Dimension reduction classification method for hydrous minerals on surface of Mars

Abstract

The invention provides a dimensionality reduction classification method for water-containing minerals on the surface of Mars, and the method comprises the steps: firstly, obtaining Mars hyperspectral remote sensing image data of a target region; then, carrying out data preprocessing, data dimension reduction processing and feature extraction, and determining a plurality of diagnostic spectral feature parameters capable of distinguishing types of hydrous minerals; and determining the similarity corresponding to the plurality of diagnostic spectral characteristic parameters according to the reference spectral characteristic parameters. Furthermore, a plurality of screening spectral characteristic parameters meeting a similarity threshold are determined according to the similarity. And finally, according to the plurality of screening spectral characteristic parameters, determining a target spectral characteristic parameter through an adaptive enhancement algorithm. According to the method, the hyperspectral remote sensing technology and the adaptive enhancement algorithm are combined, and the target spectral characteristic parameters capable of accurately and effectively distinguishing different types of hydrous minerals are determined, so that the to-be-classified Mars hyperspectral remote sensing image data are subjected to hydrous mineral classification, and the accuracy of classifying the hydrous minerals on the surface of the Mars can be improved.

Description

Technical Field

[0001] The invention relates to the technical field of hyperspectral remote sensing, and in particular to a dimensionality reduction classification method for water-containing minerals on the surface of Mars. Background Art

[0002] The water-bearing minerals on the surface of Mars record some key evidence of the early water environment activity on Mars. The water contained in them is mainly in the form of water molecules (H2O) or H + OH - 、H3O + It exists in the form of plasma, specifically in the form of layered silicates, hydrous sulfates and carbonate minerals. Identifying and distinguishing the types of hydrous minerals on the surface of Mars not only helps to analyze the geological and hydrological environment of Mars, but also provides a reference for the subsequent resource acquisition and landing site selection of Mars probes.

[0003] Usually, hyperspectral remote sensing technology can be used to classify hydrous minerals on the surface of Mars. However, the existing remote sensing classification methods for hydrous minerals on the surface of Mars cannot accurately identify hydrous minerals on Mars that are similar in spectral absorption depth, absorption width, etc., and it is difficult to assign and adjust the weight values ​​of various diagnostic spectral characteristic parameters. Most of them have problems of poor generalization and low precision, which makes it impossible to accurately classify hydrous minerals on the surface of Mars. Summary of the invention

[0004] The technical problem to be solved by the present invention is the inability to accurately classify the water-containing minerals on the surface of Mars.

[0005] In order to solve the above technical problems, the present invention provides a dimensionality reduction classification method for water-containing minerals on the surface of Mars, which specifically adopts the following technical solutions:

[0006] The present invention provides a dimensionality reduction classification method for hydrous minerals on the surface of Mars, comprising: first, obtaining Mars hyperspectral remote sensing image data of a target area, wherein the target area includes multiple types of hydrous minerals. Then, data preprocessing is performed on the Mars hyperspectral remote sensing image data to obtain preprocessed Mars hyperspectral remote sensing image data. Data dimension reduction processing is performed on the preprocessed Mars hyperspectral remote sensing image data to obtain reduced-dimensional Mars hyperspectral remote sensing image data. Next, feature extraction is performed on the reduced-dimensional Mars hyperspectral remote sensing image data to determine multiple diagnostic spectral feature parameters, and the diagnostic spectral feature parameters are used to determine the type of hydrous minerals. According to the multiple diagnostic spectral feature parameters and the corresponding reference spectral feature parameters, the similarities corresponding to the multiple diagnostic spectral feature parameters are determined. Further, according to the similarities corresponding to the multiple diagnostic spectral feature parameters, multiple screening spectral feature parameters that meet the similarity threshold are determined. Finally, according to the multiple screening spectral feature parameters, a target spectral feature parameter is determined by an adaptive enhancement algorithm, and the target spectral feature parameter is used to classify the hydrous minerals in the Mars hyperspectral remote sensing image data to be classified.

[0007] This method combines hyperspectral remote sensing technology with an adaptive enhancement algorithm. By performing data preprocessing, data dimension reduction, feature extraction, feature parameter calculation, and feature screening on Mars hyperspectral remote sensing image data, the target spectral feature parameters that can accurately and effectively distinguish different types of hydrous minerals are determined. Finally, the target spectral feature parameters are used to classify the hydrous minerals in the Mars hyperspectral remote sensing image data to be classified, thereby improving the accuracy of classifying hydrous minerals on the surface of Mars.

[0008] In an optional implementation, the above-mentioned data dimensionality reduction processing is performed on the preprocessed Mars hyperspectral remote sensing image data to obtain the reduced-dimensional Mars hyperspectral remote sensing image data, including: using principal component analysis to perform data dimensionality reduction processing on the preprocessed Mars hyperspectral remote sensing image data to obtain the reduced-dimensional Mars hyperspectral remote sensing image data.

[0009] In this implementation method, by performing data dimensionality reduction processing on the preprocessed Mars hyperspectral remote sensing image data, the spectral information in the Mars hyperspectral remote sensing image data that can effectively identify water-containing minerals can be concentrated in a few principal components, thereby highlighting the spectral information of water-containing minerals, achieving compression and dimensionality reduction of remote sensing image data, and improving the efficiency of subsequent determination of target spectral characteristic parameters.

