A method and system for upscaling lunar DEMs based on expert knowledge and random forest models

Abstract

This application provides a lunar DEM upscaling method and system based on expert knowledge and a random forest model, belonging to the field of digital processing technology. Experts annotate different types of terrain and landform regions at an initial resolution (high resolution) in the initial lunar DEM data, and quantize these annotated regions to generate second raster data. The second raster data contains labels for different types of terrain and landforms at a target resolution, thus quantizing expert knowledge into labels at the target resolution. Based on this, training sample datasets corresponding to different types of terrain and landforms are constructed using the initial lunar DEM data. The feature importance of different training sample datasets output by the random forest model is then multiplied with the feature vectors of the training samples to transfer expert knowledge from the initial lunar DEM data to the upscaled DEM data. This scheme can improve the practicality of lunar DEM upscaling methods in multi-scale lunar mapping.

Description

Technical Field

[0001] This application relates to the field of digital processing technology, and in particular to a method and system for upscaling lunar DEMs based on expert knowledge and a random forest model. Background Technology

[0002] The study of lunar landforms is the scientific foundation for understanding the formation and evolution of the Moon. Currently, the vast majority of lunar landform studies use lunar DEM data. In lunar landform research, scale (spatial resolution) is particularly important. At different scientific scales, the types of lunar landforms are different, which has a significant impact on the results of subsequent studies. In other words, different research results may be obtained based on lunar DEM data of different resolutions.

[0003] From the perspective of data availability, currently available DEM data includes: lunar terrain data with a resolution of 100m / pixel generated based on data from the Wide Angle Camera (WAC) on the US Lunar Reconnaissance Orbiter (LRO), and higher resolution DEM data generated based on data from the Terrain Camera (TC) on the Japanese SELENE satellite. However, the resolution of the currently available lunar DEM data is very limited.

[0004] To enrich research on lunar features, it is necessary to address the problem of how to obtain lunar DEM data at different scales (spatial resolution) from existing limited data. One solution is to scale the data, for example, by using upscaling techniques to obtain lunar DEM data at different scales. Upscaling refers to using a method to obtain corresponding low-resolution data from high-resolution data, such as obtaining a 100m resolution lunar DEM from 10m resolution lunar DEM data.

[0005] Existing technologies also provide some methods for upscaling Earth DEM data, such as regional averaging and bilinear interpolation. However, these methods cannot be directly applied to upscaling lunar DEM data because they often ignore the complexity of the lunar surface. The mapping established between low-resolution and high-resolution data often fails to reflect the complex nonlinear relationship between the two. Furthermore, traditional DEM data suffers from the loss of key topographic information (such as the boundaries and central peaks of lunar impact craters) during scale conversion, especially upscaling, which fails to meet the needs of practical applications. For example, it can lead to insufficient accuracy in the path planning of exploration equipment (such as lunar rovers).

[0006] Therefore, there is a need to provide an improved technical solution that addresses the shortcomings of the existing technology. Summary of the Invention

[0007] The purpose of this application is to provide a lunar DEM upscaling method and system based on expert knowledge and a random forest model to solve or alleviate the problems existing in the above-mentioned prior art.

[0008] To achieve the above objectives, this application provides the following technical solution:

[0009] Firstly, this application provides a lunar DEM upscaling method based on expert knowledge and a random forest model, including:

[0010] Experts marked different types of terrain and landform regions at the initial resolution in the initial lunar DEM data, and quantized the marked regions to generate second raster data;

[0011] Wherein, the second raster data is raster data at the target resolution, the target resolution is the resolution after upscaling, and the target resolution is lower than the initial resolution; the second raster data contains labels for different types of terrain and landforms at the target resolution.

[0012] Based on the second raster data and the initial lunar DEM data, training sample datasets corresponding to different types of terrain and landforms are constructed.

[0013] Based on the training sample dataset, construct random forest models corresponding to different types of terrain and landforms, train the random forest models until convergence, and output the feature importance corresponding to the feature vectors of each training sample dataset.

[0014] The feature importance is multiplied by the corresponding feature vectors of each training sample dataset and then post-processed to obtain the upscaling result of the lunar DEM at the target resolution.

[0015] In conjunction with the first aspect, in some possible implementations, the method further includes, before experts label different types of terrain and landform regions at the initial resolution in the initial lunar DEM data:

[0016] Based on the scope of the study area, the lunar DEM data at the initial resolution is cropped to obtain the initial lunar DEM data.

[0017] In conjunction with the first aspect, in some possible implementations, the step of having experts mark different types of terrain and landform regions at the initial resolution in the initial lunar DEM data includes:

[0018] At the initial resolution, experts used GIS software to mark the vector ranges of different types of terrain and landform areas using GIS software.

[0019] All pixels of the initial lunar DEM data covered by the vector range are assigned a value of 1, and the remaining pixels are assigned a value of 0, to obtain different types of terrain and landform regions at the initial resolution, and these are recorded as the first raster data.

[0020] In conjunction with the first aspect, in some possible implementations, the labeled area is quantized to generate second raster data, as follows:

[0021] Based on the area covered by the first raster data, generate unassigned raster data according to the target resolution;

[0022] Traverse all cells in the unassigned raster data and assign labels to each cell to generate a second raster data containing labels for different types of terrain features.

[0023] In conjunction with the first aspect, in some possible implementations, the different types of terrain and landform regions include: the interior region of a lunar impact crater, the edge region of a lunar impact crater, and other terrain and landform regions;

[0024] Correspondingly, all cells in the unassigned raster data are traversed, and labels are assigned to each cell to generate second raster data containing labels for different types of terrain features, specifically:

[0025] The cell currently traversed in the unassigned raster data is recorded as the first cell. Based on the spatial positional correspondence between the unassigned raster data and the first raster data, the second cell set corresponding to the first cell is determined.

[0026] The second set of pixels is the set of pixels in the first raster data that corresponds to the first pixel.

[0027] If all pixels in the second pixel set have a value of 1, then the first pixel is assigned the label of the interior region of the lunar impact crater.

[0028] If some pixels in the second pixel set have a value of 1, then the first pixel is labeled as the label of the lunar impact crater edge region.

[0029] If all pixels in the second pixel set have values ​​other than 1, then the label of the first pixel is marked as the label of other terrain and landform areas.

