Mars surface mineral characteristic wave band screening method

Abstract

The invention provides a Mars surface mineral characteristic wave band screening method which comprises the following steps: acquiring a Mars mineral standard hyperspectral data set, dividing the standard hyperspectral data set into a training hyperspectral data set and a testing hyperspectral data set according to a preset division proportion, and respectively carrying out data preprocessing; and constructing a waveband screening supervised learning network model, performing iterative training according to the preprocessed training hyperspectral data set to obtain an optimized waveband screening supervised learning network model, and determining a waveband screening result. And carrying out wave band screening on the preprocessed test hyperspectral data set according to a wave band screening result, and determining test classification accuracy. And under the condition that the test accuracy meets a preset accuracy condition, performing band screening on the to-be-screened hyperspectral data set of the Mars minerals according to the optimized band screening supervised learning network model to obtain a target screening data set. In this way, the Mars surface mineral characteristic wave bands can be accurately screened, and therefore the accuracy of Mars surface mineral recognition and classification is improved.

Description

Technical Field

[0001] The present invention relates to the field of hyperspectral remote sensing technology, and in particular to a method for screening characteristic bands of minerals on the surface of Mars. Background Art

[0002] Hyperspectral remote sensing technology is the primary means of identifying and classifying minerals on the Martian surface. Although hyperspectral remote sensing images of Mars have high spectral dimensions, they contain a significant amount of noise and redundant bands, with only a small fraction of the bands containing mineral information. In practical applications of Martian surface mineral identification and classification, using full-band classification is insufficient to extract mineral information and can lead to overfitting and other issues, resulting in suboptimal classification results. However, band filtering of hyperspectral remote sensing images can effectively extract important information from the spectral data, thereby improving classification results.

[0003] Currently, unsupervised learning-based band screening methods are commonly used, including clustering, sorting, and search-based methods. While these unsupervised learning band screening methods can fully extract mineral spectral and structural features, they are relatively independent of spectral classification processing. This means that the extracted features do not contain mineral classification information. Consequently, they cannot accurately screen for characteristic bands of Martian surface minerals in hyperspectral data, resulting in reduced accuracy in identifying and classifying Martian surface minerals. Summary of the Invention

[0004] The technical problem to be solved by the present invention is that it is impossible to accurately screen the characteristic bands of Martian surface minerals in hyperspectral data, resulting in a decrease in the accuracy of Martian surface mineral identification and classification.

[0005] To solve the above technical problems, the present invention provides a method for screening characteristic bands of Martian surface minerals, which specifically adopts the following technical solutions:

[0006] The present invention provides a method for screening Martian surface minerals using characteristic bands. The method comprises the following steps: first, obtaining a standard hyperspectral dataset of Martian minerals and dividing the dataset into a training hyperspectral dataset and a test hyperspectral dataset according to a preset partitioning ratio. Then, data preprocessing is performed on the training hyperspectral dataset and the test hyperspectral dataset, respectively, to obtain a preprocessed training hyperspectral dataset and a preprocessed test hyperspectral dataset. Secondly, a band screening supervised learning network model is constructed. The band screening supervised learning network model is iteratively trained based on the preprocessed training hyperspectral dataset to obtain an optimized band screening supervised learning network model. Furthermore, a band screening result is determined based on the optimized band screening supervised learning network model. Band screening is performed on the preprocessed test hyperspectral dataset based on the band screening result to obtain a test screening dataset. Mineral classification is performed based on the test screening dataset, and the test classification accuracy is determined. Then, if the test accuracy meets a preset accuracy condition, a hyperspectral dataset of Martian minerals to be screened is obtained. Finally, band screening is performed on the hyperspectral dataset to be screened based on the optimized band screening supervised learning network model to obtain a target screening dataset.

[0007] This method uses a supervised learning network model to filter bands of Martian surface hyperspectral data, achieving dimensionality reduction on the raw data. This method leverages the contextual information of the hyperspectral data and the categorization of different mineral spectra. While extracting the spectral and structural features of Martian hyperspectral data, it also uses spectral categorization as a basis for band filtering. This ensures that the selected bands contain mineral categorization information, further facilitating mineral classification. Furthermore, this method allows for more targeted band filtering of hyperspectral data in diverse application scenarios, demonstrating strong adaptability. This improves the accuracy of Martian surface mineral identification and classification.

