Methods, apparatus, electronic devices and storage media for generating images of weathered rocks on Mars

Abstract

This application specifically relates to a method, apparatus, electronic device, and storage medium for generating images of Martian weathered rocks. The method includes: acquiring Martian weathered rock samples; generating a Martian rock dataset based on the samples; transferring knowledge from a pre-defined CGAN model to the Martian rock dataset to obtain a transfer-optimized Martian rock dataset; generating microscopic images of Martian weathered rocks based on the transfer-optimized dataset; comparing the microscopic images with the samples; and evaluating the generation quality of the microscopic images based on the comparison results. Thus, by using transfer learning from an Earth weathered rock database to construct a conditional generative adversarial network and generate microscopic images of Martian weathered rocks, this method solves the problem of significant limitations in characterizing Martian rock features in related technologies, providing technical support for Mars probe landing site selection, drilling strategy formulation, and in-situ resource utilization.

Description

Technical Field

[0001] This application relates to the fields of digital rock cores, artificial intelligence (AI) and numerical simulation technology, and in particular to a method, apparatus, electronic device and storage medium for generating images of weathered rocks on Mars. Background Technology

[0002] The microstructure of Martian rocks directly affects their physical and mechanical properties, and is of significant scientific value for understanding Martian surface weathering, geological evolution, and planetary resource assessment. However, due to limitations in Mars exploration missions, Martian rock research in related technologies has mainly focused on mineral composition analysis, while systematic studies on their microstructure, physical and mechanical properties, and weathering mechanisms remain relatively scarce. The lack of microscopic image data limits the quantitative analysis of the pore structure, fracture evolution, and mechanical properties of Martian rocks, thus affecting data interpretation, landing site selection, and resource development strategies for Mars exploration missions. Traditional laboratory testing methods are difficult to apply directly due to the extreme scarcity of Martian rock samples and limited experimental conditions. While analogical studies based on weathered rocks on Earth provide a reference, the significant differences between Earth and Martian environmental conditions (such as gravity, temperature, and atmospheric composition) limit their ability to characterize Martian rock properties. Therefore, developing a microscopic image generation method suitable for Martian rocks is crucial for Martian rock mechanics research.

[0003] Among related technologies, generative AI technologies such as generative adversarial networks, variational autoencoders, and diffusion models have made significant progress in areas such as Earth rock image generation, pore structure modeling, and rock physical parameter prediction.

[0004] However, research in related technologies is mainly based on training on Earth rock samples, and the generalization ability of the models depends on a large amount of data. Martian rock samples are extremely limited, making it difficult for existing AI models to accurately reflect the characteristics of Martian weathered rocks when data is insufficient, which urgently needs to be addressed. Summary of the Invention

[0005] This application provides a method, apparatus, electronic device, and storage medium for generating images of weathered rocks on Mars, in order to solve the problem that there are significant limitations in the description of Martian rock characteristics in related technologies, and to provide technical support for the selection of landing sites for Mars probes, the formulation of drilling strategies, and the utilization of in-situ resources.

[0006] The first aspect of this application provides a method for generating images of weathered rocks on Mars, including the following steps:

[0007] Obtain samples of weathered Martian rocks;

[0008] A Martian rock dataset is generated based on the Martian weathered rock samples. The knowledge of the preset CGAN model is then transferred to the Martian rock dataset to obtain a transfer-optimized Martian rock dataset.

[0009] Microscopic images of Martian weathered rocks are generated based on the migration-optimized Martian rock dataset. These microscopic images are then compared with the Martian weathered rock samples, and the generation quality of the microscopic images is evaluated based on the comparison results.

[0010] Optionally, prior to obtaining Martian weathered rock samples, the method further includes:

[0011] Obtain various Earth rock samples;

[0012] Microscopic and macroscopic tests were performed on the various Earth rock samples to obtain their microscopic and macroscopic parameters, respectively. The microscopic and macroscopic parameters were then standardized to obtain the processed microscopic and macroscopic parameters.

[0013] A multi-scale database of Earth's weathered rocks is established based on the processed microscopic and macroscopic physical and mechanical parameters.

[0014] Optionally, before obtaining Martian weathered rock samples, the method further includes:

[0015] Based on the aforementioned various Earth rock samples, multiple rock feature tags are generated for the various Earth rock samples; wherein, the rock feature tags include mineral composition tags, weathering degree tags, and microstructure feature tags;

[0016] A pre-defined CGAN model is constructed based on a conditional generative adversarial network and the rock feature labels.

[0017] Optionally, the preset CGAN model includes a CGAN model generator and a CGAN model discriminator, wherein,

[0018] The CGAN model generator is obtained based on random vectors and the weathering degree labels. The CGAN model generator is used to generate rock and mineral crystal models based on the random vectors, the rock feature labels and a preset loss function.

[0019] The discriminator of the preset CGAN model is obtained based on the rock and mineral crystal model and the weathering degree label. The discriminator of the preset CGAN model is used to discriminate the image generated by the rock and mineral crystal model.