[0010] In an optional implementation, the feature extraction of the Mars hyperspectral remote sensing image data after dimensionality reduction is performed to determine multiple diagnostic spectral feature parameters, including: extracting water molecule spectral information corresponding to the characteristic wavelength of water molecules in the Mars hyperspectral remote sensing image data after dimensionality reduction. Extracting spectral information of iron ion hydroxyl coordination groups and spectral information of magnesium ion hydroxyl coordination groups corresponding to the characteristic bands of iron ion hydroxyl coordination groups and magnesium ion hydroxyl coordination groups in the Mars hyperspectral remote sensing image data after dimensionality reduction. Extracting spectral information of aluminum ion hydroxyl coordination groups corresponding to the characteristic wavelength of aluminum ion hydroxyl coordination groups in the Mars hyperspectral remote sensing image data after dimensionality reduction. Extracting spectral information of polyhydrate sulfate corresponding to the characteristic wavelength of polyhydrate sulfate in the Mars hyperspectral remote sensing image data after dimensionality reduction. Extracting spectral information of monohydrate sulfate corresponding to the characteristic wavelength of monohydrate sulfate in the Mars hyperspectral remote sensing image data after dimensionality reduction. Extracting spectral information of carbonate corresponding to the characteristic band of carbonate in the Mars hyperspectral remote sensing image data after dimensionality reduction. The multiple diagnostic spectral characteristic parameters include: water molecule spectral information, iron ion hydroxyl coordination group spectral information, magnesium ion hydroxyl coordination group spectral information, aluminum ion hydroxyl coordination group spectral information, polyhydrate sulfate spectral information, monohydrate sulfate spectral information and carbonate spectral information.

[0011] In this implementation, water molecule spectral information, iron ion hydroxyl coordination group spectral information, magnesium ion hydroxyl coordination group spectral information, aluminum ion hydroxyl coordination group spectral information, polyhydrate sulfate spectral information, monohydrate sulfate spectral information and carbonate spectral information are extracted as diagnostic spectral feature parameters, which can effectively identify ferromagnesian (Fe / Mg) layered silicate minerals, aluminum (Al) layered silicate minerals, hydrated sulfate minerals and carbonate minerals.

[0012] In an optional implementation, the characteristic wavelengths of the water molecules are: 1.4μm and 1.9μm. The characteristic wavelengths of the hydroxyl coordination groups of iron ions and magnesium ions are: 2.28μm-2.29μm and 2.32μm-2.33μm. The characteristic wavelengths of the hydroxyl coordination groups of aluminum ions are: 2.17μm and 2.21μm. The characteristic wavelengths of polyhydrated sulfate are: 1.9μm and 2.4μm. The characteristic wavelengths of monohydrated sulfate are: 2.1μm and 2.4μm. The characteristic wavelengths of carbonate are: 2.33μm-2.37μm and 2.52μm-2.57μm.

[0013] In the present implementation, the above-mentioned characteristic wavelengths and characteristic bands can be used to effectively perform feature extraction to determine the spectral information of water molecules, the spectral information of hydroxyl coordination groups of iron ions, the spectral information of hydroxyl coordination groups of magnesium ions, the spectral information of hydroxyl coordination groups of aluminum ions, the spectral information of polyhydrated sulfate, the spectral information of monohydrated sulfate and the spectral information of carbonate.

[0014] In an optional implementation, the above-mentioned feature extraction of the Mars hyperspectral remote sensing image data after dimensionality reduction to determine multiple diagnostic spectral feature parameters includes: feature extraction of the Mars hyperspectral remote sensing image data after dimensionality reduction by an internal maximum volume method to determine multiple diagnostic spectral feature parameters.

[0015] In this implementation, the feature extraction of Mars hyperspectral remote sensing image data after dimensionality reduction is performed through the internal maximum volume method, which can improve the effectiveness and accuracy of feature extraction.

[0016] In an optional implementation, the above-mentioned multiple diagnostic spectral characteristic parameters also include one or more of the following: absorption depth, inverse absorption depth, broad absorption characteristics and double absorption characteristics corresponding to water molecules, iron ion hydroxyl coordination groups, magnesium ion hydroxyl coordination groups, aluminum ion hydroxyl coordination groups, polyhydrated sulfate, monohydrated sulfate and carbonate, respectively.

[0017] In the present implementation, the accuracy of classification of hydrous minerals can be improved by using absorption depth, inverse absorption depth, broad absorption feature and double absorption feature as diagnostic spectral feature parameters.

[0018] In an optional implementation, the determining of the similarities corresponding to the plurality of diagnostic spectral characteristic parameters according to the plurality of diagnostic spectral characteristic parameters and the corresponding reference spectral characteristic parameters includes: performing spectral angle matching according to the plurality of diagnostic spectral characteristic parameters and the corresponding reference spectral characteristic parameters to determine the spectral angle matching results corresponding to the plurality of diagnostic spectral characteristic parameters. Determining the corresponding similarities according to the spectral angle matching results corresponding to the plurality of diagnostic spectral characteristic parameters.

[0019] In this implementation, the accuracy of determining the similarity of diagnostic spectral feature parameters can be improved through spectral angle matching.