[0008] In one optional implementation, the aforementioned band-selection supervised learning network model includes a weight learning unit, a band-selection unit, and a category learning unit. The weight learning unit is connected to the band-selection unit, which in turn is connected to the category learning unit. The weight learning unit consists of a normalization layer (LN), a fully connected layer (FCN), and a Relu activation function, arranged in sequence; the category learning unit consists of a convolutional layer (CONV), a pooling layer, a fully connected layer (FCN), and a Relu activation function, arranged in sequence.

[0009] In this implementation, the band screening supervised learning network model with the above structure can realize the screening and category classification of the Martian mineral hyperspectral dataset.

[0010] In an optional implementation, the above-mentioned iterative training of the band screening supervised learning network model based on the preprocessed training hyperspectral dataset to obtain an optimized band screening supervised learning network model includes: first, inputting the preprocessed training hyperspectral dataset into the weight learning unit to obtain a band weight combination. Then, inputting the band weight combination into the band screening unit to sort the band weight combination in descending order of weight, and determining the training band sub-dataset, wherein the data bands in the training band sub-dataset are: the bands corresponding to the preset number of weights before weight sorting. Further, inputting the training band sub-dataset into the category learning unit for mineral classification, and determining the training mineral classification results. Finally, constructing a loss function based on the training mineral classification results, iteratively optimizing the parameters so that the loss function converges, and obtaining the optimized band screening supervised learning network model.

[0011] In an optional implementation, the band weight in the above band weight combination is expressed as:

[0012] bw=f(x;θ a );

[0013] Where bw represents the band weight, f represents the weight learning unit, x represents the data in the preprocessed training hyperspectral dataset, θ a Represents the parameters of the weight learning unit;

[0014] Specifically, the expansion of f is:

[0015]

[0016] θ a The expression is:

[0017]

[0018] Among them, z i ,i=1,2…na represents the intermediate variable, σ i ,i=1,2…na represents the ReLU activation function of the weight learning unit, H i ,i=1,2…na represents na fully connected layers, norm() represents layer normalization, w i ,i=1,2…na represents the weight of the fully connected layer, b i ,i=1,2…na represents the bias of the fully connected layer, and na is the number of layers of weight learning units.

[0019] In an optional implementation, the output result expression of the above-mentioned category learning unit is:

[0020]

[0021] in, represents the output result of the category learning unit, f represents the category learning unit, x′ represents the data in the training band sub-dataset, θ c Parameters representing the category learning unit;

[0022] Specifically, the expansion of g is:

[0023]

[0024] θ c The expression is:

[0025]

[0026] Among them, s i ,i=1,2…nc represents the intermediate variable, α i ,i=1,2…nc represents the ReLU activation function of the category learning unit, h′ i ,i=1,2…nc represents nc fully connected layers, w′ i ,i=1,2…nc represents the weight of each fully connected layer, b′ i ,i=1,2…nc represents the bias of each fully connected layer, and nc is the number of layers of the category learning unit.

[0027] In an optional implementation, the above loss function is a cross entropy loss function.

[0028] In one optional implementation, performing mineral classification based on the test screening dataset and determining the test classification accuracy includes: first, performing mineral classification using a category learning unit based on the test screening dataset to obtain a mineral classification test result. Then, determining the test classification accuracy based on the mineral classification test result and the mineral classification standard result.

[0029] In an optional implementation, the expression of the above test classification accuracy is:

[0030]

[0031] in,

[0032]

[0033] Among them, ACC represents the test classification accuracy, n′ represents the size of the preprocessed test hyperspectral dataset, Indicates the mineral classification test results, y′ i Indicates the results of mineral classification standards.

[0034] In an optional implementation, the data preprocessing is performed on the training hyperspectral dataset and the test hyperspectral dataset respectively, including at least one of the following: clipping, smoothing, continuum removal, and normalization.

