF) data elements representing reflections of an RF scan from a land area in one or more polarizations; analyze the RF data elements to generate a synthetic aperture radar (SAR) data structure, the SAR data structure including one or more polarization layers each representing the one or more polarizations, each polarization layer including a plurality of patch data elements representing a respective plurality of sub-areas of the scanned land area; and apply a machine learning (ML) model to the SAR data structure to predict values ​​of mineral concentration bins in the at least one sub-area of ​​the scanned land area.

[0011] In some embodiments, each polarization layer includes data representing at least one of: (a) the amplitude of the RF scan reflection; and (b) the phase of the RF scan reflection.

[0012] In some embodiments, the method further includes calculating one or more local incidence angle values ​​representing local incidence angles of RF reflections from a plurality of sub-regions based on the RF data elements, generating a local incidence angle map representing the calculated local incidence angle values, and further applying an ML model to the local incidence angle map to predict mineral concentration bin values.

[0013] In some embodiments, the method further includes calculating a digital elevation map representing the elevation of a sub-region of the scanned land area based on the RF data elements, and further applying an ML model to the digital elevation map to predict mineral concentration bin values.

[0014] In some embodiments, the method further includes receiving light spectrum data elements representing a representation of the scanned land area in at least one of an infrared (IR) band, a visible spectrum band, and an ultraviolet (UV) band, and further applying an ML model to the light spectrum data elements to predict mineral concentration bin values.

[0015] In some embodiments, the ML model includes at least one convolutional neural network (CNN) model including one or more input channels each configured to receive input selected from one or more polarization layers, a local incidence angle map, a digital elevation map, and an optical spectrum data element.

[0016] In some embodiments, the ML model further includes at least one binary classifier model adapted to receive the output of the CNN model for at least one sub-region of the scanned land area and to calculate the probability that a mineral concentration in the at least one sub-region is associated with a range of concentrations as defined by particular mineral concentration bins.

[0017] In some embodiments, the range of radio frequencies is selected from the list consisting of X-band, C-band, S-band, L-band, and P-band.

[0018] In some embodiments, the one or more polarizations are selected from (i) horizontal transmit-horizontal receive (HH) linear polarization, (ii) horizontal transmit-vertical receive (HV) linear polarization, (iii) vertical transmit-horizontal receive (VH) linear polarization, (iv) vertical transmit-vertical receive (VV) linear polarization, (v) right-hand transmit-right-hand receive (RR) circular polarization, (vi) right-hand transmit-left-hand receive (RL) circular polarization, (vii) left-hand transmit-right-hand receive (LR) circular polarization, and (viii) left-hand transmit-left-hand receive (LL) circular polarization.

[0019] In some embodiments, the method further includes receiving a training dataset, the training dataset including at least one training SAR data structure corresponding to the training land area and at least one annotation representing bin values ​​of mineral concentrations in the training land area, and training an ML model based on the training dataset to predict values ​​of mineral concentrations in the training land area.

[0020] In some embodiments, the training data set further includes a training local incidence angle map corresponding to the training land area.

[0021] In some embodiments, the training data set further includes a training digital elevation map corresponding to the training land area.

[0022] In some embodiments, the training data set further includes training light spectrum data elements that describe at least a portion of a training land area.

[0023] The subject matter which is regarded as the invention is particularly pointed out and distinctly claimed in the concluding portion of this specification. However, the invention, both as to organization and method of operation, together with its objects, features, and advantages, may best be understood by reference to the following detailed description when read in conjunction with the accompanying drawings. [Brief explanation of the drawings]

[0024] [Figure 1]FIG. 1 is a block diagram illustrating a computing device that may be included in a system for predicting mineral concentrations according to some embodiments. [Figure 2A] FIG. 1 is a block diagram illustrating a system for predicting mineral concentrations, according to some embodiments. [Figure 2B] FIG. 1 is a block diagram illustrating aspects of training an ML model of a system for predicting mineral concentrations, according to some embodiments. [Figure 3] 1 is a flow diagram illustrating a method for predicting mineral concentrations, according to some embodiments. DETAILED DESCRIPTION OF THE INVENTION

[0025] It will be appreciated that for simplicity and clarity of illustration, elements shown in the figures have not necessarily been drawn to scale. For example, the dimensions of some elements may be exaggerated relative to other elements for clarity. Further, where considered appropriate, reference numerals may be repeated among the figures to indicate corresponding or similar elements.

[0026] Those skilled in the art will understand that the present invention may be embodied in other specific forms without departing from its spirit or essential characteristics. Accordingly, the foregoing embodiments are to be considered in all respects as illustrative and not limiting of the invention described herein. The scope of the invention is, therefore, indicated by the appended claims, rather than the foregoing description, and all changes that come within the meaning and range of equivalency of the claims are therefore intended to be embraced within their scope.

[0027] In the following detailed description, numerous specific details are set forth to provide a thorough understanding of the present invention. However, it will be understood by those skilled in the art that the present invention 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 present invention. Some features or elements described with respect to one embodiment may be combined with features or elements described with respect to other embodiments. For clarity, discussion of the same or similar features or elements may not be repeated.

[0028] For example, although embodiments of the invention are not limited in this respect, discussions utilizing terms such as "processing," "computing," "calculating," "determining," "establishing," "analyzing," "checking," etc. may refer to operations and / or processes of a computer, computing platform, computing system, or other electronic computing device that manipulates and / or transforms data represented as physical (e.g., electronic) quantities in the computer's registers and / or memory into other data similarly represented as physical quantities in the computer's registers and / or memory, or other information non-transitory storage media capable of storing instructions for performing operations and / or processes.