[0030] In conjunction with the first aspect, based on the second raster data and the initial lunar DEM data, training sample datasets corresponding to different types of terrain and landforms are constructed, specifically as follows:

[0031] The second raster data is divided into multiple raster data subsets according to different types of terrain and landforms;

[0032] Each pixel in each of the aforementioned raster data subsets is traversed, and the values ​​of multiple pixels corresponding to the currently traversed pixel are extracted from the initial lunar DEM data to form a feature vector.

[0033] The feature vector is combined with the label of the currently traversed cell to obtain the training sample dataset corresponding to each of the raster data subsets.

[0034] In conjunction with the first aspect, in some possible implementations, the feature importance is multiplied by the corresponding feature vectors of each training sample dataset and then post-processed to obtain the upscaling result of the lunar DEM at the target resolution, including:

[0035] The importance of the features is normalized to obtain the normalized feature importance.

[0036] The normalized feature importance is multiplied by the corresponding feature vectors of each training sample dataset to obtain the DEM value set corresponding to different types of terrain and landforms after upscaling.

[0037] The DEM value sets corresponding to different types of terrain and landforms after upscaling are shaped to generate upscaled lunar DEM results at the target resolution.

[0038] In conjunction with the first aspect, in some possible implementations, the DEM value sets corresponding to different types of terrain and landforms after upscaling are shaped to generate upscaled results of the lunar DEM at the target resolution, including:

[0039] The DEM value sets corresponding to the different types of terrain and landforms after upscaling are rearranged to obtain the restored raster data;

[0040] According to different types of terrain and landforms, different subsets of raster data are used to perform masking operations on the restored raster data. The masking results of different types of terrain and landforms are superimposed to obtain the upscaled results of the lunar DEM at the target resolution.

[0041] Secondly, this embodiment provides a lunar DEM upscaling system based on expert knowledge and a random forest model, including:

[0042] The knowledge quantification unit is configured to have experts mark different types of terrain and landform areas at the initial resolution in the initial lunar DEM data, and quantify the expert knowledge to generate second raster data.

[0043] Wherein, the second raster data is raster data at the target resolution, the target resolution is the resolution after upscaling, and the target resolution is lower than the initial resolution; the second raster data contains labels for different types of terrain and landforms at the target resolution.

[0044] The sample generation unit is configured to construct training sample datasets corresponding to different types of terrain and landforms based on the second raster data and the initial lunar DEM data.

[0045] The model training and output unit is configured to construct random forest models corresponding to different types of terrain and landforms based on the training sample dataset, train the random forest models until convergence, and output the feature importance corresponding to the feature vectors of each training sample dataset.

[0046] The result generation unit is configured to multiply the feature importance with the corresponding feature vectors of each training sample dataset and perform post-processing to obtain the upscaling result of the lunar DEM at the target resolution.

[0047] Thirdly, this embodiment provides an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the lunar DEM upscaling method described in any of the above embodiments.

[0048] This application provides a method and system for upscaling lunar DEMs based on expert knowledge and a random forest model. Experts annotate different types of terrain and landform regions at an initial resolution (high resolution) in the initial lunar DEM data, and quantize these annotated regions to generate second raster data. The second raster data contains labels for different types of terrain and landforms at a target resolution (low resolution), essentially quantizing expert knowledge into labels at the target resolution. Based on this, training sample datasets for different types of terrain and landforms are constructed using the initial lunar DEM data. The feature importance of different training sample datasets output by the random forest model is then multiplied with the feature vectors of the training samples, thereby transferring expert knowledge from the initial lunar DEM data to the upscaled DEM data. In short, this technical solution quantifies expert knowledge into labels at the target resolution and uses a random forest model to transfer expert knowledge, thereby establishing a nonlinear mapping relationship between the initial resolution lunar DEM data and the target resolution lunar DEM data. This ensures that the key topographic and geomorphic information expressed by the lunar DEM data at the initial resolution is preserved at the target resolution. This not only alleviates the problem of acquiring lunar DEM data at different scales, but also ensures that the upscaled lunar DEM data can meet the needs of practical applications, thus improving the practicality and applicability of the lunar DEM upscaling method in multi-scale lunar mapping. Attached Figure Description

[0049] Figure 1 A schematic diagram of key topographical information for lunar mapping.

[0050] Figure 2 A flowchart illustrating a lunar DEM upscaling method based on expert knowledge and a random forest model, provided for some embodiments of this application.

[0051] Figure 3 This application provides a technical logic block diagram of a lunar DEM upscaling method based on expert knowledge and a random forest model, as provided in some embodiments.

[0052] Figure 4 The diagram below illustrates the structure of a lunar DEM upscaling system based on expert knowledge and a random forest model, provided for some embodiments of this application.

[0053] Figure 5 A schematic diagram of the structure of an electronic device provided for some embodiments of this application.

[0054] Explanation of reference numerals in the attached figures:

[0055] 1-Pit rim, 2-Pit wall, 3-Pit bottom, 4-Central peak, 5-Lunar crust. Detailed Implementation

[0056] To facilitate understanding of the technical solution of this disclosure, please refer to the following: Figure 1 Explain the key topographic and geomorphological information in lunar mapping.

[0057] Lunar impact craters are the most common and prominent topographical units and geological features on the lunar surface. They are numerous, varied in shape, and exhibit a ring-shaped structure of varying sizes and uneven clustering. A typical impact crater profile is shown below. Figure 1 As shown. From Figure 1As can be seen, lunar impact craters consist of: rim 1 (crater edge), wall 2, floor 3, and central peak 4. The rim 1 refers to the prominent edge surrounding the crater, often ring-shaped or elliptical, slightly higher than the surrounding terrain. The wall 2 refers to the steep inner wall extending from the rim 1 down to the floor 3. The floor 3 is the lowest, flat or slightly undulating area within the crater, located between the wall 2 and the central peak, or directly filling the entire floor 3. The central peak 4 refers to the raised ridge or peak formed in the central region of larger impact craters (typically exceeding 15-20 kilometers in diameter). Located above the lunar crust 5, impact craters not only affect the lunar surface appearance but also significantly impact lunar exploration. Currently, lunar exploration is still unmanned deep space exploration. Before exploration, lunar topography mapping of the exploration area is often conducted to plan the route of exploration equipment (such as lunar rovers). If the lunar topography mapping does not retain key topographic information about impact craters, it may prevent exploration equipment from avoiding obstacles. Furthermore, the shape of the impact crater not only affects obstacle avoidance by the exploration equipment, but also affects the choice of path and travel efficiency. In other words, if the location and shape of the impact crater cannot be accurately reflected in the lunar topography map, the path planning may need to choose a more circuitous route, increasing exploration time and energy consumption, which may result in it not being able to accurately reach the target location, affecting the quality of lunar exploration data and exploration results.

