Dynamic cloudless image synthesis method and device based on multi-source remote sensing data

Abstract

The embodiment of the invention provides a dynamic cloudless image synthesis method and device based on multi-source remote sensing data, and is applied to the technical field of remote sensing images. The method comprises the following steps: acquiring multi-source remote sensing data to construct a global progress management space file, and creating a unit state vector file and a unit source list text file for each grid unit in the global progress management space file; acquiring newly added multi-source remote sensing data and constructing a to-be-processed data list; dynamically calculating, updating and synthesizing a corresponding virtual grid file according to a unit state vector file and a unit source list text file of a to-be-updated grid unit in the global progress management space file and a to-be-processed data list; and when the cloud-free coverage rate of the grid units in the global progress management space file reaches a preset cloud-free coverage rate, generating a cloud optimization format file according to the virtual grid file. On the basis, an automatic production system which deeply integrates spatialized state management, intelligent data scheduling and high-quality seamless synthesis is constructed.

Description

Technical Field

[0001] This disclosure relates to the field of image processing technology, and more particularly to the field of remote sensing image technology, specifically to a method and apparatus for dynamic cloudless image synthesis based on multi-source remote sensing data. Background Technology

[0002] Various satellite constellations are continuously acquiring images of Earth at an unprecedented frequency, forming massive, multi-source, and multi-temporal remote sensing data archives at the petabyte (PB) and even exabyte (EB) levels. When utilizing these massive data archives, a common and fundamental technical goal is to generate standardized synthetic image products that cover a wide area, are seamless, and cloud-free. These products are the key data foundation for subsequent geospatial information extraction and analysis.

[0003] However, efficiently fusing spatiotemporally discrete raw data products into high-quality, large-scale homogeneous imagery still faces significant technical challenges. Current technological approaches can be broadly categorized as follows: I. Manual / Semi-Automatic Processing Based on Professional Software: This method relies on professional remote sensing or geographic information systems (GIS), specifically commercial or open-source professional remote sensing / GIS software such as ENVI, ERDAS, ArcGIS, and QGIS. While this approach allows for precise operations within small areas, the workflow requires technicians to manually or with the aid of software tools to perform image screening, sorting, color correction, seam line editing, and final quality checks. The entire process is operator-led, with the software providing computational and visualization assistance. However, when processing large-scale data, its lengthy process, labor-intensive nature, and poor reproducibility make it unscalable. II. Script-based Automated Batch Processing: This approach uses scripts, such as Python's GDAL library or IDL language, to automate pre-defined mosaicking processes. Typically, these scripts can perform automated operations such as cropping, stitching, color balancing, and output on a given range and pre-selected image set, aiming to improve processing efficiency. However, while this method improves automation to some extent, its processing logic is usually linear and static, lacking intelligence to handle complex scenarios. For example, they struggle to dynamically select the optimal pixels in massive overlapping images, cannot perform efficient incremental updates when new data arrives, and lack mechanisms for fine-grained state tracking and error recovery for large-scale processing tasks such as framing and block partitioning. III. Online processing services based on cloud platforms: Some cloud service platforms, such as Google Earth Engine, provide online image compositing functions. Users can use API calls to execute predefined compositing algorithms on datasets hosted by the platform, such as taking the maximum / minimum / median value in time sequence, to quickly generate preview-level or analysis-level image products; however, the information chain of its production process is broken, and the results cannot be traced.

[0004] Therefore, there is a certain technological gap in the current field of remote sensing data processing: a lack of a systematic methodology for intelligent scheduling, automated processing, and refined management of massive, heterogeneous data streams. Current technologies cannot effectively solve a series of intertwined engineering and algorithmic challenges encountered in large-scale, dynamic synthesis tasks, such as data selection, process control, quality assurance, and efficiency bottlenecks.

[0005] Specifically, the current process of transforming massive, heterogeneous remote sensing data products into an authoritative, dynamic "unified map" faces the following five core technological challenges: (1) The scheduling and management of massive heterogeneous data: How to establish a scalable production management and control system in a continuously growing, petabyte-level multi-source remote sensing data product archive that can automatically manage massive data files, avoid redundant processing, and finely track the production status of tens of thousands of geographic units; that is, there is a lack of a unified production management and scheduling architecture. All current solutions lack a top-level, spatialized status management and task scheduling mechanism. They cannot macroscopically monitor and microscopically track the production progress, data coverage completeness, quality status, etc. of tens of thousands of independent work units decomposed into a large area, resulting in chaotic and uncontrollable processes during large-scale production; (2) Large-scale dynamic update and optimal pixel selection problem: How to abandon the current static production mode of "selecting images first and then embedding" and establish a dynamic, incremental update mechanism? This mechanism needs to be able to automatically and intelligently select the "optimal" observation value from different time phases and different sensors for each pixel on the map from the massive amount of data added daily and archived in history, such as cloudless, highest quality, and latest time phase, and seamlessly integrate it into the "one map"; that is, the update mechanism is static and inefficient. The current solution is essentially a "one-off" production mode. When new, higher quality data is available, they lack an incremental, intelligent update engine to "integrate" the effective parts of these new data into the existing product. Usually, only large-scale or even complete re-production can be selected, which is costly and slow to respond. (3) Computational efficiency and resource bottleneck issues: Current image mosaicking methods generate a large number of large intermediate files during processing, which makes disk I / O an efficiency bottleneck and consumes a staggering amount of storage resources. How to design a new computing architecture that can minimize or even eliminate the dependence on physical intermediate files during large-scale synthesis and achieve lightweight and high efficiency in the computing process; that is, the computing mode consumes huge resources. The current mosaicking process relies heavily on generating a large number of physical intermediate files on the disk, such as cropped and color-balanced images, which leads to huge storage and I / O overhead and a long computing chain. When processing TB or even PB level data, the efficiency bottleneck of this mode is extremely prominent. (4) Product quality and visual consistency issues: How to fundamentally solve the quality problems such as seam marks, color differences, and "holes" or "flaws" caused by tiny clouds / shadows or missing data when splicing data from different sources, different time phases, and different imaging conditions, so as to ensure that the final "one-page map" achieves a high degree of continuity, consistency and integrity in both visual and physical aspects; that is, product quality is highly dependent on human intervention. Whether it is the consistency of color or the natural transition of the seam area, the processing effect of existing automated algorithms is often difficult to meet the application standards. The final quality assurance still relies heavily on time-consuming and labor-intensive manual optimization by professional personnel, which makes it difficult to guarantee the uniformity of product quality on a large scale. (5) Issues of end-to-end traceability and reliability of results: How to ensure that the final "one-map" product is not a "black box" result of unknown origin; that is, it is necessary to establish a complete traceability mechanism to ensure that any pixel in the product can be accurately and quickly traced back to its original data source, such as which satellite, which scene image, and which acquisition time, thereby providing a solid guarantee for the product's authority, application reliability, and subsequent scientific analysis; that is, if the information chain of the production process is broken, the results cannot be traced. In the complex synthesis process, the information chain about which original scene each pixel in the final image product specifically comes from and when and where it was acquired is usually broken. This untraceability of the source seriously weakens the authority and reliability of the final results and also brings obstacles to subsequent scientific verification.

[0006] It is evident that all of the aforementioned existing technical solutions, regardless of which one, have fundamental limitations in their design concepts when dealing with large-scale, dynamic, and high-quality image compositing tasks.

