Multi-source data assimilation method based on remote sensing

Abstract

The invention discloses a multi-source data assimilation method based on remote sensing in the technical field of remote sensing data, and the method comprises the following steps: S100: data preprocessing: carrying out the preprocessing of multi-source remote sensing data, including radiation correction, geometric correction and cloud detection, so as to eliminate the noise and error in the data; and S200, feature extraction: extracting useful feature information from the preprocessed data, including spectral features and texture features, so as to facilitate subsequent data analysis and application. And the prediction capability is improved: the model prediction capability can be improved through data assimilation, so that a model result is continuously close to an observation value, and the prediction accuracy is improved. The observation data of different sources and resolutions are fused through data assimilation, information more accurate and richer than single data can be obtained, fuzziness is reduced, and interpretation reliability is improved.

Description

Technical Field

[0001] The invention relates to the technical field of remote sensing data, and in particular to a multi-source data assimilation method based on remote sensing. Background Art

[0002] Remote sensing refers to non-contact, long-distance detection technology. It generally refers to the detection of the radiation and reflection characteristics of electromagnetic waves of objects using sensors / remote sensors. Remote sensing is a science and technology that uses remote sensors, which are sensitive to electromagnetic waves, to detect target objects far away from the target and without contacting the target object. It obtains the information of electromagnetic waves reflected, radiated or scattered (such as electric field, magnetic field, electromagnetic waves, seismic waves, etc.), and extracts, determines, processes, analyzes and applies them.

[0003] Data assimilation refers to a method of integrating new observation data during the dynamic operation of a numerical model, taking into account the temporal and spatial distribution of data and the errors of the observation field and background field. It is a method of automatically adjusting the model trajectory by continuously integrating direct or indirect observation information of different sources and different resolutions of discrete distribution in time and space through a data assimilation algorithm within the dynamic framework of the process model, so as to improve the estimation accuracy of the dynamic model state and enhance the model prediction capability. Data assimilation is a data processing technology that originally originated from numerical weather forecasting and provides the initial field for numerical weather forecasting. It has been widely used in the field of atmosphere and ocean.

[0004] With the continuous development of remote sensing technology, the acquisition and processing of multi-source data has become a hot issue in the field of remote sensing. Traditional data processing methods often cannot effectively integrate data from different sources, resulting in information loss and increased errors. Therefore, how to effectively assimilate multi-source data has become an urgent problem to be solved. Summary of the invention

[0005] The purpose of the present invention is to provide a multi-source data assimilation method based on remote sensing to solve the problems raised in the above background technology.

[0006] To achieve the above object, the present invention provides the following technical solution: a multi-source data assimilation method based on remote sensing, comprising the following steps:

[0007] S100: Data preprocessing: Preprocess multi-source remote sensing data, including radiation correction, geometric correction, and cloud detection, to eliminate noise and errors in the data;

[0008] S200: Feature extraction: extract useful feature information from the preprocessed data, including spectral features and texture features, for subsequent data analysis and application;

[0009] S300: Data fusion: Fuse the extracted feature information, using appropriate fusion algorithms, weighted average, and principal component analysis to integrate the advantages of multi-source data;

[0010] S400: Data verification: multiple verification methods are used to verify the data, and feedback is given for wrong data. If the data verification is correct, the next step is entered;

[0011] S500: Assimilation model construction: Based on the fused data, construct an appropriate assimilation model, including Kalman filtering and variational method, to achieve optimal estimation of multi-source data;

[0012] S600: Result output: Based on the results of the assimilation model, the final remote sensing data products are output, including land cover maps and vegetation index maps.

[0013] As a further solution of the present invention: the data preprocessing in step S100 includes the following sub-steps:

[0014] Data collection: Collect data through various means and determine the data source;

[0015] Preliminary screening: Check the completeness of the data to ensure that the required data is not missing and convert all data into the unified format required in the processing flow;

[0016] Data cleaning: Use filters or smoothing techniques to reduce random noise in the data. For missing data, use interpolation, mean value filling and ignoring methods to identify and process outliers in the data.