[0029] Although embodiments of the invention are not limited in this respect, the terms "plurality" and "a plurality," as used herein, may include, for example, "multiple" or "two or more." The terms "plurality" or "a plurality" may be used throughout this specification to describe two or more components, devices, elements, units, parameters, etc. The term "set," as used herein, may include one or more items.

[0030] Unless explicitly stated, the method embodiments described herein are not constrained to a particular order or sequence. Furthermore, some of the described method embodiments or elements thereof may occur or be performed simultaneously, contemporaneously, or in parallel.

[0031] In the context of describing the claimed invention, the term "concentration bin" refers to a "range of concentration values," and therefore these terms may be used interchangeably.

[0032] In the context of describing the claimed invention, "predicting mineral concentrations in a land area" may refer to "predicting mineral concentrations in the top layer and below the surface of a land area."

[0033] In some embodiments of the present invention, the ML model may be an artificial neural network (ANN).

[0034] A neural network (NN) or artificial neural network (ANN), e.g., a neural network implementing machine learning (ML) or artificial intelligence (AI) functions, may refer to an information processing paradigm that may include nodes, called neurons, organized into layers with links between the neurons. The links may transmit signals between neurons and may be associated with weights. A NN may be configured or trained for a specific task, e.g., pattern recognition or classification. Training a NN for a specific task may involve adjusting these weights based on examples. Each neuron in an intermediate or final layer may receive an input signal, e.g., a weighted sum of output signals from other neurons, and may process the input signal using a linear or nonlinear function (e.g., an activation function). The results of the input and intermediate layers may be transferred to other neurons, and the results of the output layer may be provided as the output of the NN. Typically, neurons and links in a NN are represented by mathematical constructs, such as activation functions and matrices of data elements and weights. A processor, e.g., a CPU or a graphics processing unit (GPU), or a dedicated hardware device, may perform the relevant calculations.

[0035] It will be apparent to those skilled in the art that various ML models can be implemented without departing from the essence of the present invention. It should also be understood that in some embodiments, the ML model may be a single ML model or a set (ensemble) of ML models that collectively achieve the same functionality as a single one. Therefore, the above variations should be considered equivalent in light of the scope of the present invention.

[0036] In some respects, the following description of the claimed invention is presented in accordance with the task of locating lithium deposits (e.g., lithium carbonate, lithium oxide, etc.) or any other mineral, such as other metal deposits. Such specific embodiments are provided so that the description is illustrative enough and are not intended to limit the scope of protection claimed by the present invention. Those skilled in the art will appreciate that the implementation of the claimed invention in accordance with such a task is provided as a non-exclusive example, and that other actual implementations may be covered by the claimed invention.

[0037] As is known, artificial intelligence and machine learning techniques are highly useful for solving tasks where the connections and dependencies between input data and target output data are complex and uncertain, at least for humans to clearly define. Accordingly, as described in further detail herein, the proposed invention incorporates a combination of specific ML and SAR techniques. Such a combination is claimed to be effective in predicting mineral concentrations in specific land areas. Such a combination helps achieve an improved technical effect of increasing the efficiency and reliability of applying SAR remote sensing techniques for mineral deposit exploration purposes. The claimed technical effect has been further demonstrated during the actual implementation of the invention.

[0038] Reference is now made to FIG. 1, which is a block diagram illustrating a computing device that may be included within one embodiment of a system for predicting mineral concentrations, according to some embodiments.

[0039] Computing device 1 may include a processor or controller 2, which may be, for example, a central processing unit (CPU) processor, chip, or any suitable computing or calculation device, an operating system 3, a memory device 4, instruction code 5, a storage system 6, input devices 7, and output devices 8. Processor 2 (or one or more controllers or processors, possibly across multiple units or devices) may be configured to perform methods described herein and / or to execute or function as various modules, units, etc. More than one computing device 1 may be included in a system according to embodiments of the present invention, and one or more computing devices 1 may function as components thereof.

[0040] Operating system 3 may be or include any code segment (e.g., similar to instruction code 5 described herein) designed and / or configured to perform tasks including coordinating, scheduling, arbitrating, supervising, controlling, or otherwise managing the operation of computing device 1, such as scheduling the execution of software programs or tasks, or enabling software programs or other modules or units to communicate. Operating system 3 may be a commercial operating system. Note that operating system 3 may be an optional component, for example, in some embodiments, a system may include a computing device that does not require or include an operating system 3.

[0041] The memory device 4 may be or include, for example, random access memory (RAM), read-only memory (ROM), dynamic RAM (DRAM), synchronous DRAM (SD-RAM), double data rate (DDR) memory chips, flash memory, volatile memory, non-volatile memory, cache memory, buffer, short-term memory unit, long-term memory unit, or other suitable memory or storage unit. The memory device 4 may be or include multiple, possibly different, memory units. The memory device 4 may be a non-transitory computer or processor-readable medium or a non-transitory computer storage medium, such as RAM. In one embodiment, the non-transitory storage medium, such as the memory device 4, a hard disk drive, or another storage device, may store instructions or code that, when executed by a processor, cause the processor to perform the methods described herein.