[0020] Optionally, the preset loss function is:

[0021] L gen =Ladv +λ1L mineral-comp +λ2L mineral-shape +λ3L weathering-prod +λ4L porosity +λ5L SSIM ;

[0022] Among them, L adv Let λ1 be the adversarial loss of CGAN, and L be the weighted hyperparameter of the mineral composition loss. mineral-comp λ² represents the mineral composition loss, λ² represents the weighted hyperparameter of the mineral morphology loss, and L represents the mineral composition loss. mineral-shap λ3 represents the loss in mineral morphology, and L represents the weighted hyperparameter for the loss in weathering degree. weathering-prod For the loss due to weathering degree, λ4 is the weighted hyperparameter for the loss of porosity percentage, and L porosity λ5 is the porosity percentage loss, and L is the weighted hyperparameter of the structural similarity loss. SSIM This is the structural similarity loss.

[0023] Optionally, the comparison results include structural similarity, perceptual loss, peak signal-to-noise ratio, and Frechet Inception Distance. The evaluation of the generation quality of the Martian weathered rock microscopic image based on the comparison results includes:

[0024] Based on a preset quantitative evaluation method, the structural similarity, the perceptual loss, the peak signal-to-noise ratio, and the Frechet Inception Distance are compared with their corresponding thresholds to obtain multiple comparison results.

[0025] Based on the multiple comparison results, it is verified whether the microscopic image of Martian weathered rocks meets the preset validity conditions, and the generation quality of the microscopic image of Martian weathered rocks is evaluated according to the verification results.

[0026] A second aspect of this application provides an apparatus for generating images of weathered rocks on Mars, comprising:

[0027] The acquisition module is used to acquire Martian weathered rock samples;

[0028] The transfer module is used to generate a Martian rock dataset based on the Martian weathered rock samples, and to transfer the knowledge of the preset CGAN model to the Martian rock dataset to obtain a transfer-optimized Martian rock dataset.

[0029] The generation module is used to generate microscopic images of Martian weathered rocks based on the migration-optimized Martian rock dataset, compare the microscopic images of Martian weathered rocks with the Martian weathered rock samples, and evaluate the generation quality of the microscopic images of Martian weathered rocks based on the comparison results.

[0030] Optionally, prior to obtaining the Martian weathered rock sample, the acquisition module is further configured to:

[0031] Obtain various Earth rock samples;

[0032] Microscopic and macroscopic tests were performed on the various Earth rock samples to obtain their microscopic and macroscopic parameters, respectively. The microscopic and macroscopic parameters were then standardized to obtain the processed microscopic and macroscopic parameters.

[0033] A multi-scale database of Earth's weathered rocks is established based on the processed microscopic and macroscopic physical and mechanical parameters.

[0034] Optionally, before obtaining the Martian weathered rock sample, the acquisition module is further configured to:

[0035] Based on the aforementioned various Earth rock samples, multiple rock feature tags are generated for the various Earth rock samples; wherein, the rock feature tags include mineral composition tags, weathering degree tags, and microstructure feature tags;

[0036] A pre-defined CGAN model is constructed based on a conditional generative adversarial network and the rock feature labels.

[0037] Optionally, the preset CGAN model includes a CGAN model generator and a CGAN model discriminator, wherein,

[0038] The CGAN model generator is obtained based on random vectors and the weathering degree labels. The CGAN model generator is used to generate rock and mineral crystal models based on the random vectors, the rock feature labels and a preset loss function.

[0039] The discriminator of the preset CGAN model is obtained based on the rock and mineral crystal model and the weathering degree label. The discriminator of the preset CGAN model is used to discriminate the image generated by the rock and mineral crystal model.

[0040] Optionally, the preset loss function is:

[0041] L gen =L adv +λ1L mineral-comp +λ2L mineral-shape +λ3L weathering-prod +λ4L porosity +λ5L SSIM ;

[0042] Among them, L adv Let λ1 be the adversarial loss of CGAN, and L be the weighted hyperparameter of the mineral composition loss. mineral-compλ² represents the mineral composition loss, λ² represents the weighted hyperparameter of the mineral morphology loss, and L represents the mineral composition loss. mineral-shap λ3 represents the loss in mineral morphology, and L represents the weighted hyperparameter for the loss in weathering degree. weathering-prod For the loss due to weathering degree, λ4 is the weighted hyperparameter for the loss of porosity percentage, and L porosity λ5 is the porosity percentage loss, and L is the weighted hyperparameter of the structural similarity loss. SSIM This is the structural similarity loss.

[0043] Optionally, the comparison results include structural similarity, perceptual loss, peak signal-to-noise ratio, and Frechet Inception Distance, and the generation module is specifically used for:

[0044] Based on a preset quantitative evaluation method, the structural similarity, the perceptual loss, the peak signal-to-noise ratio, and the Frechet Inception Distance are compared with their corresponding thresholds to obtain multiple comparison results.

