A scheduling and distribution management system for remote sensing image data

Abstract

This invention relates to the field of remote sensing image data processing technology, and discloses a remote sensing image data scheduling and distribution management system, comprising: a central management platform and an analysis agent deployed at the remote sensing image data source. The platform is responsible for distributing structured analysis primitives as task instructions, and the analysis agent performs analysis locally at the data source according to the instructions, generates analysis result data containing only analysis conclusions, and sends it back. This invention avoids the overall transmission of massive amounts of raw image data by precisely delivering analysis tasks to the data source and executing them locally. This transforms the bottleneck of physical bandwidth in traditional systems into logical-level analysis definitions, changing the working mode of the entire system from data transport to distributed information extraction, and improving the immediacy of data acquisition and the effectiveness of resource utilization.

Description

Technical Field

[0001] This invention relates to a scheduling and distribution management system for remote sensing image data, belonging to the field of remote sensing image data processing technology. Background Technology

[0002] Currently, the acquisition and application process of remote sensing image data generally adopts a request-schedule-download-analysis approach. This approach treats remote sensing satellites or ground data centers as networked data warehouses, and its core task is to optimize the transmission link and efficiency of massive raw image data files to the user end. However, as remote sensing technology evolves towards wide-area, continuous, and high-frequency dynamic monitoring applications, the inherent limitations of this approach, which is based on transmitting the data itself, are becoming increasingly prominent. Especially when dealing with large-scale, routine monitoring tasks, it has formed a common bottleneck that restricts the development of the industry.

[0003] When this approach is applied to scenarios requiring continuous data analysis, such as nationwide agricultural growth monitoring or large-scale land subsidence tracking, it has inherent design flaws. For example, since users are usually only concerned with local information in the image that carries specific changes or states, rather than all pixels of the entire image, the system inevitably has to schedule and transmit a large amount of redundant data that has no direct value to the analysis target in each task. This not only causes the satellite-to-ground and ground-to-ground communication bandwidth resources to be occupied in a long-term and inefficient manner, but also creates huge data storage, management and computing pressure on the user end, which in turn lengthens the response cycle of the entire information acquisition chain and keeps the application cost high.

[0004] Specifically, existing technologies suffer from the following shortcomings: 1. To acquire localized effective information, the system must bear the enormous bandwidth and storage overhead of transmitting the entire image; 2. The method of aggregating all raw data to a ground center for processing makes ground computing facilities the performance bottleneck of the entire business process, making it difficult to flexibly handle large-scale concurrent analysis requests; 3. The lengthy data download process from task submission to obtaining the final analysis results leads to significant time delays, weakening the timeliness value of remote sensing information in many scenarios. Therefore, how to construct a completely new data processing and distribution process, avoiding redundant data transmission based on raw image files, and shifting the focus of data scheduling and distribution from the data itself to effective information, thereby systematically improving the response speed and economic efficiency of remote sensing data applications without relying on unlimited bandwidth growth, becomes the technical problem to be solved by this invention. Summary of the Invention

[0005] This invention provides a scheduling and distribution management system for remote sensing image data. Its main purpose is to solve the problems of redundant transmission of massive amounts of data, response delays, and high application costs caused by the use of download-and-analyze method in the prior art.

[0006] To achieve the above objectives, the present invention provides a remote sensing image data scheduling and distribution management system, including a central management platform and at least one analysis agent deployed at the remote sensing image data source.

[0007] The central management platform is configured to: receive structured analysis primitives that include the analysis task type and scope as determined by the user; distribute the analysis primitives as task instructions to the analysis agents covering the scope; and receive analysis result data containing only the analysis conclusions returned by the analysis agents.

[0008] The analysis agent is configured to: receive task instructions and directly acquire raw remote sensing image data locally at the remote sensing image data source to execute the analysis task defined by the analysis primitives; generate analysis result data containing only the analysis conclusions based on the execution results of the analysis task; the analysis agent is also configured to perform an executability pre-check step on the raw remote sensing image data before executing the analysis task. The executability pre-check step includes: determining whether there are preset unexecutable conditions based on the pixel value distribution or brightness saturation of the raw remote sensing image data; and if unexecutable conditions are determined to exist, suspending the analysis task and generating a negative knowledge package containing the type of unexecutable condition as the reason for failure and sending it back to the central management platform.

[0009] The central management platform is further configured to: update a global avoidance data model based on the received negative knowledge package, and apply the global avoidance data model to filter or delay the new analysis primitives when scheduling them.