[0017] Data correction: Radiation correction: perform radiation calibration on remote sensing data to eliminate sensor errors and atmospheric influences, and perform geographic correction on remote sensing images to ensure that the images correspond to the actual geographic location;

[0018] Data fusion: Data of different resolutions are fused to the same spatial scale through resampling and other methods, time series data are analyzed, time labels are adjusted, and time consistency is ensured;

[0019] Feature extraction: extract reflectance and vegetation index spectral features from remote sensing data, calculate gray-level co-occurrence matrix and local binary pattern texture features, and perform edge detection on objects to extract shape features;

[0020] Data normalization: Standardize the data to a range of 0-1 to reduce the impact of different dimensions, use maximum-minimum normalization or Z-score normalization to make the data comparable;

[0021] Data enhancement: rotation transformation: increase sample diversity by rotating the image, flip transformation: flip the image horizontally or vertically, translation transformation: translate the image along the x or y direction, scaling transformation: change the size of the image;

[0022] Data segmentation: Divide the data set into a training set and a test set, with a common ratio of 70% training set and 30% test set;

[0023] Data encoding: Convert categorical data into a numerical form that the model can understand, converting class labels into integers.

[0024] As a further solution of the present invention: the feature extraction in step S200 also includes spectral feature extraction, texture feature extraction, shape feature extraction, spatial relationship feature extraction, time feature extraction, deep learning feature extraction, feature selection, and feature fusion.

[0025] As a further solution of the present invention: the data fusion in step S300 also includes data processing, feature extraction, data-level fusion, feature-level fusion, model-level fusion, fusion strategy, post-fusion processing, and quality assessment.

[0026] As a further solution of the present invention: the data verification in step S400 includes data source check, data format and structure verification, data range and logic check, data accuracy verification, data timeliness verification, data consistency verification, data quality assessment, data clarity and preprocessing, verification result recording, and user feedback.

[0027] As a further solution of the present invention: the step S500 of assimilation model construction includes the following sub-steps:

[0028] Model selection: Select a suitable assimilation algorithm based on data characteristics and research objectives, determine the basic architecture of the model, and select the appropriate number of neural network layers and nodes;

[0029] Model design: Design the input layer of the model according to the fused data features, design the hidden layer of the model, including the number of layers, the number of neurons in each layer, and the activation function; design the output layer of the model according to the prediction target, including the design of continuous output for regression problems and the design of discrete output for classification problems;

[0030] Parameter initialization: Set initial values ​​for the model's weight parameters, use random initialization or pre-trained weights, and set initial values ​​for the model's bias parameters;

[0031] Loss function definition: Choose an appropriate error metric, including mean square error for regression problems and cross entropy for classification problems, and add regularization terms to prevent overfitting;

[0032] Optimization algorithm selection: Select an appropriate optimization algorithm and adjust the hyperparameters of the optimization algorithm, including learning rate, batch size, and number of iterations;

[0033] Training and validation: Prepare training sets using preprocessed and fused data, prepare validation sets to evaluate model performance and make model selection, and apply cross-validation techniques to evaluate the generalization ability of the model;

[0034] Model training: Divide the dataset into small batches for training to fit within memory constraints, and use an early stopping strategy to avoid overfitting, i.e. stop training when the validation error no longer decreases;

[0035] Model testing: Use an independent test set to perform a final evaluation of the model and calculate the model's performance metrics;

[0036] Model tuning: adjust hyperparameters and model structure based on the model’s performance on the validation set;

[0037] Model deployment: Save the trained model for subsequent use and load the model in actual applications for prediction and analysis.

[0038] Compared with the prior art, the present invention has the following beneficial effects: Improving prediction capability: Data assimilation can improve the prediction capability of the model, making the model results continuously approach the observed values, thereby improving the accuracy of the forecast.

[0039] Optimize observation data quality: By integrating observation data from different sources and resolutions through data assimilation, more accurate and richer information can be obtained than with a single data set, reducing ambiguity and improving interpretation reliability.

[0040] Enhanced model simulation accuracy: Data assimilation provides macro-monitoring information and the ability to dynamically reflect the growth and development process of crops by integrating remote sensing information and crop growth models, which is conducive to achieving complementary advantages and enhancing application potential.

[0041] Promoting scientific research: The combination of deep learning and data assimilation provides a new way to solve complex problems in the Earth system and promotes in-depth scientific research.

