Mars surface mineral identification method based on Transform

Abstract

The invention provides a Mars surface mineral identification method based on Transform. The Mars surface mineral identification method comprises the following steps: dividing a preprocessed Mars mineral standard hyperspectral data set into a training data set, a verification data set and a test data set; and then, a mineral recognition Transform model is constructed. And training the mineral identification Transform model according to the training data set and the verification data set to obtain an optimized mineral identification Transform model. And according to the test data set, the optimized mineral identification Transform model is tested, and the test classification accuracy is determined. And finally, under the condition that the test accuracy meets a preset accuracy condition, inputting to-be-identified Mars mineral hyperspectral data into the optimized mineral identification Transform model, and outputting to obtain a Mars mineral classification identification result. Therefore, the accuracy of classification and identification of Mars minerals can be effectively improved.

Description

Technical Field

[0001] The present invention relates to the field of remote sensing technology, and in particular to a method for identifying Martian surface minerals based on Transformer. Background Art

[0002] Hyperspectral remote sensing technology is the main means of identifying minerals on the surface of Mars. Hyperspectral remote sensing data contains rich spectral and spatial information, and has the characteristics of unified image and spectrum. Different types of minerals have their own unique spectral characteristics. Mineral types and mineral components can be identified based on the diagnostic spectral characteristics of minerals. Therefore, Mars hyperspectral images have been widely used in the research of mineral identification on the surface of Mars.

[0003] At present, methods based on machine learning and deep learning are usually used, such as artificial neural networks (ANN), convolutional neural networks (CNN), support vector machines (SVM), etc. Although these methods have shown good results in identifying minerals on the Martian surface, they still have certain limitations. For example, the extraction of global band information is insufficient, and the long-distance dependency between spectral bands cannot be fully considered. The existence of these limitations leads to the existing Martian mineral classification models having problems such as low classification accuracy and low degree of automation, making it difficult to achieve large-scale application. Summary of the invention

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

[0005] In order to solve the above technical problems, the present invention provides a method for identifying Martian surface minerals based on Transformer, which specifically adopts the following technical solutions:

[0006] The present invention provides a method for identifying minerals on the surface of Mars based on Transformer, comprising: first, obtaining a standard hyperspectral data set of Martian minerals. Then, preprocessing the standard hyperspectral data set of Martian minerals to obtain a preprocessed standard hyperspectral data set of Martian minerals. Secondly, the preprocessed standard hyperspectral data set of Martian minerals is divided into: a training data set, a verification data set, and a test data set according to a preset division ratio. Further, a mineral identification Transformer model is constructed, and the mineral identification Transformer model is used to classify and identify Martian surface minerals. Then, the mineral identification Transformer model is trained according to the training data set and the verification data set to obtain an optimized mineral identification Transformer model. The optimized mineral identification Transformer model is tested according to the test data set, and the test classification accuracy is determined. Finally, when the test accuracy meets the preset accuracy condition, the hyperspectral data of the Martian minerals to be identified is input into the optimized mineral identification Transformer model, and the Martian mineral classification and identification results are obtained as output.

[0007] This method is based on the Transformer model, which uses deep learning methods to capture the internal correlation of Martian mineral standard hyperspectral data, focusing on the valuable information in the Martian mineral standard hyperspectral data while ignoring unimportant information. In this way, the mineral identification Transformer model can learn richer feature representations in the Martian mineral standard hyperspectral data, thereby effectively improving the accuracy of Martian mineral classification and identification.

[0008] In an optional implementation, the mineral identification Transformer model includes: an embedding mapping function, an encoder, a fully connected layer, and an output layer. The embedding mapping function is composed of a band embedding function and a position embedding function, and the encoder is composed of a plurality of encoder layers connected in series.

[0009] In this implementation, the mineral identification Transformer model with the above structure can realize mineral classification and identification of the Martian mineral hyperspectral dataset.