[0010] Preferably, the analysis agent is further configured to: pre-set at least one virtual standard corresponding to the ground permanent calibration field, consisting of standard spectral feature vectors; when the operation trajectory of the analysis agent overlaps with the location of the ground permanent calibration field, autonomously trigger a self-verification task, compare the spectral features of the acquired real-time calibration image with the virtual standard locally at the remote sensing image data source end to generate a credibility token containing validity status and generation timestamp; and when transmitting the analysis result data back, attach the credibility token to the analysis result data and transmit it back together.

[0011] Preferably, the analysis agent is further configured to: when there are multiple analysis primitives for the same geographic area in the task queue of the analysis agent, compare the predefined computational processes of the multiple analysis primitives to identify shared computational steps; construct a temporary fusion execution process based on the comparison results, merge the shared computational steps locally at the remote sensing image data source end and execute them only once to generate intermediate data; and continue to execute the non-shared computational steps in each analysis primitive based on the intermediate data.

[0012] Preferably, the analysis agent is further configured to: pre-load an onboard context map consisting of geographic partitions and corresponding parameter compensation models; before executing the analysis task, the analysis agent queries the onboard context map to obtain the corresponding parameter compensation model based on its current covered geographic location, and uses the model to adjust or replace the values ​​of one or more control parameters in the received analysis primitives.

[0013] Preferably, the analysis agent is further configured to: generate intermediate data describing the continuous changes in pixel values ​​during the execution of the analysis task; calculate an information entropy index characterizing the scene stability of the original remote sensing image data based on the pixel value distribution of the intermediate data before generating the analysis result data; and append the information entropy index to the analysis result data and transmit it back together when transmitting the analysis result data.

[0014] Preferably, the information entropy index Calculated using the following formula: in, This represents the total number of different pixel values ​​in the intermediate data. For the first pixel value, pixel value The probability of it appearing in the intermediate data.

[0015] Preferably, the analysis agent is further configured to: after executing the analysis task, capture one or more process metadata during the execution of the analysis task, the process metadata including execution time and peak memory usage; calculate a complexity index locally at the remote sensing image data source end based on the process metadata and a preset benchmark value to characterize the scene complexity of the original remote sensing image data; and append the complexity index to the analysis result data and send it back together when sending back the analysis result data.

[0016] Preferably, the analysis primitives are selected from a standard library containing basic remote sensing analysis models.

[0017] Preferably, the analysis results data are in the format of a vector file, a sparse matrix file, or a text statistical file.

[0018] Preferably, the analysis agent is a software module deployed on the payload processor of the remote sensing satellite or on a ground data center server, and the analysis agent is also configured to mark the raw remote sensing image data used to perform the analysis task as deletable or degraded storage after generating the analysis result data.

[0019] Compared with the prior art, the beneficial effects of the present invention are:

[0020] 1. By using structured analysis primitives that include the analysis task type and the area of ​​effect as the scheduling core, the direct request and transmission of raw image data files in the traditional method is replaced. As a result, the workflow of the system has been fundamentally changed. The analysis agent at the data source end executes the analysis task locally and only sends back the analysis result data containing the analysis conclusion. This method avoids the back-and-forth transmission of massive amounts of raw remote sensing image data between the central management platform and the data source end. This transforms the bandwidth bottleneck and data storage pressure that restrict the performance of traditional systems into a logical problem of defining and scheduling the analysis task itself. As a result, the operating efficiency of the entire system is no longer limited by the physical volume of the data body.

[0021] 2. By analyzing the collaboration of multiple mechanisms within the agent, the final analysis results are endowed with multi-dimensional intrinsic quality assurance. Specifically, before executing the analysis task, the agent dynamically compensates the control parameters of the analysis primitives locally through an onboard context map, ensuring the suitability of the analysis logic to the specific geographical environment. At the same time, by comparing with a pre-set virtual standard, a credibility token is generated in the form of a self-verification task, providing direct proof of the health status of the original data source on which the analysis is based. This method of combining pre-analysis parameter adaptation with data source quality self-verification ensures that every set of analysis results received by the end user naturally contains a dual confirmation of the rationality of its generation process and the reliability of the data source, avoiding decision-making errors caused by parameter mismatch or sensor performance drift.

[0022] 3. Through intelligent perception and scheduling of the task processing process, the system demonstrates resource utilization flexibility and overall operational intelligence under complex working conditions. When faced with multiple concurrent analysis primitives targeting the same geographical area, the analysis agent can identify and merge shared calculation steps by comparing the calculation process, combining multiple repetitive calculation processes into one, significantly reducing the system load under sudden, high-concurrency requests. In addition, by performing executability pre-checks before executing analysis tasks, the system can identify unexecutable conditions such as cloud cover and abort invalid tasks. At the same time, the cause of failure is sent back as a negative knowledge package. The central management platform updates the global avoidance data model based on this information, thereby proactively filtering or delaying requests to known invalid areas in subsequent task scheduling. This feedback loop, which transforms task failures into global scheduling experience, enables the system to learn from execution errors and optimize itself.

