Mars mineral abundance inversion method and system

Abstract

The invention provides a Mars mineral abundance inversion method and system. The Mars mineral abundance inversion method comprises the steps of firstly obtaining target Mars hyperspectral reflectivity data; and then generating simulated Mars hyperspectral image data based on the target Mars hyperspectral reflectivity data. Secondly, training the constructed mineral abundance inversion deep learning model based on the simulated Mars hyperspectral image data, and determining the trained mineral abundance inversion deep learning model; thirdly, acquiring to-be-processed Mars hyperspectral image data, and performing preprocessing; and finally, performing mineral abundance inversion on the preprocessed to-be-processed Mars hyperspectral image data through the trained mineral abundance inversion deep learning model to obtain a mineral abundance inversion result. Therefore, the weight distribution of each wave band of the hyperspectral image can be enhanced through the mineral abundance inversion deep learning model, and the non-linear modeling capability and the wave band characteristic response capability of abundance inversion are improved, so that the accuracy and the efficiency of Mars mineral abundance inversion can be effectively improved.

Description

Technical Field

[0001] This invention relates to the field of hyperspectral remote sensing technology, specifically to a method and system for inverting the mineral abundance of Mars. Background Technology

[0002] Mars exploration is one of the core missions in the field of deep space exploration. Martian mineral abundance inversion is a key technology for analyzing the composition of surface materials, and it is of great significance for resource exploration, geological evolution research, and manned mission planning. With the continuous advancement of Mars exploration technology, the acquired Martian hyperspectral imagery data is gradually becoming an important data source for Martian mineral resource research. Martian mineral abundance information plays a crucial role in deeply understanding its geological evolution history, finding potential mineral resources, and providing support for future Mars exploration missions.

[0003] In the field of Mars mineral abundance inversion, relevant methods include least squares and linear mixture models. These methods estimate the abundance distribution of different minerals by analyzing the spectral characteristics in remote sensing data.

[0004] However, with the increase in hyperspectral image resolution and the surge in data volume, especially for highly complex hyperspectral remote sensing data, the limitations of existing Martian mineral abundance inversion methods are significant. These limitations include: insufficient data processing capabilities, making it difficult to handle high-dimensional, large-scale data; difficulty in modeling nonlinear features, particularly nonlinear spectral mixing features; insufficient identification of band value and inadequate feature utilization; weak cross-layer information transfer, affecting inversion accuracy; and the potential for results to deviate from reality by ignoring unmixing physical constraints. Furthermore, problems such as unclear directionality, poor sparsity, and weak interpretability limit their application in the complex Martian environment. Consequently, these technologies cannot efficiently and accurately invert Martian mineral abundance. Summary of the Invention

[0005] The technical problem to be solved by this invention is the inability to efficiently and accurately perform Martian mineral abundance inversion.

[0006] To address the aforementioned technical problems, this invention provides a method and system for inverting Martian mineral abundance, specifically employing the following technical solution: In a first aspect, the present invention provides a method for inverting Martian mineral abundance. The method includes: first, acquiring target Martian hyperspectral reflectance data, which includes hyperspectral reflectance data of multiple target Martian minerals; then, generating simulated Martian hyperspectral image data based on the target Martian hyperspectral reflectance data; next, constructing a deep learning model for mineral abundance inversion and training the model based on the simulated Martian hyperspectral image data to determine the trained model; further, acquiring Martian hyperspectral image data to be processed; performing data preprocessing on the Martian hyperspectral image data to obtain preprocessed Martian hyperspectral image data; and finally, performing mineral abundance inversion on the preprocessed Martian hyperspectral image data based on the trained deep learning model to obtain the mineral abundance inversion result.

[0007] This method first acquires hyperspectral reflectance data of the target Mars. Then, it generates simulated Martian hyperspectral image data based on this data. Next, it trains a deep learning model for mineral abundance inversion using the simulated Martian hyperspectral image data, determining the trained model. Then, it acquires and preprocesses the Martian hyperspectral image data to be processed, obtaining preprocessed data. Finally, it uses the trained deep learning model to perform mineral abundance inversion on the preprocessed image data to determine the obtained results. This deep learning model strengthens the weight allocation of each band in the hyperspectral image during mineral abundance inversion, giving the abundance estimation nonlinear jump connectivity, significantly improving the nonlinear modeling capability and band feature response capability of abundance inversion. It also improves the accuracy and training efficiency of mineral abundance inversion, exhibiting higher adaptability and stronger practical effects, thus effectively improving the accuracy and efficiency of Martian mineral abundance inversion.

