Lunar rock abundance inversion method and system combining SAR and DEM data, storage medium and electronic equipment

Abstract

The invention discloses a lunar rock abundance inversion method and system combining SAR and DEM data, a storage medium and an electronic device, and the method comprises the following steps: employing a reduced polarization Mini-RF SAR image, and extracting Stokes parameters and sub-parameters thereof; a polarization decomposition technology is applied to extract scattering mechanism characteristics, and various polarization decomposition components such as surface scattering, secondary scattering and volume scattering are obtained. And performing correlation evaluation on the multi-source characteristic parameters and DiViner rock abundance by applying a correlation analysis method, and screening out characteristic parameters with relatively high correlation with the rock abundance. And in combination with DEM data, terrain related parameters are calculated to represent topographic relief features. And an SSA optimized RF model is constructed, and SAR and DEM features are trained. The trained model is migrated and applied to a permanent shadow area of the South Pole of the moon, and a rock abundance distribution diagram of the PSR area is obtained through inversion. According to the method, the rock abundance inversion precision and stability in complex terrains and low-illumination areas are effectively improved, and reliable data support is provided for polar region landing point selection and resource evaluation.

Description

Technical Field

[0001] The present application relates to the field of deep space exploration and remote sensing, and in particular to a lunar rock abundance inversion method that combines SAR and DEM data, a lunar rock abundance inversion method, a system, a storage medium, and an electronic device. Background Art

[0002] Lunar soil not only records information about important events such as crustal evolution, early volcanic activity, and meteorite impacts, but its study helps deepen our understanding of the Moon, the Earth-Moon system, and the inner solar system. Rocks are one of the important components of lunar soil and are also the main feature of the lunar surface. Therefore, they have always been the focus of research. Understanding the physical properties and spatial distribution patterns of lunar rocks is crucial for the selection of landing sites and the evolution of craters. Rock abundance is defined as the surface rock area fraction within a resolution unit. It can be used to characterize the degree of rock coverage on the lunar surface. However, it cannot directly reflect the undulations of the lunar surface topography. However, rock accumulation is essentially a form of topography, and rock abundance data has profound significance for displaying the complexity of local topography on the lunar surface.

[0003] Currently, lunar rock abundance inversion relies primarily on two data sources: one is based on differential analysis of thermophysical properties using the LRO Diviner thermal radiometer, and the other is rock detection based on high-resolution optical imagery. However, thermal infrared and optical remote sensing methods still have limitations: the former is significantly affected by thermal inertia and radiation properties, and it is difficult to obtain effective data at high latitudes and in PSRs; the latter relies on good lighting conditions and high image resolution, limiting its application in complex terrain or low-light areas. Polarimetric synthetic aperture radar provides an effective means of studying the physical properties of planetary surfaces.

[0004] Although initial progress has been made in radar interpretation methods based on physical models, challenges remain in handling remote sensing features and complex nonlinear relationships, particularly in parameter setting and model adaptability, which rely heavily on prior knowledge. Machine learning methods, with their robust learning efficiency, offer an effective solution to these nonlinear problems.

[0005] To overcome the bottlenecks of existing methods in inverting rock abundance at high latitudes and PSRs, this study introduced the SSA-RF model to carry out rock abundance inversion, which is significantly innovative and of practical significance. Compared with traditional physical models and empirical methods, the SSA-RF model can more effectively explore the nonlinear mapping relationship between multi-source remote sensing data under a data-driven framework, significantly improving the model's adaptability to complex terrain conditions and multi-scale features. By simulating the intelligent foraging and early warning behavior of sparrow colonies, the SSA algorithm comprehensively optimizes the hyperparameter configuration in the random forest model, effectively enhancing the model's global search capability and predictive stability, and providing technical support for high-precision rock abundance estimation. This method provides new ideas and reliable support for revealing the distribution characteristics of polar rocks and assisting in the screening of landing sites in future lunar exploration missions, and has important scientific research value and application prospects. Summary of the Invention

[0006] The purpose of this application is to provide a lunar rock abundance inversion method and system combining SAR and DEM data to address or alleviate the challenges faced by the above-mentioned existing technologies.

[0007] In order to achieve the above objectives, this application proposes the following technical solutions:

[0008] This application provides a lunar rock abundance inversion method that combines SAR and DEM data, including: Step S101: Extracting Stokes parameters and their sub-parameters using reduced polarization Mini-RF SAR images. Step S102: Applying polarization decomposition techniques to extract scattering mechanism characteristics, obtaining multiple polarization decomposition components such as surface scattering, secondary scattering, and volume scattering. Step S103: Using correlation analysis methods to evaluate the correlation between multi-source characteristic parameters and Diviner rock abundance, screening for characteristic parameters with strong correlations with rock abundance and reducing redundant input. Step S104: Calculating terrain-related parameters, including slope and local height difference, in combination with DEM data to characterize topographic relief. Step S105: Constructing a SSA-optimized RF model and training it on SAR and DEM features. Step S106: Applying the trained SSA-RF model to invert rock abundance in the lunar mare region and comparing it with Diviner data for verification. The trained model is then transferred and applied to the permanently shadowed region of the lunar south pole to invert a rock abundance distribution map of the PSR region.

[0009] Preferably, step S101 is specifically as follows: after a series of processing of the Mini-RF data, detailed Stokes1 to 4 parameters can be obtained. This process first involves orthorectification and projection, which is performed in ISIS to convert the PDS data into the ISIS format. The "spiceinit" command can automatically process the input ISIS data, and use the spacecraft position, pointing, spacecraft shape and direction, relative position of the sun and other information according to the corresponding area to calculate the ground position and radar beam incident direction and other information, thereby correcting the geographic location of the input data and outputting it. The "cam2map" command projects the ISIS data onto a map. Subsequently, the mask tool of ArcGIS is used to crop out the study area, and the geographic registration of the image is completed to ensure its consistency with data such as DEM. Thereafter, the SAR data and DEM data are downsampled to the same resolution as the rock abundance, and sub-parameters such as CPR, δ, and χ are calculated. The Stokes parameters can be expressed by the following formula:

[0010] S1=<|E HL | 2 +|E VL | 2 <

[0011] S2=<|E HL | 2 -|E VL | 2 >

[0012]

[0013] In these equations, where E is the electric field, the subscripts H and V indicate the received horizontal and vertical polarization echoes. Brackets "<>" denote the overall average, Re and Im represent the real and imaginary values ​​of the complex cross product amplitude, respectively, and "*" denotes conjugation. Here, S1 indicates a measure of the overall average power of the received signal. The S2 and S3 parameters calculate the linearly polarized power, while S4 calculates the polarized power, whether left-hand circularly polarized or right-hand circularly polarized. Left-hand circularly polarized power is indicated by the negative sign on the S4 parameter.

