Mineral alteration feature extraction method based on hyperspectrum

Abstract

The invention relates to the technical field of ground feature component analysis, and discloses a hyperspectrum-based mineral alteration feature extraction method, which comprises the following steps: acquiring hyperspectral remote sensing image data of a to-be-processed region, and carrying out atmospheric correction and noise reduction preprocessing; extracting an initial end member spectrum after preprocessing; constructing an objective function of nonlinear spectral unmixing, wherein the objective function comprises a data fidelity term; an abundance constraint term; punishing a physical consistency regularization term of a deviation between a spectrum generated by the physical model and a current end member spectrum based on a differentiable radiation transfer model; solving the objective function to obtain a final end member spectrum and an abundance matrix corresponding to the end member spectrum; and identifying altered minerals by utilizing the final end member spectrum, and extracting mineral alteration characteristics of the to-be-processed region in combination with an abundance matrix corresponding to the end member spectrum. According to the method, the space-spectrum composite kernel function is introduced, the nonlinear unmixing problem is solved in the high-dimensional feature space, and the complex nonlinear mixing effect existing in the surface environment can be known.

Description

Technical Field

[0001] The present invention relates to the technical field of ground feature composition analysis, and in particular to a method for extracting mineral alteration features based on hyperspectral analysis. Background Art

[0002] Hyperspectral remote sensing technology offers the unique ability to combine image and spectrum in mineral exploration and geological mapping. By analyzing the spectral curves reflected by ground objects, surface mineral composition can be identified, thereby delineating favorable areas for prospecting. However, due to the limited spatial resolution of remote sensing sensors, a single pixel often contains multiple ground object components, resulting in a mixed pixel, which seriously affects the accuracy of mineral identification.

[0003] Spectral unmixing technology is key to resolving the problem of mixed pixels. Its goal is to decompose mixed pixels into a series of spectra of pure components (endmembers) and their proportions (abundances) within the pixel. Unmixing methods are often based on the linear spectral mixture model (LSMM), which assumes that a pixel spectrum is a linearly weighted sum of the endmember spectra. However, in complex geological environments, due to factors such as topographical fluctuations, multiple scattering between mineral particles, and the close mixing of minerals at the microscale, the relationship between pixels and endmembers often exhibits nonlinearity. This leads to significant errors in the unmixing results of linear models, making it difficult to meet the requirements for high-precision alteration information extraction.

[0004] Kernel-based and bilinear-model-based methods improve unmixing performance to a certain extent. However, most methods treat unmixing as a purely mathematical inversion problem, ignoring the specific combination patterns of minerals in nature. For example, some minerals often coexist, while others are mutually exclusive, and fail to effectively incorporate this geological prior knowledge into the model. Furthermore, existing methods rarely consider the physical transmission process of spectral signals in the imaging link, resulting in the unmixed endmember spectra potentially lacking clear physical meaning. The spectra of the same mineral can vary due to slight changes in lighting conditions, atmospheric conditions, and its own physical and chemical properties, resulting in endmember variability. Most current models are unable to model this variability and often simplify the mineral's spectrum into a fixed, unchanging spectrum, which is inconsistent with reality. Therefore, how to construct an unmixing framework that can simultaneously consider nonlinear effects and integrate geology is a technical challenge that needs to be urgently addressed in the field of high-precision mineral mapping. Summary of the Invention

[0005] The present invention provides a hyperspectral-based mineral alteration feature extraction method to solve the problem in the above-mentioned prior art that it is difficult to construct an unmixing framework that can simultaneously consider nonlinear effects and fusion geology and can effectively deal with end-member spectral variability.

