Method and device for determining thickness of lunar soil, storage medium and program product
By using independently trained external and internal crater detection models, combined with spatial inclusion relationships and repose angles, the problem of low efficiency and accuracy in determining lunar regolith thickness was solved, and automated and high-precision inversion of lunar regolith thickness was achieved.
Patent Information
- Application Number
- CN202610799305.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-06-04
- Publication Date
- 2026-08-25
AI Technical Summary
Existing technologies for determining lunar regolith thickness are inefficient and lack precision, and it is difficult to achieve automated and high-precision inversion.
Independently trained outer crater detection models and inner crater detection models were used to obtain the center position and diameter information, respectively. The outer and inner craters were matched by spatial inclusion relationship, and the lunar regolith thickness was determined by combining the angle of repose.
It improved the accuracy of internal and external crater detection and diameter measurement precision, reduced the mismatch rate, realized automated inversion of lunar regolith thickness, and improved estimation efficiency and accuracy.
Smart Images
Figure CN122636705A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of planetary geology and remote sensing data processing technology, and in particular to a method, apparatus, storage medium, and program product for determining lunar regolith thickness. Background Technology
[0002] The lunar regolith, or lunar regolith, is a layer of loose sediment covering the lunar bedrock. It serves as a recorder of the Moon's geological history, containing crucial information about the Moon's material composition, space environment, and the evolution of the solar system. The thickness of the lunar regolith is one of the most critical parameters for studying it; its accurate inversion is essential for understanding lunar geological evolution, assessing lunar resource potential, and ensuring the safety of future lunar exploration and landing missions.
[0003] Currently, lunar regolith thickness inversion based on impact crater morphology is the mainstream method in academia. The basic principle is that when an impact crater forms, its morphology is affected by the lunar regolith thickness. By measuring the geometric parameters of a specific type of impact crater (such as a concentric impact crater), the local lunar regolith thickness can be estimated.
[0004] However, the aforementioned methods heavily rely on manual visual interpretation to identify and measure the morphological parameters of impact craters, which is not only inefficient but also highly subjective. With the increasing volume of high-resolution lunar remote sensing data, how to quickly and accurately determine lunar regolith thickness from massive amounts of data has become a pressing technical problem in this field. Summary of the Invention
[0005] This invention provides a method, apparatus, storage medium, and program product for determining lunar soil thickness, in order to address the shortcomings of low efficiency and accuracy in determining lunar soil thickness in the prior art, and to improve the efficiency and accuracy of determining lunar soil thickness.
[0006] This invention provides a method for determining the thickness of lunar regolith, comprising the following steps.
[0007] The remote sensing images to be detected are input into the trained concentric circular impact crater outer crater detection model and the trained concentric circular impact crater inner crater detection model, respectively, to obtain the outer crater detection results and the inner crater detection results. The outer crater detection results and the inner crater detection results include the center position and diameter information. The outer crater detection model is trained based on the outer crater sample set, and the inner crater detection model is trained based on the inner crater sample set. Based on the spatial inclusion relationship between the outer crater detection results and the inner crater detection results, the outer crater and the inner crater are matched to obtain multiple concentric impact crater combinations; The lunar regolith thickness corresponding to each concentric impact crater assembly is determined based on the outer crater diameter, inner crater diameter, and angle of repose of each concentric impact crater assembly.
[0008] In one possible implementation, the methods for obtaining the outer pit sample set and the inner pit sample set include: Acquire lunar surface remote sensing images with solar incidence angles within a preset range and spatial resolution better than a preset resolution threshold; Based on the criteria for distinguishing between concentric impact craters and their inner and outer craters, the boundaries of the outer and inner craters are marked on lunar surface remote sensing images to generate sample labels. Based on lunar surface remote sensing images and sample labels, generate outer crater sample sets and inner crater sample sets.
[0009] In another possible implementation, the training methods for the external pit detection model and the internal pit detection model include: The outer pit sample set and the inner pit sample set are respectively input into the initial detection model for training. During the training process, the model parameters are updated by strengthening the iteration strategy to obtain the trained outer pit detection model and the trained inner pit detection model. Enhanced iterative strategies include: In each round of training iteration, the best model weights from the previous round of training are inherited as the initial weights. The best model weights are the model weights that make the model performance optimal among multiple model weights in the previous round of training. The current model is tested using a validation set to obtain validation results, which include: detection accuracy information and / or diameter error information. Detection accuracy information indicates the accuracy of the model's detection, and diameter error information is the error between the model's predicted crater diameter and the actual diameter. If the verification result does not meet the first preset detection condition, proceed to the next iteration. The first preset detection condition includes: the detection accuracy information is greater than the first preset accuracy threshold and the diameter error information is less than the first preset error threshold. If the verification results meet the first preset detection conditions, training is stopped, and the first detection model and the second detection model are obtained. The first detection model is the trained external pit detection model, and the second detection model is the trained internal pit detection model.
[0010] In another possible implementation, the method also includes: Multiple experimental sample sets are obtained. Each experimental sample set corresponds to a landform type. Each experimental sample set includes multiple experimental remote sensing images. The solar incidence angles of the multiple experimental remote sensing images are within a preset range and are distributed in a gradient according to fixed angular intervals. The first and second detection models were trained based on multiple experimental sample sets to obtain the trained external pit detection model and the trained internal pit detection model.