[0058] As described in the background section, the available lunar DEM data is very limited in the prior art. For example, the more readily available lunar DEM data includes lunar DEMs with 10-meter and 100-meter resolutions. Other resolutions of lunar DEMs have not yet been developed into mature data products and cannot meet the needs of multi-scale lunar mapping. When using traditional Earth DEM upscaling methods to upscale lunar DEM data, it is easy to cause the loss of key information such as the boundaries and central peaks of lunar impact craters. This often leads to changes in the morphology of the impact craters after upscaling, such as the disappearance of the central peak and the blurring of the crater walls and rims.

[0059] In view of this, this disclosure provides a lunar DEM upscaling method and system based on expert knowledge and a random forest model. This method fully considers the complexity of lunar surface topography and geomorphology, and establishes a nonlinear mapping relationship between high-resolution DEM data and low-resolution DEM data by combining expert knowledge, which is difficult to quantify, and machine learning algorithms (random forest algorithm), so as to improve the mapping accuracy of lunar DEM data after upscaling.

[0060] The embodiments of this application will now be described with reference to the accompanying drawings.

[0061] The terms "first," "second," "third," and "fourth," etc., used in the specification, claims, and accompanying drawings of this application are used to distinguish different objects, not to describe a specific order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or apparatuses.

[0062] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0063] "Multiple" refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone. The character " / " generally indicates that the preceding and following related objects have an "or" relationship.

[0064] Example 1:

[0065] This embodiment provides a lunar DEM upscaling method based on expert knowledge and a random forest model. This method can be executed by electronic devices, such as... Figure 2 As shown, the method includes the following steps:

[0066] Step S101: Experts mark different types of terrain and landform areas at the initial resolution in the initial lunar DEM data (hereinafter also referred to as: raster A), and quantize the marked areas to generate the second raster data (hereinafter also referred to as: raster C).

[0067] The second raster data is the raster data at the target resolution, which is the resolution after upscaling and is lower than the initial resolution. The second raster data contains labels for different types of terrain and landforms at the target resolution.

[0068] Step S102: Based on the second raster data and the initial lunar DEM data, construct training sample datasets corresponding to different types of terrain and landforms;

[0069] Step S103: Based on the training sample dataset, construct random forest models corresponding to different types of terrain and landforms, train the random forest models until convergence, and output the feature importance corresponding to the feature vectors of each training sample dataset.

[0070] Step S104: Multiply the feature importance with the corresponding feature vectors of each training sample dataset and perform post-processing to obtain the upscaling result of the lunar DEM at the target resolution.

[0071] In this embodiment, expert knowledge refers to the knowledge and insights possessed by experts with rich experience, in-depth understanding and high-level skills in the field of lunar mapping. Through expert knowledge, it is possible to determine the key topographic and geomorphic information that is indispensable in lunar mapping.

[0072] In this embodiment, the initial lunar DEM data refers to the obtainable lunar DEM data with an initial resolution. In upscaling applications, the initial resolution refers to high resolution, while the upscaled resolution (target resolution) is low resolution. Traditional methods for generating low-resolution lunar DEM data from high-resolution lunar DEM data typically involve cartographic generalization to simplify, abstract, and refine the terrain and landforms to obtain a low-resolution (target resolution) lunar mapping result.

[0073] On the lunar surface, different types of terrain features can include impact craters, lunar crust, lunar maria, lunar highlands, etc.

[0074] Furthermore, in this embodiment, depending on the mapping purpose, different types of terrain and landforms can be specifically defined as detailed terrain and landforms of impact craters, including the interior region of lunar impact craters, the edge region of lunar impact craters, and other terrain and landform regions.

[0075] The interior region of a lunar impact crater can include all areas within the crater's rim. The crater rim refers to the edge of the crater. Other topographical regions encompass all areas outside the crater's interior and rim.

[0076] In this embodiment, based on expert knowledge, different types of the above-mentioned topographic and geomorphic feature regions are identified and marked on the lunar DEM at initial resolution.

[0077] Here, "annotation" refers to the process by which experts, using their expertise and tools (such as GIS software), classify, identify, and mark different types of terrain and landform regions in lunar digital elevation model (DEM) data. In this process, experts first analyze the initial lunar DEM data to determine the distribution range and boundaries of different terrain and landform features; then, they categorize the terrain and landforms into several types, such as the interior and edges of impact craters, and other terrain and landforms; based on this, they determine the specific geographical locations of each terrain and landform type and mark them using vector polygons.

[0078] Quantization of the labeled areas includes: scale transformation and assigning a corresponding label or classification value to each area. For example, pixels at the edge of an impact crater are labeled as 1, pixels inside the impact crater are labeled as 2, and other landform types are labeled as 3. Scale transformation refers to upscaling, which involves aggregating the quantization results at the initial resolution to generate raster data at the target resolution. For ease of description, this is referred to as second raster data.

[0079] The target resolution is the resolution after scaling up. Scaling up will lead to a decrease in spatial resolution. Therefore, the target resolution is lower than the initial resolution. For example, if the initial resolution is 10 meters, the target resolution can be 25 meters, 50 meters, or 100 meters, or other resolutions. This embodiment does not limit the values ​​of the initial resolution and the target resolution.

[0080] It's important to note that in geographic information data, resolution (spatial resolution) reflects the data's ability to describe geographic spatial details. Higher resolution means each pixel covers a smaller area of ​​the Earth's surface, representing more refined geographic information. For example, 1-meter resolution DEM data means each pixel covers a 1×1 meter area, while 10-meter resolution DEM data means each pixel covers a 10×10 meter area. Relatively speaking, 1-meter resolution is high resolution, and 10-meter resolution is low resolution.

