Construction method and application of Mars geology large model based on multi-source data fusion

Abstract

The invention belongs to the related technical field of large models, and discloses a construction method and application of a Mars geology large model based on multi-source data fusion, and the method comprises the steps: (1) carrying out the time alignment of various types of Mars geology heterogeneous data through employing a sliding window and an interpolation method, meanwhile, performing spatial alignment on various types of Mars geological heterogeneous data by adopting GIS spatial registration and feature point matching; performing vectorization conversion on the aligned data in time and space; (2) identifying the data entities subjected to quantitative conversion, extracting a relationship between the data entities to generate a triple, storing the triple in a database, and then constructing a Mars geological knowledge map based on the database; and (3) taking the Mars geological knowledge map as structured background information, and constructing a large-scale machine learning model based on vectorized conversion data. According to the invention, the alignment accuracy and efficiency are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field related to large models, and more specifically, relates to a method for constructing a large Mars geology model based on multi-source data fusion and its application. Background Art

[0002] The study of Mars geology is an important part of planetary science, aiming to reveal the formation history, geological structure and its evolution process of Mars. With the progress of space technology and exploration equipment, humans have accumulated a large amount of data on Mars, including high-resolution remote sensing images, analysis results of detector samples, and Earth observation simulation experiment data, etc. These data cover various aspects such as the surface morphology, mineral composition, and internal structure of Mars, constituting a huge information library. In recent years, the development of artificial intelligence and big data processing technologies has provided new opportunities for analyzing these complex data sets, but existing research mainly focuses on single-type data and lacks the ability to comprehensively analyze multi-source heterogeneous data.

[0003] The main challenge in current Mars geology research lies in how to efficiently integrate massive data from different sources with various formats and extract valuable information from it. Specifically, the multi-source heterogeneity of Mars geology data brings the following technical problems:

[0004] Time alignment problem: The data acquisition frequency of Mars exploration missions is low, the time span is large (possibly up to several decades), and the timestamps of data from different sources are not completely consistent. Existing technologies mainly use fixed time windows for alignment, which is difficult to adapt to the particularity of the large time span and low acquisition frequency of Mars data, resulting in ineffective alignment of multi-source data in the time dimension.

[0005] Space alignment problem: The spatial resolution differences of Mars geology data are huge. For example, the registration difficulty between high-resolution HiRISE images and low-resolution THEMIS thermal infrared data is relatively high. Existing technologies mainly target Earth data and lack optimization methods for the spatial resolution differences of Mars data, making it difficult to achieve precise alignment in the spatial dimension. Summary of the Invention

[0006] In view of the above defects or improvement requirements of the existing technology, the present invention provides a method for constructing a large Mars geology model based on multi-source data fusion and its application, aiming to solve the problem that it is difficult to align existing Mars heterogeneous data both in time and space.

[0007] To achieve the above object, according to one aspect of the present invention, there is provided a method for constructing a large Mars geology model based on multi-source data fusion, the method comprising the following steps:

[0008] (1) Align various types of Martian geological heterogeneous data in time using a sliding window and interpolation method, and align various types of Martian geological heterogeneous data in space using GIS spatial registration and feature point matching; then perform vectorization conversion on the data that is aligned both in time and space;

[0009] (2) Identify the data entities after quantization conversion and extract the relationships between the data entities to generate triples, and store the triples in a database, and then construct a Martian geological knowledge graph based on the database;

[0010] (3) Use the Martian geological knowledge graph as structured background information and construct a large-scale machine learning model based on the vectorized converted data.

[0011] Furthermore, extract the timestamp information of various types of Martian geological heterogeneous data, and perform standardization processing to ensure consistent timestamp formats, arrange the time series data in chronological order, and generate a time series data set.

[0012] Furthermore, a time series matching method based on a sliding window is used to process data with inconsistent timestamps, and dynamically adjust the size and step length of the sliding window.

[0013] Furthermore, a time alignment method based on interpolation is used to fill in the missing data in the time series.

[0014] Furthermore, extract the spatial coordinate information of various types of Martian geological heterogeneous data, and perform standardization processing, arrange the spatial data according to the coordinate positions, and generate a spatial data set.

[0015] Furthermore, a spatial registration method based on GIS is used to align remote sensing images and terrain data with different resolutions; a spatial matching method based on feature points is used to perform spatial alignment between image data and numerical data.

[0016] Furthermore, the large-scale machine learning model adopts a deep neural network architecture, which can process multiple modal data; during training, the LoRA fine-tuning technology is used to adjust the parameters of the predetermined layers of the large-scale machine learning model.

