A remote sensing constellation data satellite-ground integrated sub-directive processing method and system

Abstract

This invention relates to the field of remote sensing data processing technology, specifically to a satellite-ground integrated distributed processing method for remote sensing constellation data, comprising the following steps: S1: acquiring raw remote sensing images in orbit; S2: performing data preprocessing on the raw remote sensing data, including radiometric calibration, sensor correction, and geometric coarse correction; S3: performing rapid region segmentation and ground feature classification and initial screening; S4: making task coordination and distribution decisions based on the confidence level C and type Type of each initially screened target; S5: data encapsulation and downlink, encapsulating the data to be transmitted back to the ground; S6: ground system reception, receiving the data packets transmitted from satellite; S7: the ground system distributes the data to the corresponding dedicated processing algorithm for final image interpretation and processing based on the data information. This invention creatively designs a "coarse screening-distribution" satellite-ground collaborative processing flow and a multi-level processing architecture of "main satellite distribution-constellation system-ground fine interpretation".

Description

Technical Field

[0001] This invention relates to the field of remote sensing data processing technology, specifically a method and system for integrated satellite-ground processing of remote sensing constellation data. Background Technology

[0002] Currently, there are two main technical solutions for processing and transmitting on-board remote sensing data: First, on-board storage and ground processing. After acquiring wide-swath, high-resolution raw image data in orbit, the remote sensing satellite stores the data completely in its onboard memory. When the satellite flies over a ground receiving station, it transmits all the stored raw data downlink to a ground data center. Subsequently, high-performance computing servers on the ground process, analyze, and interpret the massive amounts of data to extract valuable information. Second, on-orbit intelligent processing. This solution aims to pre-process some data on the satellite platform. By carrying high-performance computing units on the satellite and running pre-set intelligent algorithm models, the raw images are analyzed in real-time in orbit, directly extracting key information or generating processed products, with only a small amount of effective results transmitted down to the ground. This technology is still in the development and verification stage.

[0003] The aforementioned existing technologies have the following obvious drawbacks: The core problem with Option 1 lies in the immense pressure it places on satellite-to-ground data transmission bandwidth. With the explosive growth of remote sensing data, the limited reception time and downlink bandwidth of domestic ground stations have become bottlenecks, resulting in low data transmission efficiency and a large backlog of valuable data onboard, unable to promptly serve time-sensitive applications such as emergency response and real-time monitoring. Simultaneously, transmitting useless or redundant background data also wastes valuable onboard storage and downlink resources.

[0004] While Option 2 theoretically alleviates transmission pressure significantly, it relies on highly reliable, high-performance, and low-power onboard computing hardware, as well as algorithms highly adapted to onboard image processing. Currently, this technology is not fully mature and faces challenges such as limited onboard computing resources, insufficient algorithm generalization capabilities, and difficulties in updating onboard models, making it difficult to meet the complex and ever-changing demands of onboard missions. Summary of the Invention

[0005] The purpose of this invention is to provide a remote sensing constellation data integrated satellite-ground split-processing method and system to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, one aspect of the present invention provides the following technical solution: A method for integrated satellite-ground processing of remote sensing constellation data includes the following steps: S1: Acquire raw remote sensing images in orbit; S2: Perform data preprocessing on the raw remote sensing data, including radiometric calibration, sensor correction, and geometric coarse correction; S3: Perform rapid region segmentation and land feature classification and initial screening; S4: Make task coordination and distribution decisions based on the confidence level C and type Type of each initial screening target; S5: Data encapsulation and downlink, encapsulating the data that needs to be transmitted back to the ground; S6: Ground system reception, receiving data packets transmitted up and down from the satellite; S7: The ground system distributes the data to the corresponding dedicated processing algorithms for final image interpretation and processing based on the data information.

[0007] Preferably, step S3 specifically includes: S31. Based on preset geographic information, macro-regional screening and functional structure-level fine segmentation are used to generate structural image slices with functional semantics. S32. Perform rapid target existence determination on structural-level image slices and remove functional structural regions with no targets or low value. S33. For the image areas that have been identified and preserved, perform land cover classification and target-level preliminary screening based on clustering algorithm, and generate a preliminary target list and its confidence level C.

[0008] Preferably, step S31 specifically includes: S311, Coordinate Mapping: Extract macro-regional boundary vertices that are relevant to the current task or have potential high value from the pre-stored geographic information database, and use the linear transformation relationship between the corrected pixel coordinates and geographic coordinates to directly convert the latitude and longitude coordinates of all boundary vertices into the pixel coordinates of the current image. S312, Boundary Fitting: The least squares method is used to perform polygon fitting and registration on the transformed pixel vertices to generate the initial macroscopic region pixel boundaries; S313. Macro-region segmentation: Based on the fitted macro-region pixel boundaries, segment out the corresponding region of interest from the whole image.