[0045] Based on the multiple comparison results, it is verified whether the microscopic image of Martian weathered rocks meets the preset validity conditions, and the generation quality of the microscopic image of Martian weathered rocks is evaluated according to the verification results.

[0046] A third aspect of this application provides an electronic device, including: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being configured to perform the Martian weathered rock image generation method as described in the above embodiments.

[0047] A fourth aspect of this application provides a computer-readable storage medium having a computer program stored thereon, which is executed by a processor to implement the Martian weathered rock image generation method as described in the above embodiments.

[0048] Therefore, this application's embodiments obtain Martian weathered rock samples, generate a Martian rock dataset, transfer knowledge from a pre-defined CGAN model to the Martian rock dataset to obtain a transfer-optimized Martian rock dataset, generate microscopic images of Martian weathered rocks, compare these microscopic images with Martian weathered rock samples, and evaluate the generation quality of the Martian weathered rock microscopic images based on the comparison results. Thus, by using transfer learning from an Earth weathered rock database to construct a conditional generative adversarial network and generate Martian weathered rock microscopic images, this addresses the significant limitations in characterizing Martian rock properties found in related technologies, providing technical support for Mars probe landing site selection, drilling strategy formulation, and in-situ resource utilization.

[0049] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0050] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein:

[0051] Figure 1 This is a flowchart of a method for generating images of weathered rocks on Mars according to an embodiment of this application;

[0052] Figure 2 This is a schematic diagram of small sample data of Martian meteorites provided in an embodiment of the present application for a method of generating images of weathered rocks on Mars;

[0053] Figure 3 This is a schematic diagram of the CGAN model construction framework for a method of generating images of weathered rocks on Mars according to an embodiment of this application;

[0054] Figure 4 This is a schematic diagram illustrating the knowledge transfer strategy of a Martian weathered rock image generation method according to an embodiment of this application;

[0055] Figure 5 This is a flowchart illustrating a method for generating images of weathered rocks on Mars according to an embodiment of this application;

[0056] Figure 6 This is a schematic diagram of an apparatus for generating images of weathered rocks on Mars according to an embodiment of this application;

[0057] Figure 7 This is a schematic diagram of the structure of an electronic device provided according to an embodiment of this application. Detailed Implementation

[0058] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.

[0059] The following describes a method, apparatus, electronic device, and storage medium for generating Martian weathered rock images according to embodiments of this application, with reference to the accompanying drawings. Addressing the significant limitations in characterizing Martian rock characteristics in related technologies mentioned in the background, this application provides a method for generating Martian weathered rock images. In this method, Martian weathered rock samples are acquired, a Martian rock dataset is generated, knowledge from a pre-defined CGAN model is transferred to the Martian rock dataset to obtain a transfer-optimized Martian rock dataset, microscopic images of Martian weathered rocks are generated, and the microscopic images are compared with Martian weathered rock samples. The generation quality of the microscopic images is evaluated based on the comparison results. Thus, by using transfer learning from an Earth weathered rock database to construct a conditional generative adversarial network and generate microscopic images of Martian weathered rocks, the significant limitations in characterizing Martian rock characteristics in related technologies are solved, providing technical support for Mars probe landing site selection, drilling strategy formulation, and in-situ resource utilization.

[0060] Specifically, Figure 1 This is a flowchart illustrating a method for generating images of weathered rocks on Mars, as provided in an embodiment of this application.

[0061] like Figure 1 As shown, the method for generating images of weathered rocks on Mars includes the following steps:

[0062] In step S101, a Martian weathered rock sample is obtained.

[0063] Specifically, such as Figure 2 As shown, Figure 2 This is a schematic diagram of small sample data of Martian meteorites, illustrating a method for generating images of Martian weathered rocks according to an embodiment of this application. The main diagenetic mineral composition of the Martian meteorite was determined using a fully automated mineral analysis system (Tescan Integrated Mineral Analyzer, TIMA). Load-displacement curves of the main diagenetic minerals were obtained based on nanoindentation, and their elastic modulus and hardness were calculated accordingly. Surface roughness and elastic modulus of intercrystalline phases of different minerals were obtained using atomic force microscopy. The three-dimensional mineral and micropore distribution of the Martian meteorite was obtained using computed tomography (CT) scanning technology.

[0064] Optionally, in some embodiments, before obtaining Martian weathered rock samples, the method further includes: obtaining multiple Earth rock samples; performing microscale and macroscale tests on the multiple Earth rock samples to obtain microscale and macroscale parameters of the multiple Earth rock samples respectively, and standardizing the microscale and macroscale parameters to obtain processed microscale and macroscale parameters; and establishing a multiscale database of Earth weathered rocks based on the processed microscale and macroscale physical and mechanical parameters.