[0008] In conjunction with the first aspect, in one alternative implementation, the aforementioned deep learning model for mineral abundance inversion includes an encoder module and a decoder module. The encoder module has a nonlinear skip connection structure. Specifically, the encoder module can be used to extract features from the input hyperspectral image data based on deep autoencoders and channel attention mechanisms to obtain hyperspectral feature data; the input hyperspectral image data can be simulated Mars hyperspectral image data, or preprocessed Mars hyperspectral image data to be processed. The decoder module can be used to perform linear and nonlinear decoding based on the hyperspectral feature data, outputting mineral abundance inversion results; the mineral abundance inversion results include: abundance map data and endmember spectral data corresponding to Martian minerals, with the endmember spectral data used to characterize the spectral reflectance curves of Martian minerals.

[0009] In conjunction with the first aspect, in one alternative implementation, the encoder module includes a channel attention module and a residual connection module. The channel attention module can be used to determine the weights corresponding to each band in the input hyperspectral image data based on a channel attention mechanism and perform feature extraction, outputting weighted feature data. The residual connection module can be used to extract features from the weighted feature data based on a nonlinear skip connection structure to obtain hyperspectral feature data.

[0010] In conjunction with the first aspect, in one alternative implementation, the aforementioned decoder module includes a linear decoder and a nonlinear decoder. The linear decoder can be used to perform a linear mapping on the hyperspectral feature data to determine the abundance information of Martian minerals, outputting linearly decoded feature data. The nonlinear decoder can be used to perform a nonlinear mapping on the linearly decoded feature data to determine the mineral abundance inversion result.

[0011] In conjunction with the first aspect, in one alternative implementation, the decoder module can also be used to generate reconstructed hyperspectral image data based on abundance map data and endmember spectral data, which is then used for inversion verification.

[0012] In conjunction with the first aspect, in one alternative implementation, the above-mentioned generation of simulated Mars hyperspectral image data based on the target Mars hyperspectral reflectance data specifically includes: modeling based on the target Mars hyperspectral reflectance data using a target hybrid model combined with preset signal-to-noise ratio parameters to generate simulated Mars hyperspectral image data.

[0013] In conjunction with the first aspect, in one alternative implementation, the aforementioned target hybrid model is one of the following models: linear hybrid model, bilinear hybrid model, or nonlinear hybrid model.

[0014] In conjunction with the first aspect, in one alternative implementation, the above-mentioned data preprocessing of the Mars hyperspectral image data to be processed includes at least one of the following: format conversion processing, masking processing, and normalization processing.

[0015] In conjunction with the first aspect, in one alternative implementation, the aforementioned target Martian hyperspectral reflectance data is either CRISM Mars Mineral Spectral Library data or USGS Mars Mineral Spectral Library data.

[0016] Secondly, this invention provides a Mars mineral abundance inversion system, comprising: a first acquisition module, a data synthesis module, a model training module, a second acquisition module, a data preprocessing module, and a mineral abundance inversion module. The first acquisition module acquires target Martian hyperspectral reflectance data, which includes hyperspectral reflectance data of multiple target Martian minerals. The data synthesis module generates simulated Martian hyperspectral image data based on the target Martian hyperspectral reflectance data. The model training module constructs a deep learning model for mineral abundance inversion and trains the model based on the simulated Martian hyperspectral image data to determine the trained model. The second acquisition module acquires Martian hyperspectral image data to be processed. The data preprocessing module preprocesses the data to be processed, obtaining preprocessed Martian hyperspectral image data. The mineral abundance inversion module performs mineral abundance inversion on the preprocessed Martian hyperspectral image data based on the trained deep learning model, obtaining the mineral abundance inversion result.

[0017] Thirdly, the present invention provides an electronic device, comprising: a memory and one or more processors; the memory being coupled to the processors; wherein the memory stores computer program code, the computer program code including computer instructions, which, when executed by the processor, cause the electronic device to perform the method provided by the first aspect and any of its alternative implementations.

[0018] Fourthly, the present invention provides a computer-readable storage medium including computer instructions that, when executed on an electronic device, cause the electronic device to perform the method provided by the first aspect and any alternative implementation thereof.