[0011] In another possible implementation, the method also includes: Multiple sets of experimental samples are input into the trained outer crater detection model and the trained inner crater detection model respectively to obtain multiple model detection results. Each model detection result corresponds to a solar incidence angle. Based on the detection results of multiple models, the range of target solar incidence angle is determined from the preset interval. The model detection results corresponding to the range of target solar incidence angle satisfy the second preset detection condition, which is higher than the first preset detection condition.
[0012] In another possible implementation, based on the spatial inclusion relationship between the outer and inner crater detection results, the outer and inner craters are matched to obtain multiple concentric circular impact crater combinations, including: Perform the first operation on each outer crater to obtain the target inner crater corresponding to the outer crater, thus obtaining a combination of multiple concentric impact craters. The first operation includes: Obtain at least one candidate inner pit corresponding to the outer pit, and determine the center-to-center distance between the center of the outer pit and the center of the candidate inner pit; Based on the distance between the centers, the diameter of the outer pit, and the diameter of the inner pit, the location information is determined. The location information is used to indicate whether the inner pit is entirely within the range of the outer pit. Based on the location information and confidence level of each candidate inner pit, the target inner pit corresponding to the outer pit is determined. The confidence level is output by the trained inner pit detection model. The target inner pit is the candidate inner pit that is located inside the outer pit and has the highest confidence level among at least one candidate inner pit.
[0013] In another possible implementation, after obtaining a combination of multiple concentric impact craters, the method further includes: Based on the location information of the outer crater center and the confidence level of the outer crater center of each concentric impact crater combination, multiple concentric impact crater combinations are deduplicated to obtain the deduplicated concentric impact crater combinations.
[0014] The present invention also provides a device for determining the thickness of lunar regolith, comprising the following modules: The processing module is used to input the remote sensing image to be detected into the trained outer pit detection model and the trained inner pit detection model respectively, and obtain the outer pit detection result and the inner pit detection result. The outer pit detection result and the inner pit detection result both include the center position and diameter information. The outer pit detection model is trained based on the outer pit sample set, and the inner pit detection model is trained based on the inner pit sample set. The processing module is also used to match the outer pit and the inner pit according to the spatial inclusion relationship between the outer pit detection results and the inner pit detection results, so as to obtain multiple concentric circle impact pit combinations. The processing module is also used to determine the lunar regolith thickness corresponding to each concentric impact crater combination based on the outer crater diameter, inner crater diameter, and angle of repose of each concentric impact crater combination.
[0015] The present invention also 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 any of the methods for determining lunar regolith thickness described above.
[0016] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any of the methods for determining lunar soil thickness described above.
[0017] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements any of the methods for determining lunar soil thickness described above.
[0018] The method, apparatus, storage medium, and program product for determining lunar regolith thickness provided by this invention extracts the regolith using independently trained outer crater detection models and inner crater detection models, respectively, and obtains detection results containing information on the center position and diameter. This avoids confusion between inner and outer crater features caused by a single model, improving the accuracy of inner and outer crater detection and diameter measurement precision. By matching outer and inner craters based on spatial inclusion relationships, correctly nested concentric circle combinations can be effectively screened, reducing the false matching rate and invalid matching. By directly determining the lunar regolith thickness based on the matched outer crater diameter, inner crater diameter, and angle of repose, the thickness inversion is automated, improving the efficiency and accuracy of thickness estimation. Attached Figure Description
[0019] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.
[0020] Figure 1 This is one of the flowcharts illustrating the method for determining lunar soil thickness provided by the present invention.
[0021] Figure 2 This is the second flowchart illustrating the method for determining lunar soil thickness provided by the present invention.
[0022] Figure 3 This is the third flowchart illustrating the method for determining lunar soil thickness provided by this invention.
[0023] Figure 4 This is a schematic diagram of the structure of the device for determining lunar soil thickness provided by the present invention.
[0024] Figure 5 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation
[0025] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.
[0026] The terms used in the description of this invention include and have, and any variations thereof, and are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or modules is not limited to the steps or modules listed, but may optionally include other steps or modules not listed, or may optionally include other steps or modules inherent to such processes, methods, products, or apparatus.
[0027] Furthermore, in this invention, the terms "exemplary" or "for example" are used to indicate examples, illustrations, or descriptions. Any embodiment or design described as exemplary or for example in this invention should not be construed as being more preferred or advantageous than other embodiments or design solutions. Rather, the use of "exemplary" or "for example" is intended to present concepts in a specific manner.
[0028] Before providing a detailed explanation of the embodiments of the present invention, the technical terms involved in the embodiments of the present invention will be introduced first.
[0029] (1) Outer crater: refers to the complete ring boundary of the entire concentric impact crater, which is usually represented as a complete circular or near-circular structure in remote sensing images.
[0030] (2) Inner pit: A core small pit inside the outer pit of the concentric impact crater, forming a concentric circle structure with the outer pit. Its diameter is usually much smaller than that of the outer pit.
[0031] (3) Angle of repose: that is, the angle of repose of a material, which refers to the maximum stable angle that the slope and the horizontal plane of a loose material (such as lunar soil) can form in a natural accumulation state. It is used to describe the internal friction characteristics of lunar soil.