[0023] 4. By introducing accompanying information mining into the analysis agent, this system also adds rich context describing the inherent characteristics of the analysis results data. For example, after executing the analysis task, by capturing process metadata such as execution time and memory usage, a complexity index characterizing the scene's complexity can be calculated. Or, before generating the final binary analysis conclusion, by calculating the information entropy index of intermediate data describing continuous changes between pixels, the overall stability of the scene can be quantitatively assessed. These indices are added as metadata to the analysis results data, allowing users to perceive whether the scene corresponding to the conclusion is simple or complex, stable or subject to general weak disturbances, while obtaining the analysis conclusion. This provides direct objective evidence from the data source for subsequent in-depth interpretation and decision-making classification. Attached Figure Description

[0024] Figure 1 This is a schematic diagram of the overall architecture of a remote sensing image data scheduling and distribution management system according to the present invention;

[0025] Figure 2 This is a schematic diagram illustrating the task fusion optimization effect of the agent in this invention.

[0026] Figure 3 This is a flowchart illustrating the system operation of the present invention based on negative knowledge package feedback and complexity index calculation.

[0027] The objectives, features, and advantages of this invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0028] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments; it should be understood that the specific embodiments described herein are merely illustrative of the present invention and are not intended to limit the scope of protection of the present invention.

[0029] This invention provides a remote sensing image data scheduling and distribution management system. Its overall architecture consists of a central management platform as the task scheduling and decision-making hub and at least one analysis agent deployed at the remote sensing image data source. The core working mode of the system is that the central management platform abstracts and encapsulates user requirements into standardized structured analysis primitives for distribution. The analysis agent performs analysis locally at the data source according to the instructions and only sends back the refined analysis results containing the analysis conclusions to the central management platform, thereby realizing the transformation from the traditional data transportation mode to a distributed information extraction mode.

[0030] In a typical application aimed at large-scale dynamic monitoring of vegetation growth, the system's operation process unfolds as follows: First, the user delineates a geographically defined area of ​​action on a digital map through the user interface of the central management platform. Then, from a pre-defined standard library containing basic remote sensing analysis models such as vegetation index calculation, water body extraction, and change detection, the user selects the required analysis task type, such as normalized vegetation index calculation. Based on this, the central management platform generates a structured analysis primitive, a standardized data structure, such as an XML or JSON object containing a unique task identifier, the name of the selected analysis model, and the geographical coordinates of the area of ​​action, expressed in GeoJSON format. Subsequently, the platform queries its internally maintained dynamic topology map, which records the current service range of each analysis agent, and distributes this analysis primitive as a task instruction to the analysis agent currently covering the specified area of ​​action.

[0031] The analysis agent, a software module deployed on the payload processor of a remote sensing satellite or a ground data center server, does not immediately perform calculations upon receiving task instructions. Instead, it prioritizes initiating an executability pre-check step. Given the objective challenge of frequent image quality degradation due to atmospheric phenomena such as clouds, fog, and haze in remote sensing operations, thus hindering effective analysis, this pre-check step is configured as a deterministic algorithm: The analysis agent first acquires a low-resolution quick view of the corresponding original remote sensing image data and calculates the pixel value distribution or brightness saturation of its brightness channel. If the statistical results show a brightness value higher than a preset threshold (e.g., at 8-bit quantization, the proportion of pixels with a brightness value higher than 220 exceeds a preset area ratio), the analysis agent will proceed. For example, if the failure rate is 85%, the system determines that there is a pre-defined unexecutable condition of large-area cloud coverage. Accordingly, the analysis agent will suspend the subsequent execution of the analysis task and generate a negative knowledge package containing the unexecutable condition type as the reason for failure, which will be sent back to the central management platform. After receiving such a negative knowledge package, the central management platform will update a global avoidance data model based on its content. This model is essentially a geospatial index with a timestamp and a no-scheduling mark, which is used to automatically filter or delay new analysis primitive requests sent to the same geographic area within a certain time window, such as 24 hours. In this way, an individual execution failure is transformed into a global experience that improves the subsequent scheduling intelligence of the entire system.