[0010] In an optional implementation, the above-mentioned training of the mineral identification Transformer model according to the training data set and the verification data set to obtain the optimized mineral identification Transformer model includes: first, inputting the training data set and the verification data set into the embedding mapping function, and outputting a first embedding representation corresponding to the training data set, and a second embedding representation corresponding to the verification data set. Then, the first embedding representation and the second embedding representation are input into the encoder, and the first encoder output result corresponding to the first embedding representation and the second encoder output result corresponding to the second embedding representation are output. Further, the first encoder output result and the second encoder output result are input into the fully connected layer, and the first fully connected layer output result corresponding to the first encoder output result and the second fully connected layer output result corresponding to the second encoder output result are output. Next, the first fully connected layer output result and the second fully connected layer output result are input into the output layer, and the first predicted classification recognition result corresponding to the first fully connected layer output result and the second predicted classification recognition result corresponding to the second fully connected layer output result are output. According to the first predicted classification recognition result, a loss function is constructed, and the hyperparameters of the mineral identification Transformer model are iteratively optimized so that the loss function converges to obtain the hyperparameter training result. Finally, the hyperparameter training results are optimized and adjusted according to the second prediction classification recognition results to obtain the optimized mineral recognition Transformer model.

[0011] In an optional implementation, the expression of the above embedded mapping function is:

[0012] Em(x)=Eb(x)+Ep(x);

[0013] In the formula, Em(x) represents the embedding mapping function, Eb(x) represents the band embedding function, Ep(x) represents the position embedding function, and x represents the input hyperspectral data. The expression of Eb(x) is:

[0014] Eb(x)=W(x;θ e );

[0015] Where W represents a linear mapping, θ e Represents the parameters of band embedding. The expression of Ep(x) above is:

[0016]

[0017] Where p represents the position of the band in the hyperspectral data x, p = 0, 1, ..., 232; 2i represents the even dimension of embedding, 2i+1 represents the odd dimension of embedding, and d represents the total dimension of embedding.

[0018] In an optional implementation, the function expression of the above encoder is:

[0019] xn = Encoder(xe);

[0020] Where xn represents the output of the encoder, and xe represents the output of the embedding mapping function.

[0021] Among them, the expansion of Encoder(xe) is:

[0022]

[0023] In the formula, t (i) ,i=1,2…N represents the output result of the i-th encoder layer, θ n (i) is the parameter of the i-th encoder layer, en (i) represents the i-th encoder layer.

[0024] In an alternative implementation, the expression of the above fully connected layer is:

[0025] xl=D{LN[γ(L(xn,θ f ))]};

[0026] In the formula, xl represents the output result of the fully connected layer, θ f is the parameter of the fully connected layer, γ represents the Relu activation function, L represents the fully connected layer, LN represents layer normalization, and D represents the Dropout function.

[0027] In an alternative implementation, the expression of the above output layer is:

[0028]

[0029] In the formula, represents the output result of the output layer, and S represents the Softmax activation function.

[0030] In an optional implementation, the above loss function is a cross entropy loss function, and the hyperparameters of the mineral identification Transformer model include: the number of encoder layers, the model dimension, and the number of multi-head attention branches.

[0031] In an optional implementation, the expression of the above test classification accuracy is:

[0032]

[0033] in,

[0034]

[0035] Among them, ACC represents the test classification accuracy, n ′ represents the size of the test dataset, Indicates the mineral classification test results, y i ′ Represents the results of mineral classification standards.

[0036] In an optional implementation, the above-mentioned Martian mineral standard hyperspectral data set includes: iron / magnesium montmorillonite spectral data, magnesium carbonate spectral data and fayalite spectral data; the preset division ratio is: training data set: verification data set: test data set is 8:1:1. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 A schematic diagram of the principle of a method for identifying Martian surface minerals based on Transformer provided in an embodiment of the present invention;

[0038] Figure 2 A schematic diagram of a flow chart of a method for identifying Martian surface minerals based on Transformer provided in an embodiment of the present invention;

[0039] Figure 3 A schematic diagram of the structure of a mineral identification Transformer model provided in an embodiment of the present invention;

[0040] Figure 4 A schematic diagram of the mineral mapping results provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0041] 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.