[0006] The hyperspectral-based mineral alteration feature extraction method of the present invention comprises the following steps: Obtain hyperspectral remote sensing image data of the area to be processed, perform atmospheric correction and noise reduction preprocessing on the hyperspectral remote sensing image data; use endmember extraction algorithm to extract initial endmember spectra from the preprocessed hyperspectral remote sensing image data; Constructing an objective function for nonlinear spectral unmixing, the objective function includes: a data fidelity term that maps pixels to a high-dimensional feature space using a spatial-spectral composite kernel function; an abundance constraint term that includes a group sparsity regularization based on a mineral paragenesis prior and a non-negative abundance, a sum of one, and a mineral paragenesis prior; and a physical consistency regularization term that penalizes deviations between a spectrum generated by a physical model and a current endmember spectrum based on a differentiable radiative transfer model; An alternating optimization algorithm is used to iteratively solve the objective function, and the endmember abundance matrix and the endmember spectrum are alternately updated during the iteration, wherein the endmember spectrum is updated according to a manifold learning strategy, and the spectrum is constrained to be optimized on a preset low-dimensional manifold; When the iteration converges or reaches the preset termination condition, the final end-member spectrum and the abundance matrix corresponding to the end-member spectrum are obtained; the altered minerals are identified using the final end-member spectrum, and the mineral alteration characteristics of the area to be processed are extracted in combination with the abundance matrix corresponding to the end-member spectrum.

[0007] Preferably, the spatial-spectral composite kernel function is composed of a weighted combination of a Gaussian radial basis spatial kernel function and a polynomial spectral kernel function.

[0008] Preferably, the abundance constraint term of the group sparse regularization based on the mineral symbiosis prior includes: According to the prior knowledge of mineral paragenesis, the end-members to be extracted with similar geological genesis or spatial association relationship are divided into different groups; the L2 norm is applied to the end-member abundance vector in each group, and the L2 norm results of all groups are summed to obtain the L2,1 mixed norm, which is used as the group sparse regularization term.

[0009] Preferably, the physical consistency regularization term based on the differentiable radiation transfer model that penalizes the deviation between the spectrum generated by the physical model and the current end-member spectrum includes: using the differentiable radiation transfer model, inputting the mineral physical parameters corresponding to the end-member spectrum of the current iteration into the differentiable radiation transfer model to generate a physical model spectrum; calculating the difference between the physical model spectrum and the current end-member spectrum as the physical consistency regularization term.

[0010] Preferably, the alternating optimization algorithm is used to iteratively solve the objective function, including: in each iteration, fixing the end member spectrum, using the multiplicative update rule to solve the abundance matrix to ensure the non-negativity of the abundance update, and performing a projection operation on the abundance vector of each pixel after each update so that the abundance vector satisfies the constraint that the sum is one.

[0011] Preferably, the update of the end-member spectrum is based on a manifold learning strategy, and the spectrum is constrained to be optimized on a preset low-dimensional manifold, including: using reference spectra selected from a standard mineral spectrum library and the initial end-member spectrum to jointly construct a training set, and using a manifold learning algorithm to learn the low-dimensional manifold structure of the training set data; when iteratively updating the end-member spectrum, the gradient of the objective function with respect to the end-member spectrum is projected onto the tangent space of the low-dimensional manifold, and updated along the projected gradient direction.

[0012] Preferably, the identifying of altered minerals using the final end-member spectrum includes: A spectral similarity measurement method is used to calculate the similarity between the final end-member spectrum and the reference spectrum in the standard mineral spectrum library; for each end-member spectrum, the spectrum is identified as the mineral with the highest similarity in the reference spectrum and above a preset similarity threshold.

[0013] Preferably, the method of obtaining hyperspectral remote sensing image data of the area to be processed and performing atmospheric correction and noise reduction preprocessing on the hyperspectral remote sensing image data includes: collecting hyperspectral image data of the mining area through the infrared imaging spectrometer AVIRIS, using the FLAASH model to perform atmospheric correction to obtain surface reflectance data, and using the low-rank tensor approximation LRTA algorithm to perform noise reduction on the reflectance data.

[0014] Preferably, the endmember extraction algorithm is a vertex component analysis algorithm VCA.

[0015] Preferably, the alternating optimization algorithm is the alternating direction method of multipliers ADMM.