[0014] CPR, m, δ, and χ can be expressed by the following formulas:

[0015]

[0016] Among them, σ SC Represents the energy power of the co-directional circularly polarized echo, σ OC Represents the energy power of the reverse circular polarization echo.

[0017] Preferably, step S102 is specifically as follows: In radar scattering, the role of rock is mainly reflected in the multiple scattering process. Specifically, when the radar wave is incident on the rock surface, the signal after the initial scattering is not only scattered on the rock surface, but may also undergo secondary or higher-order reflections between the rock and the adjacent rock or between the rock and the lunar soil interface. This multiple scattering process can significantly enhance the echo signal. In order to fully explore and utilize the polarization feature information of full polarization, a variety of MATLAB tools are used for encoding to achieve a variety of polarization decomposition methods. These polarization decomposition methods include: m-δ decomposition, m-χ, H-α decomposition and model-based three-component decomposition. Each decomposition model has its own unique analysis angle and advantages, and can reveal the polarization characteristics of the surface and its cover from different levels.

[0018] By analyzing the relative phase δ of the fully polarized radar signal, the relative contributions of surface scattering and secondary scattering to the echo are distinguished. Specifically, the sensitivity of δ to the two scattering modes, combined with the polarization degree m, allows for the quantitative division and identification of the dominant scattering mechanism in the signal. Three scattering components can be calculated:

[0019]

[0020] m represents the polarization intensity of the signal and is used to distinguish different scattering types, such as surface scattering and double scattering; while χ reflects the polarization shape of the scattered wave, especially for even scattering (such as secondary reflections from lunar craters). m-χ decomposition can effectively identify different scattering sources on the lunar surface, especially for describing topographic features such as craters, providing more accurate results than traditional methods. Three scattering components can be calculated:

[0021]

[0022] H-α decomposition is an eigenvalue-based polarimetric SAR target decomposition method used to analyze the target's scattering mechanism. This method calculates the eigenvalues ​​and eigenvectors of the polarimetric coherence matrix to extract two key parameters: polarimetric entropy (H) and mean scattering angle (α), thereby describing the target's scattering characteristics. This can be expressed using the following two equations:

[0023]

[0024] The core idea of ​​the three-component decomposition model is to regard the total scattering as the superposition of two principal components: one is the volume scattering term representing the depolarization characteristics, and the other is the polarization scattering term with high directionality. The following decomposition model can be obtained:

[0025] m v =0.5S1(1-m)

[0026] ms =2S1m

[0027] Among them, m v represents the scattering process of volume scattering, m s Represents the deterministic scattering process (surface scattering and secondary scattering). The scattering angle α of the main polarization mechanism is obtained by inverting from the reduced data s , and combined with geometric factors (such as cos2α s ), the polarization scattering term is further divided into surface scattering and secondary scattering components, thereby constructing a three-component decomposition model suitable for simplified polarization systems:

[0028]

[0029] Through these models, we use the Stokes parameters and sub-parameters of SAR data and the parameters obtained by polarization decomposition as alternative input features. Considering that the calculation methods of the volume scattering components of the three decomposition methods are basically the same, we select the surface scattering, secondary scattering components and one of the volume scattering components obtained by the three polarization decomposition methods as features.

[0030] Preferably, step S103 specifically includes: starting from multiple dimensions, comprehensively using three statistical methods, namely, the Pearson correlation coefficient, the Spearman rank correlation coefficient, and the Kendall correlation coefficient, to systematically evaluate the correlation between each input feature and the target variable.

[0031] Preferably, step S104 is specifically as follows: two terrain parameters, slope and local height difference, are introduced to quantitatively describe the local undulation characteristics and surface inclination of the lunar surface. They can not only physically reflect the geometric response characteristics of the radar signal, but also provide a basis for distinguishing the confusion between terrain effects and real physical properties. By integrating slope and height difference as feature inputs into the inversion model, it helps to enhance the model's perception of surface roughness, geometric structure and scattering mechanism diversity, thereby improving the accuracy and stability of rock abundance inversion, especially in areas with significant terrain undulation. The slope and height difference DEM data are calculated, and the slope is obtained by constructing a plane formed by the target pixel and its eight neighboring pixels, and calculating the angle between its normal vector and the horizontal plane. The height difference is the average value of the elevation difference between the target pixel and the surrounding eight neighborhood pixels.

[0032] Preferably, step S105 is specifically as follows: in order to improve the prediction accuracy and robustness of the RF model in the inversion of rock abundance in the lunar sea region, SSA is used to optimize the key parameters of the random forest model, including the number of decision trees and the maximum tree depth.

[0033] The SSA algorithm is an emerging intelligent optimization algorithm inspired by the foraging behavior and early warning mechanism of sparrows. It combines global optimization and local fine-grained search capabilities. The optimization process of the SSA algorithm mainly includes the following steps:

[0034] Initialization stage: A certain number of sparrow individuals are randomly generated, each of which represents a different set of random forest parameter combinations to form an initial population.

[0035] Discoverer search mechanism: A subset of sparrows in the population are designated as "discoverers," responsible for searching for the optimal parameter region globally. These discoverers update their positions based on a fitness function (the model's prediction error on the validation set) to guide the overall search.

[0036] Joiner-following mechanism: The remaining sparrow individuals act as "joiners" and perform local follow-up searches based on the discoverer's location. At the same time, a certain amount of random disturbance is introduced to maintain population diversity, thereby avoiding falling into local optimality.

[0037] Early warning escape strategy: When some individuals detect "risk" (that is, the current search falls into a local optimum), the position rapid jump mechanism will be triggered to realize global jump search to enhance the global optimization ability of the algorithm.

[0038] Elite retention and update mechanism: In each iteration, the individual with the best current fitness is selected as the elite solution and compared with the solutions of other individuals. The global optimal solution is updated only when the new individual has better performance, thereby maintaining the continuous optimization ability of the search.

[0039] Termination condition: When the maximum number of iterations is reached or the global optimal solution has not been significantly improved within several rounds of iterations, the algorithm terminates and outputs the optimal hyperparameter combination.

[0040] The final random forest inversion model was constructed based on the hyperparameters optimized by the SSA algorithm. Considering the random forest model's sensitivity to high-dimensional redundant features, this method incorporates correlation analysis and importance ranking results before model construction to remove low-contribution features, further improving the model's generalization and inversion accuracy.