[0042] Hyperspectral remote sensing technology is the main means of identifying minerals on the surface of Mars. Hyperspectral remote sensing data contains rich spectral and spatial information, and has the characteristics of unified image and spectrum. Different types of minerals have their own unique spectral characteristics. Mineral types and mineral components can be identified based on the diagnostic spectral characteristics of minerals. Therefore, Mars hyperspectral images have been widely used in the research of mineral identification on the surface of Mars.

[0043] At present, methods based on machine learning and deep learning are usually used, such as artificial neural networks (ANN), convolutional neural networks (CNN), support vector machines (SVM), etc. Although these methods have shown good results in identifying minerals on the Martian surface, they still have certain limitations. For example, the extraction of global band information is insufficient, and the long-distance dependency between spectral bands cannot be fully considered. The existence of these limitations leads to the existing Martian mineral classification models having problems such as low classification accuracy and low degree of automation, making it difficult to achieve large-scale application.

[0044] In order to solve the above problems, the present application embodiment provides a method for identifying Martian surface minerals based on Transformer. Figure 1 The schematic diagram of the principle of the Mars surface mineral identification method based on Transformer provided in the embodiment of the present invention is as follows: Figure 1 As shown, the method can divide the preprocessed Mars mineral standard hyperspectral data set into a training data set, a verification data set and a test data set. Then, the optimized mineral identification Transformer model is trained by the training data set and the verification data set to obtain the optimized mineral identification Transformer model. Furthermore, the optimized mineral identification Transformer model is tested by the test data set. Finally, when the test classification accuracy meets the preset accuracy condition, the Mars mineral hyperspectral data to be identified is subjected to mineral classification and identification by the optimized mineral identification Transformer model, and the Mars mineral classification and identification results are output. In this way, the mineral identification Transformer model can learn richer feature representations in the Mars mineral standard hyperspectral data, thereby effectively improving the accuracy of Mars mineral classification and identification.

[0045] The solution provided by the embodiment of the present application is introduced below in conjunction with the accompanying drawings.

[0046] For details, see Figure 2 , which is a flow chart of a method for identifying Martian surface minerals based on Transformer according to an embodiment of the present invention, such as Figure 2 As shown, the Mars surface mineral identification method based on Transformer provided by the present invention includes the following steps S101-S107:

[0047] S101. Obtain the standard hyperspectral dataset of Martian minerals.

[0048] In the embodiment of the present application, the Martian mineral standard hyperspectral dataset may be a compact reconnaissance imaging spectrometer for Mars (CRI SM) dataset.

[0049] In one implementation, the Martian mineral standard hyperspectral dataset can be obtained from a standard spectral library obtained through CRI SM analysis (MICA).

[0050] In some embodiments, the Martian mineral standard hyperspectral data set includes: ferromagnesian montmorillonite spectral data, magnesium carbonate spectral data, and fayalite spectral data.

[0051] S102. Preprocess the Martian mineral standard hyperspectral dataset to obtain a preprocessed Martian mineral standard hyperspectral dataset.

[0052] In some embodiments, data preprocessing is performed on the Martian mineral standard hyperspectral dataset, including at least one of the following: clipping, smoothing, continuum removal, and standardization.

[0053] Specifically, the clipping process is to intercept the hyperspectral data of the target spectral wavelength. For example, the spectral wavelength range in the MICA spectral library is 0-4μm. Through the clipping process, the 1.0-2.6μm part can be intercepted as the preprocessed Mars mineral standard hyperspectral data set. The smoothing process is to perform noise reduction on the Mars mineral standard hyperspectral data set to remove the peaks in the hyperspectral data. The continuum removal process is to remove the continuum of the hyperspectral data to highlight the absorption characteristics. The standardization process is to normalize the value of the hyperspectral data to near 0, so that the hyperspectral data in the Mars mineral standard hyperspectral data set is more regular.