[0016] The beneficial effects of the present invention are as follows: by introducing a spatial-spectral composite kernel function, the present invention places the nonlinear unmixing problem in a high-dimensional feature space for solution, and is able to understand the complex nonlinear mixing effects that are prevalent in the actual surface environment. By using group sparse regularization based on the law of mineral symbiosis as an abundance constraint and integrating geological prior knowledge into the unmixing model, the abundance inversion results are more consistent with geological reality, reducing physically unreasonable mineral combinations and improving the reliability of the results. In addition, by constructing a physical consistency regularization term based on a differentiable radiation transfer model and adopting a manifold learning strategy to update the endmember spectrum, on the one hand, it is ensured that the extracted endmember spectrum has a clear physical meaning, and on the other hand, it solves the endmember variability problem caused by changes in light, atmosphere and the mineral's own properties, making the endmember extraction closer to the actual situation. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 A schematic flow chart of a method for extracting mineral alteration features based on hyperspectral data provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0018] The embodiments of the present invention are described in detail below, and examples of the embodiments are shown in the accompanying drawings. The embodiments described below with reference to the accompanying drawings are exemplary and intended to be used to explain the present invention, but should not be understood as limiting the present invention.

[0019] like Figure 1 As shown, the hyperspectral-based mineral alteration feature extraction method provided by the embodiment of the present invention specifically includes the following steps: S1, obtains the hyperspectral remote sensing image data of the area to be processed, performs atmospheric correction and noise reduction preprocessing on the hyperspectral remote sensing image data; uses the endmember extraction algorithm to extract the initial endmember spectrum from the preprocessed hyperspectral remote sensing image data.

[0020] Hyperspectral imagery data from the mining area was collected using the AVIRIS airborne visible / infrared imaging spectrometer. Atmospheric correction was performed using the FLAASH model to eliminate the effects of atmospheric scattering and absorption on the surface spectra, resulting in surface reflectance data. The reflectance data were denoised using the low-rank tensor approximation (LRTA) algorithm to remove sensor noise and bad lines while maintaining the spatial and spectral structure of the hyperspectral imagery.

[0021] The HFC virtual dimension method is used to estimate the number of endmembers P in the hyperspectral image data. The vertex component analysis algorithm VCA is used to extract P initial endmember spectra from the preprocessed hyperspectral image data, and these initial endmember spectra constitute the initial endmember spectrum matrix.

[0022] S2, constructs the objective function of nonlinear spectral unmixing, which includes: a data fidelity term that maps pixels to a high-dimensional feature space using a spatial-spectral composite kernel function; an abundance constraint term that includes a non-negative abundance, a sum of one, and a group sparse regularization based on mineral paragenesis priors; and a physical consistency regularization term that penalizes the deviation between the spectrum generated by the physical model and the current end-member spectrum based on a differentiable radiation transfer model.

[0023] The data fidelity term uses a composite kernel function that is the product of a spatial Gaussian kernel and a spectral radial basis function kernel. This kernel is used to measure the error between the reconstructed and original pixels in the high-dimensional feature space. Within the abundance constraint, the group sparsity regularization term divides geologically coexisting minerals into groups. For example, kaolinite, alunite, and montmorillonite are classified into the clay mineral group, and hematite and goethite into the iron oxide group. By applying an L2,1 norm penalty to the abundance vector within each group, the abundance of a group of minerals is simultaneously selected or suppressed. The physical consistency regularization term is based on the differentiable PROSAIL model. The regularization term calculates the Euclidean distance between the spectrum generated by the PROSAIL model based on the physical parameters of the current endmember and the endmember spectrum estimated by the algorithm, using this as a penalty term. Abundance constraints include non-negative abundance, a sum of unity, and group sparsity regularization based on mineral coexistence priors. The term "abundance is non-negative" means that the proportion of any mineral in a pixel must be greater than or equal to zero; the term "sum is one" means that in the same pixel, the sum of the abundance ratios of all considered minerals and any other background materials must equal 1; the group sparsity regularization based on the mineral paragenesis prior is a constraint determined based on the mineral paragenesis. For example, in chalcanthite-induced alteration, chlorite and epidote often appear together. In sericite-induced alteration, quartz, sericite, and pyrite are a common combination; it is rare to find epidote and sericite occurring in large quantities in the same small rock sample.

[0024] In an optional embodiment, the spatial-spectral composite kernel function is formed by a weighted combination of a Gaussian radial basis spatial kernel function and a polynomial spectral kernel function.