[0032] If the executability pre-check passes successfully, the analysis agent will perform a localized parameter compensation step before executing the core analysis task to ensure a high degree of adaptation between the analysis parameters and the specific geographic environment. Considering that different land surface types, atmospheric conditions, and solar altitude angles can all affect the control parameters in the remote sensing analysis model, the analysis agent has a pre-built onboard context map consisting of geographic partitions and corresponding parameter compensation models. This map is a key-value database in terms of data structure, where the key is a normalized geographic grid code, and the value is a set of parameter compensation models optimized for that region. The model can be a set of constants that can be directly replaced, or it can be a function that can be dynamically calculated based on real-time auxiliary data such as solar altitude angle. Before executing the analysis task, the analysis agent queries the onboard context map based on its current covered geographic location to obtain the corresponding parameter compensation model, and uses this model to adjust or replace one or more control parameters in the received analysis primitives, such as atmospheric correction coefficients or soil adjustment factors, thereby ensuring the suitability of the analysis logic to the specific geographic context.

[0033] Furthermore, in scenarios involving high concurrency requests, such as when multiple users simultaneously issue different analysis requests for a specific hotspot, the analysis agent is configured to perform a one-step task fusion optimization to avoid redundant consumption of computing resources. When multiple analysis primitives targeting the same geographic area exist in its task queue, the analysis agent will parse and compare the predefined computational processes of these analysis primitives to identify shared computational steps. For example, a vegetation cover analysis and a surface water analysis may share the exact same preprocessing steps such as atmospheric correction and radiometric calibration. Based on the comparison results, the analysis agent will construct a temporary fusion execution process, merging these shared computational steps locally at the remote sensing image data source and executing them only once to generate a standard preprocessed intermediate data. Then, based on this intermediate data, the non-shared specific computational steps in each analysis primitive will continue to be executed. In this way, multiple repetitive computational processes are merged into one, significantly reducing the peak computational load of the system.

[0034] During and after the execution of the analysis task, to ensure that the final analysis results data has richer quality and scene description dimensions, the analysis agent is also configured to perform a series of accompanying information mining and generation. To objectively evaluate the quality of the original data itself, the analysis agent is configured to autonomously trigger a self-verification task when the running trajectory overlaps with the location of the ground permanent calibration field. It has at least one virtual standard corresponding to the ground permanent calibration field, which is composed of standard spectral feature vectors. When the self-verification task is triggered, the analysis agent compares the spectral features of the acquired real-time calibration image with the virtual standard locally. If the spectral angular distance between the two is less than a preset threshold, a credibility token containing the validity status and generation timestamp is generated. This token will be attached when the analysis results data is returned, providing data users with direct proof of the health status of the data source.

[0035] Furthermore, to quantify the intrinsic characteristics of the image scene, the analysis agent is configured to generate intermediate data, such as a gradient map, describing the continuous changes in pixel values ​​before generating the final analysis conclusions. Based on the pixel value distribution of this intermediate data, an information entropy index characterizing the scene stability of the original remote sensing image data is calculated. The index Through formula Calculations are performed, in which, This represents the total number of different pixel values ​​in the intermediate data. For the first pixel value, pixel value The probability of occurrence in intermediate data is such that a higher information entropy index indicates rich details and widespread changes within the scene, while a lower one corresponds to a flat and homogeneous scene. At the same time, the analysis agent also captures process metadata when it executes analysis tasks, specifically including execution time and peak memory usage. Based on the process metadata, it compares it with a set of benchmark values ​​for standard scenes and calculates a complexity index locally at the remote sensing image data source end to characterize the scene complexity of the original remote sensing image data. These calculated information entropy indices and complexity indices will be added as metadata to the analysis results data and sent back together.

[0036] Finally, based on the execution results of the analysis task, the analysis agent generates analysis result data containing only the analysis conclusions. The format is limited to one or more lightweight formats such as vector files, sparse matrix files, or text statistical files to minimize the communication bandwidth required for transmission. After generating and confirming that the analysis result data has been successfully transmitted back to the central management platform, the analysis agent will also perform a storage optimization operation. That is, on the local remote sensing image data source, the original remote sensing image data used to perform this analysis task is marked as deletable or downgraded for storage, thereby dynamically releasing valuable storage resources in the on-board or ground data center.