[0042] Expanded application areas: Data assimilation is not limited to the atmosphere and ocean fields, but can also be applied to agricultural resource monitoring, crop yield prediction and other fields, and has broad application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 The present invention is a flowchart of a multi-source data assimilation method based on remote sensing. DETAILED DESCRIPTION

[0044] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0045] The terms "first", "second", etc. in the specification and claims of the present application are used to distinguish similar objects, and are not used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable under appropriate circumstances, so that the embodiments of the present application can be implemented in an order other than those illustrated or described here, and the objects distinguished by "first", "second", etc. are generally of one type, and the number of objects is not limited. For example, the first object can be one or more. In addition, "and / or" in the specification and claims represents at least one of the connected objects, and the character " / " generally indicates that the objects associated with each other are in an "or" relationship.

[0046] In the description of the present invention, it should be understood that the terms "center", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", etc., indicating the orientation or position relationship are based on the orientation or position relationship shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention.

[0047] In the description of the present invention, it should be noted that, unless otherwise clearly specified and limited, the terms "installed", "connected", and "set" should be understood in a broad sense, for example, they can be fixedly connected or set, or detachably connected or set, or integrally connected or set. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0048] See also Figure 1 In an embodiment of the present invention, a multi-source data assimilation method based on remote sensing includes the following steps:

[0049] S100: Data preprocessing: Preprocess multi-source remote sensing data, including radiation correction, geometric correction, and cloud detection, to eliminate noise and errors in the data;

[0050] S200: Feature extraction: extract useful feature information from the preprocessed data, including spectral features and texture features, for subsequent data analysis and application;

[0051] S300: Data fusion: Fuse the extracted feature information, using appropriate fusion algorithms, weighted average, and principal component analysis to integrate the advantages of multi-source data;

[0052] S400: Data verification: multiple verification methods are used to verify the data, and feedback is given for wrong data. If the data verification is correct, the next step is entered. If the data verification is qualified, the next step is entered. If the data verification is unqualified, the data is returned to step S100 to pre-process the data again.

[0053] S500: Assimilation model construction: Based on the fused data, construct an appropriate assimilation model, including Kalman filtering and variational method, to achieve optimal estimation of multi-source data;

[0054] S600: Result output: Based on the results of the assimilation model, the final remote sensing data products are output, including land cover maps and vegetation index maps.

[0055] Preferably, in this embodiment, step S100 data preprocessing involves using specific software and hardware support to perform a series of preprocessing operations. The required software and hardware support and data preprocessing methods are described in detail below:

[0056] Software and hardware support

[0057] computer:

[0058] Configuration: high-performance CPU, large-capacity memory, high-end graphics card.

[0059] Operating system: A system compatible with remote sensing data processing software, such as Windows or Linux. Storage: A high-speed SSD or a large-capacity hard disk is used to store large amounts of remote sensing data.

[0060] monitor:

[0061] Type: High resolution to show image details clearly.

[0062] Keyboard and Mouse:

[0063] Requirements: Ergonomic, ensuring operational efficiency and comfort.

[0064] Network connection:

[0065] Speed: High-speed Internet connection to facilitate data downloads and software updates.

[0066] Remote sensing data processing software:

[0067] Examples: ENVI, ERDAS Image or PCI Geomat i ca, etc.

[0068] Storage devices and peripherals:

[0069] Types: External hard drives, cloud storage services, printers, scanners, etc.

[0070] Other auxiliary equipment:

[0071] Includes: GPS locator, topographic map, etc., for obtaining ground control points.

[0072] Data preprocessing methods

[0073] Radiometric correction:

[0074] Purpose: To eliminate the sensor error and the influence of the atmosphere on the image and obtain the surface reflectance. Steps: Convert the original DN value to the radiance value and then perform atmospheric correction.

[0075] Geometric correction:

[0076] Reason: The geometric deformation of the image caused by systematic and non-systematic factors needs to be corrected. Steps: Select ground control points (GCP), establish a correction model, and perform image resampling. Image fusion:

[0077] Purpose: To combine the advantages of multispectral and panchromatic images and improve spatial resolution.

[0078] Method: Use GS fusion method or Gram-Schmidt Pan Sharpening and other techniques. Image mosaicking and cropping:

[0079] Purpose: When the study area exceeds the range of a single image, image stitching is required; remove the area outside the study area.

[0080] Method: Select a reference image for tone matching and crop the image as needed.

[0081] Cloud Detection and Shadow Processing:

[0082] Purpose: Clouds and shadows can obscure surface information and need to be detected and processed.