[0035] In one optional implementation, the Mars mineralogy standard hyperspectral dataset is the Compact Imaging Spectrometer for Mars Reconnaissance (CRISM) dataset. This dataset includes the ratio of I radiance / F incident solar radiation. The default partition ratio is 4:1: training hyperspectral dataset: testing hyperspectral dataset. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 A schematic diagram of a process for screening characteristic bands of Martian surface minerals provided by an embodiment of the present invention;

[0037] Figure 2 A schematic diagram of the structure of a band screening supervised learning network model provided in an embodiment of the present invention;

[0038] Figure 3 A schematic diagram of the band screening results provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0039] The following embodiments are described in detail, with examples illustrated in the accompanying drawings. When the following description refers to the drawings, identical numbers in different figures represent identical or similar elements unless otherwise indicated. The embodiments described in the following embodiments are not intended to represent all possible implementations consistent with the present application. They are merely examples of systems and methods consistent with certain aspects of the present application, as detailed in the claims.

[0040] Hyperspectral remote sensing technology is the primary means of identifying and classifying minerals on the Martian surface. Although hyperspectral remote sensing images of Mars have high spectral dimensions, they contain a significant amount of noise and redundant bands, with only a small fraction of the bands containing mineral information. In practical applications of Martian surface mineral identification and classification, using full-band classification is insufficient to extract mineral information and can lead to overfitting and other issues, resulting in suboptimal classification results. However, band filtering of hyperspectral remote sensing images can effectively extract important information from the spectral data, thereby improving classification results.

[0041] Currently, unsupervised learning-based band screening methods are commonly used, including clustering, sorting, and search-based methods. While these unsupervised learning band screening methods can fully extract mineral spectral and structural features, they are relatively independent of spectral classification processing. This means that the extracted features do not contain mineral classification information. This reduces the accuracy of identifying and classifying minerals on the Martian surface.

[0042] In order to solve the above problems, an embodiment of the present application provides a method for screening characteristic bands of minerals on the surface of Mars. This method adopts a supervised learning method. While extracting the spectral characteristics and structural characteristics of Martian hyperspectral data, it uses spectral category information as the basis for band screening, so that the screened bands contain mineral category information, which is more conducive to mineral classification processing, thereby improving the accuracy of identifying and classifying minerals on the surface of Mars.

[0043] The following describes the solution provided by the embodiments of the present application in conjunction with the accompanying drawings.

[0044] For details, see Figure 1 , which is a flow chart of a method for screening characteristic bands of Martian surface minerals provided by an embodiment of the present invention, such as Figure 1 As shown, the method for screening characteristic bands of Martian surface minerals based on hyperspectral data provided by the present invention includes the following steps S101-S109:

[0045] S101. Obtain a Martian mineral standard hyperspectral dataset, and divide the Martian mineral standard hyperspectral dataset into a training hyperspectral dataset and a testing hyperspectral dataset according to a preset division ratio.

[0046] In the embodiment of the present application, the Martian mineralogy standard hyperspectral dataset may be a compact reconnaissance imaging spectromer for Mars (CRISM) dataset.

[0047] In one implementation, the Mars mineral standard hyperspectral dataset can be obtained from a standard spectral library of minerals identified through CRISM analysis (MICA), which can include 24 types of minerals.

[0048] In some embodiments, the Mars mineralogy standard hyperspectral dataset includes the ratio I (radiance value) / F (incident solar radiation), which is the ratio of the radiance value to the incident solar radiation. The I / F ratio is minimally affected by background information in remote sensing images and is most pronounced for absorption features associated with mineral types. Therefore, the Mars mineralogy standard hyperspectral dataset can utilize the I / F ratio.

[0049] In some embodiments, the preset division ratio can be set according to the actual application requirements. For example, the preset division ratio can be: training hyperspectral dataset: test hyperspectral dataset is 4:1. This can improve the training effect of the band screening supervised learning network model.

[0050] S102 , performing data preprocessing on the training hyperspectral dataset and the test hyperspectral dataset respectively to obtain a preprocessed training hyperspectral dataset and a preprocessed test hyperspectral dataset.

[0051] In some embodiments, data preprocessing is performed on the training hyperspectral dataset and the test hyperspectral dataset respectively, including at least one of the following: clipping, smoothing, continuum removal, and normalization.