[0019] Understandably, the beneficial effects of the Mars mineral abundance inversion system provided in the second aspect, the electronic equipment in the third aspect, and the computer-readable storage medium in the fourth aspect can be referenced to the beneficial effects of the first aspect and any of its possible design embodiments, which will not be elaborated here. Attached Figure Description

[0020] Figure 1 A schematic flowchart illustrating the Martian mineral abundance inversion method provided in this application embodiment; Figure 2 This is a schematic diagram of the structure of the deep learning model for mineral abundance inversion provided in the embodiments of this application; Figure 3 This is a schematic diagram of the data processing flow of the deep learning model for mineral abundance inversion provided in the embodiments of this application; Figure 4This is a schematic diagram of the structure of the Mars mineral abundance inversion system provided in an embodiment of this application. Detailed Implementation

[0021] The embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described below do not represent all embodiments consistent with this application. They are merely examples of systems and methods consistent with some aspects of this application as detailed in the claims.

[0022] Mars exploration is one of the core missions in the field of deep space exploration. Martian mineral abundance inversion is a key technology for analyzing the composition of surface materials, and it is of great significance for resource exploration, geological evolution research, and manned mission planning. With the continuous advancement of Mars exploration technology, the acquired Martian hyperspectral imagery data is gradually becoming an important data source for Martian mineral resource research. Martian mineral abundance information plays a crucial role in deeply understanding its geological evolution history, finding potential mineral resources, and providing support for future Mars exploration missions.

[0023] In the field of Mars mineral abundance inversion, relevant methods include least squares and linear mixture models. These methods estimate the abundance distribution of different minerals by analyzing the spectral characteristics in remote sensing data.

[0024] However, with the increase in hyperspectral image resolution and the surge in data volume, especially for highly complex hyperspectral remote sensing data, the limitations of existing Martian mineral abundance inversion methods are significant. These limitations include: insufficient data processing capabilities, making it difficult to handle high-dimensional, large-scale data; difficulty in modeling nonlinear features, particularly nonlinear spectral mixing features; insufficient identification of band value and inadequate feature utilization; weak cross-layer information transfer, affecting inversion accuracy; and the potential for results to deviate from reality by ignoring unmixing physical constraints. Furthermore, problems such as unclear directionality, poor sparsity, and weak interpretability limit their application in the complex Martian environment. Consequently, these technologies cannot efficiently and accurately invert Martian mineral abundance.

[0025] To address the aforementioned issues, this application provides a method and system for inverting Martian mineral abundance. This method generates simulated Martian hyperspectral image data based on target Martian hyperspectral reflectance data. Then, a deep learning model for mineral abundance inversion is trained using the simulated Martian hyperspectral image data to determine the trained model. Finally, the trained deep learning model is used to invert mineral abundance in the Martian hyperspectral image data to be processed, thus obtaining the inversion result. The deep learning model is based on a deep autoencoder and channel attention mechanism, and features a nonlinear skip connection structure. This strengthens the weight allocation of each band in the hyperspectral image during mineral abundance inversion, giving the abundance estimation nonlinear skip connection characteristics. This significantly improves the nonlinear modeling capability and band feature response capability of abundance inversion, while also increasing the accuracy and training efficiency of mineral abundance inversion, and exhibiting higher adaptability and stronger practical effects.

[0026] The solutions provided in the embodiments of this application will be described below with reference to the accompanying drawings.

[0027] For details, see Figure 1 The above is a flowchart illustrating the Martian mineral abundance inversion method provided in this application embodiment. Figure 1 As shown, the Martian mineral abundance inversion method provided by this invention includes the following steps S101-S106: S101. Obtain hyperspectral reflectance data of the target Mars.

[0028] In this embodiment of the application, the target Martian hyperspectral reflectance data includes hyperspectral reflectance data of multiple Martian target inversion minerals. The Martian target inversion minerals are those whose abundance is to be determined.

[0029] For example, target Martian hyperspectral reflectance data is typically stored in CSV file format. The first column of the CSV file is the wavelength data, and each subsequent column represents the reflectance information corresponding to different Martian target inverted minerals.

[0030] In some embodiments, the target Martian hyperspectral reflectance data may be: data from the Compact Reconnaissance Imaging Spectroradiometer for Mars (CRISM) Martian Mineral Spectral Library, or data from the USGS Martian Mineral Spectral Library.