[0065] Understandably, this involves acquiring Earth rock samples of varying weathering degrees and rock types, and conducting tests at both micro and macro scales. At the micro scale, microscopic parameters such as mineral composition, microstructure, diagenetic minerals, and intergranular phase mechanical properties are obtained through microscopic rock experiments. At the macro scale, macroscopic physical and mechanical parameters such as elastic modulus, hardness, compressive strength, tensile strength, fracture toughness, and wave velocity are obtained from Earth rock samples of different weathering degrees through macroscopic experiments such as uniaxial compression, Brazil splitting, and wave velocity testing. Combined with data standardization, a multi-scale database of Earth's weathered rocks is established. Image enhancement and feature extraction methods (such as edge detection, Fourier transform, and morphological analysis) are used to standardize the data, improving the quality of training data, reducing noise impact, and providing data support for subsequent weathering geological knowledge extraction, feature label construction, and AI model development.

[0066] Optionally, in some embodiments, before obtaining Martian weathered rock samples, the method further includes: generating multiple rock feature labels for multiple Earth rock samples based on multiple Earth rock samples; wherein the rock feature labels include mineral composition labels, weathering degree labels, and microstructure feature labels; and constructing a preset CGAN model based on a conditional generative adversarial network and the rock feature labels.

[0067] Understandably, based on multi-scale weathering studies, weathering significantly affects the mineral composition, distribution, and microporosity characteristics of rocks. Extracting rock characteristics includes mineral composition, i.e., extracting the main mineral components from rock samples, combining this with nanoindentation testing to obtain mineral crystal mechanical parameters, and classifying minerals as weather-resistant or easily weathered. Mineral composition labels use techniques such as X-ray diffraction and scanning electron microscopy to quantitatively analyze the mineral composition of rock samples (including weather-resistant minerals, easily weathered minerals, and weathering products), obtaining their volume and mass fractions. Extracting rock characteristics includes weathering products... The study focuses on identifying weathering products (such as kaolinite and hematite) formed during the weathering process and analyzing their formation and distribution patterns at different stages of weathering. Weathering degree labels are based on the weathering grading standards of the International Society for Rock Mechanics, marking the weathering degree of each rock sample to ensure a correspondence between the selected weathering characteristics and the physical and mechanical properties of the rock. Rock features extracted include microstructure, recording the evolution of microstructures such as mineral distribution, mineral morphology, and micropores during weathering. Microstructure feature labels are based on CT scanning technology to extract microstructural features such as porosity and fracture density. Furthermore, the orientation distribution of mineral crystals is analyzed using microscopic techniques such as electron backscatter diffraction, serving as "orientation" labels to describe the directional characteristics of mineral arrangement within the rock. Meanwhile, a structural similarity index is introduced to measure the similarity of local details of rock and mineral crystals; based on the extraction of weathering geological knowledge, rock feature labels with real physical properties are constructed; and based on these rock feature labels, network parameters are initialized, real samples and labels are input, fake samples G(z|y) are generated, discriminator parameters are updated, generator parameters are updated, and a pre-set CGAN model is constructed based on a conditional generative adversarial network.

[0068] Optionally, in some embodiments, the preset CGAN model includes a CGAN model generator and a CGAN model discriminator. The CGAN model generator is obtained based on random vectors and weathering degree labels, and is used to generate a rock and mineral crystal model based on random vectors, rock feature labels, and a preset loss function. The preset CGAN model discriminator is obtained based on the rock and mineral crystal model and weathering degree labels, and is used to discriminate the image generated by the rock and mineral crystal model.

[0069] Optionally, in some embodiments, the preset loss function is:

[0070] L gen =L adv +λ1L mineral-comp +λ2L mineral-shape +λ3L weathering-prod +λ4L porosity +λ5L SSIM ;

[0071] Among them, L adv Let λ1 be the adversarial loss of CGAN, and L be the weighted hyperparameter of the mineral composition loss. mineral-comp λ² represents the mineral composition loss, λ² represents the weighted hyperparameter of the mineral morphology loss, and L represents the mineral composition loss. mineral-shap λ3 represents the loss in mineral morphology, and L represents the weighted hyperparameter for the loss in weathering degree. weathering-prod For the loss due to weathering degree, λ4 is the weighted hyperparameter for the loss of porosity percentage, and L porosity λ5 is the porosity percentage loss, and L is the weighted hyperparameter of the structural similarity loss. SSIM This is the structural similarity loss.

[0072] It is understandable that, such as Figure 3 As shown, Figure 3 This diagram illustrates the CGAN model construction framework for a Martian weathered rock image generation method according to one embodiment of this application. This embodiment employs a Conditional Generative Adversarial Network (CGAN) structure. The generator G is input to random noise and weathering degree labels. Conditional information is introduced into the generator network through a concatenation vector injection mechanism. A preset loss function is used to generate mineral crystal images corresponding to different weathering stages. The generator input includes weathering degree labels (wj): x'i = G(z, wj). Here, G is the generator, and x'i is the generated weathered rock mineral crystal model. The generator generates rock mineral crystal models with physical properties and structures consistent with the input feature labels based on these conditional labels and random noise z. The discriminator D receives real samples or generated images and their corresponding labels. It uses the PatchGAN architecture to discriminate local regions of the image, improving detail fidelity. Mineral composition labels and microstructure feature labels participate in the loss function calculation. The loss function specifically includes adversarial loss, mineral composition loss, orientation loss, weathering product and porosity ratio loss, and structural similarity index (SSIM) loss, to generate a mineral crystal model with consistent physical properties. The discriminator input is the generated rock sample x'. i and the real sample x i Composed of, and supplemented with conditional tags w j , can be represented as: D(x) i ,w j ).