[0052] Specifically, the cropping 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 cropping process, the 1.0-2.6μm part can be intercepted as the preprocessed training hyperspectral dataset and the preprocessed test hyperspectral dataset. The smoothing process is to perform noise reduction on the training hyperspectral dataset and the test hyperspectral dataset 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 features. The standardization process is to normalize the values ​​of the hyperspectral data to near 0 to make the hyperspectral data in the training hyperspectral dataset and the test hyperspectral dataset more regular.

[0053] S103. Construct a band screening supervised learning network model.

[0054] In some embodiments, Figure 2 A schematic diagram of the structure of the band screening supervised learning network model provided by the embodiment of the present invention is shown as follows: Figure 2 As shown, the band screening supervised learning network model includes: a weight learning unit, a band screening unit, and a category learning unit. The weight learning unit is connected to the band screening unit, and the band screening unit is connected to the category learning unit.

[0055] Specifically, the weight learning unit is composed of the normalization layer LN, the fully connected layer FCN and the Relu activation function arranged in sequence; the category learning unit is composed of the convolution layer CONV, the pooling layer Pooling, the fully connected layer FCN and the Relu activation function arranged in sequence.

[0056] like Figure 2 As shown, the weight learning unit can be used to input the preprocessed training hyperspectral dataset into the weight learning unit to determine the band weight combination. The band screening unit can be used to sort the band weight combinations in descending order of weight and determine the band screening results, i.e., the band sub-datasets. The classification learning unit can be used to determine the mineral classification results based on the band sub-datasets.

[0057] S104. Iteratively train the band screening supervised learning network model according to the preprocessed training hyperspectral data set to obtain an optimized band screening supervised learning network model.

[0058] In some embodiments, Figure 2 Taking the band screening supervised learning network model shown in FIG. 1 as an example, S104 may specifically include the following steps S1041-S1044:

[0059] S1041. Input the preprocessed training hyperspectral data set into the weight learning unit to obtain a band weight combination.

[0060] For example, the preprocessed training hyperspectral dataset is represented by D, then D∈R nXb Where n is the number of data in the preprocessed training hyperspectral dataset, and b is the number of spectral channels, that is, the number of bands, which can be 232, for example.

[0061] After the preprocessed data in the training hyperspectral dataset D enters the weight learning unit, it will pass through multiple combined structures in sequence. Each combined structure consists of a normalization layer LN, a fully connected layer FCN and a ReLU activation function.

[0062] In some embodiments, the band weight in the band weight combination is expressed as:

[0063] bw=f(x;θ a );

[0064] Where bw represents the band weight, f represents the weight learning unit, x represents the data in the preprocessed training hyperspectral dataset, θ a Represents the parameters of the weight learning unit.

[0065] Specifically, the expansion of f is:

[0066]

[0067] θ a The expression is:

[0068] Among them, z i ,i=1,2…na represents the intermediate variable, σ i ,i=1,2…na represents the ReLU activation function of the weight learning unit, h i ,i=1,2…na represents na fully connected layers, norm() represents layer normalization, w i ,i=1,2…na represents the weight of the fully connected layer, b i , i=1,2…na represents the bias of the fully connected layer. The weight and bias together constitute the parameter θ of the weight learning unit a , na is the number of layers of weight learning units. Specifically, na is an adjustable hyperparameter.

[0069] S1042. Input the band weight combination into the band screening unit to sort the band weight combination in descending order of weight, and determine a training band sub-dataset, where the data bands in the training band sub-dataset are: bands corresponding to a preset number of weights before weight sorting.

[0070] Specifically, the band screening unit may sort the band weight combinations in descending order of weight, and use the data of the bands corresponding to the first preset number of weights with the largest weights as the training band sub-datasets.

[0071] For example, the band weight obtained by the weight learning unit can be expressed as:

[0072] bw=[bw1,bw2,…bw b ];

[0073] Sort the elements in bw from largest to smallest. After sorting, their index in bw is:

[0074] i j ,j=1,2…k,i j ∈[1,b];

[0075] Furthermore, the transformation matrix T∈R can be constructed bXk , the elements of T satisfy:

[0076]

[0077] The result of band screening, that is, the training band sub-dataset output by the band screening unit is:

[0078] D′=D·T∈R nXk ;

[0079] After band screening, each training data is represented by x′, x′∈R 1Xk . Use B′ to represent the set of selected bands.

[0080] S1043. Input the training band sub-dataset into the category learning unit to perform mineral classification and determine the training mineral classification result.