[0054] S103, dividing the preprocessed Mars mineral standard hyperspectral dataset into: a training dataset, a verification dataset, and a test dataset according to a preset division ratio.

[0055] Specifically, to facilitate the training and optimization of the mineral identification Transformer model, the embodiment of the present application can divide the preprocessed Martian mineral standard hyperspectral dataset into: a training dataset, a verification dataset, and a test dataset.

[0056] In some embodiments, the preset division ratio may be: training data set: validation data set: test data set is 8:1:1. In this way, the mineral identification Transformer model can be effectively trained through the training data set with a relatively large proportion, so as to improve the accuracy of the mineral identification Transformer model in classifying and identifying minerals on the surface of Mars.

[0057] In one implementation, the training dataset can be 1express, Among them, n 1 is the number of hyperspectral data contained in the training dataset, k 1 is the dimension, i.e. the number of spectral channels. 1 express, Therefore, the training dataset can be expressed as The representation of the validation and test sets is consistent with that of the training set.

[0058] In one implementation, the annotation of the training data set (i.e., the classification recognition result corresponding to the training data set) can be expressed as Y 1 express, Among them, 4 are the four types of mineral labels after one-hot encoding: categories 1, 2, 3, and 4 represent iron / magnesium montmorillonite, magnesium carbonate, fayalite, and background pixels, respectively. In the annotation of the training data set, a single data annotation is y 1 Represents, y∈R 1 X 4 Therefore, the annotation of the training dataset can be expressed as

[0059] It is understandable that the representation of the validation dataset and the test dataset may be the same as that of the training dataset, which will not be described in detail here.

[0060] S104. Construct a mineral identification Transformer model.

[0061] The mineral identification Transformer model is used to classify and identify minerals on the surface of Mars. In some embodiments, Figure 3 The schematic diagram of the structure of the mineral identification Transformer model provided by the embodiment of the present invention is as follows: Figure 3 As shown in the figure, the mineral identification Transformer model includes: embedding mapping function, encoder, fully connected layer and output layer. Among them, the embedding mapping function is composed of band embedding function and position embedding function, and the encoder is composed of multiple encoder layers connected in series.

[0062] Specifically, the Martian mineral hyperspectral data can be input into the embedding mapping function, the output of the embedding mapping function (i.e., the embedding representation) is input into the encoder, the output of the encoder is input into the fully connected layer, and the output of the fully connected layer is input into the output layer. Finally, the output layer outputs the predicted classification recognition result.

[0063] In some embodiments, the expression of the above embedded mapping function is:

[0064] Em(x)=Eb(x)+Ep(x);

[0065] Where Em(x) represents the embedding mapping function, Eb(x) represents the band embedding function, Ep(x) represents the position embedding function, and x represents the input hyperspectral data.

[0066] Among them, the expression of Eb(x) is:

[0067] Eb(x)=W(x;θ e );

[0068] Where W represents a linear mapping, θ e Parameters representing band embedding;

[0069] The expression of Ep(x) is:

[0070]

[0071] Where p represents the position of the band in the hyperspectral data x, that is, the index of the band, p = 0, 1, ..., 232; 2i represents the even dimension of embedding, 2i + 1 represents the odd dimension of embedding, and d represents the total dimension of embedding, that is, the model dimension.

[0072] In some embodiments, the function expression of the above encoder is:

[0073] xn = Encoder(xe);

[0074] In the formula, xn represents the output result of the encoder, and xe represents the output result of the embedding mapping function;

[0075] Among them, the expansion of Encoder(xe) is:

[0076]

[0077] In the formula, t (i) ,i=1,2…N represents the output result of the i-th encoder layer, θ n (i) is the parameter of the i-th encoder layer, en (i) represents the i-th encoder layer.