[0032] The lunar regolith is a loose layer on the lunar surface composed of rock debris, dust, and impact ejecta. Its formation occurred over billions of years through meteorite impacts and volcanic activity, during which solar wind and cosmic ray particles were trapped. Therefore, it contains crucial information about the lunar geological evolution, space environment, and the history of the solar system. Lunar regolith thickness is one of the core parameters for studying lunar geological processes. Current inversion techniques mainly rely on impact crater morphology parameters from optical images combined with empirical formulas to calculate the thickness. Methods for estimating lunar regolith thickness using the geometric features of crater edges and bottoms can be extended to more complex terrains and lighting conditions using high-resolution imagery.
[0033] However, existing methods still have significant shortcomings: deep neural networks are poorly adapted to the complex lunar surface environment due to drastic changes in lighting conditions. The image features of outer and inner craters are highly similar, making single-model training prone to feature confusion and leading to decreased detection accuracy. Furthermore, the lack of a deduplication mechanism for multiple overlapping images results in data redundancy and redundant computation. These limitations make it difficult for current technology to meet the requirements for high-precision, automated lunar regolith thickness inversion at a full lunar scale.
[0034] To address the aforementioned technical problems, this invention provides a method for determining lunar regolith thickness. This method extracts regolith using independently trained outer and inner crater detection models, obtaining detection results containing center position and diameter information separately. This avoids confusion between inner and outer crater features caused by a single model, improving the accuracy of inner and outer crater detection and diameter measurement precision. By matching outer and inner craters based on spatial inclusion relationships, correctly nested concentric circle combinations can be effectively filtered out, reducing false matching rates and invalid matches. By directly determining lunar regolith thickness based on the matched outer crater diameter, inner crater diameter, and angle of repose, the thickness inversion is automated, improving the efficiency and accuracy of thickness estimation.
[0035] The following is combined Figures 1 to 5 The present invention describes a method, apparatus, storage medium, and program product for determining lunar soil thickness.
[0036] The execution entity of the method for determining lunar soil thickness provided by this invention can be a lunar soil thickness determining device, which can be an electronic device. Furthermore, the device can also be the central processing unit (CPU) of the electronic device, or a determining module within the electronic device for determining lunar soil thickness. This invention uses the execution of the lunar soil thickness determining device as an example to illustrate the method for determining lunar soil thickness provided by this invention.
[0037] The electronic device can be a terminal or a server. The server can be a single physical server, or a server cluster consisting of multiple servers. Alternatively, the server cluster can be a distributed cluster. Alternatively, the server can be a cloud server. This invention does not limit the specific implementation of the server.
[0038] Figure 1 This is one of the flowcharts illustrating the method for determining lunar regolith thickness provided by the present invention, such as... Figure 1 As shown, the method includes the following: Step 101: Input the remote sensing image to be detected into the trained external pit detection model and the trained internal pit detection model respectively to obtain the external pit detection results and the internal pit detection results.
[0039] The outer crater detection model is a model of the outer crater of a concentric impact crater, and the inner crater detection model is a model of the inner crater of a concentric impact crater. The remote sensing image to be detected can be a high-resolution image acquired by the Narrow Angle Camera (NAC) aboard the Lunar Reconnaissance Orbiter (LRO).
[0040] In one possible implementation, the detected remote sensing images are input into two independent models, which then output corresponding detection results. Both the outer crater detection results and the inner crater detection results include at least the center position and diameter information of the impact crater.
[0041] For example, each detection result includes: center coordinates (e.g., latitude and longitude in lunar geodetic coordinate system), diameter (in meters), and confidence level (a probability value between 0 and 1).
[0042] It should be noted that the image coordinate system can adopt the Moon2000 lunar geodetic coordinate system, and the spatial projection method is sinusoidal projection, in order to reduce area distortion when stitching large areas together.
[0043] In this embodiment of the invention, the external pit detection model is pre-trained based on the external pit sample set, and the internal pit detection model is pre-trained based on the internal pit sample set.
[0044] The following section introduces the methods for obtaining the outer pit sample set and the inner pit sample set.
[0045] In one possible implementation, lunar surface remote sensing images are acquired where the solar incidence angle is within a preset range and the spatial resolution is better than a preset resolution threshold. Based on the criteria for distinguishing between impact craters and their interior / exterior boundaries, the boundaries of the outer and inner craters are marked on the lunar surface remote sensing images, generating sample labels. Based on the lunar surface remote sensing images and the sample labels, outer crater sample sets and inner crater sample sets are generated.
[0046] It should be understood that to ensure sample quality, the original images need to be screened. For example, the preset range could be 20° to 70°. Under this lighting condition, the ring structure of the impact crater is relatively clear. The preset resolution threshold could be 1 meter per pixel, that is, images with a spatial resolution better than 1 meter are preferred. Alternatively, the coverage of the lunar surface remote sensing image can be limited to within 60° north and south latitude on the Moon.
[0047] The criteria for distinguishing between inner and outer craters include, but are not limited to, the clarity of the crater rim, geometric integrity, and the ratio of the inner and outer crater diameters. Specifically, crater rim clarity requires the outer crater to have a complete annular boundary, a flat inner crater bottom, and the inner crater to have a clear boundary located inside the bottom of the outer crater. Geometric integrity requires both the outer and inner craters to be approximately circular, with the inner crater entirely within the outer crater's area and not touching its inner wall. The ratio of the inner and outer crater diameters requires the ratio of the outer crater diameter to the inner crater diameter to be within a reasonable empirical range; that is, the inner crater size is significantly smaller than the outer crater size but not so small as to be unmeasurable. Only impact craters that simultaneously meet all three conditions are labeled as either outer or inner craters for constructing the sample set.