[0083] Method: Cloud detection algorithm is used to identify clouds, and corresponding algorithms are used to try to restore the areas covered by clouds or eliminate the affected data.

[0084] Atmospheric correction:

[0085] Purpose: To reduce the effects of atmospheric scattering and absorption and obtain the true surface reflectivity.

[0086] Method: Use atmospheric correction models such as FLAASH and combine terrain and meteorological data to optimize the results.

[0087] With the support of the above-mentioned hardware and software and the application of preprocessing methods, remote sensing data can be effectively prepared for further analysis and assimilation processing. This not only improves the quality and availability of the data, but also lays a solid foundation for subsequent steps.

[0088] Preferably, in this embodiment, the feature extraction step S200 is intended to extract useful feature information from the preprocessed data, such as spectral features, texture features, etc., for subsequent data analysis and application. The required software and hardware support and specific feature extraction steps and methods will be described in detail below.

[0089] Software and hardware support

[0090] computer:

[0091] Configuration: high-performance CPU, large-capacity memory, high-end graphics card.

[0092] Operating system: A system compatible with various remote sensing data processing and machine learning software, such as Windows or Linux.

[0093] Storage: High-speed SSD or large-capacity hard disk is used to store large amounts of remote sensing data.

[0094] monitor:

[0095] Type: High resolution to show image details clearly.

[0096] Keyboard and Mouse:

[0097] Requirements: Ergonomic, ensuring operational efficiency and comfort.

[0098] Network connection:

[0099] Speed: High-speed Internet connection to facilitate data downloads and software updates.

[0100] Remote sensing data processing software:

[0101] Examples: ENVI, ERDAS Image or PCI Geomat i ca, etc.

[0102] Storage devices and peripherals:

[0103] Types: External hard drives, cloud storage services, printers, scanners, etc.

[0104] Other auxiliary equipment:

[0105] Includes: GPS locator, topographic map, etc., for obtaining ground control points.

[0106] Deep Learning Frameworks and Libraries:

[0107] Examples: TensorFlow, PyTorch, Keras, etc., for building and training feature extraction models.

[0108] Computing resources:

[0109] Requirements: Sufficient GPU resources to support large-scale data parallel processing and model training.

[0110] Platform: such as NVID IA CUDA, cuDNN, etc.

[0111] Feature extraction steps and methods

[0112] Data import and organization:

[0113] Import the preprocessed remote sensing data into feature extraction software or machine learning framework.

[0114] Ensure that the format and structure of the data meet the requirements for subsequent processing.

[0115] Spectral feature extraction:

[0116] Methods: Spectral features were extracted using band statistics and principal component analysis (PCA).

[0117] Purpose: Reduce data dimensionality while retaining key information.

[0118] Texture feature extraction:

[0119] Method: Texture features were extracted using gray-level co-occurrence matrix (GLCM), local binary pattern (LBP) and other algorithms.

[0120] Purpose: To capture texture information in images and enhance the accuracy of object classification.

[0121] Shape feature extraction:

[0122] Method: Shape features are extracted through edge detection, morphological analysis and other techniques.

[0123] Purpose: To identify and distinguish land features with specific shapes.

[0124] Spatial relationship feature extraction:

[0125] Methods: Spatial relationship features were extracted using spatial autocorrelation and neighborhood analysis.

[0126] Purpose: To understand the spatial distribution and relationships between objects.

[0127] Temporal feature extraction:

[0128] Methods: Remote sensing images of the same area at different times were analyzed to extract change detection features.

[0129] Purpose: To monitor the changing trends and dynamic processes of land features.

[0130] Deep Learning Feature Extraction:

[0131] Method: Use deep learning models such as convolutional neural network (CNN) and recurrent neural network (RNN) to automatically extract high-level abstract features.

[0132] Advantages: Able to learn complex feature representations and improve the accuracy and automation of feature extraction.

[0133] The support of the above-mentioned hardware and software and the application of feature extraction steps and methods can effectively extract useful feature information from preprocessed remote sensing data. These features will provide important input for subsequent data analysis and application, and lay the foundation for the successful implementation of the entire multi-source data assimilation method.

[0134] Preferably, the data fusion step S300 aims to fuse the extracted feature information, using appropriate fusion algorithms, such as weighted average, principal component analysis, etc., to integrate the advantages of multi-source data. The required software and hardware support and data fusion method will be described in detail below.