[0042] The instruction code 5 may be any executable code, such as an application, program, process, task, or script. The instruction code 5 may be executed by the processor or controller 2, possibly under the control of the operating system 3. For example, the instruction code 5 may be an application that can predict mineral concentrations by applying an ML model to the SAR data structure, as described further herein. For clarity, a single instruction code 5 is shown in FIG. 1 , but systems according to some embodiments of the present invention may include multiple executable code segments or modules similar to the instruction code 5 that can be loaded into the memory device 4 and cause the processor 2 to perform the methods described herein.

[0043] Storage system 6 may be or include, for example, flash memory as known in the art, memory within or embedded in a microcontroller or chip as known in the art, a hard disk drive, a CD-recordable (CD-R) drive, a Blu-ray Disc (BD), a Universal Serial Bus (USB) device, or other suitable removable and / or fixed storage unit. Various types of input and output data may be stored in storage system 6 and loaded from storage system 6 to memory device 4, where the data may be processed by processor or controller 2. In some embodiments, some of the components shown in FIG. 1 may be omitted. For example, memory device 4 may be a non-volatile memory having the storage capacity of storage system 6. Thus, although shown as a separate component, storage system 6 may be incorporated into or included in memory device 4.

[0044] Cache memory 9 may be or may include, for example, a tier 1 (L1) cache module, a tier 2 (L2) cache module, and / or a tier 3 (L3) cache memory module, as known in the art. Cache memory 9 may include, for example, instruction cache memory space and / or data cache memory space and may be configured to cooperate with one or more processors (e.g., element 2) and / or one or more processing cores to execute at least one method according to embodiments of the present invention. Cache memory 9 may typically be implemented on the same die or chip as processor 2 and, therefore, may be characterized by a memory bandwidth that may be higher than the memory bandwidth of memory device 4 and storage system 6.

[0045] Input device(s) 7 may be or include any suitable input device, component, or system, such as a detachable keyboard or keypad, a mouse, etc. Output device(s) 8 may include one or more (possibly detachable) displays or monitors, speakers, and / or any other suitable output device. Any applicable input / output (I / O) devices may be connected to computing device 1 as indicated by blocks 7 and 8. For example, a wired or wireless network interface card (NIC), a universal serial bus (USB) device, or an external hard drive may be included in input device(s) 7 and / or output device(s) 8. It will be appreciated that any suitable number of input devices 7 and output devices 8 may be operably connected to computing device 1 as indicated by blocks 7 and 8.

[0046] Systems according to some embodiments of the present invention may include components such as, but not limited to, multiple central processing units (CPUs) or any other suitable general-purpose or specific processors or controllers (e.g., similar to element 2), multiple input units, multiple output units, multiple memory units, and multiple storage units.

[0047] Reference is now made to FIG. 2A, which illustrates a system 100 for predicting mineral concentrations, according to some embodiments.

[0048] According to some embodiments of the present invention, system 100 may be implemented as software modules, hardware modules, or any combination thereof. For example, system 100 may be or include a computing device, such as element 1 of FIG. 1. Additionally, system 100 may be adapted to execute one or more modules of instruction code (e.g., element 5 of FIG. 1) for requesting, receiving, analyzing, calculating, and generating various data to predict, via ML model 140, values ​​of mineral concentration bins in subregions of a scanned land area, as described in further detail herein.

[0049] As shown in Figures 2A and 2B, arrows may represent the flow of one or more data elements to and from system 100 and / or between modules or elements of system 100. For purposes of clarity, some arrows have been omitted from Figures 2A and 2B.

[0050] In some embodiments, system 100 may include a synthetic aperture radar (SAR) processing module 110. In some embodiments, SAR processing module 110 may be configured to acquire radio frequency (RF) data elements 200A representing reflections of RF scans from a land area in one or more polarizations from a radar mounted on a mobile platform. Alternatively, RF data elements 200A may be provided by a third party input data supplier (e.g., input data supplier 200), for example, via a network connection.

[0051] As is known, chemical compounds, including, but not limited to, lithium, such as lithium carbonate, lithium oxide, etc., may have rotational absorption spectra in the microwave band. Therefore, to reliably detect these materials and quantify their abundance, in some embodiments, the range of radio frequencies used for scanning is selected from the list consisting of X-band, C-band, S-band, L-band, and P-band.

[0052] In some embodiments, the SAR processing module 110 may be further configured to analyze each RF data element 200A to generate a SAR data structure 110A. To increase the reliability of the predictions further made by the ML model 140, the SAR processing module 110 may be configured to include one or more polarization layers 111A, each representing one or more polarizations, in the SAR data structure that is further used as an input to the ML model 140. The SAR processing module 110 may be configured to select one or more polarizations from: (i) horizontal transmit-horizontal receive (HH) linear polarization, (ii) horizontal transmit-vertical receive (HV) linear polarization, (iii) vertical transmit-horizontal receive (VH) linear polarization, (iv) vertical transmit-vertical receive (VV) linear polarization, (v) right-hand transmit-right-hand receive (RR) circular polarization, (vi) right-hand transmit-left-hand receive (RL) circular polarization, (vii) left-hand transmit-right-hand receive (LR) circular polarization, and (viii) left-hand transmit-left-hand receive (LL) circular polarization. The SAR processing module 110 may be configured to include in each polarization layer 111A a plurality of patch data elements representing a respective plurality of sub-regions of the scanned land area.

[0053] Because the reflected wave function of an RF scan is generally complex, SAR processing module 110 may be configured to convert the wave function into amplitude and phase difference components so that further application of ML model 140 can effectively provide reliable predictions. Accordingly, in some embodiments, SAR processing module 110 may be configured to include in each polarization layer 111A data representing at least one of the amplitude of the RF scan reflection and the phase of the RF scan reflection. In some embodiments, system 100 may be configured to use a polarization layer 111A containing data representing the amplitude of the RF scan reflection and a polarization layer 111A containing data representing the phase of the RF scan reflection as separate input channels of ML model 140.