[0041] Preferably, step S106 specifically includes the following steps: To improve the model's regression prediction performance in rock abundance inversion tasks, a random forest is used as the core algorithm. RF has strong nonlinear modeling capabilities and excellent generalization performance, enabling stable and reliable prediction performance in remote sensing inversion scenarios with limited sample sizes and high feature dimensions. This method integrates a large number of decision trees to model the complex relationship between input features and target variables. Each tree is independently trained on different sample and feature subsets, effectively avoiding overfitting.

[0042] During model training, RF fully exploits the multi-source information features contained in the input data, such as radar backscatter intensity and topographic relief, and establishes a nonlinear mapping relationship with the rock abundance in the lunar mare region. Compared to traditional regression models, RF has significant advantages in handling highly correlated and nonlinear relationships between features. It can automatically assess the importance of each variable, providing a guarantee for improved inversion accuracy. Furthermore, the RF model is highly robust to outliers and noise, making it suitable for handling the irregular distributions and observational errors common in remote sensing data.

[0043] Based on the optimized SSA-RF model, not only was the rock abundance modeling task successfully completed in the lunar mare region, but the model was also migrated to the lunar South Pole PSR. This effectively addressed the modeling challenges caused by the scarcity of measured data in the polar regions, providing important technical support for inferring rock distribution characteristics and landing site selection in the lunar polar regions. Model performance was evaluated using the coefficient of determination, mean absolute error, and root mean square error.

[0044] Based on reduced polarimetric Mini-RF SAR imagery and combined with digital elevation model (DEM) data, this method extracts Stokes parameters, their derived sub-parameters, and various polarimetric decomposition feature parameters. The proposed method first preprocesses the SAR data to extract sub-parameter features, including circular polarization ratio (CPR), polarizability (m), relative phase (δ), and ellipticity (χ). Four reduced polarimetric decomposition methods, including m-δ, m-χ, H-α, and a model-based three-component decomposition, are then applied to extract component information such as surface scattering, secondary scattering, and volume scattering, comprehensively reflecting the response characteristics of different scattering mechanisms on the lunar surface to radar waves. Secondly, multiple statistical methods (Pearson, Spearman, and Kendall correlation coefficients) are applied to evaluate the correlation between each characteristic parameter and rock abundance, screening for the optimal feature combination highly correlated with rock abundance, thereby reducing redundant dimensions and improving modeling efficiency. Applying the optimal features extracted from SAR data, combined with terrain parameters (slope and local elevation difference), and using rock abundance products obtained by the Diviner radiometer as samples, a random forest (SSA-RF) model optimized with the Sparrow Search Algorithm was constructed to invert rock abundance in the lunar mare. The proposed model effectively handles the complex nonlinear relationships between multi-source data, such as radar and topography, enhancing the model's adaptability to terrain disturbances and local scattering anomalies, significantly improving the spatial prediction accuracy of rock abundance. Finally, the trained model was transferred and applied to a portion of the permanently shadowed region (PSR) at the lunar South Pole. Mini-RF SAR data and DEM elevation data were used for inversion, resulting in a rock abundance distribution map for the region, providing key support for the selection of polar landing sites for probes.

[0045] The present application also provides a lunar rock abundance inversion system that combines SAR and DEM data, including:

[0046] The data processing unit is configured to: perform data preprocessing on Mini-RF image data, including orthorectification, projection, masking and georeferencing, extract Stokes parameters and sub-parameters, mask DEM data and Diviner rock abundance data, and resample SAR data and DEM data to rock abundance data resolution;

[0047] The characteristic parameter extraction unit is configured to: apply the reduced polarization Stokes image to obtain sub-parameters such as CPR, m, δ, and χ, and use four polarization decomposition methods to polarize the target image to extract the characteristic parameters;

[0048] The characteristic parameter optimization unit is configured to: apply a correlation analysis method to perform a correlation analysis between rock abundance and characteristic parameters to obtain characteristic parameters with a strong correlation with rock abundance, thereby reducing data redundancy;

[0049] The network hyperparameter optimization unit is configured to apply the SSA algorithm to optimize the hyperparameters of the RF random forest to improve the interpretability of the machine learning model;

[0050] The rock abundance inversion unit is configured as follows: constructing the SSA-RF model to estimate the rock abundance of the lunar seas, comparing it with the Diviner rock abundance data, and migrating the model to the Antarctic PSR for rock abundance inversion.

[0051] An embodiment of the present application also provides a computer-readable storage medium having a computer program stored thereon, wherein the program is a lunar rock abundance inversion method combining SAR and DEM data as described above.

[0052] An embodiment of the present application also provides an electronic device, comprising: a memory, a processor, and a program stored in the memory and executable on the processor, wherein when the processor executes the program, it implements any of the above-described methods for inverting lunar rock abundance by combining SAR and DEM data.

[0053] Beneficial effects:

[0054] The embodiment of the present application proposes a lunar rock abundance inversion method that combines SAR and DEM data. By integrating Mini-RF reduced polarimetric SAR data with elevation information, multi-source features including scattering mechanism parameters, Stokes sub-parameters, and terrain features are constructed, further enriching the types of remote sensing information on which rock abundance inversion depends. In order to improve the efficiency of the model's utilization of effective features, correlation analysis is used in the method to eliminate redundant information to ensure that there is a clear physical and statistical correlation between the input features and rock abundance. During the model construction process, SSA is introduced to intelligently adjust the structural parameters of the random forest model, so that the model has better stability and adaptability when facing multi-dimensional remote sensing features and complex terrain conditions. The established SSA-RF model not only shows good consistency and robustness in the inversion of rock abundance in the lunar sea region, but also has strong migration and application capabilities. With the help of this model, the rock abundance in the lunar South Pole PSR can be estimated, providing a practical remote sensing analysis method for the material composition analysis of these special areas and subsequent scientific exploration missions. This technical approach effectively combines the advantages of multi-source remote sensing features and machine learning methods, achieving rock abundance estimation while ensuring the simplicity of the model structure, reflecting high engineering applicability and mission expansion potential. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] The drawings and descriptions that constitute part of this application are intended to provide a further understanding of this application. The exemplary embodiments and descriptions of this application are intended to explain this application and do not constitute an improper limitation on this application. Among them:

[0056] Figure 1 A schematic flow chart of a lunar rock abundance inversion method combining SAR and DEM data according to some embodiments of the present application;

[0057] Figure 2 A schematic diagram of the structure of a system for inverting lunar rock abundance using SAR and DEM data according to some embodiments of the present application;

[0058] Figure 3 A schematic diagram of a research area for a system for inverting lunar rock abundance using SAR and DEM data according to some embodiments of the present application;