[0081] In some embodiments, the output result of the category learning unit is expressed as:

[0082]

[0083] in, represents the output result of the category learning unit, f represents the category learning unit, x′ represents the data in the training band sub-dataset, θ c Represents the parameters of the category learning unit.

[0084] Specifically, the expansion of g is:

[0085]

[0086] θ c The expression is:

[0087]

[0088] Among them, s i ,i=1,2…nc represents the intermediate variable, α i ,i=1,2…nc represents the ReLU activation function of the category learning unit, h′ i ,i=1,2…nc represents nc fully connected layers, w′ i ,i=1,2…nc represents the weight of each fully connected layer, b i ′,i=1,2…nc represents the bias of each fully connected layer. The weights and biases of the fully connected layers constitute the parameters of the class learning unit, and nc is the number of layers in the class learning unit. Specifically, nc is an adjustable hyperparameter.

[0089] S1044. Construct a loss function based on the training mineral classification results, iteratively optimize the parameters to make the loss function converge, and obtain an optimized band screening supervised learning network model.

[0090] In some embodiments, the loss function is a cross-entropy loss function.

[0091] Specifically, the cross entropy loss function can be expressed as:

[0092]

[0093] Wherein, c represents the number of categories. For example, in the embodiment of the present application, c may be 24. i Represents the category label in the classification result of the training data. The model parameters after training are used express.

[0094] S105. Determine the band screening result based on the optimized band screening supervised learning network model.

[0095] For example, the band screening result can be expressed as:

[0096]

[0097] The elements of bw′ are sorted from largest to smallest, and the band data corresponding to the first preset number of the largest weights are used as the training band sub-dataset B′. For example, the preset number can be 50. The preset number can be set according to the needs of the actual application and is not specifically limited in this application.

[0098] S106 , performing band screening on the preprocessed test hyperspectral dataset according to the band screening result to obtain a test screening dataset.

[0099] For example, the preprocessed test hyperspectral dataset can be represented by S, S∈R n′Xb Where n′ is the number of data in the preprocessed test hyperspectral dataset, b is the number of spectral channels, that is, the number of bands, for example, it can be 232. Each test data is represented by x, x∈R 1Xb . Use B to represent the set of original bands,

[0100] According to the training band sub-dataset B′, construct the transformation matrix T∈R bXk , then the test screening data set S′ after band screening is:

[0101] S′=S·T∈R nXk ;

[0102] In which, each test data in S′ is represented by x′, x′∈R 1Xk .

[0103] S107. Classify minerals according to the test screening data set and determine the test classification accuracy.

[0104] Specifically, first, mineral classification can be performed using the category learning unit based on the test screening dataset to obtain the mineral classification test results. Then, the test classification accuracy is determined based on the mineral classification test results and the mineral classification standard results.

[0105] For example, the above x′ is passed through the trained category learning unit to obtain the mineral classification test result

[0106]

[0107] Will By comparing the result y′ with the mineral classification standard, the test classification accuracy can be obtained. In one implementation, the mineral classification standard result y′ can be obtained through the MICA standard spectrum library.

[0108] In some embodiments, the expression for the test classification accuracy is:

[0109]

[0110] in,

[0111]

[0112] Among them, ACC represents the test classification accuracy, n′ represents the size of the preprocessed test hyperspectral dataset, Indicates the mineral classification test results, y′ i Indicates the standard result of mineral classification. For example, the standard result of mineral classification can be a mineral category label, which can be obtained through the MICA standard spectrum library.

[0113] S108. When the test accuracy meets a preset accuracy condition, a hyperspectral dataset of Martian minerals to be screened is obtained.

[0114] Specifically, taking the above-mentioned test classification accuracy ACC as an example, the higher the ACC is, the more conducive the band screening result is to mineral classification processing.

[0115] In some embodiments, corresponding preset accuracy conditions can be set based on different application scenarios and application requirements. For example, corresponding preset accuracy conditions can be set based on different Martian regions. Preset accuracy conditions can be, for example, greater than 85%, greater than 90%, etc. If the test accuracy meets the preset accuracy conditions, it can be determined that the optimized band screening supervised learning network model can meet the application requirements. Furthermore, a hyperspectral dataset of Martian minerals to be screened can be obtained for band screening.