[0031] S102. Generate simulated Mars hyperspectral image data based on the target Mars hyperspectral reflectance data.

[0032] Then, based on the target Martian hyperspectral reflectance data obtained from S101, simulated Martian hyperspectral image data can be generated for subsequent training of a deep learning model for mineral abundance inversion.

[0033] In one implementation, based on the target Mars hyperspectral reflectance data, not only can the above-mentioned simulated Mars hyperspectral image data be generated, but also a false-color image corresponding to the simulated Mars hyperspectral image data can be generated to display the visual effect of the simulated Mars hyperspectral image data through the false-color image.

[0034] In some embodiments, the specific method for generating simulated Mars hyperspectral image data in step S102 may include: Based on the target Mars hyperspectral reflectance data, a target hybrid model is used in conjunction with preset signal-to-noise ratio parameters to generate simulated Mars hyperspectral image data.

[0035] Specifically, based on the target Martian hyperspectral reflectance data, the abundance of different Martian minerals can be modeled and images generated through the target fusion model. By using different preset signal-to-noise ratio parameters, simulated Martian hyperspectral image data that conforms to different noise environments can be generated as a dataset for training the mineral abundance inversion deep learning model.

[0036] For example, the preset signal-to-noise ratio (SNR) parameters mentioned above can be 10, 20, and 30, etc. The specific values ​​of the preset SNR parameters can be preset according to actual application requirements, and this application does not impose specific limitations on them.

[0037] In some embodiments, the target mixture model is one of the following models: Linear Mixed Model (LMM), Bilinear Mixed Model (BiLMM), or Nonlinear Mixed Model (NLMM).

[0038] Specifically, the linear mixture model assumes that mineral spectra are only linearly superimposed and have no interactions. It is the simplest and most efficient in computation, but has low accuracy and is suitable for simple mixture scenarios. The bilinear mixture model adds simple quadratic interactions between two endmembers to the linear mixture model. Its complexity and accuracy are moderate and it is suitable for scenarios with basic interactions. The nonlinear mixture model considers complex nonlinear interactions between minerals. It is computationally complex, but has the highest accuracy and is suitable for complex real-world scenarios.

[0039] S103. Construct a deep learning model for mineral abundance inversion, and train the deep learning model for mineral abundance inversion based on simulated Mars hyperspectral image data to determine the trained deep learning model for mineral abundance inversion.

[0040] Next, a deep learning model for mineral abundance inversion based on a deep autoencoder and channel attention mechanism can be constructed. Then, the deep learning model is trained using simulated Martian hyperspectral image data generated by S102 to improve its accuracy and robustness in mineral abundance inversion. The trained deep learning model is then determined for use in retrieving mineral abundance from the Martian hyperspectral image data to be processed. Specifically, the deep learning model is trained, i.e., the model weight parameters are trained. The trained deep learning model includes the trained model weight parameters (e.g., in a .sth file).

[0041] In some embodiments, Figure 2 This is a schematic diagram of the structure of the deep learning model for mineral abundance inversion provided in the embodiments of this application, as shown below. Figure 2 As shown, the mineral abundance inversion deep learning model 200 includes an encoder module 210 and a decoder module 220. The encoder module 210 has a nonlinear skip connection structure.

[0042] Figure 3 This is a schematic diagram of the data processing flow of the deep learning model for mineral abundance inversion provided in the embodiments of this application, as shown below. Figure 3 As shown, the mineral abundance inversion deep learning model is architecturally divided into an input layer, hidden layers, and an output layer, and functionally divided into an encoder module and a decoder module. Specifically, the encoder module can be used to extract features from the input hyperspectral image data based on deep autoencoders and channel attention mechanisms to obtain hyperspectral feature data. The input hyperspectral image data can be either simulated Mars hyperspectral image data or preprocessed Mars hyperspectral image data to be processed.

[0043] The decoder module can perform linear and nonlinear decoding based on hyperspectral feature data, outputting mineral abundance inversion results. These results include abundance map data and endmember spectral data corresponding to Martian minerals. Specifically, the abundance map data can be used to characterize the abundance value of each Martian mineral in different pixels, and the endmember spectral data can be used to characterize the spectral reflectance curves of the Martian minerals.