[0017] The present invention also provides an application of a large-scale machine learning model in obtaining Martian geological information, and the large-scale machine learning model is constructed by using the construction method of the Martian geological large model based on multi-source data fusion as described above.

[0018] The present invention also provides a construction system of a Martian geological large model based on multi-source data fusion, and the system includes a memory and a processor, the memory stores a computer program, and when the processor executes the computer program, it executes the construction method of the Martian geological large model based on multi-source data fusion as described above.

[0019] The present invention also provides a computer-readable storage medium storing machine-executable instructions, which, when called and executed by a processor, cause the processor to implement the method for constructing a large-scale Mars geology model based on multi-source data fusion as described above.

[0020] Generally speaking, compared with the prior art through the above technical solutions conceived by the present invention, the method for constructing a large-scale Mars geology model based on multi-source data fusion provided by the present invention and its applications mainly have the following beneficial effects:

[0021] 1. The present invention realizes time alignment through a dynamic sliding window and interpolation method to solve the problems of Mars data sparsity and noise; realizes spatial alignment through GIS spatial registration and feature point matching to solve the problem of significant differences in the spatial resolution of Mars data. Compared with the prior art, the present invention significantly improves the accuracy and efficiency of data alignment, providing a high-quality data basis for subsequent analysis.

[0022] 2. The present invention solves the problem of extensive, diverse and large-scale Mars geology data sources lacking effective integration by fusing a knowledge graph and a large model; constructs a detailed Mars geology knowledge graph using the high-quality data after multi-source data alignment processing, providing rich background information for the large model; enhances the reasoning ability of the large model through the knowledge graph, significantly improving the analysis accuracy and precision of the model for complex Mars geology phenomena, and providing intelligent and professional technical support for Mars geology research. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 is a flowchart of a method for constructing a large-scale Mars geology model based on multi-source data fusion provided by the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0024] In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0025] The present invention provides a method for constructing a large-scale Mars geology model based on multi-source data fusion. The construction method has excellent multi-source data processing capabilities, especially for the massive and complex Mars geology data. It not only effectively collects scattered data resources, significantly improving the accuracy of interpreting Mars geology problems, but also, with the assistance of professional knowledge graphs and large models, achieves more accurate acquisition of Mars information, providing strong support for related scientific research, education promotion, and space exploration of Mars geology.

[0026] Please refer to Figure 1 , the construction method mainly includes the following steps:

[0027] Step 1, use a sliding window and interpolation method to align various types of Mars geology heterogeneous data in time, and at the same time use GIS spatial registration and feature point matching to align various types of Mars geology heterogeneous data in space; then perform vectorization conversion on the data that is aligned both in time and space.

[0028] Among them, the Mars geology heterogeneous data includes but is not limited to remote sensing data, image datasets, historical literature materials, video files, etc. The remote sensing data used in this embodiment mainly comes from advanced orbital spacecraft and other space observation equipment to provide a series of detailed information about the Mars surface and its environment. For example, the Compact Reconnaissance Imaging Spectrometer (CRISM) carried by NASA's Mars Reconnaissance Orbiter (MRO) can provide detailed mineral composition information; the Thermal Emission Imaging System (THEMIS) reveals the physical properties of surface materials through the thermal infrared band; the Mars Orbiter Laser Altimeter (MOLA) generates a high-precision terrain model; in addition, there are also valuable data provided by the Mars Mineral Spectral Analyzer (MoRIC) carried by Tianwen-1, China's first Mars exploration mission, and the High Resolution Stereo Camera (HRSC) on the European Space Agency's Mars Express.

[0029] The image dataset not only covers the high-definition image annotation dataset taken by the HiRISE camera, but also includes a diverse dataset designed specifically for weakly supervised landform analysis on Mars. In addition, an ML-ready dataset is specially prepared for training and validating machine learning models that can accurately identify frost features in HiRISE images.

[0030] By collecting published academic research papers, reports, and technical documents, a historical literature library containing a large amount of Mars geology background information is established. To further enrich the knowledge system of the large-scale Mars geology model, the present invention also incorporates a variety of video resources, including but not limited to conference videos, expert lectures, educational courses in the field of planetary science, etc.

[0031] Before alignment, it also includes operations such as cleaning, annotating, standardizing, and quality assessment on the original Martian geological data to ensure that all data has high quality and consistency for subsequent processing by machine learning algorithms, thereby improving the reliability and efficiency of the model.