[0073] Furthermore, L adv The adversarial loss of CGAN is calculated using the following formula:

[0074]

[0075] Among them, L adv As for the losses inflicted by CGAN, To determine the true data distribution p data The expected value, logD(x|y), is the predicted log probability of the discriminator D classifying the input sample x as a true sample given condition y. To determine the noise distribution p z The expected value, D(G(z|y)), is the probability output of the discriminator D under given condition y, which classifies the sample G(z|y) generated by the generator G as a real sample.

[0076] L mineral-comp To account for the loss of mineral composition and ensure that the mineral composition of the formed rock conforms to the laws of weathering, the calculation formula is as follows:

[0077]

[0078] in, To generate the predicted volume fraction of the i-th type of mineral in the image, M i denoted as the volume fraction of the i-th type of mineral in a real geological sample.

[0079] L mineral-shap To account for mineral morphology loss, angular variance and cosine similarity are used to measure the difference in orientation between the generated mineral crystals and the real samples. The calculation formula is as follows:

[0080]

[0081] in, To generate the orientation distribution of mineral crystals, O i For the orientation distribution of real mineral crystals, λ θ and λ cos V is a hyperparameter. θ To control for the variance of orientation, S cos To control for similarity in orientation, weights are adjusted.

[0082] Specifically, to ensure that the overall distribution and local details of the rock mineral structure conform to reality during the generation of the weathered rock AI model, an Lo loss function is introduced. SSIM It is used to control the local details (such as mineral crystal boundaries) of the generated rock and mineral crystal patterns; the preset loss function of the generator and discriminator consists of two parts: adversarial loss and physical property loss.

[0083] In step S102, a Martian rock dataset is generated based on Martian weathered rock samples, and the knowledge of the preset CGAN model is transferred to the Martian rock dataset to obtain a transferred and optimized Martian rock dataset.

[0084] Specifically, based on Martian weathered rock samples and data standardization processing, the small data constraints for Martian meteorites include the introduction of microscopic and macroscopic data, including mineral composition, microstructure, mineral crystals, and the mechanical properties of intergranular phases. For example... Figure 4 As shown, Figure 4 This diagram illustrates the knowledge transfer strategy of a Martian weathered rock image generation method according to an embodiment of this application. Through transfer learning, weathering knowledge learned by the source domain Generative Adversarial Network (GAN) is applied to the target domain (Martian weathered rock samples). This allows for the generation of Martian weathered rock images with realistic physical properties even with very limited Martian rock image data. Based on a knowledge transfer mechanism, weathering features from the source domain (Earth rock samples) are effectively transferred to the target domain (Martian weathered rock samples) through parameter transfer in the conditional batch normalization (BN) layer, addressing the problem of scarce Martian data. Based on a knowledge sharing strategy, statistical features are shared among small sample categories of Martian rocks, improving the model's generalization ability with limited data. Regularization constraints for similarity and residuals are introduced into the GAN training objective to prevent overfitting and ensure the stability and diversity of the generated results.

[0085] In step S103, a microscopic image of Martian weathered rocks is generated based on the migration-optimized Martian rock dataset, and the microscopic image of Martian weathered rocks is compared with a sample of Martian weathered rocks. The generation quality of the microscopic image of Martian weathered rocks is evaluated based on the comparison results.

[0086] Specifically, the CGAN model optimized by knowledge transfer was used to generate microscopic images of Martian weathered rocks, and these images were compared with actual Martian meteorite samples to verify the physical validity and detail fidelity of the generated images.

[0087] Optionally, in some embodiments, the comparison results include structural similarity, perceptual loss, peak signal-to-noise ratio (PSNR), and Frechet Inception Distance. Evaluating the generation quality of the Martian weathered rock microscopic image based on the comparison results includes: comparing structural similarity, perceptual loss, PSNR, and Frechet Inception Distance with corresponding thresholds using a preset quantitative evaluation method to obtain multiple comparison results; verifying whether the Martian weathered rock microscopic image meets preset validity conditions based on the multiple comparison results, and evaluating the generation quality of the Martian weathered rock microscopic image based on the verification results.

[0088] Understandably, the following quantitative metrics are used to evaluate model performance when comparing generated and real images: A mean Structural Similarity Index Measure (SSIM) of 0.87 indicates that the image structure is consistent; a mean Learned Perceptual Image Patch Similarity (LPIPS) value below 0.12 indicates that the model has good detail reproduction capabilities; a mean Peak Signal-to-Noise Ratio (PSNR) value above 30dB indicates that the generated image has high visual quality compared to the original image, with less noise and better image clarity; and a mean Frechet Inception Distance (FID) value below 25 indicates that the distribution difference between the generated and real images is small, and the generated image is visually close to the real image.