[0007] Based on this, developing a systematic method that can automatically, intelligently, and efficiently synthesize multi-source remote sensing data into large-scale images or even "cloudless maps" is not only an inevitable trend in technological development, but also an urgent need to meet the needs of refined governance in fields such as natural resources, ecological environment, agriculture, and emergency response. Summary of the Invention

[0008] This disclosure provides a method, apparatus, device, and storage medium for dynamic cloudless image synthesis based on multi-source remote sensing data.

[0009] According to a first aspect of this disclosure, a method for dynamic cloudless image synthesis based on multi-source remote sensing data is provided. The method includes: Acquire multi-source remote sensing data, construct a global progress management spatial file, and create a cell status vector file and a cell source list text file for each grid cell in the global progress management spatial file; Acquire newly added multi-source remote sensing data and construct a list of data to be processed; Based on the cell state vector file and cell source list text file of the grid cells to be updated in the global progress management space file, as well as the data list to be processed, the corresponding virtual raster file is dynamically calculated, updated, and synthesized. When the cloud-free coverage of the grid cells in the global progress management space file reaches the preset cloud-free coverage, a cloud-optimized format file is generated based on the virtual raster file.

[0010] As described above and in any possible implementation, a further implementation is provided, wherein acquiring multi-source remote sensing data, constructing a global progress management spatial file, and creating a cell state vector file and a cell source list text file for each grid cell in the global progress management spatial file includes: Acquire multi-source remote sensing data and generate corresponding standardized geographic grids; Based on the multi-source remote sensing data and the standardized geographic grid, a global progress management spatial file is constructed; the attribute fields of each grid cell in the global progress management spatial file include a unique map sheet code, cloud-free coverage percentage, processing status, last update date, and final product storage path; Create a cell status vector file and a cell source list text file for each grid cell; the cell status vector file initially includes a polygon representing the complete range of the grid cell and marks the attribute fields as uncovered, which is used for dynamic calculation and updating of cloudless coverage; the cell source list text file is used to record the source information of each scene of the original data image that constitutes the composite cloudless product of the grid cell.

[0011] In addition to the aspects and any possible implementations described above, a further implementation is provided in which the acquisition of new multi-source remote sensing data and the construction of a data list to be processed include: Newly added multi-source remote sensing data is obtained based on a registry with a preset unique identifier, and the data is identified and filtered to generate an incremental dataset. The incremental dataset includes image data files of each scene and corresponding cloud mask files. Perform multiple validation processes on the incremental dataset; The processed incremental dataset is structured to create a list of data to be processed.

[0012] As described above and in any possible implementation, a further implementation is provided, wherein the step of dynamically calculating and updating, and synthesizing the corresponding virtual raster file based on the cell state vector file and cell source list text file of the grid cells to be updated in the global progress management space file, and the data list to be processed, includes: When the candidate image of the grid cell to be updated in the global progress management spatial file has a spatial overlap with each scene in the data list to be processed, the cloud mask file of the candidate image is vectorized to generate an effective contribution polygon. Based on the geometric difference operation, the uncovered area polygon of the grid cell is dynamically calculated and updated according to the effective contribution polygon. The cell source list text file of the grid cell is updated synchronously, and the corresponding virtual raster file is synthesized.

[0013] As described above and in any possible implementation, a further implementation is provided, wherein the dynamic calculation and updating of the corresponding virtual raster file based on the cell state vector file and cell source list text file of the grid cells to be updated in the global progress management space file, and the data list to be processed, further includes: When there are multiple candidate images in the global progress management spatial file that overlap spatially with each scene in the data list to be processed, the candidate images are analyzed based on preset optimization rules to determine the optimal candidate image. Then, the cloud mask file of the optimal candidate image is vectorized to generate an effective contribution polygon. Based on geometric difference operations, the uncovered area polygon of the grid cell is dynamically calculated and updated according to the effective contribution polygon. The cell source list text file of the grid cell is updated synchronously, and the corresponding virtual raster file is synthesized. The preset selection rule includes: selecting the latest temporal image and / or the image with the lowest cloud cover from the candidate images as the optimal candidate image.

[0014] As described above and in any possible implementation, a further implementation is provided, wherein the dynamic calculation and updating of the corresponding virtual raster file based on the cell state vector file and cell source list text file of the grid cells to be updated in the global progress management space file, and the data list to be processed, further includes: When the candidate images of the grid cells to be updated in the global progress management spatial file and the scenes in the data list to be processed are not unique, based on the preset pixel-by-pixel optimal decision model, the effective contribution polygons are generated by vectorizing the cloud mask files of each candidate image in an iterative manner according to the priority order of the candidate image quality evaluation values. Based on the geometric difference operation, the uncovered area polygons of the grid cell are dynamically calculated and updated according to the effective contribution polygons generated by the candidate image with the highest quality score. The cell source list text file of the grid cell is updated synchronously, and the corresponding virtual raster file is synthesized. The preset pixel-by-pixel optimal decision model includes: in, Represents a multi-dimensional pixel quality evaluation value. This represents the weighting coefficient of the basic quality weighted score. This represents the weighted score for basic quality. This indicates a pixel-based pure reward item. Indicates macro-level reward items, Indicates pollution penalties. Indicates sharpness score, This represents the weighting coefficient for the sharpness score. Indicates transparency. This represents the weighting coefficient for transparency scores. Indicates cloud proximity score. represents the weighting coefficient of cloud proximity score, and N(⋅) represents the normalization function.

[0015] As described above and in any possible implementation, a further implementation is provided, wherein generating a cloud-optimized format file based on the virtual raster file when the cloud-free coverage rate of the grid cells in the global progress management spatial file reaches a preset cloud-free coverage rate includes: When the cloud-free coverage of the grid cells in the global progress management spatial file reaches the preset cloud-free coverage, a cloud-optimized format file is generated based on the virtual raster file according to the preset optimization algorithm; the preset optimization algorithm includes an automatic seam line optimization algorithm based on ground feature perception and / or an image filling algorithm based on convolution kernel.

[0016] According to a second aspect of this disclosure, a dynamic cloudless image synthesis device based on multi-source remote sensing data is provided. The device includes: The module is used to acquire multi-source remote sensing data, construct a global progress management spatial file, and create a cell status vector file and a cell source list text file for each grid cell in the global progress management spatial file. The module is also used to acquire new multi-source remote sensing data and build a list of data to be processed; The calculation module is used to dynamically calculate and update the corresponding virtual raster file based on the cell state vector file and cell source list text file of the grid cells to be updated in the global progress management space file and the data list to be processed; The generation module is used to generate a cloud-optimized format file based on the virtual raster file when the cloud-free coverage of the grid cells in the global progress management spatial file reaches a preset cloud-free coverage.

[0017] According to a third aspect of this disclosure, an electronic device is provided. The electronic device includes a memory and a processor, wherein the memory stores a computer program, and the processor executes the program to implement the method described above.

[0018] According to a fourth aspect of this disclosure, a computer-readable storage medium is provided having a computer program stored thereon that, when executed by a processor, implements the method described above.

[0019] This application provides a method for dynamic cloudless image synthesis based on multi-source remote sensing data. It acquires multi-source remote sensing data, constructs a global progress management spatial file, and creates a unit status vector file and a unit source list text file for each grid cell within the global progress management spatial file. It then acquires new multi-source remote sensing data and constructs a data list to be processed. Based on the unit status vector file, unit source list text file, and data list to be processed of the grid cells to be updated in the global progress management spatial file, it dynamically calculates and updates the data, synthesizing the corresponding virtual raster file. When the cloudless coverage rate of the grid cells in the global progress management spatial file reaches a preset cloudless coverage rate, it generates a cloud-optimized format file based on the virtual raster file. Based on this, this disclosure provides a method for large-scale dynamic cloudless image synthesis based on multi-source remote sensing data. Through the above method, it constructs an automated production system that deeply integrates spatialized status management, intelligent data scheduling, and high-quality seamless synthesis. This method completely overturns the current "human-finded data, manual stitching" work mode, realizing a paradigm shift of "data-driven, automatic synthesis."