[0032] Data processing aims to convert unstructured and semi-structured multi-source data into a unified digital representation for subsequent processing by machine learning algorithms. Specifically, data processing is divided into three main parts: data preprocessing, multi-source data alignment, vectorization transformation, and quality assessment. Among them:

[0033] Data preprocessing includes using big data processing technologies such as ETL (Extract, Transform, Load) tools, content recognition technologies, and OCR (Optical Character Recognition) to clean, verify, and classify the collected data, and eliminate invalid data, abnormal data, and untrue data to ensure that all data input into the model has high quality.

[0034] Multi-source data alignment includes precisely aligning and fusing Martian geological data in the time and space dimensions to solve the problems of inconsistent acquisition time and space range of Martian data and the difficulty of achieving efficient data alignment and fusion in existing technologies. Extract the timestamp information of multi-source data and perform standardization processing to ensure consistent timestamp formats, arrange the time series data in chronological order, and generate a time series dataset.

[0035] Efficiently align and fuse Martian geological data in the time and space dimensions. Through the timestamp matching algorithm, align data from different sources on the time axis to ensure the time consistency of the data. Through the spatial coordinate mapping algorithm, register data with different resolutions in space to ensure the spatial consistency of the data.

[0036] The timestamp matching algorithm includes:

[0037] A time series matching method based on a sliding window, used to process data with inconsistent timestamps, dynamically adjust the size and step of the sliding window to adapt to the particularity of the large time span and low acquisition frequency of Martian data.

[0038] Dynamically adjust the size and step of the sliding window according to the time span and acquisition frequency of the data. For example: for data with a large time span (such as several months or years), use a larger sliding window (such as 3 months or 1 year). For data with a small time span (such as several days or weeks), use a smaller sliding window (such as 1 day or 1 week).

[0039] Within the sliding window, perform timestamp matching on data from different sources. For example, match the timestamp of remote sensing images with the timestamp of meteorological data. For data with inconsistent timestamps, use the nearest neighbor matching method to select the data point with the closest time for matching to ensure the accuracy of the matching results.

[0040] A time alignment method based on interpolation is used to fill in the missing data in the time series. Through interpolation calculation and smoothing processing, ensure the accuracy of time alignment.

[0041] For the missing data points in the time series, use the interpolation method to fill them. Specifically, it can be divided into: Linear interpolation: For the missing values in the continuous time series, calculate and fill them through linear interpolation. Periodic interpolation: For data with periodic characteristics (such as meteorological data), use the periodic interpolation method to fill in the missing values.

[0042] Finally, conduct a quality assessment on the matching results, eliminate the data points with large matching errors, and improve the accuracy of time series matching by iteratively optimizing the size and step length of the sliding window.

[0043] Extract the spatial coordinate information of multi-source data and perform standardization processing. Arrange the spatial data according to the coordinate positions to generate a spatial dataset. Convert the data in different coordinate systems into a unified coordinate system. For example, perform a projection transformation on the local coordinate system of the HiRISE image and the global coordinate system of the THEMIS thermal infrared data to ensure spatial consistency.

[0044] Perform resolution matching on high-resolution data (such as HiRISE images) and low-resolution data (such as THEMIS thermal infrared data). Specifically, it includes: Downsampling: Downsample the high-resolution data to the same resolution as the low-resolution data. Interpolation: Perform interpolation processing on the low-resolution data to align it spatially with the high-resolution data. Then, perform error correction on the registration results to reduce the registration error and ensure the accuracy of spatial alignment.

[0045] The spatial coordinate mapping algorithm includes:

[0046] A spatial registration method based on the Geographic Information System (GIS) is used to align remote sensing images and terrain data with different resolutions.

[0047] A spatial matching method based on feature points is used to spatially align image data with numerical data. By extracting and matching feature points, improve the accuracy of spatial alignment.

[0048] Use feature extraction algorithms (such as SIFT, SURF) to extract feature points from image data. For example, extract significant terrain feature points (such as craters, mountain edges, etc.) from HiRISE images and THEMIS thermal infrared data.

[0049] Describe and encode the extracted feature points to generate a feature point data set. Use the nearest neighbor matching algorithm or the RANSAC (Random Sample Consensus) algorithm to match the feature points of data from different sources.

[0050] According to the feature point matching results, calculate the spatial alignment parameters (such as rotation matrix, translation vector), and perform alignment calculations on the spatial data to ensure the spatial consistency of data from different sources.