[0116] S109. Perform band screening on the Martian mineral hyperspectral dataset to be screened based on the optimized band screening supervised learning network model to obtain a target screening dataset.

[0117] Specifically, the hyperspectral dataset of Martian minerals to be screened can be input into the optimized band screening supervised learning network model, and then pass through the weight learning unit and the band screening unit in sequence, and finally output the target screening dataset.

[0118] The Martian surface mineral characteristic band screening method based on hyperspectral data provided by an embodiment of the present invention uses a supervised learning network model to perform band screening of Martian surface hyperspectral data, achieving dimensionality reduction processing of the original data. This method fully utilizes the contextual information of the hyperspectral data and the category information of different mineral spectra. While extracting the spectral characteristics and structural characteristics of the Martian hyperspectral data, it uses the spectral category information as the basis for band screening, ensuring that the filtered bands contain mineral category information, which is more conducive to mineral classification processing. At the same time, the method can also perform more targeted band screening of hyperspectral data in different application scenarios, with strong adaptability. In this way, the accuracy of identifying and classifying Martian surface minerals can be improved.

[0119] In some embodiments, in order to verify the effectiveness of the method for screening characteristic bands of Martian surface minerals based on hyperspectral data provided by the embodiments of the present invention, the CRISM Martian mineral standard hyperspectral dataset was obtained through the MICA standard spectral library, and divided into a training hyperspectral dataset and a test hyperspectral dataset according to a 4:1 ratio, and then clipped, smoothed, continuum removed, and normalized. Then, a method was constructed as follows: Figure 2 The band screening supervised learning network model shown specifically includes: a weight learning unit, a band screening unit and a category learning unit. Then, the preprocessed training hyperspectral dataset is input into the band screening supervised learning network model for iterative training. Specifically, it includes constructing a loss function, iteratively optimizing parameters, and gradually reducing the loss function until convergence to complete model training. In this embodiment, the loss function is a cross-entropy loss function, and the number of training rounds is set to 20. Next, the preprocessed test hyperspectral dataset is input into the trained optimized band screening supervised learning network model. A subset of the Martian mineral spectral bands of size 50 is output from the band screening unit to obtain a screening dataset. Figure 3 A schematic diagram of the band screening results provided by an embodiment of the present invention, such as Figure 3 As shown, the original spectral curve is curve B, and the curve segment on curve B that overlaps with region A is the spectral curve corresponding to the band screening result. The band screening method provided by the embodiment of the present invention can effectively screen the characteristic bands of Martian surface minerals, thereby improving the accuracy of Martian surface mineral identification and classification.

[0120] 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.

[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 to indicate or imply relative importance or implicitly specify the number of the technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include at least one of such features. In the description of the present invention, "plurality" means at least two, for example, two, three, etc., unless otherwise specifically defined.

[0122] In the present invention, unless otherwise specified or limited, the terms "installed," "connected," "connect," "fixed," etc. should be understood in a broad sense. For example, they can refer to fixed connection, detachable connection, or integration; mechanical connection, electrical connection; direct connection, or indirect connection through an intermediate medium; internal communication between two components, or interaction between two components, unless otherwise specified. Those skilled in the art will understand the specific meanings of the above terms in the present invention based on specific circumstances.

[0123] In the description of this specification, the reference terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" mean 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 can be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art can combine and combine different embodiments or examples described in this specification and features of 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 overall concept of this application and do not constitute a limitation on the scope of protection of this application. For those skilled in the art, any other implementation methods expanded based on the scheme of this application without expending creative work shall fall within the scope of protection of this application.

Claims

1. A method for screening characteristic bands of Martian surface minerals, characterized in that: include: Acquire a Martian mineral standard hyperspectral dataset, and divide the Martian mineral standard hyperspectral dataset into a training hyperspectral dataset and a test hyperspectral dataset according to a preset division ratio; Preprocessing the training hyperspectral dataset and the testing hyperspectral dataset respectively to obtain a preprocessed training hyperspectral dataset and a preprocessed testing hyperspectral dataset; Construct a band screening supervised learning network model; Iteratively training the band screening supervised learning network model according to the preprocessed training hyperspectral data set to obtain an optimized band screening supervised learning network model; Determining a band screening result according to the optimized band screening supervised learning network model; Performing band screening on the preprocessed test hyperspectral data set according to the band screening result to obtain a test screening data set; Classifying minerals according to the test screening data set and determining the test classification accuracy; When the test accuracy meets a preset accuracy condition, obtaining a hyperspectral data set of Martian minerals to be screened; The Martian mineral hyperspectral data set to be screened is subjected to band screening according to the optimized band screening supervised learning network model to obtain a target screening data set.