[0020] It should be understood that the description in the Summary of the Invention is not intended to limit the key or essential features of the embodiments of this disclosure, nor is it intended to restrict the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0021] The above and other features, advantages, and aspects of the embodiments of this disclosure will become more apparent from the accompanying drawings and the following detailed description. The drawings are provided for a better understanding of the invention and are not intended to limit the scope of this disclosure. In the drawings, the same or similar reference numerals denote the same or similar elements, wherein: Figure 1 A flowchart is shown for a dynamic cloudless image synthesis method based on multi-source remote sensing data according to an embodiment of the present disclosure; Figure 2 An automated production system for a dynamic cloudless image synthesis method based on multi-source remote sensing data according to embodiments of the present disclosure is shown; Figure 3 A block diagram of a dynamic cloudless image synthesis apparatus based on multi-source remote sensing data according to an embodiment of the present disclosure is shown; Figure 4 A block diagram of an exemplary electronic device capable of implementing embodiments of the present disclosure is shown. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of the embodiments of this disclosure clearer, the technical solutions of the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this disclosure, and not all embodiments. Based on the embodiments of this disclosure, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this disclosure.

[0023] Furthermore, the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.

[0024] This disclosure provides a method for synthesizing large-scale dynamic cloudless images based on multi-source remote sensing data. Through the above method, an automated production system is constructed that deeply integrates spatial state management, intelligent data scheduling, and high-quality seamless synthesis. This method completely subverts the current operation mode of "humans finding data and manually stitching together" and realizes a paradigm shift of "data-driven and automatic synthesis".

[0025] Figure 1 A flowchart of a dynamic cloudless image synthesis method 100 based on multi-source remote sensing data according to an embodiment of the present disclosure is shown.

[0026] In box 110, acquire multi-source remote sensing data, construct a global progress management spatial file, and create a cell status vector file and a cell source list text file for each grid cell in the global progress management spatial file.

[0027] In some embodiments, to achieve accurate tracking, management, and automated scheduling of massive data processing, a structured and scalable project workspace and status management system can be constructed.

[0028] In some embodiments, a global progress management spatial file can be constructed by acquiring multi-source remote sensing data to build a spatially indexed two-level status management system and project initialization, that is, to build a two-layer linked spatialized production and management system. This step provides a deterministic and traceable management framework for the entire automated process of the dynamic cloudless image synthesis method based on multi-source remote sensing data.

[0029] In some embodiments, the acquisition of multi-source remote sensing data, the construction of a global progress management spatial file, and the creation of a cell state vector file and a cell source list text file for each grid cell in the global progress management spatial file specifically include: Acquire multi-source remote sensing data and generate corresponding standardized geographic grids; Based on multi-source remote sensing data and standardized geographic grids, a global progress management spatial file is constructed. The attribute fields of each grid cell in the global progress management spatial file include the unique map sheet code, cloud-free coverage percentage, processing status, last update date, and final product storage path. Create a cell status vector file and a cell source list text file for each grid cell. The cell status vector file initially includes a polygon representing the complete range of the grid cell and marks the attribute fields as uncovered, which is used for dynamic calculation and updating of cloudless coverage. The cell source list text file is used to record the source information of each original data image that constitutes the cloudless product of the grid cell.

[0030] In some embodiments, when acquiring multi-source remote sensing data, a standardized, hierarchical directory structure can be automatically created based on the path information defined in the external project configuration file. This structure predefines dedicated storage locations for different types of data assets, such as configuration parameters, runtime logs, data lists, intermediate state files, and final products, thereby ensuring the standardization of data organization and the determinism of access throughout the entire process.

[0031] In some embodiments, the acquired multi-source remote sensing data may be multi-source vector data acquired based on the user's target area vector boundary.

[0032] In some embodiments, the standardized geographic grid corresponding to multi-source remote sensing data can be a set of seamless, non-overlapping standardized geographic grids that divide the target area into a set of target area vector boundaries and standard map sheet coding specifications provided by the user.

[0033] It should be noted that this grid is not only a carrier of geographical division, but is also defined as the basic grid unit for state management, task scheduling and parallel computing in this disclosure, namely, the sectional unit.

[0034] In some embodiments, based on the aforementioned standardized geographic grid and combined with multi-source remote sensing data, a global progress management spatial file for macro-monitoring can be created and initialized. This global progress management spatial file is a geographic vector file, where each polygon feature precisely corresponds to a grid cell, and includes, but is not limited to, the following key attribute fields: unique map sheet code, cloud-free coverage percentage, processing status, last update date, and final product storage path, thereby forming a production status dashboard for macro-monitoring and management. The cloud-free coverage percentage is initialized to 0.00%, and the processing status can include "Pending," "Processing," "Completed," and "Error," etc., with its initial value being "Pending."

[0035] In some embodiments, the system can traverse all grid cells, creating an independent subdirectory for each grid cell as its dedicated workspace. Within this space, two core micro-state files are instantiated: one is a spatial vector file representing the precise coverage status of the cell, i.e., the cell state vector file, which initially contains a polygon representing the complete extent of the grid cell, and its attribute is marked as "NODATA". This is updated during subsequent dynamic coverage calculations. The other is a cell source image list file, i.e., a cell source list text file, a structured text file with predefined headers, such as a comma-separated value (CSV) document. This file is used to accurately record the source information of each original data image that constitutes the cloudless product of the grid cell in subsequent processes, and is the key to achieving pixel-level source traceability.

[0036] As can be seen, by dividing the target area into a unified geographic grid according to standard map sheet specifications, a two-layer linked management structure is initialized based on this grid. At the macro-level management layer, a global progress management space file is created for visually monitoring the cloud-free coverage and production status of each grid cell, such as pending, processing, and completed macro-indicators. At the micro-level management layer, an independent workspace is created for each grid cell, containing two core files: a cell status vector file for accurately depicting the geometry of the "uncovered" areas within the cell; and a cell source list text file for recording the source information of each image element that constitutes the final product. Based on this, the current problem of lacking unified production scheduling and process management is solved, providing a foundation for achieving large-scale parallel computing and pixel-level source traceability.

[0037] In box 120, retrieve the newly added multi-source remote sensing data and build a list of data to be processed.

[0038] In some embodiments, to achieve event-driven incremental data preprocessing and scheduling, an efficient and robust data receiving and verification front-end can be built by acquiring new multi-source remote sensing data and constructing a list of data to be processed. This ensures that only incremental data that meets the requirements of quality and timeliness can be injected into the subsequent production process, thereby achieving efficient and stable operation of the system.

[0039] In some embodiments, the acquisition of new multi-source remote sensing data and the construction of a data list to be processed specifically include: New multi-source remote sensing data is obtained based on a pre-defined unique identifier registry, and then identified and filtered to generate an incremental dataset. The incremental dataset includes image data files for each scene and corresponding cloud mask files. Perform multiple validations on the incremental dataset; The processed incremental dataset is structured to create a list of data to be processed.

[0040] In some embodiments, the preset unique identifier registry can be set according to the user's actual needs.