[0081] In this embodiment, the target resolution, i.e., the resolution after scaling up, needs to be determined first through the target determination module. The target resolution is determined based on the mapping scale, which in turn is determined according to the research needs. Simultaneously, the topographic mapping area, i.e., the study area, also needs to be determined based on the research needs. Specifically, the following steps can be followed:

[0082] First, define the scope of the study area: Based on the research objectives, determine the lunar region that needs to be mapped. For example, the study area can be based on a specific geographical region of the Moon (such as a specific range of latitude and longitude) or a research mission (such as lunar rover path planning), and existing maps or coordinates (e.g., specific latitude and longitude or patch data) can be used to define the study area.

[0083] Next, determine the mapping scale: The mapping scale is used to reflect the ratio between the actual land surface and the map. The choice of scale depends on the purpose of the study, the data resolution, and the application requirements of the map.

[0084] Next, determine the target resolution based on the mapping scale: there is a direct relationship between scale and resolution. Smaller scales (i.e., covering a larger area of ​​the land) typically require lower resolution, while larger scales require higher resolution to display more detail. Resolution is the actual ground distance represented by each pixel, usually expressed in meters per pixel. The larger the scale, the finer the resolution, and the clearer the terrain features displayed. Based on the scale relationship, the target resolution can be estimated using the following formula:

[0085]

[0086] For example, for a mapping requirement of 1:1,000,000, that is, the target scale is 1:1,000,000, then according to the formula, the estimated DEM data after scaling up is 100m (based on the fact that the highest visual resolution of the human eye is 0.1mm).

[0087] After determining the target resolution, second raster data can be generated according to the target resolution. In this embodiment, in addition to the change in resolution, the second raster data also contains labels for different types of terrain features at the target resolution. That is, by labeling the terrain features, they are quantified into labels at the target resolution. The expert's experience in classifying and labeling terrain features is transformed into a digital and standardized form and mapped to the pixels of the raster data to form the second raster data. This transforms expert knowledge into a computable and scalable form, laying the foundation for the subsequent transfer and use of expert knowledge.

[0088] Step S102 aims to combine the second raster data containing expert knowledge with the initial lunar DEM data to generate training samples for the random forest model. Since the second raster data contains labels for different types of terrain features at the target resolution, it can be used to provide label values ​​for the training samples. Features are extracted from the initial lunar DEM data to form corresponding feature vectors. Then, the labels and feature vectors are combined to generate the training sample dataset.

[0089] This embodiment constructs training sample datasets corresponding to different types of terrain features. In other words, the training sample datasets are organized according to different terrain features, with each terrain feature type corresponding to a separate training sample dataset. For example, the training sample datasets may include: training sample datasets for the interior regions of lunar impact craters, training sample datasets for the edge regions of lunar impact craters, and training sample datasets for other terrain feature regions. Organizing the training sample datasets according to different terrain feature types and providing independent datasets allows each terrain feature type to be trained separately. This facilitates deep learning of single-class features by the model, helps adjust the weights of each terrain feature type during training, avoids class imbalance in the training data, and improves the targeting of model training. Simultaneously, this data organization format facilitates integration with expert knowledge; experts can easily review and update training sample datasets for a single class without needing to view and reconstruct training sample datasets for all classes.

[0090] In this embodiment, step S103 aims to construct independent models for different types of terrain and landforms. By training the models until convergence, it ensures that the models can fully learn the features of each landform and outputs the feature importance corresponding to each training sample dataset to quantify the contribution of each input feature to the prediction result. Since the input features are extracted from the initial lunar DEM data and the labels are generated through expert knowledge, the feature importance can reflect the result of the model learning after the expert knowledge is quantified.

[0091] In step S104, the feature importance is multiplied with the feature vectors of each training sample dataset to transfer the parameters learned based on expert knowledge from the initial DEM data to the DEM value set at the target resolution. Through post-processing operations, the DEM value set is reorganized into a raster data form with spatial location, thereby obtaining the upscaling result of the lunar DEM at the target resolution.

[0092] It should be noted that in this embodiment, multiplying the feature importance with the corresponding feature vectors of each training sample dataset means taking the inner product between the feature importance and the feature vector, that is, multiplying the corresponding elements of the two vectors and summing them.

[0093] Since feature importance is based on expert knowledge and reflects the contribution of each feature to the model's prediction, when feature importance is multiplied by the feature vector, it actually measures the relationship between the importance of each feature in the current sample and the feature value. This allows for the quantification of the contribution of each feature to the model's prediction. In this way, expert knowledge can be quantitatively transferred from the initial resolution to each pixel of the target resolution, enabling the model to accurately estimate the pixel values ​​after upscaling.

[0094] In summary, the method provided in this embodiment annotates the initial DEM data according to the mapping purpose, quantifies expert knowledge, constructs a training sample dataset, and trains a random forest model for each type of terrain and landform separately. By multiplying the feature importance with the corresponding feature vectors of each training sample dataset, the parameters after expert knowledge learning are transferred, generating an upscaled DEM value set. Then, a post-processing module is used to generate the upscaled result of the lunar DEM at the target resolution. The above process ensures the invariance of expert knowledge during the upscaling process, providing scientifically effective DEM data for lunar landform research and mapping.

[0095] In an alternative embodiment, before experts mark different types of terrain and landform areas at the initial resolution in the initial lunar DEM data, the method further includes: cropping the lunar DEM data at the initial resolution according to the extent of the study area to obtain the initial lunar DEM data.

[0096] In this embodiment, the cropping of lunar DEM data is accomplished by the data acquisition and preprocessing module.

[0097] When acquiring lunar DEM data, to preserve high-precision topographic information in lunar mapping, a higher-resolution DEM can be selected as the initial lunar DEM data. For example, digital terrain data with a resolution of 10 meters per pixel generated using SELENE TC (Terrain Camera) data can be used as the initial lunar DEM data. Then, based on the study area's boundaries, the initial lunar DEM data is loaded into Geographic Information System (GIS) software (such as QGIS or ArcGIS) as raster data. Next, the cropping tool in the GIS software is used to crop the loaded DEM data according to the study area's boundaries. After cropping, the output DEM data is checked to ensure it meets expectations, guaranteeing that the cropped data correctly covers the study area and that no important topographic information is lost.