[0020] In an optional implementation, the above-mentioned determining the similarities corresponding to the multiple diagnostic spectral characteristic parameters based on the multiple diagnostic spectral characteristic parameters and the corresponding reference spectral characteristic parameters further includes: calculating the root mean square error based on the multiple diagnostic spectral characteristic parameters and the corresponding reference spectral characteristic parameters, and determining the root mean square error results corresponding to the multiple diagnostic spectral characteristic parameters. Determine the corresponding similarities based on the spectral angle matching results and the root mean square error results corresponding to the multiple diagnostic spectral characteristic parameters.

[0021] In this implementation, by combining the spectral angle matching and the root mean square error results, the accuracy of determining the similarity of diagnostic spectral feature parameters can be further improved.

[0022] In an optional implementation, the target spectrum characteristic parameters are determined by an adaptive enhancement algorithm according to the plurality of screening spectrum characteristic parameters, including: determining a corresponding weak classifier according to the plurality of screening spectrum characteristic parameters, the expression of the weak classifier is:

[0023]

[0024] Among them, c i (x) represents a weak classifier, f i Indicates the type, size and position of the screening spectral characteristic parameters, θ i represents the preset reference threshold, p i Indicates the direction of the inequality sign, and x indicates the detection window of the preset size of the training sample. According to the weak classifiers corresponding to multiple screening spectral feature parameters, a strong classifier is weighted and combined. The expression of the strong classifier is:

[0025]

[0026] Among them, C(x) represents a strong classifier, t=1,2,…,T is the number of weak classifiers, α t Represents the normalized weight value. The cascade classifier is determined according to the cascade structure formed by the strong classifiers in sequence, and the target spectral feature parameters are output through the cascade classifier.

[0027] In this implementation, the target spectral characteristic parameters can be accurately determined through the above-mentioned cascade classifier.

[0028] In an optional implementation, the data preprocessing of the Mars hyperspectral remote sensing image data includes at least one of the following: format conversion processing, photometric correction processing, atmospheric correction processing, and data filtering processing.

[0029] In this implementation, by preprocessing the Mars hyperspectral remote sensing image data, the accuracy of determining the target spectral characteristic parameters can be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 A schematic flow chart of a dimensionality reduction classification method for water-containing minerals on the surface of Mars provided by an embodiment of the present invention;

[0031] Figure 2 A schematic diagram of preprocessed Mars hyperspectral remote sensing image data provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0032] The following embodiments are described in detail, and examples thereof are shown in the accompanying drawings. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The implementations described in the following embodiments do not represent all implementations consistent with the present application. They are only examples of systems and methods consistent with some aspects of the present application as detailed in the claims.

[0033] The water-bearing minerals on the surface of Mars record some key evidence of the early water environment activity on Mars. The water contained in them is mainly in the form of water molecules (H2O) or H + OH - 、H3O + It exists in the form of plasma, specifically in the form of layered silicates, hydrous sulfates and carbonate minerals. Identifying and distinguishing the types of hydrous minerals on the surface of Mars not only helps to analyze the geological and hydrological environment of Mars, but also provides a reference for the subsequent resource acquisition and landing site selection of Mars probes.

[0034] Usually, hyperspectral remote sensing technology can be used to classify hydrous minerals on the surface of Mars. However, the existing remote sensing classification methods for hydrous minerals on the surface of Mars cannot accurately identify hydrous minerals on Mars that are similar in spectral absorption depth, absorption width, etc., and it is difficult to assign and adjust the weight values ​​of various diagnostic spectral characteristic parameters. Most of them have problems of poor generalization and low precision, which makes it impossible to accurately classify hydrous minerals on the surface of Mars.

[0035] In order to solve the above problems, an embodiment of the present application provides a dimensionality reduction classification method for hydrated minerals on the surface of Mars. The method combines hyperspectral remote sensing technology with an adaptive reinforcement learning algorithm, and determines target spectral characteristic parameters that can accurately distinguish different types of hydrated minerals by performing data dimensionality reduction, feature extraction, feature parameter calculation, feature screening and other processing on the hyperspectral remote sensing image data of Mars. The target spectral characteristic parameters are used to classify the hydrated minerals in the hyperspectral remote sensing image data to be classified, thereby improving the accuracy of classifying the hydrated minerals on the surface of Mars.

[0036] The solution provided by the embodiment of the present application is introduced below in conjunction with the accompanying drawings.

[0037] For details, see Figure 1 , which is a flow chart of a dimensionality reduction classification method for water-containing minerals on the surface of Mars provided by an embodiment of the present invention, such as Figure 1 As shown, the dimensionality reduction classification method for water-containing minerals on the surface of Mars provided by the present invention comprises the following steps S101-S107:

[0038] S101. Acquire Mars hyperspectral remote sensing image data of the target area.

[0039] In the embodiment of the present application, the target area includes multiple types of hydrous minerals. That is, the Mars hyperspectral remote sensing image data may include spectral data of multiple types of hydrous minerals. Among them, the hydrous minerals may include, for example, phyllosilicate minerals, hydrous sulfate minerals, carbonate minerals, etc.

[0040] In one implementation, the Mars hyperspectral remote sensing image data may be acquired by a compact reconnaissance imaging spectromer for Mars (CRISM).

[0041] S102. Preprocessing the Mars hyperspectral remote sensing image data to obtain preprocessed Mars hyperspectral remote sensing image data.

[0042] In the embodiment of the present application, by performing data preprocessing on Mars hyperspectral remote sensing image data, the accuracy of determining the target spectral characteristic parameters can be improved.