[0051] Finally, conduct a quality assessment on the alignment results to verify the accuracy and consistency of the aligned data. By comparing the data distributions before and after alignment, ensure the reliability of the alignment results.

[0052] Dynamic update:

[0053] Adjust the alignment strategy in real time according to the addition of new data to ensure the timeliness and accuracy of data alignment. For example, integrate new data through incremental learning techniques and use automated control methods to ensure the accuracy and consistency of new data.

[0054] The vectorization transformation involves converting the preprocessed and aligned data into a machine-readable vector form. For image data, use a convolutional neural network (CNN) to extract feature vectors; for text data, use word embedding models such as Word2Vec or BERT to generate semantic vectors; for numerical data, normalization or standardization processing may be required.

[0055] Quality assessment is a comprehensive inspection of the transformed data, including aspects such as integrity, accuracy, and consistency, to ensure the consistency and reliability of the data, thereby improving the reliability and efficiency of the entire model.

[0056] This step also includes a labeling step, that is, adding labels to each piece of data. These labels can be used to train supervised machine learning models and for entity association in knowledge graphs, thereby enhancing the logical connections between data and further supporting complex queries and advanced analysis functions.

[0057] In step two, identify the data entities after quantization transformation and extract the relationships between the data entities to generate triples, and store the triples in a database, and then construct a Mars geological knowledge graph based on the database.

[0058] The triples include, but are not limited to, attributes such as the relationship type between entities, timestamp, confidence score, geographical coordinates, etc.

[0059] Identify the obtained data entities, deeply extract the relationships between them, generate knowledge triples, and store them in a high-performance database. Utilize these knowledge triples to construct a detailed Mars geological knowledge graph, providing rich background information for subsequent large model training. Specifically, this step is divided into the following sub-steps:

[0060] Entity recognition: Analyze the text data through natural language processing (NLP) techniques and machine learning algorithms to automatically identify and label entities such as Mars place names, various mineral names, and Mars exploration missions.

[0061] Relationship extraction: Based on rule matching, pattern recognition, or deep learning models, extract the relationships between entities from the text, such as the association between Mars geological structures and specific minerals, or the results discovered by different detectors at specific Mars locations.

[0062] Triple generation: Construct the identified entities and their relationships into knowledge triples. Each triple should contain at least two entities and one relationship between them. To enrich the description, the knowledge triples can also include but are not limited to the following attributes: the type of relationship between entities, timestamp (indicating the time of data collection or event occurrence), confidence score (reflecting the accuracy of relationship extraction), geographical coordinates (used to locate specific positions on Mars), etc.

[0063] Knowledge storage: Use a high-performance graph database (such as Neo4j) to store the generated knowledge triples. Graph databases can effectively manage complex relationship structures, support fast querying and advanced analysis, and are very suitable for constructing a Mars geological knowledge graph.

[0064] Step three, use the Mars geological knowledge graph as structured background information and construct a large-scale machine learning model based on the vectorized transformed data.

[0065] A large-scale machine learning model, named Mars GPT, adopts a deep neural network architecture and combines advanced technologies such as reinforcement learning. It is optimized specifically for massive multi-source data, can maintain high performance when processing a large amount of complex data, and at the same time ensures the generalization ability and prediction accuracy of the model.

[0066] In this embodiment, a large-scale machine learning model (abbreviated as Mars GPT) that can process various modal data such as text, images, and numerical values is designed and trained. This model adopts a deep neural network architecture and combines advanced technologies such as reinforcement learning. Through optimization algorithms and parameter fine-tuning, the accuracy of predicting and analyzing Mars geological characteristics is improved. Specifically, this step is divided into the following sub-steps:

[0067] Data Preparation and Model Selection: Preprocess the generated knowledge triples and other multi-source data. Text data needs to be converted into word vectors or sentence embeddings, image data needs to be resized and normalized, and numerical data may need to be standardized. At the same time, select the latest open-source pre-trained model as the base model (for text data, BERT or its variants can be used; for image data, EfficientNet or ResNet can be used).

[0068] Model Fine-tuning and Reinforcement Learning: According to the characteristics of the Mars geological feature analysis task, use the LoRA (Low-Rank Adaptation) fine-tuning technique to adjust the parameters of specific layers of the model. To improve the model's decision-making ability in complex environments, introduce a reinforcement learning mechanism. By simulating Mars exploration scenarios and setting a reward function, let the model learn how to make optimal predictions with limited information.