[0048] During annotation, the complete annular edge of the outer pit and the core pit edge of the inner pit are delineated with polygons or circles respectively, and a label file corresponding to the image is generated. The label indicates the category (outer pit or inner pit) and location information (such as center coordinates and diameter) of each annotated object.
[0049] For example, lunar surface remote sensing images are paired with their corresponding label files to form training samples. All image-label pairs containing outer crater labels constitute the outer crater sample set, and all image-label pairs containing inner crater labels constitute the inner crater sample set. Typically, to facilitate model processing, the sample images are uniformly cropped to a fixed size, such as 512×512 pixels.
[0050] In some embodiments, lunar surface remote sensing images can be preprocessed before acquisition. Specifically, the original images and their associated spacecraft attitude, orbit, and camera geometry parameters can be imported. Then, radiometric correction is performed on the original images to eliminate sensor response differences. Next, corrections for stripe noise and echo effects are performed on the original images. The corrected images are then orthorectified using a pre-defined digital elevation model and uniformly projected onto a pre-defined coordinate system (such as a sinusoidal projection coordinate system).
[0051] Step 102: Based on the spatial inclusion relationship between the outer pit detection results and the inner pit detection results, match the outer pit and the inner pit to obtain multiple concentric impact pit combinations.
[0052] In one possible implementation, after obtaining the detection results of the outer and inner craters, they need to be paired. Not all outer craters contain inner craters, nor are all inner craters contained within a suitable outer crater. This step, based on spatial geometric logic, determines whether an inner crater belongs to a certain outer crater. Specifically, if the entire circular region of an inner crater is contained within the circular region of an outer crater, then the two impact craters are considered a concentric circle impact crater combination.
[0053] In some embodiments, multiple concentric impact crater combinations can be deduplicated based on the location information of the outer crater center and the confidence level of the outer crater center of each concentric impact crater combination, to obtain a deduplicated concentric impact crater combination.
[0054] Multiple imaging operations by the lunar orbiter can result in the same area being covered by multiple images, meaning the same impact crater might be detected multiple times in different images. To eliminate this redundancy, this embodiment employs a deduplication strategy based on spatial indexing and confidence priority.
[0055] In one example, the strategy operates as follows: First, extract the outer crater centers of all concentric impact crater combinations to construct a point set. Then, build a spatial tree index, such as a K-Dimensional Tree (KDTree), to manage these center points. Next, set a neighborhood threshold (e.g., 500 meters) for each center point and perform queries within that neighborhood. If the distance between two or more centers is less than the threshold (e.g., 500 meters), they are considered to represent the same physical impact crater. When such duplicates occur, retain the combination that makes up the concentric circle combination with the highest average confidence between the outer and inner craters, while removing other duplicate combinations. The final retained combinations are the deduplicated result, ensuring that each physical impact crater appears only once in the final output.
[0056] Step 103: Determine the lunar regolith thickness corresponding to each concentric impact crater combination based on the outer crater diameter, inner crater diameter, and angle of repose of each concentric impact crater combination.
[0057] For each successfully matched concentric impact crater combination, its morphological parameters reflect the thickness of the lunar regolith.
[0058] In one possible implementation, the lunar regolith thickness corresponding to the concentric impact crater combination satisfies the following formula: In one example, the formula for calculating lunar regolith thickness used in step 103 is as follows: in, To calculate the thickness of the lunar regolith, Let be the diameter of the outer crater in the concentric impact crater assembly. The diameter of the inner crater in the concentric impact crater assembly. The angle of repose is the material's stationary angle, a known angle value (e.g., 31°). These are empirical constants derived from experimental data, for example, in one embodiment, It can take the value 1.2.
[0059] It is understandable that the ratio of the outer crater diameter to the inner crater diameter reflects the proportion of the impact crater penetration depth. This ratio is then combined with a constant k and multiplied by a scale factor related to the outer crater size. Finally, after geometric corrections related to the angle of repose, the absolute thickness of the lunar regolith was estimated.
[0060] In some embodiments, after determining the lunar regolith thickness corresponding to each concentric impact crater combination, a continuous lunar regolith thickness distribution is obtained by spatial interpolation.
[0061] For example, each concentric impact crater combination is treated as a discrete sampling point, with its location defined by the coordinates of the outer crater's center and its attribute value being the calculated lunar regolith thickness. Then, spatial interpolation algorithms, such as ordinary kriging interpolation, are used to spatially extrapolate these discrete points, thereby generating a continuous lunar regolith thickness raster map covering the entire study area. The resulting continuous thickness distribution not only visually presents the spatial variation trend of lunar regolith thickness but also provides more comprehensive and detailed basic data support for tasks such as landing site selection, lunar geological evolution analysis, and resource assessment.
[0062] Based on the above technical solution, by employing independently trained outer and inner crater detection models for extraction, and obtaining detection results containing information on the center position and diameter, the confusion between inner and outer crater features caused by a single model is avoided, thus improving the accuracy of inner and outer crater detection and the precision of diameter measurement. By matching outer and inner craters according to spatial inclusion relationships, correctly nested concentric circle combinations can be effectively filtered out, reducing the false matching rate and invalid matches. By directly determining the lunar regolith thickness based on the matched outer crater diameter, inner crater diameter, and angle of repose, the thickness inversion is automated, improving the efficiency and accuracy of thickness estimation.