[0037] Example 1: In an emergency response scenario for a large-scale sudden geological disaster, such as regional ground subsidence caused by a strong earthquake, the emergency command center needs to quickly acquire millimeter-level deformation information of a large affected area within a short period after the earthquake to delineate the core subsidence zone, assess the risk of secondary disasters, and guide the deployment of rescue forces. In the traditional working mode, even if remote sensing satellites acquire raw image data covering the disaster area, this data still needs to wait in line for downlink transmission before being transmitted to the analysis center via a bandwidth-limited terrestrial network. The latency of this process is often calculated in days, reducing the timeliness value of the analysis results for emergency response. In this scenario, the technical solution of this invention operates as follows: Users in the emergency command center define their analysis needs through the central management platform as a... A structured analysis primitive targeting the entire disaster area and the mission type of interferometric radar surface subsidence analysis was released. The central management platform then distributed mission instructions to all remote sensing satellites with analysis agents deployed before flying over the disaster area. After a remote sensing satellite carrying an analysis agent entered the airspace over the disaster area and acquired the first raw remote sensing image data, its analysis agent immediately started the mission in the onboard payload processor. It first performed an executability pre-check step on the image data. When it found that some sub-regions could not be effectively interferometrically processed because the radar signal coherence was lower than the preset threshold due to post-earthquake dust or cloud cover, the analysis agent stopped the calculation for these sub-regions and generated a negative knowledge package containing the geographic coordinates of the area and the reason for failure as low coherence. This package was then transmitted back to the central management platform via a narrowband link.

[0038] Upon receiving the negative knowledge packet, the central management platform updates its global avoidance data model in real time, guiding subsequent scheduling decisions. This avoids redistributing the same tasks for these verified invalid areas to other satellites. Furthermore, it automatically generates observation tasks for different sensor types in these areas, such as optical imaging, to be executed when weather conditions permit, thus forming a dynamic, cross-platform collaborative observation loop. Meanwhile, when the analysis agent of another satellite is performing the same task in another airspace, it automatically adjusts the control parameters of the phase unwrapping algorithm in the interferometric processing based on the parameter compensation model for the local mountainous terrain stored in the onboard scenario map, improving the solution stability under complex terrain.

[0039] Within hours, the emergency command center began receiving analysis results from various analysis agents. These were not raw image files, but lightweight vector files containing only information such as the boundaries of the subsidence area, the subsidence rate, and the location of the maximum subsidence point. Furthermore, each analysis result included a complexity index and an information entropy index generated by the analysis agent. Command center technicians noticed that while the average subsidence of some areas was not high, their complexity indices were significantly higher than the baseline. This directly indicated that the subsidence pattern in these areas was not uniform, but rather contained numerous small-scale and irregular fractures or distortions. This description of the inherent characteristics of the scene directly provided by the data source allowed the command center to raise the risk assessment level for these areas and prioritize dispatching geological expert teams for verification. The system's focus thus shifted from overcoming physical bandwidth limitations to transmitting massive amounts of data to efficiently transferring structured information about what happened and where between distributed nodes. By submitting analysis tasks to the data source and executing them locally, the architecture fundamentally changed the information acquisition process, resolving the technical contradictions previously constrained by physical transmission bottlenecks at a new logical level.

[0040] Example 2: To objectively verify the effectiveness of the technical solution of this invention in reducing data transmission load and improving the immediacy of information acquisition, this comparative experiment was designed. Its purpose is to simulate a network environment with transmission bottlenecks that conforms to typical remote sensing application conditions, and to quantitatively compare the key performance indicators of the method provided by this invention and the traditional download-then-analyze method when completing the same analysis task. The experimental platform consists of two servers and a network simulator. One server deploys the central management platform software, and the other server serves as the remote sensing image data source and deploys the analysis agent software module. The two servers are connected through the network simulator, and the communication bandwidth of this link is set to 100Mbps. This value is used... The experiment used typical transmission conditions representing a medium-load commercial remote sensing satellite X-band downlink or a long-distance terrestrial private network. A control group and an experimental group were set up. The control group used the traditional method, where the central management platform initiated a file transfer request to the data source to download the complete raw remote sensing image data locally before performing analysis. The experimental group used the method of this invention, where the central management platform distributed structured analysis primitives to the analysis agent at the data source, completed the analysis locally, and only returned the analysis results. The experiment performed a series of preset typical analysis tasks on a standard dataset covering multispectral, hyperspectral, and synthetic aperture radar images, and recorded the key performance parameters of the control and experimental groups respectively.

[0041] After the experiment started, the data transmission volume and time in the two environments showed systematic differences. In the experimental group, since the analysis and calculation were completed locally at the data source, the amount of data transmitted over the network was only the lightweight analysis result data, ranging from tens of KB to several MB in size. In the control group, the amount of data transmitted over the network was the original remote sensing image data itself, ranging from several GB to hundreds of GB in size. This difference in order of magnitude was directly reflected in the transmission time. Under the set bandwidth of 100Mbps, the transmission time of the analysis result data in the experimental group was within seconds, while the transmission time of the original remote sensing image data in the control group varied from several minutes to several hours depending on the file size. Table 1 shows the performance data comparison of some representative tasks.