[0069] Model Performance Optimization: Adopt a distributed training framework (such as Horovod or TensorFlow Distributed) to accelerate the training process and ensure that the model can maintain high performance when processing massive multi-source data. In addition, further optimize the model parameters through hyperparameter search (such as Bayesian optimization or random search) to ensure the generalization ability and prediction accuracy of the model.

[0070] Large-scale machine learning models are specifically optimized for massive multi-source data. The models can not only maintain high performance when processing a large amount of complex data, but also provide accurate and reliable prediction results when facing new Mars geological data.

[0071] The present invention also provides an application of a large-scale machine learning model in obtaining Mars geological information. The large-scale machine learning model is constructed by using the construction method of the Mars geological large model based on multi-source data fusion as described above.

[0072] When the large-scale machine learning model is used, after receiving an input (a question related to Mars geology), it will first try to retrieve an answer from the pre-constructed knowledge graph. If there is a ready answer in the knowledge graph, it will be directly returned to the user; if no appropriate answer is found, the large-scale machine learning model will then perform reasoning to generate the most likely answer, that is, the corresponding Mars geological information for the input.

[0073] The present invention also provides a construction system of a Mars geological large model based on multi-source data fusion. The system includes a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it executes the construction method of the Mars geological large model based on multi-source data fusion as described above.

[0074] The present invention also provides a computer-readable storage medium storing machine-executable instructions, which, when called and executed by a processor, cause the processor to implement the method for constructing a large-scale Mars geological model based on multi-source data fusion as described above.

[0075] Those skilled in the art can easily understand that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent replacements, and improvements made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for constructing a large-scale Mars geological model based on multi-source data fusion, characterized in that, The method includes the following steps: (1) Use a sliding window and interpolation method to align various types of Martian geological heterogeneous data in terms of time, and at the same time use GIS spatial registration and feature point matching to align various types of Martian geological heterogeneous data in terms of space; then perform vectorization conversion on the data that is aligned both in time and space; (2) Identify the data entities after quantization conversion and extract the relationships between the data entities to generate triples, and store the triples in a database, and then construct a Martian geological knowledge graph based on the database; (3) Use the Martian geological knowledge graph as structured background information, and construct a large-scale machine learning model based on the vectorized converted data.

2. The method for constructing a large-scale Martian geology model based on multi-source data fusion according to claim 1, wherein: Extract the timestamp information of various types of Martian geological heterogeneous data, and perform standardization processing to ensure that the timestamp formats are consistent, arrange the time series data in chronological order, and generate a time series data set.

3. The method for constructing a large-scale Mars geological model based on multi-source data fusion according to claim 2, wherein: A time series matching method based on a sliding window is used to process data with inconsistent timestamps, and dynamically adjust the size and step length of the sliding window.

4. The construction method of the Mars geological large model based on multi-source data fusion according to claim 3, characterized in that: A time alignment method based on interpolation is used to fill in the missing data in the time series.

5. The construction method of the Mars geological large model based on multi-source data fusion according to claim 1, characterized in that: Extract the spatial coordinate information of various types of Martian geological heterogeneous data, and perform standardization processing, arrange the spatial data according to the coordinate positions, and generate a spatial data set.

6. The construction method of the Mars geological large model based on multi-source data fusion according to claim 5, characterized in that: A spatial registration method based on GIS is used to align remote sensing images and terrain data with different resolutions; A spatial matching method based on feature points is used to perform spatial alignment between image data and numerical data.

7. The method for constructing a large-scale Mars geological model based on multi-source data fusion according to any one of claims 1-6, characterized in that: The large-scale machine learning model adopts a deep neural network architecture, which can process multiple modalities of data; during training, the LoRA fine-tuning technology is used to adjust the parameters of the predetermined layers of the large-scale machine learning model.

8. Application of a large-scale machine learning model in obtaining Martian geological information, characterized in that: The large-scale machine learning model is constructed by using the construction method of the Martian geological large model based on multi-source data fusion according to any one of claims 1-7.

9. A construction system for a large-scale Mars geological model based on multi-source data fusion, characterized in that: The system includes a memory and a processor, the memory stores a computer program, and when the processor executes the computer program, it executes the construction method of the Martian geological large model based on multi-source data fusion according to any one of claims 1-7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores machine-executable instructions, and when the machine-executable instructions are called and executed by the processor, the machine-executable instructions cause the processor to implement the construction method of the Martian geological large model based on multi-source data fusion according to any one of claims 1-7.

Images