[0059] Figure 4 A schematic diagram of a systematic correlation analysis of a lunar rock abundance inversion method combining SAR and DEM data according to some embodiments of the present application;

[0060] Figure 5 A schematic diagram showing a performance comparison between a proposed method model and a traditional model for a system for lunar rock abundance inversion combining SAR and DEM data according to some embodiments of the present application;

[0061] Figure 6 A schematic diagram showing a comparison of rock abundance inversion results in one study area using different models of a system for a lunar rock abundance inversion method combining SAR and DEM data according to some embodiments of the present application;

[0062] Figure 7 A schematic diagram of rock abundance inversion in the lunar mare study area, provided by a system for lunar rock abundance inversion method combining SAR and DEM data according to some embodiments of the present application;

[0063] Figure 8 A schematic diagram of rock abundance inversion in the Antarctic PSR research area, provided by a system for lunar rock abundance inversion method combining SAR and DEM data according to some embodiments of the present application; DETAILED DESCRIPTION

[0064] The present application will be described in detail below with reference to the accompanying drawings and in conjunction with the embodiments. Each example is provided by way of explanation of the present application and does not limit the present application. In fact, it will be clear to those skilled in the art that modifications and variations can be made in the present application without departing from the scope or spirit of the present application. For example, a feature shown or described as part of one embodiment can be used in another embodiment to produce yet another embodiment. Therefore, it is expected that the present application includes such modifications and variations within the scope of the appended claims and their equivalents.

[0065] Exemplary Methods

[0066] Figure 1 Schematic diagram of a process for inverting lunar rock abundance by combining SAR and DEM data according to some embodiments of the present application; Figure 1 As shown in FIG, the method for inverting lunar rock abundance by combining SAR and DEM data includes:

[0067] Step S101, after a series of processing of the Mini-RF data, detailed Stokes1 to 4 parameters can be obtained. This process first involves orthorectification and projection, which is performed in the Imager and Spectrometer Integrated Software (ISIS) to convert the PDS data into the ISIS format. The "spiceinit" command can automatically process the input ISIS data, and use the spacecraft position, pointing, spacecraft shape and direction, relative position of the sun and other information according to the corresponding area to calculate the ground position and radar beam incident direction and other information, thereby correcting the geographic location of the input data and outputting it. The "cam2map" command projects the ISIS data onto a map. Subsequently, the mask tool of ArcGIS is used to crop out the study area, and the geographic registration of the image is completed to ensure its consistency with data such as DEM. Thereafter, the SAR data and DEM data are downsampled to the same resolution as the rock abundance, and sub-parameters such as CPR, δ, and χ are calculated. The Stokes parameters can be expressed by the following formula:

[0068] S1=<|E HL | 2 +|E VL | 2 >

[0069] S2=<|E HL | 2 -|E VL | 2 >

[0070]

[0071] In these equations, where E is the electric field, the subscripts H and V indicate the received horizontal and vertical polarization echoes. Brackets "<>" denote the overall average, Re and Im represent the real and imaginary values ​​of the complex cross product amplitude, respectively, and "*" denotes conjugation. Here, S1 indicates a measure of the overall average power of the received signal. The S2 and S3 parameters calculate the linearly polarized power, while S4 calculates the polarized power, whether left-hand circularly polarized or right-hand circularly polarized. Left-hand circularly polarized power is indicated by the negative sign on the S4 parameter.

[0072] CPR, m, δ, and χ can be expressed by the following formulas:

[0073]

[0074] Among them, σ SC Represents the energy power of the co-directional circularly polarized echo, σ OC Represents the energy power of the reverse circular polarization echo.

[0075] Step S102: In the radar scattering mechanism, rocks primarily affect backscattered signals by inducing multiple scattering processes. When radar waves strike the rock surface, in addition to primary scattering, some of the energy will be reflected multiple times between the rock and surrounding rocks, or between the rock and the lunar soil interface. These high-order scattering phenomena typically enhance the echo intensity, becoming an important indicator of the presence of rock. To more comprehensively extract the scattering characteristics contained in the radar's full polarization information, this study utilized a variety of MATLAB tools to perform diverse polarization decomposition encodings, including m-δ decomposition, m-χ, H-α decomposition, and a three-component decomposition based on a physical model. Different polarization decomposition models characterize the scattering behavior of the surface and its covering from multiple perspectives, contributing to a deeper identification of the physical properties and structural characteristics of lunar surface materials.

[0076] By analyzing the relative phase δ of the fully polarized radar signal, the relative contributions of surface scattering and secondary scattering to the echo are distinguished. Specifically, the sensitivity of δ to the two scattering modes, combined with the polarization degree m, allows for the quantitative division and identification of the dominant scattering mechanism in the signal. Three scattering components can be calculated:

[0077]

[0078] m represents the polarization intensity of the signal and is used to distinguish different scattering types, such as surface scattering and double scattering; while χ reflects the polarization shape of the scattered wave, and is particularly good at distinguishing even-order scattering (such as secondary reflections from lunar craters). m-χ decomposition can effectively identify different scattering sources on the lunar surface, especially for describing terrain features such as craters, providing more accurate results than traditional methods. Three scattering components can be calculated: (The corresponding characters of the components are explained if possible)

[0079]

[0080] Among them, P s represents single scattering, P d represents secondary scattering, P v represents volume scattering. These parameters depend mainly on δ and m. A positive value of δ indicates the dominance of single scattering, while a negative value indicates the dominance of secondary scattering.

[0081] H-α decomposition is an eigenvalue-based polarimetric SAR target decomposition method used to analyze the target's scattering mechanism. This method calculates the eigenvalues ​​and eigenvectors of the polarimetric coherence matrix to extract two key parameters: polarimetric entropy (H) and mean scattering angle (α), thereby describing the target's scattering characteristics. This can be expressed using the following two equations:

[0082]

[0083] The core idea of ​​the three-component decomposition model is to regard the total scattering as the superposition of two principal components: one is the volume scattering term representing the depolarization characteristics, and the other is the polarization scattering term with high directionality. The following decomposition model can be obtained:

[0084] m v =0.5S1(1-m)

[0085] m s =2S1m

[0086] Among them, m v represents the scattering process of volume scattering, m s Represents the deterministic scattering process (surface scattering and secondary scattering). The scattering angle α of the main polarization mechanism is obtained by inverting from the reduced data s , and combined with geometric factors (such as cos2α s ), the polarization scattering term is further divided into surface scattering and secondary scattering components, thereby constructing a three-component decomposition model suitable for simplified polarization systems:

[0087]

[0088] Through these models, we use the Stokes parameters and sub-parameters of SAR data and the parameters obtained by polarization decomposition as alternative input features. Considering that the calculation methods of the volume scattering components of the three decomposition methods are basically the same, we select the surface scattering, secondary scattering components and one of the volume scattering components obtained by the three polarization decomposition methods as features.