[0043] In some embodiments, data preprocessing is performed on Mars hyperspectral remote sensing image data, including at least one of the following: format conversion processing, photometric correction processing, atmospheric correction processing, and data filtering processing.

[0044] Specifically, format conversion processing can be used to convert Mars hyperspectral remote sensing image data into a target format. Photometric correction processing is used to correct brightness differences caused by illumination and optical systems in Mars hyperspectral remote sensing image data, so that pixel reflectances between different regions remain consistent, and to eliminate factors such as uneven illumination and shadows in remote sensing images. Atmospheric correction processing is used to eliminate the effects of factors such as atmosphere and illumination on the reflection of objects in Mars hyperspectral remote sensing image data, and to obtain real physical model parameters such as object reflectance, radiation, and surface temperature. Data filtering processing is used to remove noise data, erroneous data, etc. in Mars hyperspectral remote sensing image data.

[0045] Exemplary performance parameters of the acquired Mars hyperspectral remote sensing image data are shown in Table 1:

[0046] Table 1 Performance parameters of Mars hyperspectral remote sensing image data

[0047]

[0048] The above Mars hyperspectral remote sensing image data are processed by format conversion, photometric correction, atmospheric correction and data filtering respectively.

[0049] Specifically, first convert the CRISM data in IMG format (or other formats) into CAT format, perform photometric correction on the format conversion result, and complete the first-order correction of the illegal solar incident radiance based on the ratio of the Mars hyperspectral remote sensing image to the cosine value of the solar incidence angle. Then, based on the ratio of the photometric correction result to the peak spectrum of Olympus Mons, eliminate the influence of the Martian atmosphere on the Mars hyperspectral remote sensing image data to obtain the corresponding atmospheric correction result. Finally, the atmospheric correction result is subjected to isolated noise peak removal, column deviation correction and spectral smile effect correction, and the result of data filtering is used as the preprocessed Mars hyperspectral remote sensing image data. Figure 2 , which is a schematic diagram of preprocessed Mars hyperspectral remote sensing image data provided by an embodiment of the present invention.

[0050] S103, performing data dimensionality reduction processing on the preprocessed Mars hyperspectral remote sensing image data to obtain dimensionality reduced Mars hyperspectral remote sensing image data.

[0051] In the embodiment of the present application, by performing data dimensionality reduction processing on the pre-processed Mars hyperspectral remote sensing image data, the spectral information that can effectively identify water-containing minerals in the Mars hyperspectral remote sensing image data can be concentrated in a few principal components, thereby highlighting the spectral information of water-containing minerals and achieving compression and dimensionality reduction of remote sensing image data. In this way, on the basis of eliminating redundant information in the pre-processed Mars hyperspectral remote sensing image data, the original Mars hyperspectral remote sensing image data can be represented more, and the efficiency of subsequent determination of target spectral characteristic parameters can be improved.

[0052] In some embodiments, the principal component analysis method may be used to perform data dimensionality reduction processing on the preprocessed Mars hyperspectral remote sensing image data to obtain dimensionality reduced Mars hyperspectral remote sensing image data.

[0053] For example, principal component analysis is used to perform data dimensionality reduction processing, and the reduced-dimensional Mars hyperspectral remote sensing image data can be expressed as:

[0054]

[0055] Among them, X represents the preprocessed Mars hyperspectral remote sensing image data, x1, x2,…, x n represents the spectral information in the preprocessed Mars hyperspectral remote sensing image data, P represents the Mars hyperspectral remote sensing image data after dimensionality reduction, p1, p2, …, p m The spectral information in the Mars hyperspectral remote sensing image data after dimension reduction is shown, and m <n。p1,p2,…,p m are x1,x2,…,x n The first, second, ...mth principal component of α ij(i=1,2,…,n;j=1,2,…,m) represents the transformation matrix elements, corresponding to the eigenvectors, and respectively represent the eigenvectors corresponding to the m largest eigenvalues ​​in the correlation matrix composed of the original variables (i.e., the preprocessed Mars hyperspectral remote sensing image data).

[0056] S104. Perform feature extraction on the dimension-reduced Mars hyperspectral remote sensing image data to determine a plurality of diagnostic spectral feature parameters.

[0057] In the embodiment of the present application, the diagnostic spectral characteristic parameter can be used to determine the type of water-containing minerals. That is, the diagnostic spectral characteristic parameter is a unique wavelength characteristic parameter that can distinguish the mineral components. For example, some mineral components will produce a unique absorption peak at a specific wavelength that is different from other minerals, and the specific wavelength characteristic parameter at that location can be used as a diagnostic spectral characteristic parameter.

[0058] For example, taking the mineral classification of ferromagnesian (Fe / Mg) phyllosilicate minerals, aluminum (Al) phyllosilicate minerals, hydrous sulfate minerals and carbonate minerals as an example, the spectral characteristic parameters of ferromagnesian (Fe / Mg) phyllosilicate minerals, aluminum (Al) phyllosilicate minerals, hydrous sulfate minerals and carbonate minerals can be extracted from the dimensionality-reduced Mars hyperspectral remote sensing image data.

[0059] Specifically, in some embodiments, the above S104 performs feature extraction on the Mars hyperspectral remote sensing image data after dimensionality reduction to determine a plurality of diagnostic spectral feature parameters, which may specifically include:

[0060] Extract the water molecule spectral information corresponding to the characteristic wavelength of water molecules in the Mars hyperspectral remote sensing image data after dimensionality reduction.