[0063] In this embodiment, step 102 (matching the outer pit and the inner pit according to the spatial inclusion relationship between the outer pit detection results and the inner pit detection results to obtain multiple concentric impact crater combinations) specifically includes: performing a first operation on each outer pit to determine its corresponding target inner pit, and finally converging all successfully matched outer pit-inner pit pairs to form multiple concentric impact crater combinations.
[0064] The first operation will be described below. Figure 2 This is a second schematic flowchart of the method for determining lunar regolith thickness provided by the present invention, as shown below. Figure 2 As shown, step 102 of this method includes the following: Step 201: Obtain at least one candidate inner pit corresponding to the outer pit, and determine the center distance between the center of the outer pit and the center of the candidate inner pit.
[0065] In one possible implementation, for any one of the multiple outer pits, all inner pit detection results are iterated through, and inner pits whose center positions are located within the region of that outer pit are selected as candidate inner pits. Then, for each candidate inner pit, the Euclidean distance between them is calculated based on the center coordinates of both.
[0066] For example, the center-to-center distance between the center of the outer pit and the center of the candidate inner pit satisfies the following formula: Where d represents the distance between the centers of the circles, ( ) is used to represent the coordinates of the center of the outer pit, ( , () is used to represent the coordinates of the center of the inner pit.
[0067] Step 202: Determine the location information based on the distance between the centers, the diameter of the outer pit, and the diameter of the inner pit.
[0068] The location information is used to indicate whether the inner pit is entirely within the range of the outer pit.
[0069] For example, the decision condition can be expressed by an inequality. For instance, the decision condition can be set as: .in, The diameter of the inner pit. The diameter of the outer pit.
[0070] It is understandable that the distance between the centers plus the radius of the inner pit is... This ensures that the inner pit is completely inside the outer pit and does not touch the boundary. For a candidate inner pit, if its center distance d satisfies this condition, its location information is marked as inner; otherwise, it is marked as outer.
[0071] Step 203: Determine the target inner pit corresponding to the outer pit based on the location information and confidence level of each candidate inner pit.
[0072] The confidence score is the output of the trained pit detection model, representing the degree of certainty with which the model believes the detection result is a real pit.
[0073] In this step, based on the center distance, the diameter of the outer pit, and the diameter of the inner pit, it is determined that there are multiple candidate inner pits corresponding to the outer pit. The candidate inner pit with the highest confidence is selected as the target inner pit for that outer pit. If there are no matching inner pits, the outer pit is considered to have no matching inner pit and does not constitute a concentric circle combination.
[0074] Based on the above technical solution, by calculating the center-to-center distance between the outer crater center and the candidate inner crater center, and combining the outer and inner crater diameters to determine whether the inner crater is entirely located inside the outer crater, nested combinations that conform to spatial inclusion relationships can be accurately screened. This improves the geometric accuracy of concentric impact crater matching and reduces the mismatch rate of misclassifying non-nested or partially overlapping craters as valid combinations. Furthermore, by selecting the inner crater with the highest confidence level that is entirely located inside the outer crater as the target inner crater based on the confidence level of each candidate inner crater, the reliability metric of the model output is fully utilized. This improves the uniqueness of the one-to-one correspondence between inner and outer craters and the confidence level of the matching results, thus providing more accurate outer and inner crater diameter parameters for subsequent lunar regolith thickness inversion.
[0075] The training process of the external pit detection model and the internal pit detection model will be described in detail below.
[0076] In some embodiments, the outer pit sample set and the inner pit sample set can be input into the initial detection model for training, and the model parameters can be updated by a reinforcement iteration strategy during the training process to obtain the trained outer pit detection model and the trained inner pit detection model.
[0077] For example, a deep learning network architecture suitable for object detection, such as YOLOv9, can be selected, and pre-trained weights can be loaded as the initial detection model. This network model can then be trained independently using the outer pit sample set and the inner pit sample set, i.e., two models can be trained.
[0078] During training, data augmentation operations can be performed on the sample images, including random translation, multi-scale scaling, random rotation, and color space transformation, while ensuring that the sample labels and the augmented images are synchronized.
[0079] The training process is not completed in one go, but rather a reinforcement iterative strategy is used to continuously optimize the model. This strategy specifically includes the following sub-steps: Sub-step 1: In each round of training iteration, inherit the best model weights from the previous round of training as the initial weights.
[0080] Let the iteration number be i. i=1, 2, 3,... When i=1, the initial weights are the pre-trained weights. Starting from the second round, the model is not trained from scratch, but instead loads the model weight file that performed best on the validation set in the previous round (i-1 round) as the initial weights for this round. The optimal model weights are the model weights that achieved the best model performance among multiple model weights during the previous round of training.
[0081] Sub-step 2: Use the validation set to test the current model and obtain the validation results.
[0082] After each training round, the current model is evaluated using a separate validation set (data not used in that round of training). The validation results should include at least: detection accuracy information and / or diameter error information.
[0083] Among them, the detection accuracy information is used to indicate the accuracy of the model detection, and the diameter error information is the error between the model's predicted impact crater diameter and the actual diameter.
[0084] For example, detection precision information may include at least one of the following: precision, recall, or mean average precision (mAP).