[0089] This study selected typical lunar sea areas in the mid-latitudes of the moon as the research objects, covering representative areas with different degrees of weathering and diverse geological backgrounds, including craters, lobed scarps, etc. The above-mentioned areas are located on the front side of the moon, with good remote sensing observation conditions, and the surface of the lunar sea area is relatively flat, and the rock distribution shows a certain spatial heterogeneity, which is helpful for verifying the stability of the model under different backgrounds. The selection of the study area not only takes into account the data accessibility and regional representativeness, but also takes into account the possibility of future probe landing and scientific value, to ensure that the constructed model has good versatility and potential for promotion and application. Specific research areas such as Figure 3 As shown in the figure, the four areas in the yellow boxes are the sample areas selected for this study. 80% of the data in these sample areas served as the training set, and 20% served as the test set. Feature parameters were extracted using Mini-RF SAR data and DEM data within these areas, and the rock abundance data inverted using the Diviner radiometer were used as input for training the model. The four study areas in the red boxes served as the rock abundance inversion areas for this study.

[0090] Step S103 quantitatively analyzes the statistical correlations between features from multiple perspectives, employing a combination of three classic correlation assessment methods: the Pearson correlation coefficient, the Spearman rank correlation coefficient, and the Kendall correlation coefficient. Detailed examples are not provided for this step; if available, please provide them. In multi-source remote sensing data modeling and inversion research, different input features have varying degrees of influence on the target variable. Some redundant or irrelevant features may introduce noise, reducing the model's predictive accuracy. Therefore, conducting correlation analysis between features and the target variable helps identify key driving factors, improve model stability and generalization, and provide theoretical support for feature selection and variable interpretation. This study comprehensively employs the Pearson, Spearman, and Kendall correlation coefficients to assess the correlations between each feature and the target variable from different perspectives.

[0091] The Pearson correlation coefficient is widely used to measure the degree of linear correlation between two continuous variables due to its simple calculation and intuitiveness. This method reflects the direction and strength of the following relationship between variables through the ratio of covariance to standard deviation. The coefficient value ranges from -1 to +1. The larger the absolute value, the stronger the linear correlation. Its mathematical expression is:

[0092]

[0093] In the formula, n represents the number of samples, x j and y j represent the independent variable and dependent variable of sample j respectively, and are the mean of the x and y samples respectively.

[0094] The Spearman correlation coefficient is a nonparametric method used to measure the monotonic relationship between two variables. Its basic idea is to convert the raw data into ranks before performing Pearson correlation analysis. This makes it more robust to outliers and is suitable for remote sensing data where there may be nonlinear but monotonic changes between variables. Its calculation formula is:

[0095]

[0096] In the formula, R(x) and R(y) represent the ranks of the independent variable x and the dependent variable y, respectively. and represents the mean rank.

[0097] The Kendall correlation coefficient is also a rank-based nonparametric method. Its core is to determine whether the order of any two sample pairs on two variables is consistent. Compared with the Spearman method, the Kendall coefficient is more conservative and suitable for dealing with small samples or complex data distribution situations. It is defined as:

[0098]

[0099] Where n is the total number of samples, Mc is the number of consistent pairs, and Md is the number of divergent pairs. Consistent pairs refer to the sample values ​​of two variables being taken in the same relative relationship; otherwise, they are divergent pairs. We used the three proposed correlation analysis methods to conduct correlation analysis on 22 SAR data features and Diviner rock abundance data, as shown in the following example: Figure 4 shown.

[0100] Step S104: Topography plays a crucial modulating role in the remote sensing inversion of lunar surface rock abundance. Especially when using SAR data for inversion analysis, topography significantly influences the radar signal's incident angle, reflection path, and scattering mechanism. The lunar surface is home to numerous undulating terrains, such as crater walls, the margins of the lunar mare, and impact ejecta deposits. These areas have dramatic slope variations and significant height differences, directly leading to localized anomalies in radar echo intensity and changes in polarization characteristics. This topographic influence is particularly pronounced in polar regions with low solar altitudes, permanently shadowed areas, and slopes. Ignoring these effects can lead to model bias during the inversion process, further compromising the spatial consistency and physical plausibility of the inversion results. Furthermore, topographic perturbations often lead to misjudgment of scattering types in the scattering decomposition results of polarimetric SAR data. For example, exaggerated double scattering or volume scattering components may appear on inclined slopes, obscuring the physical interpretation of certain polarization indices related to rock abundance. Therefore, relying solely on polarization parameters or backscatter intensity is difficult to accurately reflect the true variations in surface scattering mechanisms, especially in areas with complex terrain. In order to improve the model's adaptability to scattering mechanisms under different terrain conditions and enhance its tolerance to terrain disturbances, it is necessary to introduce auxiliary factors reflecting the terrain characteristics as additional input information to participate in the modeling.

[0101] Based on the above considerations, this study introduced two terrain parameters, slope and local height difference, to quantitatively describe the local undulation characteristics and surface inclination of the lunar surface. They can not only physically reflect the geometric response characteristics of the radar signal, but also provide a basis for distinguishing the confusion between terrain effects and real physical properties. By integrating slope and height difference as feature inputs into the inversion model, it helps to enhance the model's perception of surface roughness, geometric structure and the diversity of scattering mechanisms, thereby improving the accuracy and stability of rock abundance inversion, especially in areas with significant terrain undulation. Slope and height difference are calculated from DEM data. The slope is calculated by constructing a plane formed by the target pixel and its eight neighboring pixels, and calculating the angle between its normal vector and the horizontal plane. The height difference is the average of the elevation difference between the target pixel and the surrounding eight neighboring pixels.

[0102] Step S105: In order to improve the prediction accuracy and robustness of the random forest (RF) model in the inversion of rock abundance in the lunar mare region, SSA is used to optimize the key parameters of the random forest model, including the number of decision trees and the maximum tree depth.

[0103] The SSA algorithm is an emerging intelligent optimization algorithm inspired by the foraging behavior and early warning mechanism of sparrows. It combines global optimization and local fine-grained search capabilities. The optimization process of the SSA algorithm mainly includes the following steps:

[0104] Initialization stage: A certain number of sparrow individuals are randomly generated, each of which represents a different set of random forest parameter combinations to form an initial population.