[0041] In some embodiments, to prevent resource waste caused by data duplication, a persistent registry of unique identifiers for processed images can be maintained. When automatically scanning external data sources daily, the identifiers of newly discovered images are compared with this registry to accurately identify incremental datasets that have not yet been processed.

[0042] In some embodiments, the image data file and corresponding cloud mask file of each in the incremental dataset can be automatically identified according to the rules defined in the above configuration, and multiple verifications can be performed on them, including checking whether the file is complete and readable using a geospatial data access library, verifying whether its spatial reference system is consistent with the project definition, extracting the acquisition date of the image by parsing the file name or metadata, and removing outdated data whose acquisition time is not within the valid window according to the configured timeliness parameters, etc.

[0043] In some embodiments, the metadata of all verified multi-source image products, including file storage path, acquisition time, resolution, sensor type, etc., can be aggregated, formatted, and uniformly imported into a structured "list of data to be processed". This list serves as a candidate data pool, decoupling the front-end data exploration from the back-end processing in the architecture, providing a clear, stable, and deterministic source of task input for the next step of large-scale parallel computing.

[0044] As can be seen, an event-driven incremental data preprocessing and scheduling mechanism can be constructed through the above, and this process constructs an active and efficient data input and task triggering mechanism. For example, by continuously monitoring data storage through the background service, once a new remote sensing image product is found, the following automated process is triggered: (1) Incremental identification and filtering: By comparing with the global processed image registry, i.e. the preset unique identifier registry, it is ensured that only the truly new data is processed to avoid duplicate processing; (2) Data quality and availability verification: A series of automated checks are performed on the new product, including file integrity, spatial reference consistency, metadata validity, etc., to ensure the quality of the original data of the synthesized image; (3) Constructing a list of tasks to be processed: The qualified product information that has passed the verification, such as file path, cloud cover, acquisition time and other key metadata, is constructed into a structured "list of data to be processed". This list serves as the data source for subsequent steps, and its dynamic updates drive the continuous operation of the entire production process. Based on this, the current passive data filtering can be upgraded to an active, event-driven scheduling engine. This not only ensures the incrementality and high quality of the data source, but more importantly, it provides a clear and deterministic data input pool for subsequent parallel update tasks, which is the key to realizing "data-driven" automated production.

[0045] In box 130, based on the element state vector file, element source list text file, and data list to be processed of the grid elements to be updated in the global progress management space file, the corresponding virtual raster file is dynamically calculated and updated to synthesize the corresponding virtual raster file.

[0046] In some embodiments, a multi-process parallel processing architecture can be adopted, in which the map sheet units to be updated are treated as independent task packages and distributed to multiple computing cores for asynchronous concurrent processing through mechanisms such as multi-process pools. This enables dynamic updates and real-time virtual synthesis based on geometric operations, accurately calculating cloud-free coverage and effective coverage areas using vector geometry. Combined with real-time virtual computing technology, this allows for the traceability of dynamic "growth" of images and visualization of the process within a parallel processing framework.

[0047] In some embodiments, in each independent processing step, the current state of the target map sheet unit can be loaded first, namely the geometry of the "uncovered" area in the "Unit State Vector File" and the existing records in the "Unit Source List Text File".

[0048] In some embodiments, the above-mentioned dynamic calculation and updating of the corresponding virtual raster file based on the element state vector file, element source list text file, and data list of the grid elements to be updated in the global progress management spatial file, specifically includes: When there is only one candidate image with spatial overlap between the grid cell to be updated in the global progress management spatial file and each scene in the data list to be processed, the cloud mask file of the candidate image is vectorized to generate an effective contribution polygon. Based on the geometric difference operation, the uncovered area polygon of the grid cell is dynamically calculated and updated according to the effective contribution polygon. The cell source list text file of the grid cell is updated synchronously, and the corresponding virtual raster file is synthesized.

[0049] In some embodiments, for each candidate image in the data list to be processed that has spatial overlap with the grid cell to be updated, a raster-to-vector conversion can be performed on its cloud mask file in real time in memory, and a geometric difference operation can be performed with the current "uncovered" geometry of the grid cell to accurately derive the net effective contribution geometry of the image to the grid cell, i.e. the effective contribution polygon.

[0050] In some embodiments, the above-mentioned dynamic calculation and updating of the corresponding virtual raster file based on the element state vector file, element source list text file, and data list to be processed of the grid elements to be updated in the global progress management spatial file, further includes: When there are multiple candidate images with spatial overlap between the grid cell to be updated and each scene in the data list to be processed in the global progress management spatial file, the candidate images are analyzed based on the preset optimization rules to determine the optimal candidate image. Then, the cloud mask file of the optimal candidate image is vectorized to generate an effective contribution polygon. Based on the geometric difference operation, the uncovered area polygon of the grid cell is dynamically calculated and updated according to the effective contribution polygon. The cell source list text file of the grid cell is updated synchronously, and the corresponding virtual raster file is synthesized. The preset selection rules include: selecting the latest temporal image and / or the image with the lowest cloud cover from the candidate images as the optimal candidate image.

[0051] In some embodiments, preset preference rules can be set according to the user's actual needs.

[0052] In some embodiments, if multiple candidate images contribute, a preset optimization rule can be used to simply select the latest temporal and / or lowest cloud cover image data from the candidate images for "geometric update." This intelligently selects the optimal data product for the current grid cell. Then, using the cloud mask file attached to the product, the cloudless area is vectorized in memory to form an "effective contribution polygon." Through precise geometric difference operations—that is, the "uncovered" area of ​​the cell minus the effective contribution polygon—the geometry of the "uncovered" area of ​​the grid cell is updated. Simultaneously, the source image ID, time, and other information corresponding to the contributing area are written to the cell source list text file to ensure strong consistency between the status and the source record. After the cell source list text file is updated, the virtual composite result is updated immediately, and the virtual raster file (VRT) of the cell is regenerated or updated.

[0053] In some embodiments, the above-mentioned dynamic calculation and updating of the corresponding virtual raster file based on the element state vector file, element source list text file, and data list to be processed of the grid elements to be updated in the global progress management spatial file, further includes: When there are multiple candidate images with spatial overlap between the grid cell to be updated in the global progress management spatial file and each scene in the data list to be processed, based on the preset pixel-by-pixel optimal decision model, the effective contribution polygons are generated by vectorizing the cloud mask files of each candidate image in an iterative manner according to the priority order of the candidate image quality evaluation value. Based on the geometric difference operation, the uncovered area polygons of the grid cell are dynamically calculated and updated according to the effective contribution polygons generated by the candidate image with the highest quality score. The cell source list text file of the grid cell is updated synchronously, and the corresponding virtual raster file is synthesized. The preset pixel-wise optimal decision model includes: in, Represents a multi-dimensional pixel quality evaluation value. This represents the weighting coefficient of the basic quality weighted score. This represents the weighted score for basic quality. This indicates a pixel-based pure reward item. Indicates macro-level reward items, Indicates pollution penalties. Indicates sharpness score, This represents the weighting coefficient for the sharpness score. Indicates transparency. This represents the weighting coefficient for transparency scores. Indicates cloud proximity score. represents the weighting coefficient of cloud proximity score, and N(⋅) represents the normalization function.

[0054] In some embodiments, the preset pixel-by-pixel optimal decision model can be set according to the user's actual needs.