[0025] The spatial-spectral composite kernel function is designed to simultaneously exploit the spatial distribution and spectral characteristics of pixels in hyperspectral imagery. The Gaussian radial basis kernel function measures the spatial correlation of pixels by calculating the exponential decay of the spatial distance between them, assuming that geographically proximal pixels have a higher probability of similarity. For example, within a 500-by-500-pixel image region, the spatial kernel function values ​​for pixels within a 3-by-3 neighborhood of the central pixel will be significantly higher than those for pixels 100 meters away from the central pixel. The kernel function's bandwidth parameter, such as sigma, can be set to 3.0 to control the spatial range of influence.

[0026] The polynomial spectral kernel function is used to capture the complex nonlinear relationship between different spectral bands, which is crucial for distinguishing minerals with subtle spectral characteristics but different origins. For example, by setting the polynomial degree to 3, high-order interaction modeling can be performed on the spectral vectors of two pixels in 224 bands, effectively amplifying their slight differences in the depth of specific absorption valleys. Through a weighting coefficient such as 0.4, the results of the Gaussian spatial kernel and the polynomial spectral kernel are linearly combined to form a comprehensive similarity measure, so that the unmixing model can not only identify spectrally similar substances, but also take into account their spatial aggregation patterns, thereby improving the accuracy and continuity of mineral mapping.

[0027] In an optional embodiment, the abundance constraint term of the group sparse regularization based on the mineral symbiosis prior includes: dividing the end members to be extracted with similar geological genesis or spatial association relationships into different groups according to the prior knowledge of the mineral symbiosis combination; applying the L2 norm to the end member abundance vector in each group, summing the L2 norm results of all groups to obtain the L2,1 mixed norm, and the L2,1 mixed norm is used as the group sparse regularization term.

[0028] Integrating geological knowledge into mathematical models guides the determination of mineral abundance. When exploring a typical porphyry copper deposit, according to mineralization theory, sericite, pyrite, and kaolinite often occur in the core of the alteration zone, forming an alteration assemblage, while chlorite and epidote often occur in the periphery. Based on this, of the 12 end-members to be extracted, the three end-members representing sericite, pyrite, and kaolinite can be grouped into Group 1, the two end-members representing chlorite and epidote into Group 2, and the remaining end-members into their own groups.

[0029] The application of the L2,1 mixing norm achieves group-level sparsity. For any pixel in the image, the regularization term encourages the abundance coefficients of the entire group to be simultaneously zero or non-zero. For example, for a pixel located in the core of an alteration, the model will tend to assign a non-zero abundance value to each of the three end members in group 1 during the iterative solution, such as 0.3, 0.2, and 0.15, while also causing the abundance values ​​of the two end members in group 2 to approach zero. This mechanism avoids the unreasonable splitting of mineral assemblages that should logically coexist, making the unmixing results more consistent with geological laws and effectively suppressing the emergence of isolated outliers caused by noise.

[0030] In an optional embodiment, a physical consistency regularization term is used to penalize the deviation between the spectrum generated by the physical model and the current end-member spectrum based on a differentiable radiation transfer model, including: using the differentiable radiation transfer model, inputting the mineral physical parameters corresponding to the end-member spectrum of the current iteration into the differentiable radiation transfer model to generate a physical model spectrum; and calculating the difference between the physical model spectrum and the current end-member spectrum as the physical consistency regularization term.

[0031] This regularization term ensures that the endmember spectra extracted by the algorithm are not only mathematically optimal but also physically plausible. During the 50th iteration of the optimization, the algorithm generates an endmember spectrum that resembles goethite. To verify the physical authenticity, a differentiable Hapke radiative transfer model is employed. Typical physical parameters of goethite, such as an average grain size of 80 microns and single-scattering albedo data measured under laboratory conditions, are input into the Hapke radiative transfer model.

[0032] The Hapke radiative transfer model simulates the spectral curve that goethite with these physical properties should exhibit under the current imaging conditions—this is known as the physical model spectrum. The root mean square error (RMSE) between the physical model spectrum and the algorithm-generated endmember spectra is calculated. Assuming the calculated error is 0.08, this value is added as a penalty term to the overall objective function. This penalty term drives the next iteration, adjusting the endmember spectra to bring them closer to the output of the physical model. This penalty term corrects spectral shapes that are inconsistent with the physical laws of light-matter interaction due to data noise or model limitations, such as unrealistic absorption depths or peak positions.