[0085] Precision P = TP / (TP + FP), where TP is the number of correctly detected positive samples and FP is the number of incorrectly detected negative samples. Recall R = TP / (TP + FN), where FN is the number of undetected positive samples.
[0086] The diameter error information satisfies the following formula: in, The predicted outer / inner pit diameter from the model. This is the actual diameter.
[0087] Sub-step 3: Determine whether the verification result meets the first preset detection condition.
[0088] The first preset detection condition is a pre-defined standard used to judge whether the model is accurate enough. For example, the first preset detection condition can be set as follows: the average accuracy of the model on the validation set reaches above 0.85, while the average relative error of the impact crater diameter detection is controlled within 10%.
[0089] In other words, when the model's detection accuracy meets the above numerical requirements, the model performance can be considered satisfactory, and training can be stopped (execute sub-step 5). Conversely, if the average accuracy is below 0.85 or the relative error of the diameter exceeds 10%, it indicates that the model still has shortcomings and needs to enter the next round of reinforcement iterations to continue optimization by supplementing and correcting samples and adjusting training parameters (execute sub-step 4).
[0090] Sub-step 4: If the verification result does not meet the first preset detection condition, proceed to the next iteration.
[0091] When model performance is substandard, the training data will be optimized. Specifically, for results where the model performed poorly in the previous detection, such as low-confidence detections (e.g., confidence below 0.7), false positives or false negatives, and targets with large relative diameter errors, recalibration or supplementary calibration will be performed. These corrected or added samples will be added to the training set to form an enhanced training set, and then the next round (round i+1) of training will be performed, returning to sub-step 1.
[0092] Sub-step 5: If the verification results meet the first preset detection conditions, stop training and obtain the first detection model and the second detection model.
[0093] Once the model performance stably meets the first preset detection condition, the training process terminates. The first detection model trained at this point is used as the final trained external pit detection model, and the second detection model is used as the final trained internal pit detection model.
[0094] Based on the above technical solution, by inheriting the model weights that optimized the model performance in the previous training round as the initial weights in each iteration, repeated training from scratch is avoided, thus improving the convergence speed and training efficiency of model parameters. By using a validation set to obtain detection accuracy information and diameter error information, and judging whether the stopping condition is met based on preset accuracy thresholds and error thresholds, the iteration can be terminated in a timely manner after the model performance reaches the target, reducing ineffective training overhead. By incorporating diameter error information into the iteration stopping criterion, the accuracy of impact crater diameter measurement is directly optimized, improving the model's regression accuracy of diameter parameters, thereby providing higher accuracy inputs for the outer and inner crater diameters for subsequent thickness inversion.
[0095] In some embodiments, a multi-lighting sample enhancement training method may be further introduced to improve the model's generalization ability under different lighting conditions.
[0096] Figure 3 This is the third flowchart illustrating the method for determining lunar regolith thickness provided by this invention, as shown below. Figure 3 As shown, after obtaining the first detection model and the second detection model, the method further includes the following: Step 301: Obtain multiple sets of experimental samples.
[0097] One experimental sample set corresponds to a specific landform type.
[0098] For example, landform types include lunar maria, highlands, etc. For instance, multiple (e.g., 20) typical experimental zones are designated, covering different latitudes, lighting conditions, and landform features.
[0099] In this embodiment of the invention, an experimental sample set includes multiple experimental remote sensing images, the solar incidence angles of the multiple experimental remote sensing images are located within a preset range, and are distributed in a gradient according to fixed angular intervals.
[0100] For example, the preset range is 20° to 70°.
[0101] Step 302: Train the first detection model and the second detection model based on multiple experimental sample sets to obtain the trained outer pit detection model and the trained inner pit detection model.
[0102] In one possible implementation, multiple experimental sample sets are added to the outer pit sample set and the inner pit sample set to obtain a new training set. Then, the first detection model and the second detection model are trained using the new training set to obtain the trained outer pit detection model and the trained inner pit detection model.
[0103] It should be noted that the specific training process can be referred to the above embodiments, and will not be repeated here.
[0104] It should be understood that images covering different illumination gradients, obtained from all experimental sample sets, will be added to the training process. This means that the model's training data is no longer limited to samples under a single illumination condition, but includes impact crater images with rich illumination variations taken under various solar incidence angles. By learning from this diverse data, the model can adapt to the morphological differences exhibited by impact craters under different illumination conditions (such as different shadow lengths and edge contrasts), thereby obtaining a more generalized scheme for determining lunar regolith thickness.
[0105] In some embodiments, after obtaining the trained outer crater detection model and the trained inner crater detection model, multiple sets of experimental samples can be input into the trained outer crater detection model and the trained inner crater detection model respectively to obtain multiple model detection results, with one model detection result corresponding to one solar incidence angle.
[0106] In other words, for each incident angle, quantitative indicators such as the impact crater detection accuracy, recall rate, and diameter measurement error under that illumination condition can be statistically calculated.
[0107] Then, based on the detection results of multiple models, the target solar incidence angle range is determined from the preset interval. The model detection results corresponding to the target solar incidence angle range meet the second preset detection condition, which is higher than the first preset detection condition.
[0108] By analyzing and comparing the detection results corresponding to all incident angles, a range of solar incident angles with optimal performance, namely the target solar incident angle range, is determined.