[0061] The spectral information of the hydroxyl coordination group of iron ions and the spectral information of the hydroxyl coordination group of magnesium ions corresponding to the characteristic bands of the hydroxyl coordination group of iron ions and the hydroxyl coordination group of magnesium ions in the Mars hyperspectral remote sensing image data after dimensionality reduction are extracted.

[0062] The spectral information of the aluminum ion hydroxyl coordination group corresponding to the characteristic wavelength of the aluminum ion hydroxyl coordination group is extracted from the Mars hyperspectral remote sensing image data after dimensionality reduction.

[0063] The spectral information of polyhydrated sulfate corresponding to the characteristic wavelength of polyhydrated sulfate is extracted from the Mars hyperspectral remote sensing image data after dimensionality reduction.

[0064] The spectral information of sulfate monohydrate corresponding to the characteristic wavelength of sulfate monohydrate is extracted from the Mars hyperspectral remote sensing image data after dimensionality reduction.

[0065] The carbonate spectral information corresponding to the carbonate characteristic band in the Mars hyperspectral remote sensing image data after dimensionality reduction is extracted.

[0066] The above-mentioned multiple diagnostic spectral characteristic parameters may include: water molecule spectral information, iron ion hydroxyl coordination group spectral information, magnesium ion hydroxyl coordination group spectral information, aluminum ion hydroxyl coordination group spectral information, polyhydrate sulfate spectral information, monohydrate sulfate spectral information and carbonate spectral information. The spectral information may include a spectral curve.

[0067] Specifically, the spectral information of water molecules, the spectral information of iron ion hydroxyl coordination groups, and the spectral information of magnesium ion hydroxyl coordination groups can be used to identify Fe / Mg layered silicate minerals such as nontronite, chlorite, and serpentine. The spectral information of water molecules and the spectral information of aluminum ion hydroxyl coordination groups can be used to identify Al layered silicate minerals such as aluminum montmorillonite, kaolin, and illite. The spectral information of water sulfate and the spectral information of monohydrate sulfate can be used to identify hydrous sulfate minerals. The spectral information of carbonate can be used to identify carbonate minerals.

[0068] Among them, the above-mentioned characteristic wavelengths of water molecules, characteristic bands of iron ion hydroxyl coordination groups and magnesium ion hydroxyl coordination groups, characteristic wavelengths of aluminum ion hydroxyl coordination groups, characteristic wavelengths of polyhydrated sulfate, characteristic wavelengths of monohydrated sulfate and characteristic bands of carbonate can be set according to the application requirements of Mars hyperspectral remote sensing image data and data errors.

[0069] In some embodiments, the characteristic wavelengths of water molecules may be: 1.4 μm and 1.9 μm. The characteristic wavelengths of the hydroxyl coordination groups of iron ions and magnesium ions may be: 2.28 μm to 2.29 μm and 2.32 μm to 2.33 μm. The characteristic wavelengths of the hydroxyl coordination groups of aluminum ions may be: 2.17 μm and 2.21 μm. The characteristic wavelengths of polyhydrated sulfate may be: 1.9 μm and 2.4 μm. The characteristic wavelengths of monohydrated sulfate may be: 2.1 μm and 2.4 μm. The characteristic wavelengths of carbonate may be: 2.33 μm to 2.37 μm and 2.52 μm to 2.57 μm.

[0070] In some embodiments, the reduced-dimensional Mars hyperspectral remote sensing image data may be subjected to feature extraction by an internal maximum volume method to determine a plurality of diagnostic spectral feature parameters. For example, water molecule spectral information, iron ion hydroxyl coordination group spectral information, magnesium ion hydroxyl coordination group spectral information, aluminum ion hydroxyl coordination group spectral information, polyhydrate sulfate spectral information, monohydrate sulfate spectral information, and carbonate spectral information may be extracted by the internal maximum volume method.

[0071] Specifically, the internal maximum volume method can obtain the mineral end-member spectra, that is, spectral information, contained in the reduced-dimensionality Mars hyperspectral remote sensing image data by finding the simplex with the largest volume based on the convex simplex characteristics of the reduced-dimensionality Mars hyperspectral remote sensing image data in the feature space.

[0072] For example, since any geometric body in a finite-dimensional space can define a volume, we can assume that a one-dimensional geometric body G 1 The volume in represents the length, and the two-dimensional geometric body G 2 The volume in represents the area, and the three-dimensional geometric body G 3 The volume in is the volume, then G L The geometric body represented by (L≥4) also has volume.

[0073] Furthermore, suppose G k The k+1 different points in the set The matrix S is:

[0074]

[0075] Or: S = [s1-s k+1 ,s2-s k+1 ,…,s k -s k+1 ];

[0076] The volume formula of the convex simplex surrounded by the point set S is as follows:

[0077]

[0078] Where V(S) represents a set function, det(·) is a determinant operation, and abs[·] is an absolute value operation.

[0079] In this way, the above-mentioned internal maximum volume method can effectively extract features from the reduced-dimensional Mars hyperspectral remote sensing image data and determine the diagnostic spectral feature parameters.