[0089] Therefore, addressing the challenges of extremely scarce Martian rock samples, limitations of traditional experimental methods, and insufficient generalization ability of existing AI models, this application utilizes a multi-scale database of weathered rocks on Earth and constructs a generative adversarial neural network model through transfer learning. This model effectively learns the microstructural characteristics of Martian rocks from small sample data of Martian meteorites, thereby generating microscopic images of Martian weathered rocks with realistic physical meaning. This application can be used for numerical simulation of Martian weathered rocks, prediction of their physical and mechanical properties, and data support for deep space exploration missions, providing technical support for Mars probe landing site selection, drilling strategy formulation, and in-situ resource utilization.

[0090] To facilitate a better understanding of the Martian weathered rock image generation method of this application embodiments by those skilled in the art, the following is combined with... Figure 5 The embodiments shown will be described in detail.

[0091] Specifically, such as Figure 5 As shown, Figure 5This flowchart illustrates a method for generating Martian weathered rock images according to one embodiment of this application. Addressing the challenges of extremely scarce Martian samples and the difficulty in generalizing traditional experimental methods and AI models, this embodiment utilizes a conditional generative adversarial network (CGAN) constructed from an Earth weathered rock database to extract weathering labels such as mineral composition, weathering products, and microstructure. Through transfer learning, Earth weathering knowledge is transferred to the Martian rock target domain, achieving physically reliable generation of microscopic images under limited Martian meteorite data conditions. The introduction of multi-label physical constraints and regularization mechanisms significantly improves the model's structural fidelity and consistency of weathering evolution in small sample environments. This embodiment is applicable to scenarios such as Martian weathered rock structure prediction, landing area analysis, and image completion and auxiliary interpretation for deep space exploration missions, possessing the advantage of efficient modeling under data-scarce conditions. The specific steps include:

[0092] S501: Data collection and preprocessing: Collect Earth rock samples for microscopic and macroscopic testing and establish a standardized multiscale database.

[0093] S502: Extracting the mineral composition, weathering products, and microstructure of Earth rock samples to construct physical property labels.

[0094] S503: Construct a CGAN model with feature labels as conditional input, design the generator and discriminator, and define a multi-constraint loss function.

[0095] S504: Based on Martian meteorites NWA 13190 and NWA 12564, this study obtained small data on the minerals, mechanics, and porosity of Martian meteorite 13190 from the meteorite sample using techniques such as TIMA, nanoindentation, and CT scanning.

[0096] S505: The parameters of the Earth rock CGAN model are normalized by conditional batch normalization and feature sharing to introduce regularization to prevent overfitting and to perform knowledge transfer; the weathering knowledge learned by the source domain CGAN model is applied to the target domain (Mars meteorite 13190) through transfer learning.

[0097] S506: Based on the image data of Martian meteorite 13190, generate images of Martian weathered meteorite 13190 with real physical properties, and quantitatively verify the physical validity and detail fidelity through indicators such as SSIM and PSNR.

[0098] Therefore, the embodiments of this application overcome data bottlenecks. Through transfer learning and small data constraints, the GAN model can still generate high-quality microscopic images under limited Martian rock sample conditions. By combining physical constraints and weathering simulation, the rationality of the mineral structure of the generated images is improved, giving them higher physical credibility and enhancing image realism. By combining generative AI algorithms with rock physical modeling, a multi-scale Martian weathered rock database is constructed, providing data support for numerical simulation and exploration missions and supporting multi-scale analysis. The embodiments of this application can be used for Martian landing site selection, drilling analysis, and resource assessment, improving the scientific efficiency and engineering feasibility of exploration missions and enhancing the application of Mars exploration.

[0099] According to the Martian weathered rock image generation method proposed in this application, Martian weathered rock samples are obtained, a Martian rock dataset is generated, knowledge from a pre-defined CGAN model is transferred to the Martian rock dataset to obtain a transfer-optimized Martian rock dataset, microscopic images of Martian weathered rocks are generated, and the microscopic images of Martian weathered rocks are compared with Martian weathered rock samples. The generation quality of the microscopic images of Martian weathered rocks is evaluated based on the comparison results. Therefore, by using transfer learning from an Earth weathered rock database to construct a conditional generative adversarial network and generate microscopic images of Martian weathered rocks, this method solves the problem of significant limitations in characterizing Martian rock properties in related technologies, providing technical support for Mars probe landing site selection, drilling strategy formulation, and in-situ resource utilization.

[0100] Next, referring to the accompanying drawings, a Martian weathered rock image generation apparatus according to an embodiment of this application is described.

[0101] Figure 6 This is a block diagram of a Martian weathered rock image generation device according to an embodiment of this application.