[0135] Software and hardware support

[0136] computer:

[0137] Configuration: high-performance CPU, large-capacity memory, high-end graphics card.

[0138] Operating system: A system that is compatible with various remote sensing data processing and machine learning software, such as Windows or Linux.

[0139] Storage: High-speed SSD or large-capacity hard disk is used to store large amounts of remote sensing data.

[0140] monitor:

[0141] Type: High resolution to show image details clearly.

[0142] Keyboard and Mouse:

[0143] Requirements: Ergonomic, ensuring operational efficiency and comfort.

[0144] Network connection:

[0145] Speed: High-speed Internet connection to facilitate data downloads and software updates.

[0146] Remote sensing data processing software:

[0147] Examples: ENVI, ERDAS Image or PCI Geomat i ca, etc.

[0148] Storage devices and peripherals:

[0149] Types: External hard drives, cloud storage services, printers, scanners, etc.

[0150] Other auxiliary equipment:

[0151] Includes: GPS locator, topographic map, etc., for obtaining ground control points.

[0152] Deep Learning Frameworks and Libraries:

[0153] Examples: TensorFlow, PyTorch, Keras, etc., for building and training feature extraction models.

[0154] Computing resources:

[0155] Requirements: Sufficient GPU resources to support large-scale data parallel processing and model training.

[0156] Platform: such as NVID IA CUDA, cuDNN, etc.

[0157] Data fusion method

[0158] Spectral feature fusion:

[0159] Methods: Spectral features were extracted using band statistics and principal component analysis (PCA).

[0160] Purpose: Reduce data dimensionality while retaining key information.

[0161] Texture feature fusion:

[0162] Method: Texture features were extracted using gray-level co-occurrence matrix (GLCM), local binary pattern (LBP) and other algorithms.

[0163] Purpose: To capture texture information in images and enhance the accuracy of object classification.

[0164] Shape feature fusion:

[0165] Method: Shape features are extracted through edge detection, morphological analysis and other techniques.

[0166] Purpose: To identify and distinguish land features with specific shapes.

[0167] Spatial relationship feature fusion:

[0168] Methods: Spatial relationship features were extracted using spatial autocorrelation and neighborhood analysis.

[0169] Purpose: To understand the spatial distribution and relationships between objects.

[0170] Temporal feature fusion:

[0171] Methods: Remote sensing images of the same area at different times were analyzed to extract change detection features.

[0172] Purpose: To monitor the changing trends and dynamic processes of land features.

[0173] Deep Learning Feature Fusion:

[0174] Method: Use deep learning models such as convolutional neural network (CNN) and recurrent neural network (RNN) to automatically extract high-level abstract features.

[0175] Advantages: Able to learn complex feature representations and improve the accuracy and automation of feature extraction.

[0176] Through the support of the above-mentioned hardware and software and the application of feature extraction steps and methods, useful feature information can be effectively extracted from the preprocessed remote sensing data. These features will provide important input for subsequent data analysis and application, and lay the foundation for the successful implementation of the entire multi-source data assimilation method.

[0177] Preferably, step S400 data verification is a key step to ensure data accuracy and reliability, and can help identify and correct errors that may occur during data processing. The following is a specific verification method:

[0178] Data verification steps and methods

[0179] Compare with reference data:

[0180] Use reference data of known accuracy (such as ground-truth data, high-precision remote sensing products, etc.) to compare with the assimilation results.

[0181] Calculate error metrics, such as root mean square error, mean absolute error, etc., to evaluate the accuracy of the data.

[0182] Cross Validation:

[0183] If possible, perform cross-validation using different datasets or algorithms.

[0184] Compare the results obtained from different datasets or algorithms to check their consistency.

[0185] Time Series Analysis:

[0186] Analyze time series data and check the temporal consistency of the data.

[0187] Look for outliers, sudden changes, or changes in trends that may be indicators of errors in the data.

[0188] Spatial consistency check:

[0189] Check the spatial consistency of the data to ensure that there are no unreasonable jumps or differences in the data between adjacent areas.

[0190] Use spatial statistical analysis methods, such as spatial autocorrelation analysis, to identify spatial anomalies.

[0191] Expert Review:

[0192] Ask domain experts to review the data; their experience and knowledge can help identify problems in the data.

[0193] The data is corrected and improved through expert feedback.

[0194] User feedback:

[0195] If the data is already used in real-world applications, collect user feedback.