[0054] Various types of data may prove useful for performing the task of locating mineral deposits. Accordingly, system 100 may be configured to perform data enrichment, i.e., system 100 may be configured to receive, calculate, and generate various types of data, either separately from SAR data structure 110A or correspondingly as additional data for each patch data element, for further use as input channels for ML model 140. ML model 140 may then be further configured to perform a data fusion (or concatenation) function that combines data from different input channels. Supplementing ML model 140 with such data can enhance the operation of system 100, as these data may contain deeply hidden features that may be highly relevant to the target output data, i.e., mineral concentration value prediction 100A. Furthermore, the fusion of several data sources and data collection methods using well-improved system 100 in general, and ML model 140 in particular, can increase the robustness of the system to shortcomings in any one of those data sources, data types, or collection methods. As a result, providing additional input channels has the technical effect of increasing the efficiency of operation of ML model 140 and the reliability of the predictions of ML model 140. Embodiments of system 100 including such enhancements are described in further detail herein.

[0055] In some embodiments, system 100 may include a local incidence angle (LIA) processing module 120. In the context of describing the claimed invention, the term "local incidence angle" refers to the angle between the normal to the ground at a particular location and the angle at which a satellite receives a reflected signal.

[0056] The LIA processing module 120 may be configured to obtain RF data elements 200A provided by the input data supplier 200, for example, via a network connection. The LIA processing module 120 may be further configured to calculate, based on the RF data elements 200A, one or more local angle-of-incidence values ​​representing local angles of incidence of RF reflections from a plurality of sub-regions. The LIA processing module 120 may be further configured to generate an LIA map 120A representing the calculated local angle-of-incidence values.

[0057] In some embodiments, system 100 may include an elevation processing module 130. Elevation processing module 130 may be configured to obtain RF data elements 200A provided by input data supplier 200, for example, via a network connection. Elevation processing module 130 may be configured to calculate a digital elevation map 130A representing the elevation of a small area of ​​the scanned land area based on the RF data elements 200A.

[0058] Additionally, supplemental data relating to different domains (e.g., optical spectral data, geological maps, etc.) can be used to improve the training of the ML model 140. Training aspects are further described below with reference to FIG. 2B.

[0059] Additionally or alternatively, in some embodiments, system 100 may be configured to receive, during the inference stage, optical spectrum data elements (not shown in FIG. 2A ) representing a depiction of the scanned land area in at least one of the infrared (IR) band, the visible spectrum band, and the ultraviolet (UV) band, which may be provided by a third-party data supplier.

[0060] The embodiments described herein provide a non-exclusive list of data types that may be used as input channels for ML model 100. For example, in some embodiments, system 100 may additionally or alternatively be configured to receive a digital geological map representing fault lines in the Earth and use the received geological map for the scanned land area as an additional input channel for ML model 100.

[0061] In some embodiments, the system 100 may be further configured to apply an ML model 140 to the SAR data structure 110A, the LIA map 120A, and the digital elevation map 130A (and optionally the optical spectrum data elements and the geological map data elements) to predict values ​​of mineral concentration bins in at least one sub-region of the scanned land area to generate concentration value predictions 100A.

[0062] In some embodiments, the ML model 140 may include a convolutional neural network (CNN) model (e.g., a CNN-based encoder 141). According to some embodiments of the present invention, the CNN model may be implemented as an encoder (e.g., the encoder 141). Therefore, it should be understood that the terms “CNN model,” “CNN-based model,” “CNN-based encoder,” and “encoder” may refer to the same element and may be used interchangeably herein. The encoder 141 may include one or more input channels, each configured to receive input selected from one or more of the polarization layer 111A, the LIA map 120A, and the digital elevation map 130A, and may additionally or alternatively perform concatenation of the data received from the different input channels. The encoder 141 may be further configured to perform multiple convolution operations on each of the input channels and generate a feature map (an efficient representation of the input data) as output data. The encoder 141 may include a convolution block using a batch normalization layer, which has been found to provide better accuracy than a max-pooling layer. Additionally, system 100 may be configured to concatenate the output data of encoder 141 with scalar features, e.g., orbital parameters such as whether the orbital direction is ascending or descending relative to the North Pole and whether the satellite's side-looking sensors were looking left or right during the scan. In some additional or alternative embodiments (not shown), multiple ML models 140 including multiple permutations of ML models 140 (e.g., with different hyperparameter settings, different types of architectures, etc.) may be used, and system 100 may use such scalar features, e.g., in a decision tree model, to select which encoder 141 to apply.

[0063] In some embodiments, ML model 140 may include a single multi-class classifier model or respective multiple binary classifier models, such as a multi-layer perceptron (MLP) classifier 142. MLP classifier 142 may be adapted to receive the output of encoder 141 for at least one sub-region of the scanned land area. MLP classifier 142 may be further configured to calculate a probability that a mineral concentration in the at least one sub-region is associated with a range of concentrations as defined by a particular mineral concentration bin (e.g., output concentration value prediction 100A).