[0105] Discoverer search mechanism: A subset of sparrows in the population are designated as "discoverers," responsible for searching for the optimal parameter region globally. These discoverers update their positions based on a fitness function (the model's prediction error on the validation set) to guide the overall search.

[0106] Joiner-following mechanism: The remaining sparrow individuals act as "joiners" and perform local follow-up searches based on the discoverer's location. At the same time, a certain amount of random disturbance is introduced to maintain population diversity, thereby avoiding falling into local optimality.

[0107] Early warning escape strategy: When some individuals detect "risk" (that is, the current search falls into a local optimum), the position rapid jump mechanism will be triggered to realize global jump search to enhance the global optimization ability of the algorithm.

[0108] Elite retention and update mechanism: In each iteration, the individual with the best current fitness is selected as the elite solution and compared with the solutions of other individuals. The global optimal solution is updated only when the new individual has better performance, thereby maintaining the continuous optimization ability of the search.

[0109] Termination condition: When the maximum number of iterations is reached or the global optimal solution has not been significantly improved within several rounds of iterations, the algorithm terminates and outputs the optimal hyperparameter combination.

[0110] The final random forest inversion model was constructed based on the hyperparameters optimized by the SSA algorithm. Considering the random forest model's sensitivity to high-dimensional redundant features, this method incorporates correlation analysis and importance ranking results before model construction to remove low-contribution features, further improving the model's generalization and inversion accuracy.

[0111] Step S106: To improve the model's regression prediction performance in the rock abundance inversion task, a random forest algorithm is used as the core algorithm. RF has strong nonlinear modeling capabilities and excellent generalization performance, enabling stable and reliable predictions in remote sensing inversion scenarios with limited sample sizes and high feature dimensionality. This method integrates a large number of decision trees to model the complex relationship between input features and the target variable. Each tree is independently trained on a different subset of samples and features, effectively avoiding overfitting.

[0112] During the model training process, RF fully exploited the multi-source information features contained in the input data, such as radar backscatter intensity, terrain undulation, etc., and established a nonlinear mapping relationship with the rock abundance in the lunar sea region. Compared with traditional regression models, RF has significant advantages in dealing with high correlation and nonlinear relationships between features. It can automatically evaluate the importance of each variable and provide a guarantee for improving the inversion accuracy. In addition, the RF model is highly robust to outliers and noise, and is suitable for processing common irregular distributions and observation errors in remote sensing data. In order to comprehensively evaluate the performance of the constructed synthetic aperture radar (SAR) rock abundance inversion model and verify the applicability and accuracy of different machine learning methods in rock abundance inversion, this study selected three typical regression models-back propagation neural network (BP), convolutional neural network (CNN) and random forest (RF) for comparative analysis, such as Figure 5 The results show that the random forest model optimized by sparrow search algorithm (SSA-RF) performs best in all three evaluation indicators, with R 2 The SSA-RF model achieved the highest value, 0.7748, indicating stronger fitting ability. Furthermore, it achieved the lowest RMSE, 0.004, with minimal prediction error and the highest accuracy, outperforming other traditional models. These results validate the effectiveness and robustness of SSA-RF in lunar rock abundance inversion and demonstrate its potential in processing complex remote sensing data.

[0113] In addition, to further evaluate the generalization performance of the inversion model, four representative machine learning methods - BP, CNN, RF and SSA-RF - were applied to another area (called study area 1) for rock abundance inversion. The inversion results were visually compared with the reference Diviner data, as shown in Figure 2. Figure 6 As shown, the results were quantitatively evaluated using three metrics: structural similarity index (SSIM), mean absolute error (MAE), and root mean square error (RMSE). Among all models, SSA-RF demonstrated the best overall performance, with the highest SSIM of 0.8935, indicating the strongest structural similarity with the reference data. It also had the lowest MAE (0.0013) and RMSE (0.0030), reflecting minimal prediction bias and error. The RF model also performed well, with SSIM, MAE, and RMSE of 0.8537, 0.0016, and 0.0039, respectively, ranking second in overall accuracy. In contrast, the BP model and CNN model performed relatively poorly. The BP model achieved an SSIM of 0.8005, a MAE of 0.0022, and an RMSE of 0.0048, while the CNN model achieved an SSIM of 0.7848, a MAE of 0.0021, and an RMSE of 0.0044. These results suggest that BP and CNN models may suffer from oversmoothing or local errors when applied to new regions. Comparative analysis in Study Area 1 confirms the good generalization and robustness of the SSA-RF model. It not only preserves more spatial detail in regions with high rock abundance but also achieves the best quantitative agreement with the reference dataset, highlighting its potential for broader application in lunar rock abundance inversion.

[0114] Based on the optimized SSA-RF model, the rock abundance modeling task in the lunar mare region was successfully completed, such as Figure 7As shown in Figure 3 , quantitative evaluation metrics such as MAE, RMSE, and SSIM were used for evaluation. The results showed that in the four selected regions, the MAE ranged from 0.0012 to 0.0023, and the RMSE ranged from 0.0023 to 0.0112. Furthermore, the SSIM values ​​for all regions exceeded 0.83, reaching a maximum of 0.9273. These results demonstrate high consistency between the inversion results and the Diviner data in terms of numerical accuracy and spatial distribution, validating the reliability and applicability of the proposed method. Specifically, in areas with relatively uniform geological backgrounds and minimal topography, the surface scattering mechanism is relatively simple, and terrain distortion and surface roughness variations have less impact on the SAR signal. This allows the inversion model to fully utilize its feature extraction and generalization capabilities, achieving higher prediction accuracy. The error metrics in these regions are significantly lower than those in areas with complex terrain, indicating higher prediction accuracy. The spatial distribution of the inverted rock abundance in these regions is highly consistent with the Diviner data and existing geological survey results, accurately depicting major tectonic units and lithologic variations. Conversely, in areas with strong variations in surface roughness, diverse scattering mechanisms, or frequent radar shadows and bright spots, SAR signals are susceptible to multipath interference and strong scattering distortion, leading to increased inversion errors and large local deviations. Despite this, the SSA-RF model is still able to maintain the overall spatial distribution trend of rock abundance, demonstrating its robustness and cross-regional adaptability even under complex surface conditions.

[0115] Finally, the model migration to the permanent shadow region (PSR) of the lunar south pole was achieved, as shown in the figure. Figure 8 This effectively solves the modeling problem caused by the scarcity of measured data in the polar regions, and provides important technical support for the speculation of rock distribution characteristics and landing site selection in the lunar polar regions.