[0109] For example, the initial preset range was 20° to 70°, but experiments showed that when the incident angle was below 30°, the image contrast was low and effective results were few. When the incident angle was above 65°, the detection accuracy and diameter accuracy dropped sharply. The model achieved the highest detection accuracy and diameter accuracy between 52° and 57°. Therefore, the target solar incident angle range was determined to be 30° to 65°, with the optimal range being 52° to 57°. The model detection results corresponding to this target solar incident angle range need to meet a second preset detection condition that is more stringent than the first preset detection condition. For example, it requires not only mAP50 > 0.9 but also an average diameter relative error of less than 5%, representing the model's performance under optimal conditions. This range will serve as the image selection criterion for subsequent automated extraction at the full-month scale, thereby maximizing the accuracy of the overall inversion results.
[0110] The apparatus for determining lunar regolith thickness provided by the present invention will be described below. The apparatus described below corresponds to the method described above for determining lunar regolith thickness. It is understood that, in order to achieve the above functions, the electronic device includes hardware structures and / or software modules corresponding to the execution of each function. Those skilled in the art should readily recognize that, based on the steps of the lunar regolith thickness determination method described in conjunction with the embodiments disclosed herein, the present invention can be implemented in hardware or a combination of hardware and computer software. Whether a function is executed in hardware or by computer software driving hardware depends on the specific application and design constraints of the technical solution. 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 present invention.
[0111] Figure 4 This is a schematic diagram of the structure of the device for determining lunar soil thickness provided by the present invention, as shown below. Figure 4 As shown, the device for determining the thickness of lunar regolith includes the following modules: a processing module 401 and an acquisition module 402.
[0112] The processing module 401 is used to input the remote sensing image to be detected into the trained outer pit detection model and the trained inner pit detection model respectively, and obtain the outer pit detection result and the inner pit detection result; wherein, the outer pit detection result and the inner pit detection result both include the center position and diameter information, the outer pit detection model is trained based on the outer pit sample set, and the inner pit detection model is trained based on the inner pit sample set; The processing module 401 is also used to match the outer pit and the inner pit according to the spatial inclusion relationship between the outer pit detection results and the inner pit detection results to obtain a combination of multiple concentric impact pits. The processing module 401 is also used to determine the lunar regolith thickness corresponding to each concentric impact crater combination based on the outer crater diameter, inner crater diameter, and angle of repose of each concentric impact crater combination.
[0113] In one possible implementation, an acquisition module is used to acquire lunar surface remote sensing images with a solar incidence angle within a preset range and a spatial resolution better than a preset resolution threshold. A processing module is used to label the outer and inner crater boundaries on the lunar surface remote sensing images according to the criteria for distinguishing between impact craters and their interiors, generating sample labels. The processing module is also used to generate outer crater sample sets and inner crater sample sets based on the lunar surface remote sensing images and the sample labels.
[0114] Figure 5 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 5 As shown, the electronic device may include a processor 510, a communications interface 520, a memory 530, and a communication bus 540. The processor 510, communications interface 520, and memory 530 communicate with each other via the communication bus 540. The processor 510 can call logical instructions in the memory 530 to execute a method for determining lunar regolith thickness. This method includes: inputting the remote sensing image to be detected into a trained outer crater detection model and a trained inner crater detection model, respectively, to obtain outer crater detection results and inner crater detection results; wherein both the outer crater detection results and the inner crater detection results include the center position and diameter information, and the outer crater detection model is trained based on an outer crater sample set, and the inner crater detection model is trained based on an inner crater sample set. Based on the spatial inclusion relationship between the outer crater detection results and the inner crater detection results, the outer crater and the inner crater are matched to obtain multiple concentric circular impact crater combinations. Based on the outer crater diameter, inner crater diameter, and angle of repose of each concentric circular impact crater combination, the lunar regolith thickness corresponding to each concentric circular impact crater combination is determined.
[0115] Furthermore, the logical instructions in the aforementioned memory 530 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.
[0116] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the method for determining lunar regolith thickness provided by the above-described methods. This method includes: inputting the remote sensing image to be detected into a trained outer crater detection model and a trained inner crater detection model, respectively, to obtain outer crater detection results and inner crater detection results; wherein both the outer crater detection results and the inner crater detection results include the center position and diameter information, and the outer crater detection model is trained based on an outer crater sample set, and the inner crater detection model is trained based on an inner crater sample set. Based on the spatial inclusion relationship between the outer crater detection results and the inner crater detection results, matching the outer crater and the inner crater is performed to obtain multiple concentric circular impact crater combinations. Based on the outer crater diameter, inner crater diameter, and angle of repose of each concentric circular impact crater combination, the lunar regolith thickness corresponding to each concentric circular impact crater combination is determined.
[0117] Furthermore, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon. When executed by a processor, this computer program implements a method for determining lunar regolith thickness provided by the aforementioned methods. This method includes: inputting a remote sensing image to be detected into a trained outer crater detection model and a trained inner crater detection model, respectively, to obtain outer crater detection results and inner crater detection results; wherein both the outer crater detection results and the inner crater detection results include center position and diameter information; the outer crater detection model is trained based on an outer crater sample set, and the inner crater detection model is trained based on an inner crater sample set. Based on the spatial inclusion relationship between the outer crater detection results and the inner crater detection results, matching the outer craters and inner craters yields multiple concentric impact crater combinations. Based on the outer crater diameter, inner crater diameter, and angle of repose of each concentric impact crater combination, determining the lunar regolith thickness corresponding to each concentric impact crater combination.