[0033] S3, using an alternating optimization algorithm to iteratively solve the objective function, and alternately updating the endmember abundance matrix and the endmember spectrum during the iteration, wherein the update of the endmember spectrum is based on a manifold learning strategy, and the spectrum is constrained to be optimized on a preset low-dimensional manifold.

[0034] An iterative solution using the alternating direction multiplier method (ADMM) is employed. When the endmember spectra are fixed, the abundance matrix is ​​updated by solving a quadratic programming problem with group sparsity regularization. When the abundance matrix is ​​fixed, the Laplace eigenmap algorithm is used to learn the spectra of minerals with the same name from a standard mineral spectral library. This constructs a low-dimensional manifold embedding space, onto which the endmember spectrum update problem is projected. The optimal solution is found using gradient descent, and the endmember spectra are then mapped back to the original spectral space.

[0035] In an optional embodiment, an alternating optimization algorithm is used to iteratively solve the objective function, including: in each iteration, fixing the end member spectrum, using a multiplicative update rule to solve the abundance matrix to ensure the non-negativity of the abundance update, and performing a projection operation on the abundance vector of each pixel after each update so that the abundance vector satisfies the constraint that the sum is one.

[0036] Because the objective function is nonconvex with respect to both endmembers and abundances, an alternating optimization strategy is employed to simplify the solution process. Within each iteration, the endmember spectra obtained from the current iteration are fixed, and only the abundance matrix is ​​optimized. For example, in the kth iteration, a multiplicative update formula is used to calculate the abundance ratios of the five minerals in each pixel using the five known endmember spectra. This multiplicative update formula ensures that the calculated abundance values, such as 0.2, 0.5, 0.1, 0.3, and 0.05, are always greater than or equal to zero, consistent with the physical requirement that abundance cannot be negative.

[0037] After the multiplicative update, while the abundance vector for a pixel is guaranteed to be non-negative, its sum may not be unity. For example, in the example above, the sum is 1.15. This violates the constraint that the sum of all component abundances must equal unity. Therefore, a projection operation is performed on this abundance vector. The projection operation scales the abundance vector [0.2, 0.5, 0.1, 0.3, 0.05] to the nearest point on the plane where the abundance vector sums to unity, resulting in a new abundance vector, such as [0.17, 0.43, 0.09, 0.26, 0.04], where the sum of all components equals 1. After completing the process of fixing the endmembers and updating the abundances for all pixels in the image, the next step is to fix the abundances and update the endmembers.

[0038] In an optional embodiment, the endmember spectrum is updated based on a manifold learning strategy, and the spectrum is constrained to be optimized on a preset low-dimensional manifold, including: using reference spectra selected from a standard mineral spectral library and the initial endmember spectrum to jointly construct a training set, and using a manifold learning algorithm to learn the low-dimensional manifold structure of the training set data; when iteratively updating the endmember spectrum, the gradient of the objective function with respect to the endmember spectrum is projected onto the tangent space of the low-dimensional manifold, and updated along the projected gradient direction.

[0039] Hyperspectral data has high dimensionality, such as 224 bands, but all possible spectral variants of a particular mineral class are actually distributed on a nonlinear, low-dimensional manifold with dimensions far lower than 224. Leveraging this property, a training set was constructed, for example, consisting of 80 spectra of epidote-family minerals from the USGS spectral library and 20 initial endmember spectra. Applying a locally linear embedding algorithm to analyze this 100-sample training set revealed that the inherent structure of the training set can be represented by a manifold of only 4 dimensions.

[0040] When iteratively updating an epidote-related endmember spectrum, the gradient of the objective function with respect to the endmember spectrum is calculated. This gradient is a 224-dimensional vector pointing in the direction of the fastest decrease in the function value. However, this direction may deviate from the learned 4-dimensional epidote manifold, causing the updated spectrum to lose its physical meaning. Therefore, this 224-dimensional gradient vector is projected onto the tangent space of the 4-dimensional epidote manifold at the current endmember position. This new projected gradient direction ensures that the update step is performed along the manifold surface, ensuring that the updated endmember spectrum remains a reasonably varied and valid epidote family spectrum.