[0042] Table 1: Performance data comparison of the two methods under different tasks.

[0043]

[0044] Referring to Table 1, the experimental data shows that the total time consumed under the method of the present invention is mainly composed of the local computing time at the data source end, while the total time consumed under the traditional method is dominated by the data transmission time. Taking the key area change detection task as an example, the total time consumed under the method of the present invention is about 2.1 seconds, while the traditional method requires about 12.7 minutes. There is a significant difference between the two. This difference stems from the core working mechanism of the present invention, which is to use the analysis task itself rather than the massive data body as the scheduling and transmission object, complete the calculation at the data source end and only return highly condensed conclusions, thereby shifting the bottleneck restricting the efficiency of the process from network physical bandwidth to local computing power.

[0045] Example 3: This example combines Figures 1 to 3 This paper describes the implementation of a remote sensing image data scheduling and distribution management system. Figure 1 As shown in the figure, this invention illustrates the overall architecture of a remote sensing image data scheduling and distribution management system. The architecture consists of two main logical parts: a central management platform and a remote sensing image data source. The core workflow begins with a user or external system defining a structured analysis primitive containing task type and area of ​​effect through the user interface of the central management platform. Upon receiving this primitive, the central management platform distributes it as a task instruction to analysis agents covering the specified area. The analysis agents deployed at the remote sensing image data source, upon receiving the instruction, first perform a crucial executability pre-check on the original remote sensing image data. If the check passes, the analysis task is executed locally at the data source, generating lightweight analysis result data containing only the analysis conclusions. This analysis result data is ultimately transmitted back to the central management platform. Simultaneously, specific information generated by the analysis agent during execution can serve as a basis for strategy optimization, feeding back to the key module of the central management platform—the global avoidance data model—to guide subsequent task distribution decisions.

[0046] like Figure 2 As shown in the figure, the horizontal axis represents the number of concurrent tasks, and the vertical axis represents the relative computation time required to complete the task. The figure contains three curves, representing three scenarios: unoptimized (independent execution), task fusion optimization, and theoretically optimal (complete sharing). As can be seen from the figure, as the number of concurrent tasks increases from 1 to 8, the relative computation time increases exponentially in the unoptimized scenario, while the growth trend of the relative computation time slows down significantly after task fusion optimization, approaching the theoretical optimal value.

[0047] like Figure 3As shown, the input structured analysis primitives are processed by the task instruction distribution module. The module's decision-making is influenced by a global avoidance data model that dynamically adjusts strategies based on negative analysis knowledge packages. The generated task instructions are sent to the analysis agent deployed at the remote sensing image data source. After performing an executability pre-check on the raw image data, the analysis agent performs local analysis execution at the source according to the instructions. During the analysis process, the system captures metadata of the analysis process and uses it to quantify the analysis complexity to calculate the complexity index. Ultimately, the generated analysis results not only contain the core analytical conclusions, but also include the complexity index. The data is then sent back to the central management platform or user terminal. If an unexecutable condition is identified during the pre-inspection or execution phase, a negative knowledge package will be generated and fed back to the global avoidance data model, thereby enabling dynamic optimization and self-learning of the system scheduling strategy.

[0048] Example 4: Before deploying the system of the present invention on a large scale and across regions, in order to ensure the adaptability of the analysis task and the consistency of the results in different geographical environments, an offline modeling and calibration process covering the entire target deployment area needs to be performed. This process aims to generate an onboard context map for the analysis agent to use when performing tasks, and to establish a benchmark value for the subsequent online computation complexity index. In an application scenario covering a specific continental plate, the onboard context map is first constructed. The construction process takes a geographic information database containing historical remote sensing image data and corresponding spatiotemporal locations as input. The data time span of this database is five years.

[0049] The first step in the process is to divide the target deployment area into multiple geographic partitions with latitude and longitude of 0.5 degrees multiplied by 0.5 degrees, using a uniform geographic grid, and assign a unique ID to each partition. Then, for each geographic partition, all historical remote sensing image data falling within that partition's range are retrieved from the database and categorized by season based on image metadata. For each season's image set, the pixel value distribution across one or more specific spectral bands is statistically analyzed. Combined with publicly available geographic environmental data for that partition, a pre-defined regression analysis model is used to calculate a parameter compensation model for adjusting the control parameters of the analysis primitives. This parameter compensation model is defined as a set of polynomial coefficients used to adjust a specific control parameter in the atmospheric correction stage, and its calculation follows... ,in, These are the adjusted parameter values. It is the solar altitude angle acquired in real time during task execution, while the coefficient... , , The values ​​are obtained by performing regression analysis on the historical data of the region for that season and are then stored in a fixed manner. Finally, the IDs of all geographic regions, along with the parameter compensation models corresponding to each season, are stored together as a key-value pair database file, i.e., an onboard context map, and are pre-installed in the memory of each analysis agent.