[0064] In some embodiments, system 100 may be configured to predict average mineral concentrations (e.g., average lithium concentrations) for various quantities. For example, in some embodiments, ML model 140 includes k−1 MLP classifiers 142, where k is the number of ranges (bins), and in the provided embodiment, k=7. Each MLP classifier 142 may be configured to predict whether the average lithium concentration within at least one subregion (at a given pixel location) is higher than a particular preset value for the concentration bin. System 100 may be further configured to combine the outputs of the MLP classifiers 142 to arrive at a resulting probability for a given subregion with respect to each of the preset concentration bins. For each MLP classifier 142, a concentration value may be calculated as follows:

[0065] For the first bin, MLP classifier 142:

number

[0066] For the MLP classifier 142 in the middle bin:

number

[0067] For the MLP classifier 142 in the last bin:

number

[0068] In the formula, P i is the calculated probability that the concentration value V is related to a preset concentration bin of a particular MLP classifier 142, and Pr is the probability that the concentration value V is less than the preset target value for the particular concentration bin.

[0069] Furthermore, to achieve the resulting concentration value prediction 100A, the system 100 may be configured to apply the estimated likelihood provided by each of the MLP classifiers 142. For example, the system 100 may be configured to sort the outputs of the MLP classifiers 142 in descending order by estimated likelihood and use the output of the MLP classifier 142 with the highest estimated likelihood as the resulting concentration value prediction 100A.

[0070] In some embodiments, system 100 may be further configured to perform training of ML model 140. Aspects of training ML model 140 are further discussed with reference to FIG. 2B.

[0071] Reference is now made to FIG. 2B, which illustrates aspects of training the ML model 140 of the system 100 for predicting mineral concentrations, according to some embodiments.

[0072] In some embodiments, system 100 may be configured to receive all necessary training data (described in more detail below) from one or more training data suppliers (eg, training data supplier 300).

[0073] In some embodiments, the training data may include input data samples 310A, supplemental data 320A for penalizing the ML model 140 (provided in additional or alternative embodiments), and output labels 330A (corresponding to the input data samples 310A). The input data samples 310A may include training SAR data structure samples 311A, training LIA map samples 312A, and training digital elevation map samples 313A.

[0074] The supplemental data 320A may include optical spectrum data elements 321A and geological map data elements 322A. The geological map data elements 322A may represent a digital geological map of the Earth's fault lines, geological units, rock types, unit ages, and geomorphological features. The optical spectrum data elements 321A may represent a depiction of the scanned land area in the infrared (IR) band and at least one of the visible and ultraviolet (UV) bands.

[0075] Output label 330A may include ground truth concentration data 331A, which may include multiple annotations representing bin values ​​of mineral concentrations in the training land area. Ground truth concentration data 331A may be primarily created by measuring lithium concentrations in multiple boreholes at different depths, for example, ranging from 0.5 meters to 10 meters.

[0076] In some additional or alternative embodiments, the system 100 may include a training data preparation module (not shown), which may be configured to receive the ground truth concentration data 331A.

[0077] The training data preparation module may be further configured to request the SAR processing module 110, the LIA processing module 120, and the elevation processing module 130 to provide input training data samples 310A including SAR data structure samples 311A, training LIA map data samples 312A, and training digital elevation map data samples 313A, respectively, corresponding to a training land area (as shown in the ground truth concentration data 331A). The SAR processing module 110, the LIA processing module 120, and the elevation processing module 130 may be configured to request respective RF data elements 200A from the input data supplier 200 corresponding to the training land area. The SAR processing module 110, the LIA processing module 120, and the elevation processing module 130 may be further configured to calculate the training SAR data structure samples 311A, training LIA map data samples 312A, and training digital elevation map data samples 313A, respectively, based on the received RF data elements 200A.

[0078] In some embodiments, additionally or alternatively, the training data preparation module may be further configured to request optical spectrum data elements 321A and geological map data elements 322A corresponding to the training land area from the training data supplier 300, and receive the requested data elements 321A and 322A.

[0079] The training data preparation module may be further configured to form a training data set including input data samples 310A, supplemental data 320A for penalizing the ML model 140, and output labels 330A corresponding to a training land area.

[0080] In some embodiments, forming the training data set may include the following steps.

[0081] The training data preparation module may be configured to generate images of the training land area based on the ground truth density data 331A, with pixels sampled into polygons of equal density bins so that all images of the training land area are covered. The training data preparation module may be configured to perform a dilation operation around each pixel to form a plurality of square patch data elements representing respective subregions of the scanned land area. In some embodiments, the training data preparation module may be configured to select and set the size of the patch data elements.

[0082] The training data preparation module may be further configured to collate each patch data element from each polarization layer 111A of the training SAR data structure sample 311A, and additionally or alternatively each of the training optical spectrum data element 321A, training LIA map sample 312A, and training digital elevation map sample 313A, to form a training dataset with a corresponding channel of image data. Thus, in the prepared training dataset, each training sample is a patch data element containing multi-channel image data centered on a pixel to which ground truth density data 331A is attributed. The training data preparation module may be further configured to sample the training dataset using a stratified sampling technique to ensure equal representation of each class and thereby reduce overfitting. The training data preparation module may be further configured to split the training dataset into a set of training samples and a set of validation samples (e.g., in a 75 / 25% ratio favoring training samples). In some embodiments, the training data preparation module may be further configured to exclude some small regions where ground truth labels for both the training and validation sample sets are available in order to assess the generalizability of the model and its robustness to the distribution of input data.

[0083] In some embodiments, the system 100 may further include an ML training module 150.