[0044] Specifically, the mineral abundance inversion deep learning model 200, by combining an encoder module 210 and a decoder module 220, can learn mineral abundance information from hyperspectral images during training. The encoder module 210 can extract features using channel attention mechanisms and residual blocks, while the decoder module 220 can output mineral abundance information and endmember spectral information through linear and nonlinear decoding, ultimately generating accurate abundance map data and endmember spectral data as the mineral abundance inversion result.

[0045] In some embodiments, such as Figure 2 and Figure 3 As shown, the encoder module 210 includes a channel attention module 211 and a residual connection module 212. The channel attention module 211 is used to determine the weights corresponding to each band in the input hyperspectral image data based on a channel attention mechanism and to extract features, outputting weighted feature data. The residual connection module 212 is used to extract features from the weighted feature data based on a nonlinear skip connection structure to obtain hyperspectral feature data.

[0046] Specifically, the channel attention module 211 can extract the correlation between different channels in hyperspectral image data based on the channel attention mechanism, so as to better focus on important feature information. The channel attention module 211 can determine the importance of bands in the hyperspectral image data according to the input hyperspectral image data and assign corresponding weights to them, thereby enhancing the discrimination ability of the mineral abundance inversion deep learning model. For example, for a hyperspectral image data of size H×W×B (e.g., hyperspectral image feature map X), where B represents the number of bands, global max pooling and global average pooling operations are first performed on each band along the channel dimension to obtain two band description vectors of size 1×1×B. These two band description vectors represent the strongest response and average response of each band in space, respectively, reflecting its contribution to endmember features in all bands. Next, the two band description vectors are input into a shared band importance extraction network. This network employs a two-layer multilayer perceptron (MLP). The first layer contains B / r neurons and uses the ReLU activation function, while the second layer recovers to B neurons. The MLP's role is to extract the importance weight of each band from the compressed band statistical features. Finally, the outputs of the two MLPs are added together and normalized using the Sigmoid activation function to obtain a 1×1×B band attention weight vector Mb(X), representing the relative importance of different bands in the current hyperspectral image data processing. This band attention weight vector is then weighted band-by-band (i.e., element-wise multiplication) with the original hyperspectral image feature map X to enhance bands with high information content and strong discriminative power, suppress redundant or noisy bands, and thus improve the accuracy of endmember identification and abundance estimation. Finally, the band attention weight vector is multiplied element-wise with the original hyperspectral image feature map X to obtain the feature map after channel attention weight allocation, i.e., the feature data after weight allocation.

[0047] The residual connection module 212 can be used to extract features from the weighted feature data based on the nonlinear jump connection structure to obtain hyperspectral feature data.

[0048] The residual connection module 212 features a nonlinear skip connection structure. This structure helps address the vanishing gradient problem, maintains feature fluidity, and improves the training efficiency of the mineral abundance inversion deep learning model. Specifically, the residual connection module 212 helps capture complex nonlinear relationships, enhancing the performance of the mineral abundance inversion deep learning model. The nonlinear skip connection structure allows input data to bypass intermediate nonlinear layers and be directly added to the output, thereby alleviating the training difficulty of deep networks, accelerating convergence, and suppressing gradient vanishing.

[0049] For example, suppose the input is The residual connection module 212 contains a series of nonlinear mappings. Then the output of residual connection module 212 The expression is: ; in, It typically consists of 1-3 linear layers (or convolutional layers) and activation functions. You can skip the intermediate mapping directly, and Addition. Activation functions (e.g., ReLU function, LeakyReLU function) can be located in the middle or at the end. In the abundance inversion processing of hyperspectral image data: the input hyperspectral image data has high spectral dimension and complex data features (sparse mixing, nonlinear interference). The residual connection module 212 can improve the stability and expressive power of abundance estimation or endmember separation.

[0050] In some embodiments, such as Figure 2 As shown, the decoder module 220 includes a linear decoder 221 and a nonlinear decoder 222. The linear decoder 221 can be used to perform linear mapping on hyperspectral feature data to determine the abundance information of Martian minerals, outputting linearly decoded feature data. The nonlinear decoder 222 can be used to perform nonlinear mapping on the linearly decoded feature data to determine the mineral abundance inversion result.