[0050] The second step of the process involves benchmarking the complexity index in a standardized hardware environment with the same computational performance as the analysis agent to be deployed. The benchmark uses a set of standard scene raw remote sensing imagery data with uniform features and a single scene type. This set of data consists of images of large desert areas, open still water surfaces, and cloudless, solid-colored ice caps. The analysis agent performs a fixed-computation benchmark analysis task on each of these standard scene images, recording the execution time and peak memory usage for each task. After testing all standard scene images, the arithmetic mean of all recorded execution times is calculated to obtain the time benchmark. And calculate the arithmetic mean of the memory peak values ​​of all records to obtain the memory baseline value. .

[0051] After completing the above offline process, the system is ready for online operation. When an analysis agent equipped with an onboard scenario map and configured with baseline values ​​is running in orbit, upon receiving a task instruction pointing to a specific area of ​​action within its coverage, it first queries the onboard scenario map based on its current location and time to obtain the corresponding parameter compensation model, and then uses this model to locally adjust the control parameters in the analysis primitives. After completing the analysis task, the analysis agent captures the actual execution time of this task. and actual peak memory usage And based on the deterministic formula Computational complexity index Among them, weighting factors and As preset constants, their values ​​are all 0.5; thus, an analysis result adapted through localized parameters is attached with metadata that objectively quantifies the complexity of its corresponding scenario, and together they are sent back to the central management platform.

[0052] Example 5: Under the condition of ensuring long-term stable operation and rational utilization of resources, the central management platform analyzes and responds strategically to the received negative knowledge packets. When three negative knowledge packets originating from the same failure cause are received for the same geographical area within 24 hours, the global avoidance data model in the central management platform will upgrade the avoidance strategy for that area. That is, the original short-term delay processing will be automatically adjusted to a task scheduling restriction period of a preset duration. This can effectively avoid the consumption of computing and communication resources caused by continuously assigning tasks to permanent or long-term ineffective areas, such as areas with continuous cloud cover or deep mountain shadow areas.

[0053] To ensure the traceable metrological basis of the credibility tokens generated autonomously by the analysis agent, a virtual standard composed of standard spectral feature vectors is pre-built within it. This standard is established through an offline, controlled calibration process. This process selects several globally recognized, long-term stable surface and atmospheric calibration fields as benchmarks. Over a period of more than one year, thousands of cloudless images of these calibration fields are acquired at different seasons and times using one or more precisely calibrated benchmark remote sensing payloads. Combined with ground-measured spectral data acquired simultaneously with the images, radiometric and atmospheric corrections are performed on the images. Finally, the multiple spectral feature vectors representing the true surface reflectance of the calibration fields are statistically averaged to generate a master standard spectral feature vector and its allowable deviation range. This master vector and deviation range together constitute the core data of the virtual standard and are embedded in the analysis agent as the benchmark for comparison when it performs subsequent self-verification tasks.

[0054] Example 6: In an engineering procedure aimed at standardizing system deployment and operation, the definition paradigm and execution protocol of structured analysis primitives in the standard library were set. According to this setting, the definition of each analysis primitive contains a computational flow description consisting of an ordered and standardized sequence of atomic computation function calls. Each atomic computation function corresponds to a basic and indivisible remote sensing data processing step. When there are multiple analysis primitives targeting the same geographic area in the task queue of the analysis agent, the internal fusion execution flow is initiated. This flow identifies shared computational steps by comparing the atomic computation function call sequences in the definitions of these analysis primitives one by one. If the first N atomic computation function call sequences of two or more analysis primitives are exactly the same, then the same function call sequence is determined to be a shared computational step. The analysis agent then merges the shared computational step and executes it only once to generate intermediate data, and continues to execute the subsequent non-shared computational steps in each analysis primitive based on the intermediate data.

[0055] The engineering procedures also include a protocol for handling failures of self-verification tasks by the analysis agent. When the analysis agent's trajectory overlaps with the location of a permanent ground calibration field and triggers a self-verification task, if the spectral characteristics of the acquired real-time calibration image exceed a preset threshold after comparison with a pre-set virtual standard, the self-verification task is deemed a failure. Accordingly, the analysis agent will not generate a credibility token containing a validity status, but will instead mark its own data source credibility as unverified in its internal state machine and attach a valid time window to this status. Within this time window, all analysis result data generated and returned by the analysis agent will be appended with a metadata flag indicating that its data source quality has not been effectively self-verified, and this flag will only be removed after the self-verification task is successfully completed on the next available calibration field.