[0084] In some embodiments, the ML training module 150 may be configured to train the ML model 140 based on the training dataset to predict mineral concentration values ​​in a training land area (e.g., concentration value prediction 100A as shown in FIG. 2A) using the included multi-channel image data.

[0085] Because some of the target minerals (those whose concentrations can be predicted by system 100, e.g., lithium) are rare earth elements (REEs), in most cases there may be very few examples of significant accumulation in known regolith. Therefore, creating a balanced dataset for supervised learning of the classification task (e.g., a dataset with equal or nearly equal numbers of labeled samples for each target class, e.g., a class corresponding to a particular concentration bin) may be impossible, making training a reliable MLP classifier nearly unattainable using traditional supervised learning methods. On the other hand, much more SAR image data (e.g., SAR data structure sample 311A) of a particular land area (or sub-area thereof) may be provided as input, ignoring the concentrations of the target minerals in that area.

[0086] The present invention provides the following solution to the above problem: It is proposed herein to perform a two-stage training procedure, where the first stage (pre-text training) may be aimed at training the encoder 141, and the second stage (downstream training) may be aimed at training the MLP classifier 142.

[0087] In some embodiments, ML training module 150 may include a pre-text training module 151 and a downstream training module 152 to perform the above two-stage procedure for training ML model 140, as described in more detail below.

[0088] The combined two-stage procedure may be referred to as self-supervised learning (SSL). The "pre-text" training is an unsupervised stage and does not require knowledge of mineral concentrations. Its main purpose is to enable the ML model 140 to "understand" the nature of the input data (which may be, for example, a concatenation of SAR data structure samples 311A, LIA map data samples 312A, and digital assessment map data samples 313A) and to construct an efficient representation 141A of the input data samples 310A, which may, for example, reduce dimensionality but focus on the most important input data features.

[0089] To do this, in some embodiments, the pre-text training module 151 may be configured to (i) concatenate the SAR data structure sample 311A, the LIA map data sample 312A, and the digital assessment map data sample 313A, thereby obtaining the input data sample 310A, and (ii) train an autoencoder 140′ model to reconstruct the input data sample 310A. The autoencoder 140′ may be of any architecture commonly known in the art, for example, it may include an encoder 141 block and a decoder 141′ block, and it may be trained using ML methods that may be known to those skilled in the art. In some embodiments, the encoder 141 may be implemented as a coding CNN-based model. The term “encoder” is intended to be understood in the broadest possible sense herein, and the present invention is not limited to that particular architecture.

[0090] After being fully trained, the autoencoder 140′ may be able to reliably reconstruct the input data samples 310A and thus compute an efficient representation 141A of the input data samples 310A in the “information bottleneck” segment of the autoencoder 140′ (the output of the encoder 141 may also be referred to herein as a “representation”). Once the training of the autoencoder 140′ is considered complete, the decoder 141′ portion may no longer be used in further training or in inferencing the ML model 140.

[0091] In the second training stage (downstream training), the ability to compute efficient representations 141A developed in the first training stage can be utilized for further supervised training of the ML model 140, in particular for training the MLP classifier 142.

[0092] In some embodiments, the downstream training module 152 may be configured to receive input data samples 310A and output labels 330A, which may be based on ground truth concentration data 331A. The downstream training module 152 may further be configured to perform training of the ML model 140, including (i) providing the input data samples 310A as input to the pre-trained encoder 141, (ii) computing, by the encoder 141, an efficient representation 141A of the input data samples 310A, and (iii) further training the MLP classifier to predict mineral concentration values ​​(e.g., concentration value prediction 100A as shown in FIG. 2A ) in the training land area based on the efficient representation 141A of the input data samples 310A and the corresponding ground truth concentration data 331A values ​​(e.g., each sample in the efficient representation 141A may correspond to a particular land area (or a subregion thereof) and be labeled with the corresponding ground truth concentration data 331A value measured in that area). The "downstream" training stage may involve supervised ML methods commonly known in the art.

[0093] Thus, in some embodiments, the ML model 140 may be configured such that the encoder 141 receives input data (e.g., training input data samples 310A (when referred to as the training phase) or inference input data samples including one or more polarization layers 111A, LIA map 120A, and digital elevation map 130A (when referred to as the inference phase)) in a multi-channel or concatenated single-channel format, computes an efficient representation 141A of the input data, and forwards it downstream to the MLP classifier 142, which can receive the representation 141A as input and compute a density value prediction 100A as output.

[0094] Therefore, the proposed configuration of the ML model 140 and the illustrated two-stage procedure for its training may contribute to improving upon the prior art by mitigating the problems caused by the lack of substantial training data, thereby increasing the efficiency of the training process and, consequently, the reliability of mineral concentration value (or bin) predictions.

[0095] Furthermore, in some additional or alternative embodiments, supplemental data 320A analysis may be included to further improve the training of ML model 140, particularly by being used as a basis for adjusting the penalty of ML model 140 for failing to provide accurate concentration predictions.

[0096] As is known, the presence of certain mineral accumulations (e.g., lithium) can be explained as follows: Lithium originates from volcanic activity, where lithium-rich magma passes through the Earth's crust through fissures. Once above the Earth's surface, this lava is deposited as volcanic rocks such as pegmatites. The rocks undergo weathering and erosion, and their components are transported to accumulation ponds, such as salt lakes (where, for example, lithium accumulates in salt water) or muddy clay deposits.

[0097] Thus, the following distinct stages can be detected: volcanic deposition, transport, and accumulation. Each of the stages has an associated temporal and spatial scale, which can be used to improve the training process.