[0098] In this embodiment, by cropping the lunar DEM data at the initial resolution, the focus of data processing is ensured, making subsequent data processing and model training more accurate and efficient.

[0099] In an optional embodiment, experts mark different types of terrain and landform regions at the initial resolution in the initial lunar DEM data, including: marking the vector range of different types of terrain and landform regions at the initial resolution in the initial lunar DEM data using GIS software; assigning a value of 1 to all pixels of the initial lunar DEM data covered by the vector range, and assigning a value of 0 to the remaining pixels, thereby obtaining different types of terrain and landform regions at the initial resolution, and recording them as the first raster data (hereinafter also referred to as: raster B).

[0100] In the above embodiments, the annotation method of first marking the vector range and then rasterizing has the following advantages: experts can flexibly adjust the boundaries of landform types according to the actual situation, which is more intuitive and accurate than operating directly in the raster, and is conducive to improving the flexibility of annotation. In addition, vector data can accurately represent the boundaries of landform areas, and the boundaries of landform areas can be accurately annotated by GIS software to ensure high data accuracy. By assigning a value of 1 to all pixels of the initial lunar DEM data covered by the vector range, a clear geographical range is provided for the labels of the upscaled lunar DEM data. This ensures that when generating raster data at the target resolution, the corresponding label values ​​of the corresponding pixels at the target resolution can be correctly and quickly assigned using the multiple pixel values ​​at the initial resolution.

[0101] In an optional embodiment, the labeled area is quantized to generate second raster data, as follows:

[0102] Step S1011: Generate unassigned raster data according to the target resolution based on the range covered by the first raster data.

[0103] It should be noted that, here, the meaning of unassigned raster data is that the resolution of the raster data has been scaled up from the initial resolution to the target resolution, but it is uncertain which terrain type each pixel in the raster data belongs to, and it needs to be determined through the label assignment process in step S1012.

[0104] In this embodiment, the unassigned raster data is generated based on the area covered by the first raster data. That is, the geographical area covered by the unassigned raster data is the same as the area covered by the first raster data. In other words, pixels outside the area covered by the first raster data do not contain expert knowledge and do not participate in the subsequent label assignment process.

[0105] Step S1012: Traverse all cells in the unassigned raster data and assign labels to each cell to generate a second raster data containing labels for different types of terrain features.

[0106] To ensure that the label assignment results accurately reflect key lunar landform information, based on the above embodiments, a further technical solution is to classify different types of terrain and landform regions, including: lunar impact crater interior regions, lunar impact crater edge regions, and other terrain and landform regions; correspondingly, all pixels in the unassigned raster data are traversed, and labels are assigned to each pixel to generate second raster data containing labels for different types of terrain and landforms, specifically:

[0107] Step S1012a: Record the currently traversed cell in the unassigned raster data as the first cell, and determine the second cell set corresponding to the first cell based on the spatial position correspondence between the unassigned raster data and the first raster data.

[0108] The second set of pixels is the set of pixels in the first raster data that corresponds to the first pixel.

[0109] During the upscaling process, the first raster data is a high-resolution raster, while the unassigned raster data is a low-resolution raster. The spatial relationship between the high-resolution and low-resolution raster data is mainly reflected in the fact that multiple cells in the high-resolution raster will be "aggregated" or "merged" into a single cell in the low-resolution raster. In other words, there is a positional mapping relationship between the high-resolution and low-resolution raster data. For each cell in the low-resolution raster data, there are multiple corresponding cells in the high-resolution raster data. Therefore, it is possible to determine the multiple cells in the first raster data that correspond to any cell in the unassigned raster data, forming the second cell set.

[0110] Step S1012b: If all pixels in the second pixel set have a value of 1, then assign the first pixel the label of the lunar impact crater interior region.

[0111] Step S1012c: If some pixels in the second pixel set have values ​​of 1, then label the first pixel as the label of the lunar impact crater edge region.

[0112] Step S1012d: If the value of all pixels in the second pixel set is not 1, then the label of the first pixel is marked as the label of other terrain and landform areas.

[0113] The purpose of steps S1012b to S1012d is to determine the value of the first pixel by using the values ​​of all pixels in the second pixel set according to a nonlinear mapping, that is, to determine the specific terrain type of the lunar surface covered by the first pixel. Specifically, if the value of all pixels in the second pixel set is 1, it indicates that the lunar surface area covered by all pixels in the set is the interior region of a lunar impact crater, and the first pixel is labeled as the interior region of a lunar impact crater; if the value of some pixels in the second pixel set is 1, it indicates that the area covered by the second pixel set is the edge of an impact crater, and the first pixel is labeled as the edge region of a lunar impact crater; if the value of all pixels in the second pixel set is not 1, that is, all pixels are 0, it indicates that the area covered by the second pixel set is another terrain region, and the first pixel is labeled as another terrain region.

[0114] Furthermore, the label values ​​for the lunar impact crater interior, edge, and other terrain and landform areas can be determined as needed. For example, the label for the lunar impact crater edge area is 1, the label for the lunar impact crater interior area is 2, and the label for other terrain and landform areas is 3. Then, the pixels in the second raster data have three possible values: 1, 2, and 3, each representing a different type of terrain and landform.

[0115] In some alternative embodiments, based on the second raster data and the initial lunar DEM data, constructing training sample datasets corresponding to different types of terrain features may specifically include the following steps:

[0116] The second raster data is divided into multiple raster data subsets according to different types of terrain and landforms.

[0117] Specifically, the second raster data is divided according to the different values ​​of each pixel. For example, the label for the edge region of a lunar impact crater is 1. Therefore, all pixels with a value of 1 are extracted to form the label data for the interior region of the lunar impact crater, and the extracted pixels are assigned a value of 1, while all other pixels are assigned a value of 0. This result is denoted as raster data subset C1. Similarly, all pixels with a value of 2 are extracted to form the label data for the edge region of the lunar impact crater, and the extracted pixels are assigned a value of 1, while all other pixels are assigned a value of 0. This result is denoted as raster data subset C2. All pixels with a value of 3 are extracted to form the label data for other terrain and landform regions, and the extracted pixels are assigned a value of 1, while all other pixels are assigned a value of 0. This result is denoted as raster data subset C3.