[0116] Exemplary Systems

[0117] Figure 2 Schematic diagram of the structure of a system for inverting lunar rock abundance using SAR and DEM data according to some embodiments of the present application; Figure 2As shown, the system of the lunar rock abundance inversion method combining SAR and DEM data includes: a data processing unit, configured to perform data preprocessing on Mini-RF image data, including orthorectification, projection, masking and georeferencing, and extracting Stokes parameters and sub-parameters. The DEM data and Diviner rock abundance data were masked, and the SAR data and DEM data were resampled to the rock abundance data resolution; the feature parameter extraction unit was configured to apply reduced polarization Stokes images to obtain sub-parameters such as CPR, m, δ, and χ, and four polarization decomposition methods were used to polarize the target image and extract feature parameters; the feature parameter optimization unit was configured to apply the correlation analysis method to perform correlation analysis between rock abundance and feature parameters to obtain feature parameters with strong correlation with rock abundance and reduce data redundancy; the network hyperparameter optimization unit was configured to apply the SSA algorithm to optimize the hyperparameters of the RF random forest to improve the interpretability of the machine learning model; the rock abundance inversion unit was configured to construct the SSA-RF model to estimate the rock abundance of the lunar seas, compare it with the Diviner rock abundance data, and migrate the model to the Antarctic PSR for rock abundance inversion.

[0118] The system of the lunar rock abundance inversion method combining SAR and DEM data provided in the embodiment of the present application can implement the steps and processes of any of the above-mentioned method embodiments for lunar rock abundance inversion combining SAR and DEM data, and achieve the same technical effects, which will not be repeated here.

[0119] A computer-readable storage medium having a computer program stored thereon. When executed by a processor, the computer program causes the device containing the computer-readable storage medium to perform the above-described lunar rock abundance inversion method combining SAR and DEM data. The computer program includes computer program code, which may be in source code form, object code form, an executable file, or some intermediate form. The computer-readable medium may include any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a mobile hard drive, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM), a random access memory, or other memory.

[0120] An electronic device includes: a memory and a processor, wherein the memory stores a program that can be run on the processor, and when the processor executes the program, it implements the lunar rock abundance inversion method combining SAR and DEM data as described above.

[0121] If the modules / units integrated in the electronic device described in this application are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the present application can implement all or part of the processes in the above-mentioned embodiment methods by instructing the relevant hardware devices to complete them through a computer program. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by a processor, it can implement the steps of the above-mentioned various method embodiments.

[0122] Furthermore, the computer-readable storage medium may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function, etc.; the data storage area may store data created according to the use of the blockchain node, etc.

[0123] Computer-readable instructions are stored in the computer-readable storage medium, and the computer-readable instructions are executed by a processor in the electronic device to implement the lunar rock abundance inversion method combining SAR and DEM data as described in any of the above embodiments.

[0124] In the several embodiments provided in this application, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the module division is merely a logical function division, and other division methods may be used in actual implementation.

[0125] The technical features of the above-mentioned embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the above-mentioned embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0126] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the patent application. It should be noted that a person of ordinary skill in the art may make various modifications and improvements without departing from the spirit of the present application, and these modifications and improvements fall within the scope of protection of the present application. Therefore, the scope of protection of the present patent application shall be determined by the appended claims.

[0127] It should be noted that the terms "including" and "having" and any variations thereof in the specification and claims of this application are intended to cover non-exclusive inclusions. For example, a process, method, system, product or apparatus that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units that are not explicitly listed or are inherent to these processes, methods, products or apparatuses.

[0128] Note that the above are only preferred embodiments of the present invention and the principles of the technology used. Those skilled in the art will understand that the present invention is not limited to the specific embodiments described herein, and that various obvious changes, readjustments, and substitutions can be made by those skilled in the art without departing from the scope of protection of the present invention. Therefore, although the present invention is described in detail through the above embodiments, the present invention is not limited to the specific embodiments described herein. Without departing from the concept of the present invention, it may also include many other effective embodiments, and the scope of the present invention is determined by the scope of the appended claims.

Claims

1. A lunar rock abundance inversion method combining SAR and DEM data is characterized by: include: Step S101: preprocess the Mini-RF image data to extract Stokes parameters and sub-parameters; Step S102, applying four simplified polarization decomposition methods, namely m-δ, m-χ, H-α and model-based three-component decomposition, to extract the components of surface scattering, secondary scattering and volume scattering and record them as characteristic parameters; Step S103: applying a correlation analysis method to perform a correlation analysis between rock abundance and characteristic parameters to obtain characteristic parameters with a strong correlation with rock abundance, thereby reducing data redundancy; Step S104: Calculate the slope and height difference based on the DEM data to extract the terrain features; Step S105: Apply the SSA algorithm to optimize the hyperparameters of the random forest RF model to improve the interpretability of the machine learning model; Step S106: Apply the SSA-RF model to estimate the rock abundance of the lunar seas, and migrate to the lunar south pole region to estimate the rock abundance within the PSR.

2. The lunar rock abundance inversion method combining SAR and DEM data according to claim 1, characterized in that: Step S101 is specifically as follows: First, orthorectification and map projection operations are completed in the integrated imaging and spectral processing system ISIS, converting the original PDS format data into an ISIS-recognizable format; On this basis, by executing the "spiceinit" command, the system automatically retrieves the aircraft attitude, orbital position, radar incidence angle, solar azimuth, and other geometric parameters corresponding to the data, thereby achieving spatial correction of the observation area and precise positioning of the radar beam direction; Next, the data was projected onto a map using the "cam2map" command to form an information image with geographic reference; The study area was extracted using the mask clipping tool in ArcGIS, and the image was georeferenced to ensure spatial consistency with subsequent DEM and other data. Subsequently, the SAR data and topographic data were uniformly downsampled according to the resolution of the rock abundance data, and relevant polariton parameters such as CPR, relative phase difference δ, and ellipticity χ were calculated. The Stokes parameters can be expressed by the following formula: S1=<|E HL | 2 +|E VL | 2 > S2=<|E HL | 2 -|E VL | 2 > Where E is the electric field, the subscripts H and V indicate the received horizontal and vertical polarization echoes; brackets "<>" denote the overall average, Re and Im represent the real and imaginary values ​​of the complex cross product amplitude, respectively, and "*" denotes conjugation. Here, S1 indicates the measurement of the total average power of the received signal; the S2 and S3 parameters calculate the linearly polarized power, and S4 calculates whether the polarized power is left-hand circularly polarized or right-hand circularly polarized. The left-hand circularly polarized power is indicated by the negative sign on the S4 parameter. CPR, m, δ, and χ can be expressed by the following formulas: Among them, σ SC Represents the energy power of the co-directional circularly polarized echo, σ OC Represents the energy power of the reverse circular polarization echo.