[0196] Users' practical experience can provide important information about the accuracy and usefulness of the data.

[0197] Sensitivity analysis:

[0198] Conduct sensitivity analysis on the assimilation model to understand how changes in different input data affect the results.

[0199] This helps identify uncertainties and potential sources of error in the data.

[0200] Error Propagation Analysis:

[0201] Analyze how errors propagate in the assimilation process.

[0202] Identify steps or parameters that may cause errors to be amplified and take steps to correct them.

[0203] Statistical tests:

[0204] Use statistical tests, such as hypothesis testing, to assess whether the data have expected statistical properties.

[0205] For example, you can use a t-test to compare whether two sets of data are significantly different.

[0206] Visual inspection:

[0207] Use data visualization tools, such as GIS software, to map data onto geographic space.

[0208] Visually inspect the spatial distribution of your data, looking for discontinuities or anomalous areas.

[0209] Software and hardware support

[0210] For the data verification step, in addition to the software and hardware support required for the above-mentioned result output, the following specific tools may also be required:

[0211] Statistical analysis software:

[0212] Statistical libraries such as R, SPSS or Python (e.g., statsmode ls, scipy) for performing statistical tests and analyses.

[0213] Data Visualization Tools:

[0214] Such as Tableau and Power BI, which are used to create interactive data visualizations to help identify problems in the data.

[0215] Database query tools:

[0216] Such as SQL Server Management Studio and pgAdmin, which are used to query and analyze data stored in the database.

[0217] Through the data verification step, the "multi-source data assimilation method based on remote sensing" of the present invention will be more perfect, ensuring the accuracy and reliability of the data and improving the quality of the final product. Data verification should become a standard part of any data processing and analysis work to ensure the validity and credibility of the results.

[0218] Preferably, step S500 of constructing an assimilation model is a key step in combining remote sensing data with numerical models, and determines the accuracy and efficiency of data fusion. In terms of software and hardware support, it is necessary to fully consider the complexity of data processing and real-time requirements to ensure the efficiency and stability of model operation. The construction method and main features of the assimilation model are as follows:

[0219] Hardware Support:

[0220] High-performance multi-core CPU, used to process complex model operations and algorithm logic.

[0221] High-performance GPUs, such as NVIDIA's A100 or H100, are used to accelerate numerical computing and deep learning algorithms.

[0222] Large-capacity, high-speed storage devices, such as SSDs or high-speed disks, to store large data sets and model parameters.

[0223] High-speed network connection ensures the speed and stability of data communication in distributed training.

[0224] Software Framework:

[0225] Use deep learning frameworks such as TensorFlow or PyTorch to build models. These frameworks provide rich API interfaces to facilitate model building and training.

[0226] Adopt parallel computing and distributed computing strategies, such as data parallelism and model parallelism, to optimize training speed and expand to multi-machine and multi-card environments.

[0227] Model construction method:

[0228] Data preprocessing, including format conversion, normalization, denoising, etc., provides standardized input for subsequent models.

[0229] Build a deep learning module and use convolutional neural network (CNN), recurrent neural network (RNN) and other structures for feature extraction and time series analysis.

[0230] The remote sensing data is integrated into the numerical model through specific assimilation algorithms (such as Kalman filtering and variational methods) to achieve the optimal estimation of state variables.

[0231] Model features:

[0232] Highly non-linear and able to capture complex data relationships.

[0233] Real-time and dynamic update capabilities can quickly reflect environmental changes.

[0234] Fault tolerance and robustness, with a certain degree of resistance to data missing and noise.

[0235] Hash system selection:

[0236] Although the system based on NVIDIA's GPU is expensive, its hardware performance and programming ecosystem are relatively mature.

[0237] Systems based on domestic AI chips need to improve the ecosystem and enhance user acceptance.

[0238] Use the idle computing power of supercomputers to train large models, such as the powerful computing power and optimized system design provided by the Sunway supercomputer.

[0239] Building an assimilation model requires comprehensive consideration of hardware selection, software framework, model building method, and model characteristics to ensure the accuracy and efficiency of data processing. When selecting hardware configuration, it is necessary to weigh multiple factors such as computing power, storage, and communication to provide strong support for model training and operation. At the same time, according to model requirements and actual application scenarios, choosing a suitable computing system can effectively improve the efficiency and effectiveness of model training.