[0118] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.
[0119] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.
[0120] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A method for determining the thickness of lunar regolith, characterized in that, The method includes: The remote sensing image to be detected is input into the trained concentric circular impact crater outer crater detection model and the trained concentric circular impact crater inner crater detection model, respectively, to obtain the outer crater detection result and the inner crater detection result; wherein, the outer crater detection result and the inner crater detection result both include the center position and diameter information, the outer crater detection model is trained based on the outer crater sample set, and the inner crater detection model is trained based on the inner crater sample set; Based on the spatial inclusion relationship between the outer crater detection results and the inner crater detection results, the outer crater and the inner crater are matched to obtain multiple concentric impact crater combinations. The lunar regolith thickness corresponding to each concentric impact crater assembly is determined based on the outer crater diameter, inner crater diameter, and angle of repose of each concentric impact crater assembly.
2. The method according to claim 1, characterized in that, The training methods for the external pit detection model and the internal pit detection model include: The outer pit sample set and the inner pit sample set are respectively input into the initial detection model for training. During the training process, the model parameters are updated by an enhancement iteration strategy to obtain the trained outer pit detection model and the trained inner pit detection model. The enhanced iteration strategy includes: In each round of iterative training, the best model weights from the previous round of training are inherited as the initial weights. The best model weights are the model weights that make the model performance optimal among multiple model weights in the previous round of training. The current model is tested using a validation set to obtain validation results, which include: detection accuracy information and / or diameter error information. The detection accuracy information is used to indicate the accuracy of the model's detection, and the diameter error information is the error between the model's predicted impact crater diameter and the actual diameter. If the verification result does not meet the first preset detection condition, proceed to the next iteration. The first preset detection condition includes: the detection accuracy information is greater than the first preset accuracy threshold and the diameter error information is less than the first preset error threshold. If the verification result meets the first preset detection condition, training is stopped, and a first detection model and a second detection model are obtained. The first detection model is the trained external pit detection model, and the second detection model is the trained internal pit detection model.
3. The method according to claim 2, characterized in that, The method further includes: Multiple experimental sample sets are obtained, each experimental sample set corresponds to a landform type, and each experimental sample set includes multiple experimental remote sensing images. The solar incidence angles of the multiple experimental remote sensing images are located within a preset range and are distributed in a gradient according to fixed angular intervals. The first detection model and the second detection model are trained based on the multiple experimental sample sets to obtain the trained external pit detection model and the trained internal pit detection model.
4. The method according to claim 3, characterized in that, The method further includes: The multiple sets of experimental samples are respectively input into the trained outer crater detection model and the trained inner crater detection model to obtain multiple model detection results, and one model detection result corresponds to one solar incidence angle; Based on the multiple model detection results, the target solar incidence angle range is determined from the preset interval. The model detection results corresponding to the target solar incidence angle range satisfy the second preset detection condition, which is higher than the first preset detection condition.
5. The method according to claim 1, characterized in that, The spatial inclusion relationship between the outer crater detection results and the inner crater detection results is used to match the outer crater and the inner crater to obtain multiple concentric circular impact crater combinations, including: Perform a first operation on each outer crater to obtain the target inner crater corresponding to the outer crater, thereby obtaining the combination of the plurality of concentric impact craters. The first operation includes: Obtain at least one candidate inner pit corresponding to the outer pit, and determine the center-to-center distance between the center of the outer pit and the center of the candidate inner pit; Based on the distance between the centers, the diameter of the outer pit, and the diameter of the inner pit, position information is determined, which is used to indicate whether the inner pit is entirely located within the range of the outer pit; Based on the location information of each candidate inner pit and the confidence level of each candidate inner pit, the target inner pit corresponding to the outer pit is determined. The confidence level is output by the trained inner pit detection model. The target inner pit is the candidate inner pit that is located entirely within the outer pit and has the highest confidence level among the at least one candidate inner pit.
6. The method according to claim 1, characterized in that, After obtaining a combination of multiple concentric impact craters, the method further includes: Based on the position information of the outer crater center of each concentric impact crater combination and the confidence level of the outer crater center, the multiple concentric impact crater combinations are deduplicated to obtain the deduplicated concentric impact crater combinations.
7. A device for determining the thickness of lunar regolith, characterized in that, The device includes: The processing module is used to input the remote sensing image to be detected into the trained concentric circular impact crater outer crater detection model and the trained concentric circular impact crater inner crater detection model, respectively, to obtain the outer crater detection result and the inner crater detection result; wherein, the outer crater detection result and the inner crater detection result both include the center position and diameter information, the outer crater detection model is trained based on the outer crater sample set, and the inner crater detection model is trained based on the inner crater sample set; The processing module is also used to match the outer pit and the inner pit according to the spatial inclusion relationship between the outer pit detection results and the inner pit detection results to obtain a combination of multiple concentric impact pits. The processing module is also used to determine the lunar regolith thickness corresponding to each concentric impact crater combination based on the outer crater diameter, inner crater diameter, and angle of repose of each concentric impact crater combination.
8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the method for determining the thickness of lunar regolith as described in any one of claims 1 to 6.
9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method for determining the thickness of lunar regolith as described in any one of claims 1 to 6.
10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the method for determining the thickness of lunar regolith as described in any one of claims 1 to 6.