[0118] In an optional implementation, the division of multiple raster data subsets can also be performed synchronously with the label assignment step. That is, in steps S1012b to S1012d, while assigning labels to various terrain types, different types of terrain are divided into corresponding raster subsets, and in each raster subset, a cell value of 1 is defined as the corresponding type of terrain to be retained, and 0 is for others.

[0119] Traverse each cell in each raster data subset, extract the values ​​of multiple cells corresponding to the currently traversed cell from the initial lunar DEM data, and form a feature vector;

[0120] The feature vectors are combined with the labels of the currently traversed pixels to obtain the training sample datasets corresponding to each raster data subset.

[0121] Information from a single pixel may be insufficient to fully describe the upscaled terrain features. By extracting the values ​​of multiple pixels from the initial lunar DEM data to form a feature vector, local spatial patterns and terrain variations can be captured, thereby improving the model's discriminative ability. Combining the feature vector with labels provides a training sample containing spatial information for each pixel. Simultaneously, each training sample contains the features of that location (i.e., the values ​​of multiple pixels) and the corresponding label (i.e., terrain type), which helps the model learn the nonlinear mapping relationship between terrain type and its spatial features. This enables the model to more accurately identify the boundaries and structures of impact craters on the lunar surface, improving model accuracy.

[0122] Furthermore, the feature importance is multiplied by the corresponding feature vectors of each training sample dataset and post-processed to obtain the upscaling results of the lunar DEM at the target resolution, including:

[0123] The feature importance is normalized to obtain the normalized feature importance.

[0124] The normalized feature importance is multiplied by the corresponding feature vectors of each training sample dataset to obtain the DEM value set corresponding to different types of terrain and landforms after upscaling.

[0125] The DEM value sets corresponding to different types of terrain and landforms after upscaling are shaped to generate upscaled results of lunar DEM at the target resolution.

[0126] Specifically, the DEM value sets corresponding to different types of terrain and landforms after upscaling are shaped to generate upscaled lunar DEM results at the target resolution, including:

[0127] The DEM value sets corresponding to different types of terrain and landforms after upscaling are rearranged to obtain the restored raster data;

[0128] According to different types of terrain and landforms, different subsets of raster data are used to perform masking operations on the restored raster data. The masking results of different types of terrain and landforms are superimposed to obtain the upscaled results of the lunar DEM at the target resolution.

[0129] The following reference Figure 3 The technical solution provided in this embodiment will be illustrated with examples.

[0130] like Figure 3 As shown, this embodiment can be implemented according to the following steps:

[0131] Step 1: Target Determination Module

[0132] First, determine the mapping area and scale based on the research needs, and then determine the resolution after scaling up based on the map scale. For example, for a 1:100w mapping requirement, the scaled-up DEM data should be 100m (converted based on the human eye's highest visual resolution of 0.1mm).

[0133] Step 2: Data Acquisition and Preprocessing Module

[0134] To preserve high-precision topographic information during mapping, high-resolution DEM data was selected as the initial DEM data. For example, digital terrain data with a resolution of 10 m / pixel generated by the SELENE TC (Terrain Camera) was used as the initial DEM data. The initial DEM data was cropped according to the study area to obtain DEM data A (initial lunar DEM data) for the study area.

[0135] Step 3: Expert Knowledge Quantification Module

[0136] According to the purpose of mapping, DEM data A is labeled and expert knowledge is quantified. Specifically: (1) At the initial DEM resolution, experts use GIS software (such as ArcGIS) to label the terrain and landform feature areas that cannot be missing after scaling up, and obtain the corresponding vector S; (2) Using DEM data A as a template, vector S is rasterized into raster B, where the raster value of the area corresponding to vector S is 1, and the rest is 0; (3) Raster B is aggregated and scaled up to obtain the raster set C (i.e., raster C) at the target resolution:

[0137] C = {C1, C2, C3}

[0138] In other words, the raster set C consists of multiple raster subsets, namely C1, C2, and C3. Aggregation upscaling yields raster C1, C2, and C3 according to the following rules:

[0139] Raster C1: During the upscaling process, if any number of elements in each cell of raster C and the corresponding cells of raster B are 1 (some are 1, but not all are 1), then the cell value of raster C after upscaling will be 1; otherwise, it will be 0.

[0140] Raster C2: During the upscaling process, if all elements in the multiple pixels corresponding to each pixel of raster C are 1, then the pixel value of raster C after upscaling is 1; otherwise, it is 0.

[0141] Raster C3: During the upscaling process, if none of the elements in each cell of raster C corresponding to the multiple cells of raster B are 1, then the cell value of raster C after upscaling is 1; otherwise, it is 0.

[0142] Step 4: Training Sample Construction Module

[0143] Based on the data obtained in the previous step, a training sample dataset is constructed. Specifically, for each pixel of grid C1, the value of grid data A within the corresponding region of each pixel is extracted. Assuming that each pixel of grid C1 corresponds to 100 pixels of grid A, an X vector (x1, x2, ..., x...) is obtained. 100 A label and a corresponding Y value, where the Y value is the label and Y is the pixel value of raster C1, specifically 0 or 1.

[0144] By iterating through each cell of the low-resolution DEM data (raster C1), a series of X vectors and Y values ​​are obtained, forming a training sample dataset DATA1.

[0145] For grids C2 and C3, obtain DATA2 and DATA3 in the same way.

[0146] Step 5: Random Forest Model

[0147] (1) Based on the training sample dataset DATA1 obtained in the previous step, construct a random forest classification model Model-1, train the model until convergence, and obtain the importance X1 corresponding to the X vector, such as (x11, x12, ..., x1...). 100 ), and normalize X1 to obtain X1_α.

[0148] (2) Based on the training sample dataset DATA2 obtained in the previous step, construct a random forest classification model Model-2, train the model until convergence, and obtain the importance X2 corresponding to the X vector, such as (x21, x22, ..., x22). 100 ), and normalize X2 to obtain X2_α.

[0149] (3) Based on the training sample dataset DATA3 obtained in the previous step, construct a random forest classification model Model-3, train the model until convergence, and obtain the importance X3 corresponding to the X vector, such as (x31, x32, ..., x3). 100 ), and normalize X3 to obtain X3_α.