[0055] In some embodiments, if multiple candidate images contribute, a multi-dimensional quality evaluation value can be constructed for each effective pixel of each candidate image overlapping with the current grid cell, based on a "pixel-by-pixel optimal decision" model with multi-dimensional quantification indicators. This evaluation value is dynamically calculated using a pre-defined weighted formula. A typical pixel quality evaluation value... It can be constructed by the following formula: in, This is a pollution penalty item. For cells identified as clouds or cloud shadows, and cells within a specified buffer distance of the cloud edge, this item is a large negative value to ensure that they are excluded first in the decision-making process. This is a macro-level reward item. For global attributes of the source image, such as resolution level and pixels with better overall scene purity, this item is a reward value of a large order of magnitude, reflecting the priority of high-quality data sources. This is a pixel purity bonus; if a pixel itself has no clouds, this basic bonus value will be obtained. It is a basic quality weighted score, which consists of multiple normalized micro-quality indicators; It is the weighting coefficient of the basic quality weighted score.

[0056] It can be constructed by the following formula: in, It is a resolution score, calculated by measuring the variance of the Laplacian operator in the neighborhood of a pixel. The results show that the larger the variance, the richer the texture details, and the higher the score. It is a transparency score, which is obtained by statistically analyzing the brightness value of the dark channel of the image. The lower the dark channel value, the better the atmospheric transparency, and the higher the score. It is cloud proximity, calculated by determining the Euclidean distance between a pixel and the nearest cloud region. After normalization, it was found that the farther away from the cloud area, the higher the score; Let N(⋅) be the weight coefficient of each item, and N(⋅) be the normalization function.

[0057] In some embodiments, the "effective contribution geometry" contributed by higher-quality rated imagery can be "stripped" from the "uncovered" region in an iterative manner, prioritizing the data and using precise geometric difference operations. The stripped portion forms a new "covered" polygon. During the same transaction performing this geometric operation, the complete phylogenetic information of the contributing source is synchronously and atomically appended to a "cell source list text file," ensuring absolute consistency between the status and the source record.

[0058] In some embodiments, after the grid cell state is updated, the virtual raster file (VRT) of that grid cell can be dynamically reconstructed immediately. Each band source of the VRT embeds a custom Python pixel processing function. This custom function is triggered in real time only when an external program requests to read any pixel of the VRT. The function synchronously reads the analysis-ready data (ARD) image pixel values ​​and their cloud mask pixel values ​​at the corresponding location and makes an immediate judgment. For example, if there are no clouds, it returns the image value; if there are clouds, it returns the preset NoData value. This "on-demand computation" virtual processing mechanism completes cloud removal and seamless mosaicking of all ARD slices in memory, avoiding the generation of large intermediate files on disk during dynamic synthesis, and enabling the real-time display of the synthesized image that incorporates the latest data.

[0059] As can be seen, to achieve dynamic updates of cloud-free coverage, source tracing, and real-time virtual composite based on geometric operations, a parallel processing architecture can be adopted. For each grid cell to be updated, a tightly coupled "state update-real-time composite" sequence is executed to synthesize a VRT. This VRT dynamically integrates all recorded source images, and users can view the composite image result with the latest data added and seamless stitching at any time by reading the VRT.

[0060] It should be noted that the above process is the core of realizing image "dynamic growth" and "process visualization", combining "state update" and "result generation" into one. This makes image updating a precise, iterative, and traceable geometric operation, ensuring strong consistency of state and absolute traceability of pixel source. At the same time, it innovatively generates virtual products in real time after each incremental update, realizing real-time visualization and dynamic interaction of the synthesis process. This is a capability that current "black box" batch processing does not have at all.

[0061] In box 140, when the cloud-free coverage of the grid cells in the global progress management space file reaches the preset cloud-free coverage, a cloud-optimized format file is generated based on the virtual raster file.

[0062] In some embodiments, a cloud-optimized format file can be generated from the virtual raster file to perform quality optimization and persistent output of the final product. When the cloud-free coverage of the grid cells meets a preset threshold, the quality optimization and persistent output of the final product will be triggered. Its role is to perform final quality refinement on the dynamic, procedural virtual results generated above and produce a deliverable physical data product.

[0063] In some embodiments, when the cloud-free coverage of the grid cells in the global progress management spatial file reaches a preset cloud-free coverage, generating a cloud-optimized format file based on the virtual raster file includes: When the cloud-free coverage of the grid cells in the global progress management spatial file reaches the preset cloud-free coverage, a cloud-optimized format file is generated based on the preset optimization algorithm and the virtual raster file. The preset optimization algorithms include: a seam line automatic optimization algorithm based on ground feature perception and / or an image filling algorithm based on convolution kernel.

[0064] In some embodiments, the preset optimization algorithm and preset cloudless coverage rate can be set according to the user's actual needs.

[0065] In some embodiments, all grid cells whose coverage meets a preset threshold and whose production status is "product to be generated" can be selected from the global progress management spatial file. For each grid cell, its final version of the cell status vector file and cell source list text file are read to accurately grasp all "covered" patches and their corresponding source remote sensing image information.

[0066] In some embodiments, before performing the final synthesis calculation, an idempotency check is first performed on the preset final output path. If the final product, such as the main TIF file, already exists, the task is considered to have been successfully completed by default, and all subsequent processing is skipped. Only derivatives such as quick views are verified or generated as needed.

[0067] In some embodiments, visual artifacts that may occur at the stitching points of images from different sources can be eliminated by analyzing and optimizing the synthesis scheme. To avoid seam lines abruptly crossing intact features such as buildings and roads, automatic seam line optimization algorithms based on feature perception can be used, such as graph theory-based minimum cost path optimization algorithms.

[0068] For example, firstly, an object-oriented image analysis method is used to quickly classify land features in overlapping image areas and construct a "cost surface," where features such as buildings are assigned high access costs, while areas such as water bodies and bare soil have lower costs. Subsequently, a dynamic programming or A* search algorithm is used to automatically calculate the optimal path with the lowest overall cost on this cost surface, which serves as the final image cropping line.

[0069] In some embodiments, for minor data holes or gaps caused by missing original data or geometric registration errors, a kernel-based image filling algorithm can be used. This algorithm achieves smooth and natural filling by weighted averaging of the effective pixels around the hole, for example using a Gaussian kernel function, ensuring the visual integrity of the final product.

[0070] In some embodiments, after padding and optional visual enhancement, VRT raster data is efficiently instantiated into a cloud-optimized final product format (Cloud Optimized GeoTIFF) optimized for cloud-based geospatial data processing scenarios, i.e., a cloud-optimized format file, and a series of high-value-added derivative products can be generated simultaneously.

[0071] For example, derivative products may include a JSON-formatted manifest file that details the mapping between each ID and the original image file path, as well as the specific meaning of the decision logic code, forming a complete data source and decision process traceability chain, and a JPEG-formatted quick view that is visually enhanced by scaling the main product and truncating it by 2% to 98% of the original product.

[0072] In some embodiments, after all products have been generated, the storage path and completion time of the final product (including all derivatives) are written back to the globally created progress management space file, and the processing status of the grid cell is marked as "completed," signifying the end of the production process for that grid cell. Finally, according to the configuration policy, all temporary files and directories generated by this task are automatically deleted, freeing up storage space.