[0102] like Figure 6 As shown, the Martian weathered rock image generation device 10 includes: an acquisition module 100, a migration module 200, and a generation module 300.

[0103] Among them, the acquisition module 100 is used to acquire Martian weathered rock samples;

[0104] The transfer module 200 is used to generate a Martian rock dataset based on Martian weathered rock samples, and to transfer the knowledge of the preset CGAN model to the Martian rock dataset to obtain a transfer-optimized Martian rock dataset.

[0105] The generation module 300 is used to generate microscopic images of Martian weathered rocks based on the migration-optimized Martian rock dataset, compare the microscopic images of Martian weathered rocks with Martian weathered rock samples, and evaluate the generation quality of the microscopic images of Martian weathered rocks based on the comparison results.

[0106] Optionally, in some embodiments, before acquiring Martian weathered rock samples, the acquisition module 100 is further configured to: acquire multiple Earth rock samples; perform microscale and macroscale tests on the multiple Earth rock samples to acquire the microscale and macroscale parameters of the multiple Earth rock samples respectively, and standardize the microscale and macroscale parameters to obtain the processed microscale and macroscale parameters; and establish a multiscale database of Earth weathered rocks based on the processed microscale and macroscale physical and mechanical parameters.

[0107] Optionally, in some embodiments, before acquiring Martian weathered rock samples, the acquisition module 100 is further configured to: generate multiple rock feature labels for multiple Earth rock samples based on multiple Earth rock samples; wherein, the rock feature labels include mineral composition labels, weathering degree labels and microstructure feature labels; and construct a preset CGAN model based on a conditional generative adversarial network and rock feature labels.

[0108] Optionally, in some embodiments, the preset CGAN model includes a CGAN model generator and a CGAN model discriminator. The CGAN model generator is obtained based on random vectors and weathering degree labels, and is used to generate a rock and mineral crystal model based on random vectors, rock feature labels, and a preset loss function. The preset CGAN model discriminator is obtained based on the rock and mineral crystal model and weathering degree labels, and is used to discriminate the image generated by the rock and mineral crystal model.

[0109] Optionally, in some embodiments, the preset loss function is:

[0110] L gen =L adv +λ1L mineral-comp +λ2L mineral-shape +λ3L weathering-prod +λ4L porosity +λ5L SSIM ;

[0111] Among them, L adv Let λ1 be the adversarial loss of CGAN, and L be the weighted hyperparameter of the mineral composition loss. mineral-comp λ² represents the mineral composition loss, λ² represents the weighted hyperparameter of the mineral morphology loss, and L represents the mineral composition loss. mineral-shap λ3 represents the loss in mineral morphology, and L represents the weighted hyperparameter for the loss in weathering degree. weathering-prod For the loss due to weathering degree, λ4 is the weighted hyperparameter for the loss of porosity percentage, and L porosity λ5 is the porosity percentage loss, and L is the weighted hyperparameter of the structural similarity loss. SSIM This is the structural similarity loss.

[0112] Optionally, in some embodiments, the comparison results include structural similarity, perceptual loss, peak signal-to-noise ratio, and Frechet Inception Distance. The generation module 300 is specifically used to: compare structural similarity, perceptual loss, peak signal-to-noise ratio, and Frechet Inception Distance with corresponding thresholds based on a preset quantitative evaluation method to obtain multiple comparison results; verify whether the microscopic image of Martian weathered rocks meets preset validity conditions based on the multiple comparison results, and evaluate the generation quality of the microscopic image of Martian weathered rocks based on the verification results.

[0113] It should be noted that the foregoing explanation of the embodiment of the Martian weathered rock image generation method also applies to the Martian weathered rock image generation apparatus of this embodiment, and will not be repeated here.

[0114] According to the Martian weathered rock image generation device proposed in this application, this application acquires Martian weathered rock samples, generates a Martian rock dataset, transfers knowledge from a preset CGAN model to the Martian rock dataset to obtain a transfer-optimized Martian rock dataset, generates microscopic images of Martian weathered rocks, compares the microscopic images of Martian weathered rocks with Martian weathered rock samples, and evaluates the generation quality of the microscopic images of Martian weathered rocks based on the comparison results. Thus, by using transfer learning from an Earth weathered rock database to construct a conditional generative adversarial network and generate microscopic images of Martian weathered rocks, this solves the problem of significant limitations in characterizing Martian rock properties in related technologies, providing technical support for Mars probe landing site selection, drilling strategy formulation, and in-situ resource utilization.

[0115] Figure 7 A schematic diagram of the structure of an electronic device provided in an embodiment of this application. The electronic device may include:

[0116] The memory 701, the processor 702, and the computer program stored on the memory 701 and executable on the processor 702.

[0117] When the processor 702 executes the program, it implements the Martian weathered rock image generation method provided in the above embodiments.

[0118] Furthermore, electronic devices also include:

[0119] Communication interface 703 is used for communication between memory 701 and processor 702.

[0120] The memory 701 is used to store computer programs that can run on the processor 702.