[0051] Specifically, the linear decoder 221 can be used to recover the abundance information of Martian minerals from the hyperspectral feature data extracted by the encoder module. The linear decoder 221 can map the high-dimensional hyperspectral feature data to the mineral abundance estimation results through a linear layer. The nonlinear decoder 222 can be used to further optimize the linear decoded feature data to generate more complex mineral abundance maps in a nonlinear manner, capturing more details.

[0052] In some embodiments, such as Figure 3As shown, the decoder module can also be used to generate reconstructed hyperspectral image data based on abundance map data and endmember spectral data. This reconstructed hyperspectral image data can be used for inversion verification. Specifically, the reconstructed hyperspectral image data can be compared with the input hyperspectral image data to verify the processing accuracy of the mineral abundance inversion deep learning model and ensure that no key information is lost during the mineral inversion process.

[0053] S104. Acquire the Mars hyperspectral image data to be processed.

[0054] Specifically, the hyperspectral image data of Mars to be processed is the hyperspectral image data of Mars to be processed for mineral inversion. The hyperspectral image data of Mars to be processed includes hyperspectral image data of multiple Martian minerals, which can correspond to the Martian target inversion minerals obtained in S101.

[0055] S105. Perform data preprocessing on the Mars hyperspectral image data to be processed to obtain the preprocessed Mars hyperspectral image data to be processed.

[0056] In some embodiments, data preprocessing is performed on the hyperspectral image data of Mars to be processed, including at least one of the following: format conversion processing, masking processing, and normalization processing.

[0057] Specifically, format conversion can transform the Martian hyperspectral image data to be processed into a data format compatible with the mineral abundance inversion deep learning model (e.g., .mat or .hdr files), facilitating processing by the model. Masking can filter and retain target regions from the Martian hyperspectral image data, removing invalid or irrelevant areas to reduce interference and focus on core data. Normalization eliminates dimensional and scale differences between different bands in the Martian hyperspectral image data, preventing the mineral abundance inversion deep learning model from biasing towards high numerical features. Furthermore, the Martian hyperspectral image data can be mapped to a fixed range (e.g., [0,1]) to make the data comparable, ensuring stable model convergence, improving accuracy, and enhancing the data's cross-scene versatility.

[0058] S106. Based on the trained mineral abundance inversion deep learning model, mineral abundance inversion is performed on the preprocessed Martian hyperspectral image data to be processed, and the mineral abundance inversion results are obtained.

[0059] Finally, the preprocessed Martian hyperspectral image data from S105 can be input into the mineral abundance inversion deep learning model trained in S103 to perform mineral abundance inversion and obtain the mineral abundance inversion results. Specifically, the mineral abundance inversion results include: abundance map data and endmember spectral data corresponding to Martian minerals.

[0060] The method for inverting Martian mineral abundance provided in the above embodiments of this application first acquires target Martian hyperspectral reflectance data. Then, simulated Martian hyperspectral image data is generated based on the target Martian hyperspectral reflectance data. Next, the constructed deep learning model for mineral abundance inversion is trained based on the simulated Martian hyperspectral image data to determine the trained model. Then, Martian hyperspectral image data to be processed is acquired and preprocessed to obtain preprocessed Martian hyperspectral image data. Finally, the trained deep learning model for mineral abundance inversion is used to perform mineral abundance inversion on the preprocessed Martian hyperspectral image data to determine the obtained mineral abundance inversion result. The deep learning model for mineral abundance inversion is constructed based on a deep autoencoder and a channel attention mechanism and has a nonlinear skip connection structure. In this way, the deep learning model for mineral abundance inversion can strengthen the weight allocation of each band of the hyperspectral image in mineral abundance inversion, making the abundance estimation have nonlinear jump connection characteristics, significantly improving the nonlinear modeling ability and band feature response ability of abundance inversion, while improving the accuracy and training efficiency of mineral abundance inversion, and having higher adaptability and stronger practical effect, thereby effectively improving the accuracy and efficiency of Martian mineral abundance inversion.

[0061] This application also provides a Martian mineral abundance inversion system, specifically... Figure 4 This is a schematic diagram of the structure of the Mars mineral abundance inversion system provided in the embodiments of this application, as shown below. Figure 4 As shown, the Mars mineral abundance inversion system 400 includes: a first acquisition module 401, a data synthesis module 402, a model training module 403, a second acquisition module 404, a data preprocessing module 405, and a mineral abundance inversion module 406.