[0073] As can be seen, by generating cloud-optimized format files based on virtual raster files, the quality optimization and persistent output of the final product can be performed. For example, when the cloud-free coverage of a grid cell reaches a preset threshold during the dynamic calculation and update synthesis process, the final product generation process for that grid cell will be automatically triggered. The first step is refined seamless processing. Based on the synthesis scheme defined by the final generated virtual raster (VRT), algorithms with pre-set parameters are invoked, such as automatic optimization of seam lines based on ground feature perception and micro-hole filling based on convolutional kernels, to repair potential data defects, thereby improving the overall visual quality and data integrity of the image. After optimizing the synthesis scheme, the system renders the final virtual synthesis result as a physical entity, generating a physically existing, high-quality raster image file, and outputting it in standardized formats such as cloud-optimized COG, thus completing all the production work for that grid cell. The purpose of this process is to transform the dynamic, procedural virtual synthesis result—the virtual raster file—into a refined, permanently archived, and distributeable physical data product. The use of seamless processing technology ensures product quality and is crucial in connecting dynamic production and final delivery.

[0074] In summary, compared with current methods that lack a scalable production management architecture, have static and inefficient update mechanisms, consume huge amounts of resources in computation, rely on manual optimization for product quality, and suffer from broken information chains in the production process, this disclosure completely overturns the traditional processing model through a completely new and systematic design. Its core advantage lies in: (1) A refined two-level status management system: This disclosure innovatively constructs a "global-unit" two-level spatial status management system. At the global level, the production status of all map sheets is macroscopically monitored and scheduled through a geographic vector file; at the micro level, an independent workspace is established for each map sheet, and another vector file is used to accurately record the "covered" or "uncovered" status of each area within it. This system is the foundation for realizing full-process automation and intelligent scheduling.

[0075] (2) Incremental dynamic update engine: This disclosure achieves incremental storage of newly added data every day by maintaining a global registry of processed scenarios. More importantly, it replaces the current raster mosaic logic with precise vector geometric operations, which can dynamically and accurately supplement the "effective contribution" part of any new data into the "uncovered" area like a "puzzle", realizing the continuous and dynamic optimization and update of the "map".

[0076] (3) "Zero Intermediate File" Virtual Processing Architecture: To address efficiency bottlenecks, this disclosure innovatively employs a virtual processing technology based on a custom pixel processing function during the final product generation stage. A virtual grid (VRT) is constructed in memory, which performs cloud culling, pixel selection, and dynamic synthesis from multiple source data products in real time, only when a pixel is requested to be read. This mechanism fundamentally eliminates the dependence on large physical intermediate files, greatly improving processing efficiency.

[0077] (4) Advanced seamlessness and quality enhancement technology: This disclosure integrates a set of advanced quality optimization algorithms. It can not only intelligently generate visually hidden seam lines along natural boundaries such as rivers and roads through a dynamic programming algorithm based on ground object perception; it can also use an image filling algorithm based on convolution kernel to smoothly and naturally repair the small data holes that may exist in the final product, ensuring the high integrity and visual aesthetics of the "one map".

[0078] (5) Pixel-level traceability of the entire chain: In the state management system disclosed herein, each “covered” geometric area is forcibly associated with its unique source data ID. This mechanism runs through the entire production process, ensuring that any pixel in the final “one-page” product can be quickly and accurately traced back to its original data scene, time phase and sensor information, and establishing a complete and reliable “digital identity file”.

[0079] To further elaborate on the above-mentioned method for dynamic cloudless image synthesis based on multi-source remote sensing data, the following example will illustrate the construction of a dynamically self-updating application-ready data map.

[0080] It is important to understand that a core technical objective in the current Earth observation technology system is to efficiently and automatically process the continuously generated, massive, and heterogeneous multi-source remote sensing image data into standardized, analysis-ready data (ARD) products that are wide-coverage, timely, seamless, and cloud-free. These products are the cornerstone of all downstream remote sensing applications. However, their automated production faces severe technical challenges. Currently, massive amounts of image data generated by different satellite systems are received daily, producing initial ARD data products with standard geometric and radiometric corrections and pixel-level cloud masking information. However, these data are spatiotemporally discrete, discontinuous, and full of data gaps, such as clouds and cloud shadows. Therefore, how to automatically and intelligently "extract" effective information from this continuously growing, multi-source heterogeneous ARD data archive and fuse it into a homogeneous image product with wide coverage, timely information, seamlessness, and cloudlessness has become a key measure of the core service capability of spatial information. Currently, relying on manual or semi-automatic software operations is not only inefficient when processing petabyte-scale dynamic data sources, but also makes it difficult to standardize the processes of selection, splicing, and color balancing, resulting in inconsistent product quality and untraceable origins. Furthermore, traditional script-based batch processing methods, due to their linear and static logic, cannot effectively manage the state and intelligently schedule a "live" data archive, making it difficult to achieve the core requirement of "dynamically replacing old data with the latest and best data."

[0081] To overcome the aforementioned bottlenecks, this disclosure proposes a novel technological paradigm aimed at solving the specific technical challenge of transforming massive ARD data archives into dynamic, cloudless synthetic products.

[0082] like Figure 2 As shown, this disclosure provides an automated production system that deeply integrates spatialized state management, event-driven data scheduling, and intelligent decision-making based on a multi-dimensional quality model, namely, an automated production system based on a dynamic cloudless image synthesis method using multi-source remote sensing data.

[0083] Specifically, in June 2025, based on this disclosure, within one month, application ready data (ARD) from multiple satellite sources, including ZY-3 and GF-1, will be continuously integrated and intelligently combined with historical archives to dynamically produce a June “map”.

[0084] At the start of the task, the system first performs automated initialization of the production environment. Based on the 1:1,000,000 standard map sheet (million-fold sheet) and the 1:50,000 standard map sheet (fifty-thousand-fold sheet), the land area is instantiated as a distributed management network consisting of tens of thousands of independent processing units. That is, vector data is initialized into workspace processing using the million-fold and fifty-thousand-fold standard map sheets, instantiated into sheet units, and incorporated into global progress management. At the same time, a macro-level production monitoring interface is activated to track the work progress of each processing unit in real time and visually, with its initial state being "pending coverage". At the micro-level, the system establishes a dedicated status profile for each processing unit. At its core is a geometric status vector file that accurately describes the geometry of the "uncovered" area, and a list of data sources used to construct the data source genealogy.

[0085] Since June 1st, the system has entered a continuous, event-driven automated operation cycle. Daily, the system's automated data ingestion and scheduling pipeline continuously monitors the data center. When a new batch of ARD products arrives, the system triggers an incremental identification and quality verification protocol, injecting the metadata of qualified data into a structured "data queue to be processed." This involves comparing the newly added multi-source remote sensing data with the data registry, conducting a complete quality review, incorporating it into the processing task, and updating the data registry based on the processing task. Subsequently, the system's parallel computing cluster inputs this valid data and accurately allocates it to the spatially related processing units (i.e., map units) within the range that have not yet reached the coverage target. Monitoring personnel can intuitively see the continuous and stable increase of the cloud-free coverage index of each processing unit through the macro-production monitoring interface.

[0086] The system's core innovative capabilities include parallel cloudless coverage calculation based on sectional units and in conjunction with pending tasks, continuously updating coverage. If the cloudless coverage is less than a threshold, it returns to the pending task for continuous updating; if the cloudless coverage is greater than the threshold, it performs seam optimization, hole filling, and cloudless composite processing to achieve a "completed" sectional status, ultimately integrating into global progress management. These capabilities have been fully validated when handling complex areas. The cloudless coverage calculation includes candidate image analysis, pixel-by-pixel optimal decision-making, and unit state and result updates. Unit state and result updates further include coverage state vectors, data source lists, and reconstructed virtual grids.