3. The lunar rock abundance inversion method combining SAR and DEM data according to claim 1, characterized in that: Step S102 is specifically as follows: By analyzing the relative phase δ of the fully polarized radar signal, the relative contributions of the two mechanisms of surface scattering and secondary scattering in the echo are distinguished. The characteristic that the δ value is sensitive to the two scattering modes is combined with the polarization degree m to achieve quantitative division and identification of the dominant scattering mechanism in the signal. The three scattering components P s 、P v 、P d Can be calculated: m represents the polarization intensity of the signal and is used to distinguish different scattering types, such as surface scattering and double scattering. χ reflects the polarization morphology of the scattered wave, and is particularly good at distinguishing even-order scattering. The m-χ decomposition can effectively identify different scattering sources on the lunar surface, especially for describing terrain features such as craters, providing more accurate results than traditional methods. Three scattering components can be calculated: The H-α decomposition method calculates the eigenvalues ​​and eigenvectors of the polarization coherence matrix and extracts two key parameters: polarization entropy H and average scattering angle α, thereby describing the scattering characteristics of the target. This is expressed by the following two equations: The three-component decomposition based on the model considers the total scattering as the superposition of two principal components: one is the volume scattering term representing the depolarization characteristics, and the other is the polarization scattering term with high directionality. The following decomposition model can be obtained: m v =0.5S1(1-m) m s =2S1m Among them, m v represents the scattering process of volume scattering, m s represents a deterministic scattering process; the scattering angle α of the main polarization mechanism is obtained by inverting from the reduced data s , and combined with the geometric factor, the polarization scattering term is further divided into surface scattering and secondary scattering components, thereby constructing a three-component decomposition model that is applicable to simplified polarization systems: Through these models, we use the Stokes parameters and sub-parameters of SAR data and the parameters obtained by polarization decomposition as alternative input features. Considering that the calculation methods of the volume scattering components of the three decomposition methods are basically the same, we select the surface scattering, secondary scattering components and one of the volume scattering components obtained by the three polarization decomposition methods as features.

4. The lunar rock abundance inversion method combining SAR and DEM data according to claim 1, characterized in that: Step S103 is specifically as follows: Starting from multiple dimensions, the correlation between each input feature and the target variable is systematically evaluated by comprehensively using three statistical methods: Pearson correlation coefficient, Spearman rank correlation coefficient and Kendall correlation coefficient.

5. The lunar rock abundance inversion method combining SAR and DEM data according to claim 1, characterized in that: Step S104 is specifically as follows: By introducing two terrain parameters, slope and local height difference, they are used to quantitatively describe the local undulation characteristics and surface inclination of the lunar surface. They not only physically reflect the geometric response characteristics of the radar signal, but also provide a basis for distinguishing the confusion between terrain effects and real physical properties. Among them, slope is calculated by constructing a plane formed by the target pixel and its eight neighboring pixels, and calculating the angle between its normal vector and the horizontal plane; height difference is the average of the elevation differences between the target pixel and the surrounding eight neighboring pixels.

6. The lunar rock abundance inversion method combining SAR and DEM data according to claim 1, characterized in that: Step S105 is specifically as follows: Initialization phase: randomly generate a certain number of sparrow individuals, each of which represents a different set of random forest parameter combinations to form an initial population; Discoverer search mechanism: A portion of the sparrows in the population are designated as "discoverers" and are responsible for searching for the optimal parameter area globally; These discoverers update their positions based on the fitness function, that is, the prediction error of the model on the validation set, to guide the overall search direction; Joiner-following mechanism: The remaining sparrows act as "joiners" and perform local follow-up searches based on the discoverer's location. A certain amount of random perturbation is introduced to maintain population diversity, thus avoiding falling into local optimality. Early warning escape strategy: When some individuals detect "risk" or the current search falls into a local optimum, a position rapid jump mechanism will be triggered to implement a global jump search to enhance the algorithm's global optimization capability. Elite retention and update mechanism: In each iteration, the individual with the best current fitness is selected as the elite solution and compared with the solutions of other individuals. The global optimal solution is updated only when the new individual has better performance, thus maintaining the continuous optimization capability of the search; Termination condition: When the maximum number of iterations is reached or the global optimal solution has not been significantly improved within several rounds of iterations, the algorithm terminates and outputs the optimal hyperparameter combination; The final random forest inversion model was constructed based on the hyperparameters optimized by the SSA algorithm. Considering that the random forest model is sensitive to high-dimensional redundant features, the correlation analysis and importance ranking results were integrated before the model was built to eliminate low-contribution features.

7. The lunar rock abundance inversion method combining SAR and DEM data according to claim 1, characterized in that: Step S106 is specifically as follows: Random forest is used as the core algorithm to improve the regression prediction performance of the model in rock abundance inversion tasks; The coefficient of determination, R 2 , Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) are used to evaluate the model performance.

8. A lunar rock abundance inversion system combining SAR and DEM data, characterized by: include: The data processing unit is configured to: perform data preprocessing on Mini-RF image data, including orthorectification, projection, masking and georeferencing, and extract Stokes parameters and sub-parameters; mask DEM data and Diviner rock abundance data, and resample SAR data and DEM data to rock abundance data resolution; The characteristic parameter extraction unit is configured to: apply the reduced polarization Stokes image to obtain sub-parameters such as CPR, m, δ, and χ, and use four polarization decomposition methods to polarize the target image to extract the characteristic parameters; The characteristic parameter optimization unit is configured to: apply a correlation analysis method to perform a correlation analysis between rock abundance and characteristic parameters to obtain characteristic parameters with a strong correlation with rock abundance, thereby reducing data redundancy; The network hyperparameter optimization unit is configured to apply the SSA algorithm to optimize the hyperparameters of the RF random forest to improve the interpretability of the machine learning model; The rock abundance inversion unit is configured as follows: constructing the SSA-RF model to estimate the rock abundance of the lunar seas, comparing it with the Diviner rock abundance data, and migrating the model to the Antarctic PSR for rock abundance inversion.

9. A storage medium storing a plurality of programs, characterized in that: The program application is loaded and executed by a processor to implement the lunar rock abundance inversion method comprising combined SAR and DEM data as described in any one of claims 1-7.

10. An electronic device comprising a storage medium and a processor; the processor being adapted to execute various programs; and a memory being adapted to store a plurality of programs; characterized in that: When the memory executes the program on the processor, the lunar rock abundance inversion method comprising combining SAR and DEM data as described in any one of claims 1-7 is implemented.

Images