[0121] The memory 701 may include high-speed RAM memory, and may also include non-volatile memory, such as at least one disk storage device.

[0122] If the memory 701, processor 702, and communication interface 703 are implemented independently, then the communication interface 703, memory 701, and processor 702 can be interconnected via a bus to complete communication between them. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of representation, Figure 7 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0123] Optionally, in a specific implementation, if the memory 701, processor 702, and communication interface 703 are integrated on a single chip, then the memory 701, processor 702, and communication interface 703 can communicate with each other through an internal interface.

[0124] The processor 702 may be a central processing unit (CPU), an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of this application.

[0125] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method for generating images of weathered Martian rocks.

[0126] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0127] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "N" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0128] Any process or method described in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more N executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0129] It should be understood that the various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, the N steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0130] Those skilled in the art will understand that all or part of the steps of the methods described in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it includes one or a combination of the steps of the method embodiments.

Claims

1. A method for generating an image of a weathered rock on Mars, characterized by, The method comprises the following steps: acquiring a Mars weathered rock sample; generating a Mars rock dataset based on the Mars weathered rock sample, migrating the knowledge of a preset CGAN model to the Mars rock dataset to obtain a Mars rock dataset after migration optimization; generating a Mars weathered rock micrograph based on the Mars rock dataset after migration optimization, comparing the Mars weathered rock micrograph with the Mars weathered rock sample, and evaluating the generation quality of the Mars weathered rock micrograph according to the comparison result; Before the step of acquiring the Mars weathered rock sample, the method further comprises the following steps: acquiring a plurality of Earth rock samples; performing microscale and macroscale tests on the plurality of Earth rock samples to obtain microscale parameters and macroscale parameters of the plurality of Earth rock samples, respectively, and performing standardization processing on the microscale parameters and the macroscale parameters to obtain processed microscale parameters and macroscale parameters; and establishing an Earth weathered rock multiscale database based on the processed microscale parameters and the macroscale parameters; Before the step of acquiring the Mars weathered rock sample, the method further comprises the following steps: generating a plurality of rock feature labels of the plurality of Earth rock samples based on the Earth weathered rock multiscale database; wherein the rock feature labels comprise mineral composition labels, weathering degree labels, and microstructure feature labels; and constructing a preset CGAN model based on a conditional generative adversarial network and the rock feature labels; The preset CGAN model comprises a CGAN model generator and a CGAN model discriminator, wherein the CGAN model generator is obtained based on a random vector and the weathering degree labels, the CGAN model generator is configured to generate a rock mineral crystal model based on the random vector, the rock feature labels, and a preset loss function; and the discriminator of the preset CGAN model is obtained based on the rock mineral crystal model and the weathering degree labels, and the discriminator of the preset CGAN model is configured to discriminate an image generated by the rock mineral crystal model; The preset loss function is: ； wherein, L adv is the adversarial loss of CGAN, 1 is a weight hyper-parameter of the mineral composition loss, L mineral-comp is the mineral composition loss, 2 is a weight hyper-parameter of the mineral morphology loss, L mineral-shap is the mineral morphology loss, 3 is a weight hyper-parameter of the weathering degree loss, L weathering-prod is the weathering degree loss, 4 is a weight hyper-parameter of the porosity proportion loss, L porosity is the porosity proportion loss, 5 is a weight hyper-parameter of the structural similarity loss, L SSIM is the structural similarity loss.

2. The method of claim 1, wherein, The comparison result comprises structural similarity, perceptual loss, peak signal-to-noise ratio, and Frechet Inception Distance, and the evaluation of the generation quality of the Mars weathered rock micrograph based on the comparison result comprises: Based on a preset quantitative evaluation method, the structural similarity, the perceptual loss, the peak signal-to-noise ratio, and the Frechet Inception Distance are compared with corresponding threshold values, respectively, to obtain a plurality of comparison results; Based on the plurality of comparison results, it is verified whether the Mars weathered rock micrograph meets a preset validity condition, and the generation quality of the Mars weathered rock micrograph is evaluated according to the verification result.

3. A Mars weathered rock image generation device characterized by comprising: The Mars weathered rock image generation method comprises the following steps: an acquisition module configured to acquire a Mars weathered rock sample; The migration module is configured to generate a Mars rock data set based on the Mars weathered rock sample, and to migrate knowledge of a preset CGAN model to the Mars rock data set to obtain a Mars rock data set after migration optimization. The generation module is configured to generate a Mars weathered rock micrograph based on the Mars rock data set after migration optimization, to compare the Mars weathered rock micrograph with the Mars weathered rock sample, and to evaluate the generation quality of the Mars weathered rock micrograph according to a comparison result.

4. An electronic device, comprising: The method comprises the following steps: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the Mars weathered rock image generation method according to any one of claims 1-2.

5. A computer-readable storage medium having stored thereon a computer program, characterized in that, The program is executed by the processor to implement the Mars weathered rock image generation method according to any one of claims 1-2.

Images