[0087] Taking a persistently data-sparse area in a basin, caused by years of cloud cover, as an example, as of June 22nd, approximately 35% of the area in this processing unit was still obscured by historical cloud cover. On that day, a ZY-3 satellite image with extremely low cloud cover arrived. Instead of performing a simple time-series replacement, the system activated its core pixel-by-pixel optimal decision engine.

[0088] For each pixel within the "uncovered" area, the system constructed a decision matrix to conduct a refined quantitative evaluation of multiple candidate pixels from this new ZY-3 image and historical archived images. The overall quality score of each candidate pixel was calculated in real time by a hierarchical weighted scoring model. The top layer of this model is a large contamination penalty, which gives absolute veto power to any pixel located within clouds, cloud shadows, or their edge buffer zones. After passing the top-level screening, a significant macroscopic reward is given to pixels from source images that have a higher resolution or are judged as "completely clean." Finally, a base quality score, composed of multiple normalized indicators weighted together, is used to rank the remaining candidates, considering dimensions including: sharpness (based on Laplacian variance), atmospheric transparency (based on dark channel analysis), and cloud proximity (based on normalized Euclidean distance).

[0089] In this calculation, the pixels of the ZY-3 image from June 22nd won due to its extremely high "currentity" weight and excellent sharpness score. Immediately afterwards, the geometric state vector of this processing unit was updated with a precise geometric difference operation, and simultaneously, the phylogenetic information of ZY-3 was appended to the data source list. After this update, the virtual raster (VRT) of this processing unit was reconstructed in real time. This mechanism allows analysts monitoring this area to immediately perform real-time quality verification by accessing this lightweight VRT file, and preview in real time the latest appearance of this sparse data area seamlessly filled with high-quality new data.

[0090] By June 28, with the vast majority of processing units on the macro-production monitoring interface switching to the "met standard" status, the system's final product delivery subsystem was automatically and in batches triggered. The system treats VRT as a "synthetic recipe" to be executed, and the minimum cost path optimization algorithm based on ground object perception is invoked to analyze the adjacency relationships of different source images in the recipe, and generate an "invisible" seam line with the lowest visual cost along natural textures such as ridges and rivers.

[0091] Ultimately, the optimized deliverable is a complete, industry-grade, standardized product suite: a core product in cloud-optimized GeoTIFF (COG) format with efficient cloud-based chunked reading capabilities; a quality control product suite providing a complete data link for scientific verification and algorithm auditing, including a source ID traceability layer, a decision logic layer, and a JSON manifest; and a visually enhanced quick view with JGW georeferenced files. Through this new paradigm of "process-driven, dynamic fusion," this disclosure successfully constructed a highly reliable, continuously updated, and pixel-level traceable "single-map" image base within the agreed timeframe.

[0092] It is evident that the technical solution disclosed in this invention aims to transform current remote sensing image products from static, one-time delivered "data files" into dynamic, reliable, and continuously updated "information services".

[0093] More importantly, compared to current technologies, this disclosure provides a dynamic, stateful, and automated production system capable of managing continuously updated dynamic data streams. The specific core differences compared to current technologies are as follows: 1. Fundamental difference in architectural paradigm: from "static, one-time operation" to "dynamic, continuous service"; this disclosure constructs a continuously running, state-memory-enabled automated production "service"; its core is event-driven and incremental updates, the system can continuously monitor and automatically ingest new ARD data, and dynamically and optimally integrate it into the existing image base; this disclosure does not "create" an image once, but rather "maintains" a continuously "growing" and self-improving live data product over the long term; 2. Different levels of sophistication in decision-making models: from "simple rule prioritization" to "multi-dimensional quantitative optimal decision-making"; this disclosure adopts a pixel-by-pixel, multi-dimensional quantitative scoring optimal decision-making model; the system calculates the comprehensive quality score of each candidate pixel based on multiple indicators such as time phase, resolution, sharpness, atmospheric transparency, and cloud proximity through a hierarchical weighted mathematical model, thereby ensuring that each pixel ultimately selected is the objective best choice among all available data; this is a decision-making mechanism that is far more refined, scientific, and consistent than simple rules; 3. Significant differences in process management and traceability capabilities: From "stateless black box" to "fully transparent and controllable"; In the process management stage, the original two-level spatialized state management system can perform macro-monitoring and precise micro-state management of tens of thousands of independent processing units, realizing parallelism, fault tolerance, and breakpoint continuation in large-scale production, which is a system engineering capability that current technology does not possess at all; In the process traceability stage, this disclosure can not only trace the source of pixels, but also trace the reason for their selection (Logic). Through a unique quality control (QC) product suite (source ID layer + decision logic layer), this disclosure completely transforms the "black box" into transparency, providing unprecedented capabilities for scientific verification and algorithm auditing of products.

[0094] 4. Leading efficiency of technical means: This disclosure uses geometric operations for state updates and combines real-time virtual grid (VRT) technology to achieve efficient memory computing with "zero intermediate files" throughout the dynamic update process, and can preview the intermediate process in real time. This is far superior to the current batch processing methods in terms of computing efficiency, system interactivity and transparency.

[0095] According to the embodiments of this disclosure, the following technical effects are achieved: This invention provides a method for synthesizing large-scale dynamic cloudless images based on multi-source remote sensing data. By acquiring multi-source remote sensing data, a global progress management spatial file is constructed, and a cell status vector file and a cell source list text file are created for each grid cell within the global progress management spatial file. New multi-source remote sensing data is then acquired, and a data list to be processed is constructed. Based on the cell status vector file, cell source list text file, and data list to be processed in the global progress management spatial file, the method dynamically calculates and updates the corresponding virtual raster file. When the cloud-free coverage of the grid cells in the global progress management spatial file reaches a preset cloud-free coverage rate, a cloud-optimized format file is generated based on the virtual raster file. This method constructs an automated production system that deeply integrates spatialized status management, intelligent data scheduling, and high-quality seamless synthesis. This method completely overturns the current "human-finded data, manual stitching" work mode, realizing a paradigm shift of "data-driven, automatic synthesis."

[0096] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this disclosure is not limited to the described order of actions, because according to this disclosure, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are all optional embodiments, and the actions and modules involved are not necessarily essential to this disclosure.

[0097] The above is an introduction to the method embodiments. The following describes the solution described in this disclosure further through device embodiments.

[0098] Figure 3 A block diagram of a dynamic cloudless image synthesis apparatus 300 based on multi-source remote sensing data according to an embodiment of the present disclosure is shown. Figure 3 As shown, the device 300 includes: Module 310 is used to acquire multi-source remote sensing data, construct a global progress management spatial file, and create a cell status vector file and a cell source list text file for each grid cell in the global progress management spatial file. Module 310 is also used to acquire new multi-source remote sensing data and build a list of data to be processed; The calculation module 320 is used to dynamically calculate and update the corresponding virtual raster file based on the element state vector file, element source list text file and data list to be processed of the grid elements to be updated in the global progress management space file. The generation module 330 is used to generate a cloud-optimized format file based on the virtual raster file when the cloud-free coverage of the grid cells in the global progress management spatial file reaches the preset cloud-free coverage.

[0099] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the described module can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0100] The acquisition, storage, and application of user personal information involved in the technical solution disclosed herein comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0101] According to embodiments of this disclosure, this disclosure also provides an electronic device, a readable storage medium, and a computer program product.