[0062] The first acquisition module 401 can be used to acquire target Martian hyperspectral reflectance data. This target Martian hyperspectral reflectance data may include hyperspectral reflectance data of inverted minerals from multiple Martian targets.

[0063] Specifically, taking the target Mars hyperspectral reflectance data in CSV file format as an example, the first acquisition module 401 can import the target Mars hyperspectral reflectance data in CSV file format.

[0064] The data synthesis module 402 can be used to generate simulated Mars hyperspectral image data based on the target Mars hyperspectral reflectance data.

[0065] The model training module 403 can be used to construct a deep learning model for mineral abundance inversion and train the deep learning model for mineral abundance inversion based on simulated Mars hyperspectral image data to determine the trained deep learning model for mineral abundance inversion.

[0066] In one implementation, the model training module 403 can display the training progress in real time during the training of the mineral abundance inversion deep learning model, and generate the corresponding model file (e.g., model weight parameter file) and end metadata after the training is completed.

[0067] The second acquisition module 404 can be used to acquire Mars hyperspectral image data to be processed.

[0068] In one implementation, the second acquisition module 404 can import the Martian hyperspectral image data to be processed, supporting .mat or .hdr formats. After importing, the second acquisition module 404 will detect whether the number of bands in the Martian hyperspectral image data to be processed is consistent with the imported target Martian hyperspectral reflectance data, and generate a corresponding false-color image for display, facilitating subsequent mineral abundance inversion processing.

[0069] The data preprocessing module 405 can be used to preprocess the Martian hyperspectral image data to be processed, and obtain the preprocessed Martian hyperspectral image data to be processed.

[0070] The mineral abundance inversion module 406 can be used to perform mineral abundance inversion on preprocessed Martian hyperspectral image data based on a trained mineral abundance inversion deep learning model, and obtain the mineral abundance inversion results. Furthermore, the mineral abundance inversion module 406 can visualize the mineral abundance inversion results.

[0071] The Mars mineral abundance inversion system provided in the above embodiments of this application first acquires target Martian hyperspectral reflectance data through a first acquisition module. Then, a data synthesis module generates simulated Martian hyperspectral image data based on the target Martian hyperspectral reflectance data. Next, a model training module trains the constructed mineral abundance inversion deep learning model based on the simulated Martian hyperspectral image data to determine the trained model. Next, a second acquisition module acquires Martian hyperspectral image data to be processed, and a data preprocessing module performs data preprocessing to obtain preprocessed Martian hyperspectral image data. Finally, a mineral abundance inversion module performs mineral abundance inversion on the preprocessed Martian hyperspectral image data based on the trained deep learning model to determine the obtained mineral abundance inversion result. The mineral abundance inversion deep learning model is constructed based on a deep autoencoder and a channel attention mechanism and has a nonlinear skip connection structure. In this way, the deep learning model for mineral abundance inversion can strengthen the weight allocation of each band of the hyperspectral image in mineral abundance inversion, making the abundance estimation have nonlinear jump connection characteristics, significantly improving the nonlinear modeling ability and band feature response ability of abundance inversion, while improving the accuracy and training efficiency of mineral abundance inversion, and having higher adaptability and stronger practical effect, thereby effectively improving the accuracy and efficiency of Martian mineral abundance inversion.

[0072] This invention also provides an electronic device, which may include a display screen, a memory, and one or more processors. The display screen, memory, and processors are coupled. The memory stores computer program code, which includes computer instructions. When the processor executes the computer instructions, the electronic device can perform various methods or steps executed in the embodiments of the above-described method for inverting the Martian mineral abundance of defects in material components. Of course, this electronic device includes, but is not limited to, the above-described display screen, memory, and one or more processors.

[0073] This invention also provides a computer-readable storage medium for storing computer instructions for running the above-described Martian mineral abundance inversion method.

[0074] Through the above description of the embodiments, those skilled in the art can clearly understand that, for the sake of convenience and brevity, only the division of the above functional modules is used as an example. In actual applications, the above 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.

[0075] In the description of this invention, it should be understood that the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0076] In this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," "linking," and "fixing," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components, unless otherwise explicitly limited. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0077] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0078] Similar parts between the embodiments provided in this application can be referred to mutually. 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 extended from the solution of this application without creative effort shall fall within the scope of protection of this application.