[0041] S4, when the iteration converges or reaches the preset termination condition, the final end-member spectrum and the abundance matrix corresponding to the end-member spectrum are obtained; the altered minerals are identified using the final end-member spectrum, and the mineral alteration characteristics of the area to be processed are extracted in combination with the abundance matrix corresponding to the end-member spectrum.

[0042] The iteration termination condition is set to one of the following three: the number of iterations reaches a preset maximum value, such as 500; the relative change rate of the objective function value in two consecutive iterations is less than a minimum threshold, such as 0.00001; or the change in the Frobenius norm of the endmember spectral matrix and the abundance matrix in two consecutive iterations is less than a preset threshold.

[0043] The final P unmixed endmember spectra were matched against the USGS standard mineral spectral library. The spectral angle matching (SAM) algorithm was used to calculate the similarity of each extracted endmember with all standard mineral spectra in the library. The extracted endmember with the highest similarity and an angle value below a preset threshold was assigned the name of the standard mineral. Based on the identified alteration minerals such as kaolinite and alunite, the corresponding abundance map was used as the mineral distribution map. By threshold segmentation and logical overlay of the abundance maps of specific mineral assemblages, such as kaolinite and alunite, characteristic areas of mineral alteration, such as high-level mud alteration zones, were delineated.

[0044] In an optional embodiment, the altered minerals are identified using the final end-member spectrum, including: using a spectral similarity measurement method to calculate the similarity between the final end-member spectrum and the reference spectrum in the standard mineral spectrum library; for each end-member spectrum, identifying the spectrum as the mineral with the highest similarity in the reference spectrum and above a preset similarity threshold.

[0045] After the algorithm converges after 100 iterations, a set of, for example, eight endmember spectra is obtained. These endmember spectra are purely mathematical vectors that need to be assigned geological names. To this end, a spectral angle matching algorithm is used to compare these eight extracted endmember spectra one by one with a reference library containing 480 standard mineral spectra. The spectral angle matching algorithm calculates the angle between two spectral vectors; smaller angles represent greater similarity.

[0046] The implementation principle of the hyperspectral-based mineral alteration feature extraction method of the embodiment of the present invention is as follows: by introducing a spatial-spectral composite kernel function, the nonlinear unmixing problem is placed in a high-dimensional feature space for solution, so as to understand the complex nonlinear mixing effects existing in the actual surface environment. The group sparse regularization based on the mineral symbiosis law is used as the abundance constraint, and the geological prior knowledge is integrated into the unmixing model, so that the abundance inversion result is more consistent with the geological reality, the physically unreasonable mineral combination is reduced, and the reliability of the result is improved. In addition, by constructing a physical consistency regularization term based on the differentiable radiation transfer model and adopting a manifold learning strategy to update the end-member spectrum, it is not only ensured that the extracted end-member spectrum has a clear physical meaning, but also solves the end-member variability problem caused by changes in light, atmosphere and the mineral's own properties, so that the end-member extraction is closer to the actual situation.

[0047] Although the embodiments of the present invention have been shown and described above, it will be understood that the above embodiments are illustrative and are not to be construed as limitations on the present invention. A person skilled in the art may change, modify, replace and modify the above embodiments within the scope of the present invention.

Claims

1. A method for extracting mineral alteration features based on hyperspectral, characterized in that: The steps include: Obtain hyperspectral remote sensing image data of the area to be processed, perform atmospheric correction and noise reduction preprocessing on the hyperspectral remote sensing image data; use endmember extraction algorithm to extract initial endmember spectra from the preprocessed hyperspectral remote sensing image data; Constructing an objective function for nonlinear spectral unmixing, the objective function includes: a data fidelity term that maps pixels to a high-dimensional feature space using a spatial-spectral composite kernel function; an abundance constraint term that includes a group sparsity regularization based on a mineral paragenesis prior and a non-negative abundance, a sum of one, and a mineral paragenesis prior; and a physical consistency regularization term that penalizes deviations between a spectrum generated by a physical model and a current endmember spectrum based on a differentiable radiative transfer model; An alternating optimization algorithm is used to iteratively solve the objective function, and the endmember abundance matrix and the endmember spectrum are alternately updated during the iteration, wherein the endmember spectrum is updated according to a manifold learning strategy, and the spectrum is constrained to be optimized on a preset low-dimensional manifold; When the iteration converges or reaches the preset termination condition, the final end-member spectrum and the abundance matrix corresponding to the end-member spectrum are obtained; the altered minerals are identified using the final end-member spectrum, and the mineral alteration characteristics of the area to be processed are extracted in combination with the abundance matrix corresponding to the end-member spectrum.