[0098] For example, in some embodiments, training of ML model 140 may target mineral concentration value predictions (e.g., subsurface lithium concentrations in areas of known accumulation) based on the highest currently known spatial resolution of RF data elements 200A (e.g., on the order of L-band SAR image pixels (1-10 meters)). Furthermore, when analyzing characteristic lengths of normal faults in geological maps that are a result of the spatial resolution of the underlying geophysical measurements (1-10 kilometers, essentially the lowest spatial resolution), areas of potential volcanic deposits can be detected. Furthermore, when analyzing the topography of watershed / drainage systems (at 10-meter to 1-kilometer resolution), routes of lithium transport from volcanic deposits to areas of accumulation can be detected and evaluated.

[0099] Additionally or alternatively, downstream training module 152 may be configured to receive supplemental data 320A for penalizing ML model 140. Downstream training module 152 may be further configured to: (i) for each particular input data sample 310A, select a corresponding supplemental data 320A sample that represents the same training land area (or a portion / sub-area thereof) at a required resolution (as described above), (ii) analyze the optical spectrum data elements 321A and the geological map data elements 322A to determine areas of potential volcanic deposits, (iii) analyze the geological map data elements, and additionally or alternatively, the digital assessment map data elements of sample 313A, to detect potential lithium transport pathways from volcanic deposits, and (iv) determine areas of potential lithium accumulation in said training land area (or a portion / sub-area thereof). The downstream training module 152 may be further configured to perform training of the ML model 140 further based on (i) the detected areas of potential volcanic deposits, (ii) the detected pathways of potential lithium transport, and (iii) the determined areas of potential lithium accumulation. In particular, the downstream training module 152 may be configured to calculate a loss function (e.g., a function of the difference between the estimated values ​​for lithium concentrations (calculated by the ML model 140 during training) and the true values ​​(e.g., the respective ground truth concentration data 331A)) and adjust the value of the loss function further based on at least one of (i) the detected areas of potential volcanic deposits, (ii) the detected pathways of potential lithium transport, and (iii) the determined areas of potential lithium accumulation. For example, the downstream training module 152 may penalize the ML model 140 more strongly if it erroneously predicts relatively low lithium concentration values ​​in the area of ​​the settling pond than if it did not.

[0100] In some embodiments, to detect areas of potential volcanic deposits, downstream training module 152 may be further configured to use data from geological map data elements 322A, such as each unit's primary and secondary rock type, unit age range, and proximity to normal faults, their length, age, and orientation (e.g., "strike and dip"). Where geological maps are not available, the required information can also be derived from geophysical measurements, including seismic reflectivity measurements, magnetic measurements, and gravimetry. To detect potential lithium transport pathways through drainage systems, downstream training module 152 may be further configured to calculate derivative data from digital elevation model (DEM) samples 322A, including normalized elevation, slope gradient, aspect, and curvature.

[0101] Applying such supplemental data 320A can therefore help model 140 effectively focus on areas that may be considered to exhibit high regolithium concentrations, consistent with local transport pathways and even volcanic deposit origins. Such refinement of supplemental data 320A can thereby further improve the training process and increase the reliability of mineral concentration value predictions (e.g., concentration value predictions 100A).

[0102] Additionally or alternatively, to penalize errors superlinearly in magnitude, in some embodiments, ML training module 150 may be configured to use cross-entropy with a focal loss having a gamma parameter of 2 as a loss function while training ML model 140. To reduce problems caused by an imbalanced training dataset, ML training module 150 may be configured to set the alpha parameter of MLP classifier 142 to be equal to the inverse of the relative abundance of each class (corresponding to the desired output of MLP classifier 142).

[0103] In some embodiments, the ML training module 150 may be configured to utilize an ADAM optimizer, and / or early stopping, and / or linear learning rate scheduling techniques while training the ML model 140.

[0104] In some embodiments, during the training phase, system 100 may include multiple permutations of ML models 140. In some embodiments, ML training module 150 may be configured to perform hyperparameter tuning of the detection threshold of each MLP classifier 142 of the multiple ML models 140 to obtain optimal thresholds in terms of recall and accuracy for each MLP classifier 142. ML training module 150 may be further configured to perform neural architecture search (NAS) to determine the architecture of ML model 140 or its components (e.g., encoder 141 and MLP classifier 142) that has the highest predictive performance, the lowest memory consumption, the lowest model size or inference time (i.e., the time required to obtain a prediction), etc. ML training module 150 may be further configured to select a target ML model 140 from the multiple permutations of ML models 140 based on the results of the NAS procedure.

[0105] Referring now to FIG. 3, a flow diagram illustrating a method for predicting mineral concentrations by at least one processor is shown, according to some embodiments.

[0106] As shown in step S1005, at least one processor (e.g., processor 2 of FIG. 1) may perform acquiring RF data elements 200A representing RF scan returns from a land area in one or more polarizations from a radar mounted on a mobile platform. Step S1005 may be performed by SAR processing module 110, LIA processing module 120, and elevation processing module 130 (as described with reference to FIG. 2A).

[0107] As shown in step S1010, at least one processor (e.g., processor 2 of FIG. 1) may perform the analysis of RF data elements 200A to generate SAR data structure 110A, where SAR data structure 110A includes one or more polarization layers 111A each representing one or more polarizations, and each polarization layer 111A includes multiple patch data elements representing respective multiple sub-regions of the scanned land area. Step S1010 may be performed by SAR processing module 110 (as described with reference to FIG. 2A).