Claims

1. A method for inverting Martian mineral abundance, characterized in that, include: Acquire target Mars hyperspectral reflectance data, which includes hyperspectral reflectance data of inverted minerals from multiple Mars targets; Simulated Mars hyperspectral image data is generated based on the target Mars hyperspectral reflectance data. A deep learning model for mineral abundance inversion is constructed, and the model is trained based on the simulated Mars hyperspectral image data to determine the trained deep learning model for mineral abundance inversion. Acquire Mars hyperspectral image data to be processed; The Mars hyperspectral image data to be processed is preprocessed to obtain the preprocessed Mars hyperspectral image data to be processed. Based on the trained mineral abundance inversion deep learning model, mineral abundance inversion is performed on the preprocessed Martian hyperspectral image data to be processed, and the mineral abundance inversion results are obtained.

2. The method according to claim 1, characterized in that, The mineral abundance inversion deep learning model includes an encoder module and a decoder module; wherein the encoder module has a nonlinear skip connection structure. The encoder module is used to extract features from the input hyperspectral image data based on deep autoencoder and channel attention mechanism to obtain hyperspectral feature data; the input hyperspectral image data is the simulated Mars hyperspectral image data, or the preprocessed Mars hyperspectral image data to be processed. The decoder module is used to perform linear and nonlinear decoding based on the hyperspectral feature data and output the mineral abundance inversion result; the mineral abundance inversion result includes: abundance map data and endmember spectral data corresponding to Martian minerals, and the endmember spectral data is used to characterize the spectral reflectance curve of the Martian minerals.

3. The method according to claim 2, characterized in that, The encoder module includes: a channel attention module and a residual connection module; wherein... The channel attention module is used to determine the weight of each band in the input hyperspectral image data based on the channel attention mechanism and perform feature extraction, and output the feature data after weight allocation. The residual connection module is used to extract features from the weighted feature data based on the nonlinear skip connection structure to obtain the hyperspectral feature data.

4. The method according to claim 2 or 3, characterized in that, The decoder module includes: a linear decoder and a nonlinear decoder; wherein, The linear decoder is used to perform linear mapping on the hyperspectral feature data to determine the abundance information of Martian minerals and output linear decoded feature data. The nonlinear decoder is used to perform nonlinear mapping on the linear decoded feature data to determine the mineral abundance inversion result.

5. The method according to claim 4, characterized in that, The decoder module is further configured to generate reconstructed hyperspectral image data based on the abundance map data and the endmember spectral data; the reconstructed hyperspectral image data is used for inversion verification.

6. The method according to claim 1, characterized in that, The step of generating simulated Martian hyperspectral image data based on the target Martian hyperspectral reflectance data includes: Based on the target Mars hyperspectral reflectance data, a target fusion model is used in conjunction with preset signal-to-noise ratio parameters to generate the simulated Mars hyperspectral image data.

7. The method according to claim 6, characterized in that, The target mixture model is one of the following models: linear mixture model, bilinear mixture model, or nonlinear mixture model.

8. The method according to claim 1, characterized in that, The data preprocessing of the Mars hyperspectral image data to be processed includes at least one of the following: format conversion processing, masking processing, and normalization processing.

9. The method according to claim 1, characterized in that, The target Martian hyperspectral reflectance data is either CRISM Mars Mineral Spectral Library data or USGS Mars Mineral Spectral Library data.

10. A Martian mineral abundance inversion system, characterized in that, include: The system comprises a first acquisition module, a data synthesis module, a model training module, a second acquisition module, a data preprocessing module, and a mineral abundance inversion module; among which, The first acquisition module is used to acquire target Mars hyperspectral reflectance data, which includes hyperspectral reflectance data of multiple Mars target inverted minerals; The data synthesis module is used to generate simulated Mars hyperspectral image data based on the target Mars hyperspectral reflectance data; The model training module is used to construct a mineral abundance inversion deep learning model, and to train the mineral abundance inversion deep learning model based on the simulated Mars hyperspectral image data, thereby determining the trained mineral abundance inversion deep learning model. The second acquisition module is used to acquire the Mars hyperspectral image data to be processed; The data preprocessing module is used to preprocess the Mars hyperspectral image data to be processed to obtain preprocessed Mars hyperspectral image data to be processed. The mineral abundance inversion module is used to perform mineral abundance inversion on the preprocessed Martian hyperspectral image data based on the trained mineral abundance inversion deep learning model, and obtain the mineral abundance inversion result.

Images