[0080] In some embodiments, the above-mentioned multiple diagnostic spectral characteristic parameters also include one or more of the following: absorption depth, inverse absorption depth, broad absorption characteristics and double absorption characteristics corresponding to water molecules, iron ion hydroxyl coordination groups, magnesium ion hydroxyl coordination groups, aluminum ion hydroxyl coordination groups, polyhydrated sulfate, monohydrated sulfate and carbonate, respectively.

[0081] Specifically, the absorption depth, reverse absorption depth, broad absorption characteristics and double absorption characteristics can be calculated based on the above-mentioned water molecule spectral information, iron ion hydroxyl coordination group spectral information, magnesium ion hydroxyl coordination group spectral information, aluminum ion hydroxyl coordination group spectral information, polyhydrate sulfate spectral information, monohydrate sulfate spectral information and carbonate spectral information.

[0082] For example, at a central wavelength of λ C The expression of the absorption depth at is:

[0083]

[0084] Among them, BD(λ C ) is the absorption depth, R C is C The reflectivity at C* is the left end point of the absorption valley (λ L ,R L ) and the right endpoint (λ K ,R K ) Connect the line C The corresponding reflectivity, R L and R K are the reflectances at the left and right ends of the absorption valley respectively. The central wavelength is λ C It can be one or more of the characteristic wavelength of water molecules, the characteristic wavelength of aluminum ion hydroxyl coordination group, the characteristic wavelength of polyhydrated sulfate, and the characteristic wavelength of monohydrated sulfate.

[0085] At the central wavelength λ C The expression of the inverse absorption depth at is:

[0086]

[0087] Among them, BH(λ C ) is the absorption depth.

[0088] At wavelength λ C1 and wavelength λ C2 The expression of the absorption depth at is:

[0089] MIN(BD(λ C1 )and BD(λ C2 ))=min[BD(λ C1 ),BD(λ C2 )];

[0090] Among them, the wavelength λ C1 and wavelength λ C2 It can be the wavelength corresponding to the double absorption peak.

[0091] Broad absorption features can be represented by weighted values ​​of each absorption depth:

[0092] BD(λ C1-CN )=D1(BD(λ C1 ))+D2(BD(λ C2 ))+…+D N (BD(λCN ));

[0093] Among them, D i is the weight of the i-th absorption depth, which can be determined according to each absorption depth, and the sum of all weights is 1.

[0094] In this way, the absorption depth, inverse absorption depth, broad absorption feature and double absorption feature can be effectively and accurately determined by the above expressions.

[0095] S105 . Determine the similarities corresponding to the multiple diagnostic spectral characteristic parameters according to the multiple diagnostic spectral characteristic parameters and the corresponding reference spectral characteristic parameters.

[0096] In the embodiment of the present application, the reference spectral characteristic parameters may be standard spectral characteristic parameters corresponding to the mineral, for example, a laboratory standard mineral spectral curve. In one implementation, the reference spectral characteristic parameters may be obtained through the USGS spectral library or the Brown University spectral library (RELAB spectral library).

[0097] In some embodiments, determining the similarities corresponding to the plurality of diagnostic spectral characteristic parameters according to the plurality of diagnostic spectral characteristic parameters and the corresponding reference spectral characteristic parameters may include:

[0098] First, spectral angle matching is performed according to a plurality of diagnostic spectral characteristic parameters and corresponding reference spectral characteristic parameters to determine spectral angle matching results corresponding to the plurality of diagnostic spectral characteristic parameters.

[0099] Then, the corresponding similarity is determined according to the spectral angle matching results corresponding to the multiple diagnostic spectral feature parameters.

[0100] Among them, the expression of the spectral angle matching result is:

[0101]

[0102] Where α is the spectral angle, n is the number of bands, X = (x1, x2, ..., x n ) and Y=(y1,y2,…,y n ) are the response values ​​of the spectral vector in the diagnostic spectral characteristic parameters and the reference spectral characteristic parameters in the n-band, respectively.

[0103] In some embodiments, determining the similarities corresponding to the plurality of diagnostic spectral characteristic parameters according to the plurality of diagnostic spectral characteristic parameters and the corresponding reference spectral characteristic parameters may further include:

[0104] A root mean square error is calculated based on a plurality of diagnostic spectral characteristic parameters and corresponding reference spectral characteristic parameters, and a root mean square error result corresponding to the plurality of diagnostic spectral characteristic parameters is determined.

[0105] The corresponding similarity is determined according to the spectral angle matching results and the root mean square error results corresponding to the multiple diagnostic spectral feature parameters.

[0106] Exemplarily, first, the first weight and the second weight can be set for the spectral angle matching result and the root mean square error result respectively according to the needs of the actual application, and the sum of the first weight and the second weight is 1. Then, the sum of the multiplication of the spectral angle matching result and the corresponding first weight and the multiplication of the root mean square error result and the corresponding second weight is used as the similarity.

[0107] S106 . Determine, according to the similarities corresponding to the plurality of diagnostic spectral characteristic parameters, a plurality of screening spectral characteristic parameters that meet a similarity threshold.

[0108] In the embodiment of the present application, a similarity threshold can be determined according to the needs of actual applications to achieve preliminary screening of multiple diagnostic spectral characteristic parameters and determine multiple screening spectral characteristic parameters to facilitate further determination of target spectral characteristic parameters.

[0109] S107, determining target spectrum characteristic parameters through an adaptive enhancement algorithm according to the multiple screening spectrum characteristic parameters.