[0108] As shown in step S1015, at least one processor (e.g., processor 2 of FIG. 1) may apply the ML model 140 to the SAR data structure 110A to predict mineral concentration bin values ​​(concentration value predictions 100A) in at least one subregion of the scanned land area. Step S1015 may be performed by the encoder 141 and the MLP classifier 142 (as described with reference to FIG. 2A).

[0109] As can be seen from the description provided, the claimed invention represents a system and method for predicting mineral concentrations in a particular land area. The claimed invention increases the efficiency and reliability of applying SAR remote sensing techniques for mineral deposit exploration purposes.

[0110] Unless explicitly stated, the method embodiments described herein are not constrained to a particular order or sequence. Furthermore, all formulas described herein are intended as examples only, and other or different formulas may be used. Furthermore, some of the described method embodiments or elements thereof may occur or be performed at the same time.

[0111] While certain features of the invention have been illustrated and described herein, many modifications, substitutions, changes, and equivalents will occur to those skilled in the art. It is, therefore, to be understood that the appended claims are intended to cover all such modifications and changes as fall within the true spirit of the invention.

[0112] Various embodiments are presented, each of which may, of course, include features from the other embodiments presented, and embodiments not specifically described may include various features described herein.

Claims

1. 1. A method for predicting mineral concentrations in a land area by at least one processor, comprising: acquiring radio frequency (RF) data elements representing reflections of an RF scan from said land area in one or more polarizations from a radar mounted on a mobile platform; parsing the RF data elements to generate a synthetic aperture radar (SAR) data structure, the SAR data structure including one or more polarization layers each representing the one or more polarizations, each polarization layer including a plurality of patch data elements representing a respective plurality of sub-regions of the scanned land area; applying a machine learning (ML) model to the SAR data structure to predict mineral concentration bin values ​​in at least one subregion of the scanned land area; A method comprising:

2. 2. The method of claim 1, wherein each polarization layer includes data representing at least one of: (a) the amplitude of the RF scan reflection; and (b) the phase of the RF scan reflection.

3. calculating one or more local angle of incidence values ​​representing local angles of incidence of the RF reflections from the plurality of sub-regions based on the RF data elements; generating a local incidence angle map representing the calculated local incidence angle values; further applying the ML model to the local incidence angle map to predict the mineral concentration bin values; 3. The method of claim 1, further comprising:

4. calculating a digital elevation map representing the elevation of the scanned land area sub-region based on the RF data elements; further applying the ML model to the digital elevation map to predict the mineral concentration bin values; The method of any one of claims 1 to 3, further comprising:

5. receiving optical spectrum data elements representing a representation of the scanned land area in at least one of an infrared (IR) band, a visible spectrum band, and an ultraviolet (UV) band; further applying the ML model to the optical spectrum data elements to predict the mineral concentration bin values; The method of claim 1 , further comprising:

6. 6. The method of claim 1, wherein the ML model comprises at least one convolutional neural network (CNN) model including one or more input channels each configured to receive input selected from the one or more polarization layers, the local incidence angle map, the digital elevation map, and the optical spectrum data elements.

7. The ML model is receiving an output of the CNN model for at least one subregion of the scanned land area; calculating a probability that the mineral concentration in the at least one sub-region is associated with a range of concentrations as defined by a particular mineral concentration bin; The method of claim 1 , further comprising at least one binary classifier model adapted to:

8. 8. The method of claim 1, wherein the range of radio frequencies is selected from the list consisting of X-band, C-band, S-band, L-band, and P-band.

9. 9. The method of claim 1, wherein the one or more polarizations are selected from: (i) horizontal transmit-horizontal receive (HH) linear polarization; (ii) horizontal transmit-vertical receive (HV) linear polarization; (iii) vertical transmit-horizontal receive (VH) linear polarization; (iv) vertical transmit-vertical receive (VV) linear polarization; (v) right-hand transmit-right-hand receive (RR) circular polarization; (vi) right-hand transmit-left-hand receive (RL) circular polarization; (vii) left-hand transmit-right-hand receive (LR) circular polarization; (viii) left-hand transmit-left-hand receive (LL) circular polarization.

10. receiving a training data set; at least one training SAR data structure corresponding to a training land area; and a plurality of annotations representing binned values ​​of mineral concentrations in the training land area. receiving, training the ML model based on the training dataset to predict mineral concentration values ​​in the training land area; 10. The method of claim 1, further comprising:

11. The method of claim 1 , wherein the training data set further comprises a training local incidence angle map corresponding to the training land area.

12. The method of claim 1 , wherein the training data set further comprises a training digital elevation map corresponding to the training land area.

13. 13. The method of claim 1, wherein the training data set further comprises training optical spectrum data elements that describe at least a portion of the training land area.

14. 1. A system for predicting mineral concentrations, comprising: a non-transitory memory device having a module of instruction code stored therein; and at least one processor associated with the memory device and configured to execute the module of instruction code, wherein upon execution of the module of instruction code, the at least one processor: acquiring radio frequency (RF) data elements representing reflections of an RF scan from a land area in one or more polarizations from a radar mounted on a mobile platform; analyzing the RF data elements to generate a synthetic aperture radar (SAR) data structure, the SAR data structure including one or more polarization layers each representing the one or more polarizations, each polarization layer including a plurality of patch data elements representing a respective plurality of sub-regions of the scanned land area; applying a machine learning (ML) model to the SAR data structure to predict mineral concentration bin values ​​in at least one subregion of the scanned land area; A system configured to:

Images