[0078] In some embodiments, the expression of the above fully connected layer is:

[0079] xl=D{LN[γ(L(xn,θ f ))]};

[0080] In the formula, xl represents the output result of the fully connected layer, θ f is the parameter of the fully connected layer, γ represents the Relu activation function, L represents the fully connected layer, LN represents layer normalization, and D represents the Dropout function.

[0081] In some embodiments, the expression of the output layer is:

[0082]

[0083] In the formula, represents the output result of the output layer, and S represents the Softmax activation function.

[0084] S105. Train the mineral identification Transformer model according to the training data set and the verification data set to obtain an optimized mineral identification Transformer model.

[0085] In some embodiments, Figure 3 Taking the mineral identification Transformer model shown in FIG. 1 as an example, S105 may specifically include the following steps S1051-S1056:

[0086] S1051. Input the training data set and the verification data set into the embedding mapping function, and output a first embedding representation corresponding to the training data set and a second embedding representation corresponding to the verification data set.

[0087] The first embedding representation and the second embedding representation can be determined according to the expression of the above embedding mapping function.

[0088] S1052: Input the first embedded representation and the second embedded representation into an encoder, and output a first encoder output result corresponding to the first embedded representation and a second encoder output result corresponding to the second embedded representation.

[0089] The output result of the first encoder and the output result of the second encoder can be determined based on the function expression of the above encoder.

[0090] S1053. Input the output result of the first encoder and the output result of the second encoder into a fully connected layer, and output a first fully connected layer output result corresponding to the output result of the first encoder and a second fully connected layer output result corresponding to the output result of the second encoder.

[0091] Among them, the output result of the first fully connected layer and the output result of the second fully connected layer can be determined according to the expression of the above fully connected layer.

[0092] S1054. Input the output result of the first fully connected layer and the output result of the second fully connected layer into the output layer, and output a first predicted classification recognition result corresponding to the output result of the first fully connected layer and a second predicted classification recognition result corresponding to the output result of the second fully connected layer.

[0093] Among them, the first predicted classification recognition result and the second predicted classification recognition result can be determined based on the expression of the above output layer.

[0094] S1055. Construct a loss function according to the first predicted classification recognition result, iteratively optimize the hyperparameters of the mineral recognition Transformer model so that the loss function converges to obtain the hyperparameter training result.

[0095] In some embodiments, the loss function may be a cross entropy loss function. Exemplarily, the cross entropy loss function may be expressed as:

[0096]

[0097] Where d represents the model dimension. i Represents the category labels in the predicted classification recognition results of the training data.

[0098] S1056. Optimize and adjust the hyperparameter training result according to the second prediction classification recognition result to obtain an optimized mineral recognition Transformer model.

[0099] Specifically, the hyperparameters of the mineral identification Transformer model include: the number of encoder layers, model dimension, and the number of multi-head attention branches.

[0100] The hyperparameter training results can be further optimized and adjusted through the second prediction classification recognition results to determine the optimized hyperparameters, thereby obtaining an optimized mineral recognition Transformer model.

[0101] S106. Testing the optimized mineral identification Transformer model according to the test data set, and determining the test classification accuracy.

[0102] Furthermore, the test data set can be input into the optimized mineral identification Transformer model to obtain the third predicted classification recognition result, which is then compared with the standard classification recognition result in the test data set to determine the test classification accuracy.

[0103] In some embodiments, the expression of the above test classification accuracy is:

[0104]

[0105] in,

[0106]

[0107] In the formula, ACC represents the test classification accuracy, n ′ represents the size of the test dataset, represents the mineral classification test result (such as the third predicted classification identification result mentioned above), y i ′Indicates the standard results of mineral classification (such as the standard classification identification results mentioned above).