[0056] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0057] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A remote sensing image data scheduling and distribution management system, characterized in that, It includes a central management platform and at least one analysis agent deployed at the remote sensing image data source. The central management platform is configured to: receive structured analysis primitives that include the analysis task type and scope as determined by the user; distribute the analysis primitives as task instructions to the analysis agents covering the scope; and receive analysis result data containing only the analysis conclusions returned by the analysis agents. The analysis agent is configured to: receive task instructions and directly acquire raw remote sensing image data locally at the remote sensing image data source to execute the analysis task defined by the analysis primitives; generate analysis result data containing only the analysis conclusions based on the execution results of the analysis task; the analysis agent is also configured to perform an executability pre-check step on the raw remote sensing image data before executing the analysis task. The executability pre-check step includes: determining whether there are preset unexecutable conditions based on the pixel value distribution or brightness saturation of the raw remote sensing image data; and if unexecutable conditions are determined to exist, suspending the analysis task and generating a negative knowledge package containing the type of unexecutable condition as the reason for failure and sending it back to the central management platform. The central management platform is further configured to: update a global avoidance data model based on the received negative knowledge package, and apply the global avoidance data model to filter or delay the new analysis primitives when scheduling them.

2. The remote sensing image data scheduling and distribution management system according to claim 1, characterized in that, The analysis agent is also configured to: pre-set at least one virtual standard corresponding to the ground permanent calibration field, which is composed of standard spectral feature vectors; when the operation trajectory of the analysis agent overlaps with the location of the ground permanent calibration field, it autonomously triggers a self-verification task, and compares the spectral features of the acquired real-time calibration image with the virtual standard locally at the remote sensing image data source end to generate a credibility token containing validity status and generation timestamp. And when sending back the analysis results data, attach the credibility token to the analysis results data and send it back together.

3. The remote sensing image data scheduling and distribution management system according to claim 1, characterized in that, The analysis agent is also configured to: when there are multiple analysis primitives for the same geographic area in the analysis agent's task queue, compare the predefined computational flows of the multiple analysis primitives to identify shared computational steps; construct a temporary fusion execution flow based on the comparison results, merge the shared computational steps locally at the remote sensing image data source end and execute them only once to generate intermediate data; and continue to execute the non-shared computational steps in each analysis primitive based on the intermediate data.

4. The remote sensing image data scheduling and distribution management system according to claim 1, characterized in that, The analysis agent is also configured to: pre-load an onboard context map consisting of geographic partitions and corresponding parameter compensation models; before executing the analysis task, the analysis agent queries the onboard context map to obtain the corresponding parameter compensation model based on its current covered geographic location, and uses the model to adjust or replace the values ​​of one or more control parameters in the received analysis primitives.

5. The remote sensing image data scheduling and distribution management system according to claim 1, characterized in that, The analysis agent is also configured to: generate intermediate data describing the continuous changes in pixel values ​​during the execution of the analysis task; calculate an information entropy index characterizing the scene stability of the original remote sensing image data based on the pixel value distribution of the intermediate data before generating the analysis result data; and append the information entropy index to the analysis result data and send it back together when transmitting the analysis result data.

6. The remote sensing image data scheduling and distribution management system according to claim 5, characterized in that, Information entropy index Calculated using the following formula: in, This represents the total number of different pixel values ​​in the intermediate data. For the first pixel value, pixel value The probability of it appearing in the intermediate data.

7. The remote sensing image data scheduling and distribution management system according to claim 1, characterized in that, The analysis agent is also configured to: after executing an analysis task, capture one or more process metadata during the execution of the analysis task, including execution time and peak memory usage; and calculate a complexity index locally at the remote sensing image data source end, based on the process metadata and a preset baseline value, to characterize the scene complexity of the original remote sensing image data. In addition, when transmitting the analysis results data back, the complexity index is appended to the analysis results data and transmitted back together.

8. The remote sensing image data scheduling and distribution management system according to claim 1, characterized in that, The analysis primitives were selected from a standard library containing basic remote sensing analysis models.

9. The remote sensing image data scheduling and distribution management system according to claim 1, characterized in that, The analysis results data can be in the format of vector files, sparse matrix files, or text statistical files.

10. A remote sensing image data scheduling and distribution management system according to claim 1, characterized in that, An analysis agent is a software module deployed on the payload processor of a remote sensing satellite or on a ground data center server. The analysis agent is also configured to mark the raw remote sensing image data used to perform the analysis task as removable or degraded storage after generating analysis result data.

Images