[0110] In the embodiment of the present application, the multiple screening spectral characteristic parameters preliminarily screened in S106 are further screened by an adaptive enhancement algorithm to determine spectral characteristic parameters with better fitting effect. That is, the spectral characteristic parameters with better identification and classification effect and higher accuracy for hydrous minerals are determined, namely, the target spectral characteristic parameters. Specifically, the target spectral characteristic parameters can be used to classify hydrous minerals in the Mars hyperspectral remote sensing image data to be classified.

[0111] In some embodiments, S107 determines the target spectrum characteristic parameters through an adaptive enhancement algorithm according to the multiple screening spectrum characteristic parameters, which may specifically include:

[0112] First, according to multiple screening spectral feature parameters, the corresponding weak classifier is determined. The expression of the weak classifier is:

[0113]

[0114] Among them, c i (x) represents a weak classifier, f i Indicates the type, size and position of the screening spectral characteristic parameters, θ i represents the preset reference threshold, p i represents the direction of the inequality sign, and x represents a detection window of a training sample of a preset size. Specifically, the size of x can be set according to the requirements of the actual application, for example, the size of x can be set to 20×20.

[0115] Then, the weak classifiers corresponding to the multiple screening spectral feature parameters are weighted and combined into a strong classifier. The expression of the strong classifier is:

[0116]

[0117] Among them, C(x) represents a strong classifier, t=1,2,…,T is the number of weak classifiers, α t Represents the normalized weight value.

[0118] Finally, the cascade classifier is determined according to the cascade structure formed by the strong classifiers in sequence, and the target spectral feature parameters are output through the cascade classifier.

[0119] In this way, multiple screening spectral characteristic parameters are used as weak classifiers of the algorithm, that is, primary classification criteria. After these primary classification criteria are weighted and combined into strong classifiers, a higher level classification standard can be obtained. Furthermore, multiple strong classifiers form a cascade classifier of a cascade structure in a first-level order, and the screening spectral characteristic parameters corresponding to the optimal weights are output through the cascade classifier, that is, the target spectral characteristic parameters.

[0120] The dimensionality reduction classification method for hydrous minerals on the surface of Mars provided by the embodiment of the present invention is adopted. The method combines hyperspectral remote sensing technology with an adaptive enhancement algorithm, and performs data preprocessing, data dimensionality reduction, feature extraction, feature parameter calculation, feature screening and other processing on the hyperspectral remote sensing image data of Mars to determine the target spectral feature parameters that can accurately and effectively distinguish different types of hydrous minerals. Finally, the hydrous minerals are classified for the hyperspectral remote sensing image data of Mars to be classified by the target spectral feature parameters, thereby improving the accuracy of the classification of hydrous minerals on the surface of Mars.

[0121] In the description of the present invention, it should be understood that the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of the features. In the description of the present invention, the meaning of "plurality" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

[0122] In the present invention, unless otherwise clearly specified and limited, the terms "installed", "connected", "connected", "fixed" and the like should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium, it can be the internal connection of two elements or the interaction relationship between two elements, unless otherwise clearly defined. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0123] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, without contradiction.

[0124] Similar parts between the embodiments provided in this application can be referenced to each other. The specific implementation methods provided above are only a few examples under the general concept of this application and do not constitute a limitation on the protection scope of this application. For those skilled in the art, any other implementation methods expanded based on the scheme of this application without creative work belong to the protection scope of this application.

Claims

1. A dimensionality reduction classification method for water-containing minerals on the surface of Mars, characterized in that: include: Acquire Mars hyperspectral remote sensing image data of a target area, wherein the target area includes multiple types of hydrous minerals; Performing data preprocessing on the Mars hyperspectral remote sensing image data to obtain preprocessed Mars hyperspectral remote sensing image data; Performing data dimensionality reduction processing on the preprocessed Mars hyperspectral remote sensing image data to obtain dimensionality reduced Mars hyperspectral remote sensing image data; Performing feature extraction on the reduced-dimensional Mars hyperspectral remote sensing image data to determine a plurality of diagnostic spectral feature parameters, wherein the diagnostic spectral feature parameters are used to determine the type of the hydrous mineral; Determining, based on the plurality of diagnostic spectral characteristic parameters and corresponding reference spectral characteristic parameters, similarities corresponding to the plurality of diagnostic spectral characteristic parameters; Determining, based on the similarities corresponding to the plurality of diagnostic spectral characteristic parameters, a plurality of screening spectral characteristic parameters that meet a similarity threshold; According to the multiple screening spectral characteristic parameters, a target spectral characteristic parameter is determined by an adaptive enhancement algorithm, and the target spectral characteristic parameter is used to classify the water-containing minerals in the Mars hyperspectral remote sensing image data to be classified.

2. The method according to claim 1, characterized in that The step of performing data dimensionality reduction processing on the preprocessed Mars hyperspectral remote sensing image data to obtain the reduced-dimensional Mars hyperspectral remote sensing image data includes: The principal component analysis method is used to perform data dimension reduction processing on the preprocessed Mars hyperspectral remote sensing image data to obtain the Mars hyperspectral remote sensing image data after dimension reduction.