[0108] S107. When the test accuracy meets the preset accuracy condition, the hyperspectral data of the Martian mineral to be identified is input into the optimized mineral identification Transformer model, and the Martian mineral classification and identification result is output.

[0109] Specifically, taking the above test classification accuracy ACC as an example, the higher the ACC is, the more accurate the classification result of the optimized mineral identification Transformer model is.

[0110] In some embodiments, corresponding preset accuracy conditions can be set according to different application scenarios and application requirements. For example, corresponding preset accuracy conditions can be set according to different Mars regions. The preset accuracy conditions can be, for example, greater than 85%, or greater than 90%, etc. When the test accuracy meets the preset accuracy conditions, it can be determined that the optimized mineral identification Transformer model can meet the application requirements.

[0111] Furthermore, when the test accuracy meets the preset accuracy conditions, the hyperspectral data of the Martian mineral to be identified can be input into the optimized mineral identification Transformer model, and the Martian mineral classification and identification results can be output.

[0112] In some embodiments, the Martian mineral classification and identification results may be mineral mapping results marked with different colors, wherein each color corresponds to a Martian mineral type.

[0113] For example, the hyperspectral data of Martian minerals to be identified is an MTRDR image derived from CRISM. The MTRDR image data is represented by X_im, and the mineral classification and identification result is represented by Y_im. The spectral data corresponding to a single pixel in the MTRDR image is represented by x_im, that is, The classification recognition result corresponding to a single pixel is represented by y_im, that is, Where m is the total number of pixels in the MTRDR image. Input X_im into the optimized mineral recognition Transformer model to obtain Y_im. The mineral category recognition results corresponding to each pixel are marked with different colors in the MTRDR image to obtain the mineral mapping results as shown in the figure. Figure 4 In the mineral mapping result, the mineral category represented by the blue area 401 is: iron / magnesium montmorillonite, the mineral category represented by the green area 402 is: magnesium carbonate, and the mineral category represented by the red area 403 is: fayalite.

[0114] It should be noted that the embodiments of the present application provide Figure 4 The colors in the figure are for illustrative purposes only and are only used to distinguish different types of Martian minerals. They are not limiting. The colors corresponding to different types of Martian minerals can be set according to actual application requirements, and this application does not make specific limitations on this.

[0115] The Transformer-based Mars surface mineral identification method provided by the embodiment of the present invention is used to capture the internal correlation of the Mars mineral standard hyperspectral data through deep learning based on the Transformer model, focusing on the valuable information in the Mars mineral standard hyperspectral data, while ignoring unimportant information. In this way, the mineral identification Transformer model can learn richer feature representations in the Mars mineral standard hyperspectral data, thereby improving the accuracy of Mars mineral classification and identification. It also has important guiding significance for the development of Mars resource exploration projects and the development and utilization of deep space resources.

[0116] Through the description of the above implementation methods, technical personnel in the relevant field can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional modules is used as an example. In actual applications, the above-mentioned functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above.

[0117] 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.

[0118] 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.

[0119] 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.

[0120] 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 method for identifying Martian surface minerals based on Transformer, characterized in that: include: Obtain the Mars Mineralogy Standard Hyperspectral Dataset; Performing data preprocessing on the Martian mineral standard hyperspectral dataset to obtain a preprocessed Martian mineral standard hyperspectral dataset; Dividing the preprocessed Mars mineral standard hyperspectral dataset into: a training dataset, a validation dataset, and a test dataset according to a preset division ratio; Constructing a mineral identification Transformer model, wherein the mineral identification Transformer model is used to classify and identify minerals on the surface of Mars; Training the mineral identification Transformer model according to the training data set and the verification data set to obtain an optimized mineral identification Transformer model; Testing the optimized mineral identification Transformer model according to the test data set and determining the test classification accuracy; When the test accuracy meets the preset accuracy condition, the hyperspectral data of the Martian mineral to be identified is input into the optimized mineral identification Transformer model, and the Martian mineral classification and identification result is output.