[0150] Step 6, Parameter Migration Module

[0151] (1) For the training dataset DATA1 constructed above, multiply each sub-element (i.e. each feature vector in the training dataset DATA1) by X1_α to obtain the set of upscaled DEM values ​​Y1;

[0152] (2) For the training dataset DATA2 constructed above, multiply each sub-element (i.e. each feature vector in the training dataset DATA2) by X2_α to obtain the set of upscaled DEM values ​​Y2;

[0153] (3) For the training dataset DATA3 constructed above, multiply each sub-element (i.e. each feature vector in the training dataset DATA3) by X3_α to obtain the set of upscaled DEM values ​​Y3;

[0154] Step 7, Post-processing module

[0155] Reshape Y1, Y2, and Y3 to restore them to raster data. Then, use rasters C1, C2, and C3 to perform masking operations on the raster data obtained from Y1, Y2, and Y3 respectively. Finally, superimpose the masked rasters into a new raster to obtain the upscaled DEM data.

[0156] In this way, through the above operations, expert knowledge is quantified, and on this basis, the upscaling operation of DEM data is completed, maintaining the invariance of expert knowledge during the upscaling process.

[0157] Example 2:

[0158] This embodiment provides a lunar DEM upscaling system based on expert knowledge and a random forest model, such as... Figure 4 As shown, the system includes: a knowledge quantification unit 401, a sample generation unit 402, a model training and output unit 403, and a result generation unit 404. Specifically:

[0159] The knowledge quantification unit 401 is configured to have experts mark different types of terrain and landform areas at the initial resolution in the initial lunar DEM data, and quantify the expert knowledge to generate the second raster data.

[0160] The second raster data is the raster data at the target resolution, which is the resolution after upscaling and is lower than the initial resolution. The second raster data contains labels for different types of terrain and landforms at the target resolution.

[0161] The sample generation unit 402 is configured to construct training sample datasets for different types of terrain and landforms based on the second raster data and the initial lunar DEM data.

[0162] The model training and output unit 403 is configured to construct random forest models corresponding to different types of terrain and landforms based on the training sample dataset, train the random forest models until convergence, and output the feature importance corresponding to the feature vectors of each training sample dataset.

[0163] The result generation unit 404 is configured to multiply the feature importance with the corresponding feature vectors of each training sample dataset and perform post-processing to obtain the upscaling result of the lunar DEM at the target resolution.

[0164] The lunar DEM upscaling system based on expert knowledge and random forest model provided in this embodiment can implement the steps and processes of the lunar DEM upscaling method based on expert knowledge and random forest model provided in any of the above embodiments, and achieve the same technical effect. It will not be described in detail here.

[0165] Example 3:

[0166] Based on the same inventive concept, this embodiment also provides an electronic device. Figure 5 This is a schematic diagram of the electronic device; such as Figure 5 As shown, the hardware structure of the electronic device may include: a processor 501, a communication interface 502, a computer-readable storage medium (also known as a memory) 503, and a communication bus 504.

[0167] The processor 501, communication interface 502, and computer-readable storage medium 503 communicate with each other via communication bus 504.

[0168] Computer-readable storage medium 503 can be configured to store one or more programs.

[0169] Optionally, the communication interface 502 can be an interface of a communication module, such as the interface of a GSM module.

[0170] The processor 501 executes one or more programs that perform the following steps:

[0171] Experts marked different types of terrain and landform regions at the initial resolution in the initial lunar DEM data, and quantized the marked regions to generate second raster data;

[0172] The second raster data is the raster data at the target resolution, which is the resolution after upscaling and is lower than the initial resolution. The second raster data contains labels for different types of terrain and landforms at the target resolution.

[0173] Based on the second raster data and the initial lunar DEM data, training sample datasets corresponding to different types of terrain and landforms were constructed.

[0174] Based on the training sample dataset, construct random forest models corresponding to different types of terrain and landforms, train the random forest models until convergence, and output the feature importance corresponding to the feature vectors of each training sample dataset.

[0175] The feature importance is multiplied by the corresponding feature vectors of each training sample dataset and then post-processed to obtain the upscaling result of the lunar DEM at the target resolution.

[0176] The processor 501 can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc., or it can be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), an off-the-shelf programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this application. The general-purpose processor can be a microprocessor or any conventional processor.

[0177] The electronic devices described in this application exist in various forms, including but not limited to:

[0178] (1) Mobile communication devices: These devices are characterized by their mobile communication capabilities and primarily aim to provide voice and data communication. These terminals include: smartphones (e.g., iPhones), multimedia phones, feature phones, and low-end phones, etc.

[0179] (2) Ultra-mobile personal computer devices: These devices fall under the category of personal computers, possessing computing and processing capabilities, and generally also have mobile internet access features. These terminals include PDAs, MIDs, and UMPCs, such as the iPad.

[0180] (3) Portable entertainment devices: These devices can display and play multimedia content. This category includes audio and video players (e.g., iPods), handheld game consoles, e-book readers, as well as smart toys and portable car navigation devices.

[0181] (4) Server: A device that provides computing services. The components of a server include a processor, hard disk, memory, system bus, etc. Servers are similar to general computer architectures, but because they need to provide highly reliable services, they have higher requirements in terms of processing power, stability, reliability, security, scalability, and manageability.

[0182] (5) Other electronic devices with data interaction functions.

[0183] It should be noted that, depending on the implementation needs, the various components / steps described in the embodiments of this application can be broken down into more components / steps, or two or more components / steps or parts of the operation of components / steps can be combined into new components / steps to achieve the purpose of the embodiments of this application.

[0184] The methods described above according to the embodiments of this application can be implemented in hardware, firmware, or as software or computer code that can be stored in a recording medium (such as CD ROM, RAM, floppy disk, hard disk, or magneto-optical disk), or as computer code downloaded over a network that is originally stored in a remote recording medium or a non-transitory machine storage medium and will be stored in a local recording medium. Thus, the methods described herein can be stored on a recording medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware (such as an ASIC or FPGA) for such software processing. It is understood that the computer, processor, microprocessor controller, or programmable hardware includes storage components (e.g., RAM, ROM, flash memory, etc.) capable of storing or receiving software or computer code. When the software or computer code is accessed and executed by the computer, processor, or hardware, the lunar DEM upscaling method based on expert knowledge and a random forest model described herein is implemented. Furthermore, when a general-purpose computer accesses the code used to implement the methods shown herein, the execution of the code transforms the general-purpose computer into a dedicated computer for executing the methods shown herein.