[0240] Preferably, the step of outputting the results in step S600 is to convert the results of the assimilation model into a format that can be analyzed, displayed, and further applied. The following are the software and hardware support and specific methods required for the result output:

[0241] Hardware Support

[0242] Processor: A high-performance CPU that processes data quickly and runs programs that produce output results.

[0243] Storage device: A hard disk or solid-state drive (SSD) large enough to store large amounts of output data.

[0244] Display device: A high-resolution and appropriately sized monitor to clearly display the output results.

[0245] Printer: A high-quality printer if results need to be printed out for physical distribution or archiving.

[0246] Software Support

[0247] GIS software: such as ArcG IS, QG IS, etc., is used to combine the output results with the geographic information system to display and analyze spatial data.

[0248] Remote sensing data processing software: such as ENVI and ERDAS Image, these software usually include functions for data processing and visualization.

[0249] General graphics processing software: such as Adobe Photoshop or Illustrator, for post-processing and beautification of results.

[0250] Programming languages ​​and libraries: Python and its data visualization libraries Matplotlib and Seaborn, which can be used to programmatically output charts and reports.

[0251] Reporting tools: such as LaTeX or Microsoft Word, for writing scientific reports or documents containing output results.

[0252] Database management system: such as MySQL and PostgreSQL, used to store and manage output results for subsequent query and use.

[0253] Specific output method

[0254] Image output:

[0255] Generate assimilated remote sensing images, such as land cover maps, vegetation index maps, etc.

[0256] Use GIS software to overlay images with map layers to provide geographic reference information.

[0257] Adjust the color, contrast, and brightness of images to ensure they are clear and readable.

[0258] Export to common image formats such as JPEG, PNG for use in reports or web presentations.

[0259] Tabular output:

[0260] Organize the results of data assimilation into tables, such as CSV and XLSX formats.

[0261] Use data processing software to perform statistics and analysis on the tables.

[0262] Export tabular data to make it easier for non-professionals to understand and use.

[0263] Chart output:

[0264] Use programming languages ​​to generate charts, such as scatter plots, line charts, bar charts, etc., to show trends and relationships in data.

[0265] Embed these charts in reports to provide intuitive data interpretation.

[0266] Print or digitally distribute these figures for conference presentations or academic communications.

[0267] Report output:

[0268] Write a detailed analysis report explaining the process, results, and significance of the data assimilation.

[0269] Include key data images, tables, and charts to support the report's conclusions.

[0270] Use reporting tools to format and ensure reports are professional and readable.

[0271] Data Sharing:

[0272] Upload the resulting dataset to an online database or data warehouse, such as Pangaea, Figshare.

[0273] Provide necessary metadata and description files to facilitate access and use by other researchers.

[0274] Follow the principles of open data and promote data reuse and scientific progress.

[0275] Through the above methods, the results of data assimilation can be effectively converted into various forms to meet the needs of different users and application scenarios.

[0276] It should be noted that, in this article, the terms "comprise", "include" or any other variant thereof are intended to cover non-exclusive inclusion, so that the process, method, article or device including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise one..." do not exclude the presence of other identical elements in the process, method, article or device including the element. In addition, it should be pointed out that the scope of the methods and devices in the embodiments of the present application is not limited to performing functions in the order shown or discussed, and may also include performing functions in a substantially simultaneous manner or in reverse order according to the functions involved, for example, the described method may be performed in an order different from that described, and various steps may also be added, omitted, or combined. In addition, the features described with reference to certain examples may be combined in other examples.

[0277] The embodiments of the present application are described above in conjunction with the accompanying drawings, but the present application is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present application, ordinary technicians in this field can also make many forms without departing from the purpose of the present application and the scope of protection of the claims, all of which are within the protection of the present application.

Claims

1. A multi-source data assimilation method based on remote sensing, characterized by: The steps include: S100: Data preprocessing: Preprocess multi-source remote sensing data, including radiation correction, geometric correction, and cloud detection, to eliminate noise and errors in the data; S200: Feature extraction: extract useful feature information from the preprocessed data, including spectral features and texture features, for subsequent data analysis and application; S300: Data fusion: Fuse the extracted feature information, using appropriate fusion algorithms, weighted average, and principal component analysis to integrate the advantages of multi-source data; S400: Data verification: multiple verification methods are used to verify the data, and feedback is given for wrong data. If the data verification is correct, the next step is entered; S500: Assimilation model construction: Based on the fused data, construct an appropriate assimilation model, including Kalman filtering and variational method, to achieve optimal estimation of multi-source data; S600: Result output: Based on the results of the assimilation model, the final remote sensing data products are output, including land cover maps and vegetation index maps.