2. The method according to claim 1, characterized in that The mineral identification Transformer model includes: an embedding mapping function, an encoder, a fully connected layer and an output layer; The embedding mapping function is composed of a band embedding function and a position embedding function, and the encoder is composed of a plurality of encoder layers connected in series.

3. The method according to claim 2, characterized in that The step of training the mineral identification Transformer model according to the training data set and the verification data set to obtain an optimized mineral identification Transformer model includes: Inputting the training data set and the verification data set into the embedding mapping function, and outputting a first embedding representation corresponding to the training data set and a second embedding representation corresponding to the verification data set; Inputting the first embedding representation and the second embedding representation into the encoder, and outputting a first encoder output result corresponding to the first embedding representation and a second encoder output result corresponding to the second embedding representation; Inputting the first encoder output result and the second encoder output result into the fully connected layer, and outputting a first fully connected layer output result corresponding to the first encoder output result and a second fully connected layer output result corresponding to the second encoder output result; Inputting the output result of the first fully connected layer and the output result of the second fully connected layer into the output layer, and outputting a first predicted classification recognition result corresponding to the output result of the first fully connected layer and a second predicted classification recognition result corresponding to the output result of the second fully connected layer; Constructing a loss function according to the first prediction classification recognition result, iteratively optimizing the hyperparameters of the mineral recognition Transformer model so that the loss function converges, and obtaining a hyperparameter training result; The hyperparameter training result is optimized and adjusted according to the second prediction classification recognition result to obtain the optimized mineral recognition Transformer model.

4. The method according to claim 2 or 3, characterized in that: The expression of the embedded mapping function is: Em(x)=Eb(x)+Ep(x); Where, Em(x) represents the embedding mapping function, Eb(x) represents the band embedding function, Ep(x) represents the position embedding function, and x represents the input hyperspectral data; Wherein, the expression of Eb(x) is: Eb(x)=W(x;θ e ); Where W represents a linear mapping, θ e Parameters representing band embedding; The expression of Ep(x) is: Where p represents the position of the band in the hyperspectral data x, p = 0, 1, ..., 232; 2i represents the even dimension of embedding, 2i+1 represents the odd dimension of embedding, and d represents the total dimension of embedding.

5. The method according to claim 2 or 3, characterized in that: The function expression of the encoder is: xn = Encoder(xe); Wherein, xn represents the output result of the encoder, and xe represents the output result of the embedding mapping function; Among them, the expansion of Encoder(xe) is: Where, t (i) ,i=1,2…N represents the output result of the i-th encoder layer, θ n (i) is the parameter of the i-th encoder layer, en (i) represents the i-th encoder layer.

6. The method according to claim 2 or 3, characterized in that: The expression of the fully connected layer is: xlnD{LN[γ(L(xn,θ f ))]}. In the formula, xl represents the output result of the fully connected layer, θ f is the parameter of the fully connected layer, γ represents the Relu activation function, L represents the fully connected layer, LN represents layer normalization, and D represents the Dropout function.

7. The method according to claim 2 or 3, characterized in that: The expression of the output layer is: In the formula, represents the output result of the output layer, and S represents the Softmax activation function.

8. The method according to claim 3, characterized in that The loss function is a cross entropy loss function, and the hyperparameters of the mineral identification Transformer model include: the number of encoder layers, the model dimension, and the number of multi-head attention branches.

9. The method according to claim 1, characterized in that: The expression of the test classification accuracy is: in, Among them, ACC represents the test classification accuracy, n ′ represents the size of the test dataset, Indicates the mineral classification test results, y i ′ Represents the results of mineral classification standards.

10. The method according to claim 1, characterized in that The standard hyperspectral data set of Martian minerals includes: ferromagnesium montmorillonite spectral data, magnesium carbonate spectral data and fayalite spectral data; the preset division ratio is: training data set: verification data set: test data set is 8:1:1.

Images