[0185] Those skilled in the art will recognize that the units and method steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application of the technical solution and the constraints involved. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the embodiments of this application.

[0186] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A lunar DEM upscaling method based on expert knowledge and a random forest model, characterized in that, include: Experts marked different types of terrain and landform regions at the initial resolution in the initial lunar DEM data, and quantized the marked regions to generate second raster data; Wherein, the second raster data is raster data at the target resolution, the target resolution is the resolution after upscaling, and the target resolution is lower than the initial resolution; the second raster data contains labels for different types of terrain and landforms at the target resolution. Based on the second raster data and the initial lunar DEM data, training sample datasets corresponding to different types of terrain and landforms are constructed. Based on the training sample dataset, construct random forest models corresponding to different types of terrain and landforms, train the random forest models until convergence, and output the feature importance corresponding to the feature vectors of each training sample dataset. The feature importance is multiplied by the corresponding feature vectors of each training sample dataset and then post-processed to obtain the upscaling result of the lunar DEM at the target resolution. Experts marked different types of terrain and landform regions at the initial resolution in the initial lunar DEM data, including: At the initial resolution, experts used GIS software to mark the vector ranges of different types of terrain and landform areas using GIS software. All pixels of the initial lunar DEM data covered by the vector range are assigned a value of 1, and the remaining pixels are assigned a value of 0 to obtain different types of terrain and landform regions at the initial resolution, and these are recorded as the first raster data. The labeled areas are quantized to generate second raster data, following these steps: Based on the area covered by the first raster data, generate unassigned raster data according to the target resolution; Traverse all cells in the unassigned raster data and assign labels to each cell to generate a second raster data containing labels for different types of terrain features.

2. The method according to claim 1, characterized in that, Before experts label different types of terrain and landform regions at the initial resolution in the initial lunar DEM data, the method further includes: Based on the scope of the study area, the lunar DEM data at the initial resolution is cropped to obtain the initial lunar DEM data.

3. The method according to claim 1, characterized in that, The different types of terrain and landform regions include: the interior region of lunar impact craters, the edge region of lunar impact craters, and other terrain and landform regions; Correspondingly, all cells in the unassigned raster data are traversed, and labels are assigned to each cell to generate second raster data containing labels for different types of terrain features, specifically: The cell currently traversed in the unassigned raster data is recorded as the first cell. Based on the spatial positional correspondence between the unassigned raster data and the first raster data, the second cell set corresponding to the first cell is determined. The second set of pixels is the set of pixels in the first raster data that corresponds to the first pixel. If all pixels in the second pixel set have a value of 1, then the first pixel is assigned the label of the interior region of the lunar impact crater. If some pixels in the second pixel set have a value of 1, then the first pixel is labeled as the label of the lunar impact crater edge region. If all pixels in the second pixel set have values ​​other than 1, then the label of the first pixel is marked as the label of other terrain and landform areas.

4. The method according to claim 1, characterized in that, Based on the second raster data and the initial lunar DEM data, training sample datasets corresponding to different types of terrain and landforms are constructed, specifically as follows: The second raster data is divided into multiple raster data subsets according to different types of terrain and landforms; Each pixel in each of the aforementioned raster data subsets is traversed, and the values ​​of multiple pixels corresponding to the currently traversed pixel are extracted from the initial lunar DEM data to form a feature vector. The feature vector is combined with the label of the currently traversed cell to obtain the training sample dataset corresponding to each of the raster data subsets.

5. The method according to claim 4, characterized in that, The feature importance is multiplied by the corresponding feature vectors of each training sample dataset and then post-processed to obtain the upscaling result of the lunar DEM at the target resolution, including: The importance of the features is normalized to obtain the normalized feature importance. The normalized feature importance is multiplied by the corresponding feature vectors of each training sample dataset to obtain the DEM value set corresponding to different types of terrain and landforms after upscaling. The DEM value sets corresponding to different types of terrain and landforms after upscaling are shaped to generate upscaled lunar DEM results at the target resolution.

6. The method according to claim 5, characterized in that, The DEM value sets corresponding to different types of terrain and landforms after upscaling are reshaped to generate upscaled lunar DEM results at the target resolution, including: The DEM value sets corresponding to the different types of terrain and landforms after upscaling are rearranged to obtain the restored raster data; According to different types of terrain and landforms, different subsets of raster data are used to perform masking operations on the restored raster data. The masking results of different types of terrain and landforms are superimposed to obtain the upscaled results of the lunar DEM at the target resolution.

7. A lunar DEM upscaling system based on expert knowledge and a random forest model, characterized in that, include: The knowledge quantification unit is configured to have experts mark different types of terrain and landform areas at the initial resolution in the initial lunar DEM data, and quantify the expert knowledge to generate second raster data. Wherein, the second raster data is raster data at the target resolution, the target resolution is the resolution after upscaling, and the target resolution is lower than the initial resolution; the second raster data contains labels for different types of terrain and landforms at the target resolution. The sample generation unit is configured to construct training sample datasets corresponding to different types of terrain and landforms based on the second raster data and the initial lunar DEM data. The model training and output unit is configured to construct random forest models corresponding to different types of terrain and landforms based on the training sample dataset, train the random forest models until convergence, and output the feature importance corresponding to the feature vectors of each training sample dataset. The result generation unit is configured to multiply the feature importance with the corresponding feature vectors of each training sample dataset and perform post-processing to obtain the upscaling result of the lunar DEM at the target resolution. Experts marked different types of terrain and landform regions at the initial resolution in the initial lunar DEM data, including: At the initial resolution, experts used GIS software to mark the vector ranges of different types of terrain and landform areas using GIS software. All pixels of the initial lunar DEM data covered by the vector range are assigned a value of 1, and the remaining pixels are assigned a value of 0 to obtain different types of terrain and landform regions at the initial resolution, and these are recorded as the first raster data. The labeled areas are quantized to generate second raster data, following these steps: Based on the area covered by the first raster data, generate unassigned raster data according to the target resolution; Traverse all cells in the unassigned raster data and assign labels to each cell to generate a second raster data containing labels for different types of terrain features.

8. An electronic device, characterized in that, include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the steps of the lunar DEM upscaling method as described in any one of claims 1 to 6.

Images