[0102] Figure 4 A block diagram of an exemplary electronic device 400 capable of implementing embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0103] Electronic device 400 includes a computing unit 401, which can perform various appropriate actions and processes according to a computer program stored in ROM 402 or a computer program loaded into RAM 403 from storage unit 408. RAM 403 can also store various programs and data required for the operation of electronic device 400. The computing unit 401, ROM 402, and RAM 403 are interconnected via bus 404. I / O interface 405 is also connected to bus 404.

[0104] Multiple components in electronic device 400 are connected to I / O interface 405, including: input unit 406, such as keyboard, mouse, etc.; output unit 407, such as various types of displays, speakers, etc.; storage unit 408, such as disk, optical disk, etc.; and communication unit 409, such as network card, modem, wireless transceiver, etc. Communication unit 409 allows electronic device 400 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0105] Computing unit 401 can be various general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of computing unit 401 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. Computing unit 401 performs the various methods and processes described above, such as method 100. For example, in some embodiments, method 100 may be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 408.

[0106] In some embodiments, part or all of the computer program may be loaded and / or installed on the electronic device 400 via ROM 402 and / or communication unit 409. When the computer program is loaded into RAM 403 and executed by computing unit 401, one or more steps of method 100 described above may be performed. Alternatively, in other embodiments, computing unit 401 may be configured to perform method 100 by any other suitable means (e.g., by means of firmware).

[0107] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0108] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0109] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0110] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including voice input, speech input, or tactile input).

[0111] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0112] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact via communication networks. Client-server relationships are created by computer programs running on the respective computers and having a client-server relationship with each other. Servers can be cloud servers, servers in distributed systems, or servers incorporating blockchain technology.

[0113] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this disclosure can be achieved, and this is not limited herein.

[0114] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A dynamic cloud-free image synthesis method based on multi-source remote sensing data, characterized in that, The method comprises the following steps: acquiring multi-source remote sensing data, constructing a global progress management spatial file, and creating a cell state vector file and a cell source list text file for each grid cell in the global progress management spatial file; acquiring new multi-source remote sensing data and constructing a to-be-processed data list; based on the cell state vector file and the cell source list text file of the to-be-updated grid cell in the global progress management spatial file and the to-be-processed data list, dynamically calculating and updating a corresponding virtual grid file; when the cloud-free coverage of the grid cell in the global progress management spatial file reaches a preset cloud-free coverage, generating a cloud-optimized format file based on the virtual grid file.

2. The method of claim 1, wherein, The method comprises the following steps: acquiring multi-source remote sensing data and generating a corresponding standardized geographic grid; based on the multi-source remote sensing data and the standardized geographic grid, constructing a global progress management spatial file; the attribute fields of each grid cell in the global progress management spatial file include a map unique code, a cloud-free coverage percentage, a processing state, a last update date, and a final product storage path; creating a cell state vector file and a cell source list text file for each grid cell; the cell state vector file initially includes a polygon representing the complete range of the grid cell and marks an attribute field as uncovered, which is used for dynamic calculation and update of the cloud-free coverage; the cell source list text file is used to record the source information of each scene original data image constituting the cloud-free product of the grid cell.

3. The method of claim 2, wherein, The method comprises the following steps: based on a preset unique identifier registry, acquiring new multi-source remote sensing data, identifying and filtering the data, generating an incremental data set, and the incremental data set includes image data files and corresponding cloud masks of each scene; performing multiple verification processing on the incremental data set; performing structured processing on the processed incremental data set to construct a to-be-processed data list.

4. The method of claim 3, wherein, The method comprises the following steps: when the to-be-updated grid cell in the global progress management spatial file and the to-be-processed data list each have a spatially overlapping candidate image unique, performing vectorization processing on the cloud mask file of the candidate image to generate an effective contribution polygon, and based on geometric difference set operation, dynamically calculating and updating the uncovered area polygon of the grid cell according to the effective contribution polygon, synchronously updating the cell source list text file of the grid cell, and synthesizing a corresponding virtual grid file.

5. The method of claim 3, wherein, The method comprises the following steps: When the candidate images of the grid cell to be updated in the global progress management spatial file and each scene of the to-be-processed data list have spatial overlap, the candidate images are analyzed based on preset preferred rules to determine an optimal candidate image, a cloud mask file of the optimal candidate image is vectorized to generate an effective contribution polygon, and based on geometric difference set operation, the effective contribution polygon is used to dynamically calculate and update the uncovered area polygon of the grid cell, the cell source list text file of the grid cell is updated synchronously, and a corresponding virtual grid file is synthesized. The preset preferred rules include: selecting an image with the latest time phase and / or the lowest cloud cover from the candidate images as the optimal candidate image.

6. The method of claim 3, wherein, The dynamic calculation and update and the synthesis of the corresponding virtual grid file based on the cell state vector file and the cell source list text file of the grid cell to be updated in the global progress management spatial file and the to-be-processed data list further include: When the candidate images of the grid cell to be updated in the global progress management spatial file and each scene of the to-be-processed data list have spatial overlap, a preset pixel-by-pixel optimal decision model is used to generate an effective contribution polygon by vectorizing a cloud mask file of each candidate image in an iterative manner according to the priority order of the quality evaluation values of the candidate images, and based on geometric difference set operation, the effective contribution polygon generated by the candidate image with the highest quality score is used to dynamically calculate and update the uncovered area polygon of the grid cell, the cell source list text file of the grid cell is updated synchronously, and a corresponding virtual grid file is synthesized. The preset pixel-by-pixel optimal decision model includes: wherein, denotes a multi-dimensional pixel quality evaluation value, denotes a weight coefficient of a basic quality weighted score, denotes a basic quality weighted score, denotes a pixel purity bonus term, denotes a macro bonus term, denotes a pollution penalty term, denotes a sharpness score, denotes a weight coefficient of a sharpness score, denotes a transparency score, denotes a weight coefficient of a transparency score, denotes a cloud proximity score, denotes a weight coefficient of a cloud proximity score, N(·) denotes a normalization function.

7. The method according to any one of claims 1 to 6, characterized in that, When the cloud-free coverage rate of the grid cell in the global progress management spatial file reaches the preset cloud-free coverage rate, the virtual grid file is used to generate a cloud-optimized format file. When the cloud-free coverage rate of the grid cell in the global progress management spatial file reaches the preset cloud-free coverage rate, a cloud-optimized format file is generated based on a preset optimization algorithm according to the virtual grid file; the preset optimization algorithm includes a seam line automatic optimization algorithm based on feature perception and / or an image filling algorithm based on a convolution kernel. 8.A dynamic cloud-free image synthesis device based on multi-source remote sensing data, characterized in that, The method includes: The construction module is configured to acquire multi-source remote sensing data, construct a global progress management spatial file, and create a cell state vector file and a cell source list text file for each grid cell in the global progress management spatial file; The construction module is further configured to acquire new multi-source remote sensing data and construct a to-be-processed data list; The calculation module is configured to dynamically calculate and update and synthesize a corresponding virtual grid file based on the cell state vector file and the cell source list text file of the grid cell to be updated in the global progress management spatial file and the to-be-processed data list. The generation module is configured to generate a cloud-optimized format file based on the virtual grid file when the cloud-free coverage rate of the grid cell in the global progress management spatial file reaches a preset cloud-free coverage rate.

9. An electronic device, comprising: The method includes: at least one processor; and a memory connected in communication with the at least one processor; The memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-7.

10. A non-transitory computer-readable storage medium having stored thereon computer instructions, wherein, The computer instructions are for causing the computer to perform the method of any one of claims 1-7.

Images