2. The multi-source data assimilation method based on remote sensing according to claim 1, characterized in that: The data preprocessing in step S100 includes the following sub-steps: Data collection: Collect data through various means and determine the data source; Preliminary screening: Check the completeness of the data to ensure that the required data is not missing and convert all data into the unified format required in the processing flow; Data cleaning: Use filters or smoothing techniques to reduce random noise in the data. For missing data, use interpolation, mean value filling and ignoring methods to identify and process outliers in the data. Data correction: Radiation correction: perform radiation calibration on remote sensing data to eliminate sensor errors and atmospheric influences, and perform geographic correction on remote sensing images to ensure that the images correspond to the actual geographic location; Data fusion: Data of different resolutions are fused to the same spatial scale through resampling and other methods, time series data are analyzed, time labels are adjusted, and time consistency is ensured; Feature extraction: extract reflectance and vegetation index spectral features from remote sensing data, calculate gray-level co-occurrence matrix and local binary pattern texture features, and perform edge detection on objects to extract shape features; Data normalization: Standardize the data to a range of 0-1 to reduce the impact of different dimensions, use maximum-minimum normalization or Z-score normalization to make the data comparable; Data enhancement: rotation transformation: increase sample diversity by rotating the image, flip transformation: flip the image horizontally or vertically, translation transformation: translate the image along the x or y direction, scaling transformation: change the size of the image; Data segmentation: Divide the data set into a training set and a test set, with a common ratio of 70% training set and 30% test set; Data encoding: Convert categorical data into a numerical form that the model can understand, converting class labels into integers.

3. The multi-source data assimilation method based on remote sensing according to claim 1, characterized in that: The feature extraction in step S200 also includes spectral feature extraction, texture feature extraction, shape feature extraction, spatial relationship feature extraction, time feature extraction, deep learning feature extraction, feature selection, and feature fusion.

4. The multi-source data assimilation method based on remote sensing according to claim 1, characterized in that: The data fusion in step S300 also includes data processing, feature extraction, data-level fusion, feature-level fusion, model-level fusion, fusion strategy, post-fusion processing, and quality assessment.

5. The multi-source data assimilation method based on remote sensing according to claim 1, characterized in that: The data verification in step S400 includes data source check, data format and structure verification, data range and logic check, data accuracy verification, data timeliness verification, data consistency verification, data quality assessment, data clarity and preprocessing, verification result recording, and user feedback.

6. The multi-source data assimilation method based on remote sensing according to claim 1, characterized in that: The step S500 of assimilation model construction includes the following sub-steps: Model selection: Select a suitable assimilation algorithm based on data characteristics and research objectives, determine the basic architecture of the model, and select the appropriate number of neural network layers and nodes; Model design: Design the input layer of the model according to the fused data features, design the hidden layer of the model, including the number of layers, the number of neurons in each layer, and the activation function; design the output layer of the model according to the prediction target, including the design of continuous output for regression problems and the design of discrete output for classification problems; Parameter initialization: Set initial values ​​for the model's weight parameters, use random initialization or pre-trained weights, and set initial values ​​for the model's bias parameters; Loss function definition: Choose an appropriate error metric, including mean square error for regression problems and cross entropy for classification problems, and add regularization terms to prevent overfitting; Optimization algorithm selection: Select an appropriate optimization algorithm and adjust the hyperparameters of the optimization algorithm, including learning rate, batch size, and number of iterations; Training and validation: Prepare training sets using preprocessed and fused data, prepare validation sets to evaluate model performance and make model selection, and apply cross-validation techniques to evaluate the generalization ability of the model; Model training: Divide the dataset into small batches for training to fit within memory constraints, and use an early stopping strategy to avoid overfitting, i.e. stop training when the validation error no longer decreases; Model testing: Use an independent test set to perform a final evaluation of the model and calculate the model's performance metrics; Model tuning: adjust hyperparameters and model structure based on the model’s performance on the validation set; Model deployment: Save the trained model for subsequent use and load the model in actual applications for prediction and analysis.

Images