2. The method for extracting mineral alteration features based on hyperspectral according to claim 1, characterized in that: The spatial-spectral composite kernel function is composed of a weighted combination of a Gaussian radial basis spatial kernel function and a polynomial spectral kernel function.

3. The method for extracting mineral alteration features based on hyperspectral according to claim 1, characterized in that: The abundance constraint term of the group sparse regularization based on the mineral symbiosis prior includes: According to the prior knowledge of mineral paragenesis, the end members to be extracted with similar geological genesis or spatial association relationship are divided into different groups; The L2 norm is applied to the endmember abundance vector within each group, and the L2 norm results of all groups are summed to obtain the L2,1 mixed norm, which is used as the group sparse regularization term.

4. The method for extracting mineral alteration features based on hyperspectral according to claim 1, characterized in that: The physical consistency regularization term based on the differentiable radiative transfer model that penalizes the deviation between the spectrum generated by the physical model and the current endmember spectrum includes: Using a differentiable radiation transfer model, inputting mineral physical parameters corresponding to the endmember spectrum of the current iteration into the differentiable radiation transfer model to generate a physical model spectrum; The difference between the physical model spectrum and the current endmember spectrum is calculated as the physical consistency regularization term.

5. The method for extracting mineral alteration features based on hyperspectral according to claim 1, characterized in that: The adopting of an alternating optimization algorithm to iteratively solve the objective function comprises: In each iteration, the endmember spectrum is fixed and the abundance matrix is ​​solved using the multiplicative update rule to ensure the non-negativity of the abundance update. After each update, the abundance vector of each pixel is projected so that the abundance vector satisfies the sum-to-one constraint.

6. The method for extracting mineral alteration features based on hyperspectral according to claim 1, characterized in that: The endmember spectrum is updated based on a manifold learning strategy, which constrains the spectrum to be optimized on a preset low-dimensional manifold, including: A training set is constructed using reference spectra selected from a standard mineral spectral library and the initial end-member spectra, and a manifold learning algorithm is used to learn the low-dimensional manifold structure of the training set data; When iteratively updating the endmember spectrum, the gradient of the objective function with respect to the endmember spectrum is projected onto the tangent space of the low-dimensional manifold, and the update is performed along the projected gradient direction.

7. The method for extracting mineral alteration features based on hyperspectral according to claim 1, characterized in that: The method of identifying altered minerals by using the final end-member spectrum includes: Calculating the similarity between the final end-member spectrum and the reference spectrum in the standard mineral spectrum library using a spectral similarity measurement method; For each end-member spectrum, the spectrum is identified as the mineral with the highest similarity to the reference spectrum and above a preset similarity threshold.

8. The method for extracting mineral alteration features based on hyperspectral according to claim 1, characterized in that: The method of obtaining hyperspectral remote sensing image data of the area to be processed and performing atmospheric correction and noise reduction preprocessing on the hyperspectral remote sensing image data includes: Hyperspectral image data of the mining area were collected by the infrared imaging spectrometer AVIRIS, and atmospheric correction was performed using the FLAASH model to obtain surface reflectance data. The low-rank tensor approximation (LRTA) algorithm was used to denoise the reflectance data.

9. The method for extracting mineral alteration features based on hyperspectral according to claim 1, characterized in that: The endmember extraction algorithm is the vertex component analysis algorithm VCA.

10. The method for extracting mineral alteration features based on hyperspectral according to claim 1, characterized in that: The alternating optimization algorithm is the alternating direction method of multipliers ADMM.

Images