2. The method according to claim 1, characterized in that The band screening supervised learning network model comprises: a weight learning unit, a band screening unit and a category learning unit; the weight learning unit is connected to the band screening unit, and the band screening unit is connected to the category learning unit; Among them, the weight learning unit is composed of a normalization layer LN, a fully connected layer FCN and a Relu activation function arranged in sequence; the category learning unit is composed of a convolutional layer CONV, a pooling layer Pooling, a fully connected layer FCN and a Relu activation function arranged in sequence.

3. The method according to claim 2, characterized in that The iterative training of the band screening supervised learning network model according to the preprocessed training hyperspectral data set to obtain an optimized band screening supervised learning network model includes: Inputting the preprocessed training hyperspectral data set into the weight learning unit to obtain a band weight combination; The band weight combination is input into the band screening unit to sort the band weight combination in descending order of weight, and determine a training band sub-data set, wherein the data bands in the training band sub-data set are: bands corresponding to a preset number of weights before weight sorting; Inputting the training band sub-data set into the category learning unit to perform mineral classification and determine the training mineral classification result; A loss function is constructed according to the training mineral classification results, and parameters are iteratively optimized to make the loss function converge, thereby obtaining the optimized band screening supervised learning network model.

4. The method according to claim 3, characterized in that The expression of the band weight in the band weight combination is: bw=f(x;θ a ); Where bw represents the band weight, f represents the weight learning unit, x represents the data in the preprocessed training hyperspectral dataset, θ a Represents the parameters of the weight learning unit; Specifically, the expansion of f is: θ a The expression is: Among them, z i , i=1,2…na represents the intermediate variable, σ i ,i=1,2…na represents the Relu activation function of the weight learning unit, h i ,i=1,2…na represents na fully connected layers, norm() represents layer normalization, w i ,i=1,2…na represents the weight of the fully connected layer, b i ,i=1,2…na represents the bias of the fully connected layer, and na is the number of layers of the weight learning unit.

5. The method according to claim 3, characterized in that: The output result expression of the category learning unit is: in, represents the output result of the category learning unit, f represents the category learning unit, x′ represents the data in the training band sub-dataset, θ c Represents the parameters of the category learning unit; Specifically, the expansion of g is: θ c The expression is: Among them, s i ,i=1,2…nc represents the intermediate variable, α i ,i=1,2…nc represents the ReLU activation function of the category learning unit, H′ i ,i=1,2…nc represents nc fully connected layers, w′ i ,i=1,2…nc represents the weight of each fully connected layer, b′ i ,i=1,2…nc represents the bias of each fully connected layer, and nc is the number of layers of the category learning unit.

6. The method according to claim 3, characterized in that The loss function is a cross entropy loss function.

7. The method according to claim 2, characterized in that The method of classifying minerals according to the test screening data set and determining the test classification accuracy rate includes: According to the test screening data set, mineral classification is performed through the category learning unit to obtain a mineral classification test result; The test classification accuracy is determined based on the mineral classification test results and the mineral classification standard results.

8. The method according to claim 7, 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 preprocessed test hyperspectral dataset, Indicates the mineral classification test results, y′ i Represents the results of mineral classification standards.

9. The method according to claim 1, characterized in that: The data preprocessing performed on the training hyperspectral data set and the test hyperspectral data set respectively includes at least one of the following: clipping, smoothing, continuum removal and standardization.

10. The method according to claim 1, characterized in that The Mars mineral standard hyperspectral dataset is the Compact Mars Reconnaissance Imaging Spectrometer CRISM dataset; the Mars mineral standard hyperspectral dataset includes the ratio I radiation value / F incident solar radiation; the preset division ratio is: training hyperspectral dataset: test hyperspectral dataset is 4:1.

Images