3. The method according to claim 1, characterized in that The feature extraction of the reduced-dimensional Mars hyperspectral remote sensing image data to determine a plurality of diagnostic spectral feature parameters includes: Extracting water molecule spectrum information corresponding to a wavelength of a water molecule characteristic wavelength from the Mars hyperspectral remote sensing image data after dimensionality reduction; Extracting the spectrum information of the hydroxyl coordination group of iron ions and the spectrum information of the hydroxyl coordination group of magnesium ions corresponding to the characteristic bands of the hydroxyl coordination group of iron ions and the hydroxyl coordination group of magnesium ions in the Mars hyperspectral remote sensing image data after dimension reduction; Extracting spectral information of the aluminum ion hydroxyl coordination group having a wavelength corresponding to the characteristic wavelength of the aluminum ion hydroxyl coordination group from the Mars hyperspectral remote sensing image data after dimensionality reduction; Extracting polyhydrate sulfate spectral information corresponding to a wavelength of polyhydrate sulfate characteristic wavelength from the Mars hyperspectral remote sensing image data after dimension reduction; Extracting the spectral information of sulfate monohydrate corresponding to the characteristic wavelength of sulfate monohydrate from the Mars hyperspectral remote sensing image data after dimension reduction; Extracting carbonate spectral information corresponding to the carbonate characteristic band in the Mars hyperspectral remote sensing image data after dimension reduction; The multiple diagnostic spectral characteristic parameters include: the water molecule spectral information, the iron ion hydroxyl coordination group spectral information, the magnesium ion hydroxyl coordination group spectral information, the aluminum ion hydroxyl coordination group spectral information, the polyhydrate sulfate spectral information, the monohydrate sulfate spectral information and the carbonate spectral information.

4. The method according to claim 3, characterized in that The characteristic wavelengths of water molecules are: 1.4μm and 1.9μm; the characteristic wavelengths of the iron ion hydroxyl coordination group and the magnesium ion hydroxyl coordination group are: 2.28μm-2.29μm and 2.32μm-2.33μm; the characteristic wavelengths of the aluminum ion hydroxyl coordination group are: 2.17μm and 2.21μm; the characteristic wavelengths of polyhydrate sulfate are: 1.9μm and 2.4μm; the characteristic wavelengths of monohydrate sulfate are: 2.1μm and 2.4μm; the characteristic wavelengths of carbonate are: 2.33μm-2.37μm and 2.52μm-2.57μm.

5. The method according to claim 1, characterized in that The feature extraction of the reduced-dimensional Mars hyperspectral remote sensing image data to determine a plurality of diagnostic spectral feature parameters includes: The feature extraction of the Mars hyperspectral remote sensing image data after dimension reduction is performed through the internal maximum volume method to determine a plurality of diagnostic spectral feature parameters.

6. The method according to claim 3, characterized in that The multiple diagnostic spectral characteristic parameters also include one or more of the following: absorption depth, inverse absorption depth, broad absorption characteristics and double absorption characteristics corresponding to water molecules, iron ion hydroxyl coordination groups, magnesium ion hydroxyl coordination groups, aluminum ion hydroxyl coordination groups, polyhydrated sulfate, monohydrated sulfate and carbonate, respectively.

7. The method according to claim 1, characterized in that The determining, based on the plurality of diagnostic spectral characteristic parameters and the corresponding reference spectral characteristic parameters, the similarities corresponding to the plurality of diagnostic spectral characteristic parameters comprises: Performing spectral angle matching according to the plurality of diagnostic spectral characteristic parameters and corresponding reference spectral characteristic parameters, and determining spectral angle matching results corresponding to the plurality of diagnostic spectral characteristic parameters; The corresponding similarity is determined according to the spectral angle matching results corresponding to the plurality of diagnostic spectral feature parameters.

8. The method according to claim 7, characterized in that The determining, based on the plurality of diagnostic spectral characteristic parameters and the corresponding reference spectral characteristic parameters, of the similarities corresponding to the plurality of diagnostic spectral characteristic parameters further includes: Calculating a root mean square error according to the plurality of diagnostic spectral characteristic parameters and corresponding reference spectral characteristic parameters, and determining a root mean square error result corresponding to the plurality of diagnostic spectral characteristic parameters; The corresponding similarity is determined according to the spectral angle matching results and the root mean square error results corresponding to the plurality of diagnostic spectral feature parameters.

9. The method according to claim 1, characterized in that: Determining the target spectrum characteristic parameters by using an adaptive enhancement algorithm according to the multiple screening spectrum characteristic parameters includes: According to the plurality of screening spectral characteristic parameters, a corresponding weak classifier is determined, and the expression of the weak classifier is: Among them, c i (x) represents a weak classifier, f i Indicates the type, size and position of the screening spectral characteristic parameters, θ i represents the preset reference threshold, p i represents the direction of the inequality sign, and x represents the detection window of the training sample of the preset size; According to the weak classifiers corresponding to the plurality of screening spectral characteristic parameters, a strong classifier is weighted and combined, and the expression of the strong classifier is: Among them, C(x) represents a strong classifier, t=1,2,…,T is the number of weak classifiers, α t Represents the normalized weight value; A cascade classifier is determined according to the cascade structure formed by the strong classifiers in sequence, and the target spectrum characteristic parameters are outputted through the cascade classifier.

10. The method according to claim 1, characterized in that The data preprocessing of the Mars hyperspectral remote sensing image data includes at least one of the following: format conversion processing, photometric correction processing, atmospheric correction processing, and data filtering processing.

Images