[0009] Preferably, step S32 specifically includes: S321. Detect whether there are abnormal pixels within the functional structure region that do not match the expected background spectral characteristics: For all pixels within the functional structure region, calculate the average value for each band to obtain the average spectral vector of the current region. The Spectral Anomaly Index (SAI) is calculated as follows: (1); In equation (1), This represents the average spectral vector for the current region. This is a pre-stored mean vector of the background spectrum of this functional structural region; If the Spectral Anomaly Index (SAI) value is greater than the spectral index of the preset area, then the area is determined to be spectrally abnormal. S322. Detect whether the spatial structure of the functional structure area has changed: Compare the structural similarity (SSI) of the current image slice with a pre-stored targetless template: (2); In equation (2), The average grayscale value of the current image slice. The grayscale mean of the image without a target template is stored in advance. The standard deviation of the grayscale of the current image slice. The standard deviation of grayscale values ​​for a pre-stored template without a target. The grayscale covariance between the current image slice and the pre-stored targetless template. , It is a constant; The Structural Similarity Index (SSI) ranges from -1 to 1. A larger value indicates greater similarity. When the value is less than the preset similarity index, the structure is considered to have undergone significant changes. S323. Determine whether there are areas with significantly increased local contrast within the functional structural region: Calculate the local contrast ratio (LCE) between the target and the background: (3); In equation (3), The standard deviation of gray levels in the region. The mean of the local gradient magnitude. and These are the weighting coefficients; S324. Calculate the confidence level of the existence of the target. : The spectral anomaly index (SAI), structural similarity index (SSI), and local contrast ratio (LCE) were normalized to the [0,1] interval, and a weighted voting mechanism was used to calculate the target existence confidence. : (4); In equation (4), , and These are the weighting coefficients, and , This is a normalization operation; If the confidence level of the existence of the target If the value is greater than the preset value, the target is determined to exist; The output of the fast target existence determination includes: Does the functional structure sub-region contain a target? Corresponding target existence confidence level ; A flag indicating whether to proceed to the next target-level processing flow.

[0010] Preferably, step S33 specifically includes: S331. For the image areas that are retained after rapid target existence determination, the K-value clustering algorithm is used to classify the land cover categories: Suppose the image data has P pixels, and each pixel has D spectral band features. Divide the P pixels into K clusters { In the expression, the sum of squared distances J from all pixels to their respective cluster centers is minimized. (5); In equation (5), It is a pixel eigenvectors, It is clustering The centroid of the cluster, where i represents the cluster number; By iteratively updating the centroids and redistributing pixels, the final clustering result is obtained; After clustering is completed, based on each cluster center of mass The spectral features of the pixels are used to attach semantic labels, shape analysis is performed on the pixel groups, regions with regular geometric features are extracted, and they are initially screened as suspected man-made targets. Their minimum bounding rectangle is calculated to form a list of initial targets. S332. By calculating the difference in spectral and texture features between the initial screening target area and the surrounding background area, and combining this with the shape regularity, the feature confidence level C of each initial screening target is calculated comprehensively. Spectral Dissimilarity (SD): Calculated by the average Mahalanobis distance between the initial screening target area and the surrounding background area, reflecting the spectral distinction between the target and the background. The calculation formula is as follows: (6); In equation (6), This represents the average spectral vector of the target region in the initial screening. This represents the average spectral vector of the 3×3 pixel wide background region surrounding the target in the initial screening. The background region's spectral covariance matrix; Texture Difference (TD): Based on the grayscale matrix, the difference in texture contrast between the target and the background is calculated, reflecting the distinguishability of texture features. The calculation formula is as follows: (7); In equation (7), The grayscale matrix contrast of the target region. The gray matrix contrast of the background area; Shape regularity SR: Combining the filling rate of the minimum bounding rectangle and the aspect ratio to quantify the regular geometric features of artificial targets. The calculation formula is as follows: (8); In formula (8), is the area of the target pixels, is the area of the target's minimum bounding rectangle, is the aspect ratio of the minimum bounding rectangle, is the typical aspect ratio of the target, is the maximum aspect ratio threshold; Normalize the above three features to the interval [0, 1] respectively, and use weighted summation to obtain the feature confidence C: (9); In formula (9), , , are the weight coefficients, and .

[0011] Preferably, step S4 specifically includes: According to the feature confidence C and type Type of the preliminarily screened target, make a judgment based on the preset rules: Rule 1: If C ≥ T_high, where T_high represents the highest confidence threshold, then jump to S5 and send the target slice back to the ground; Rule 2: If T_low ≤ C < T_high, where T_low represents the lowest confidence threshold, then query the on-board target attributes and the rule library; Matching judgment: If there is a record in the on-board target attributes and the rule library, it indicates that the preliminarily screened target of this Type can be processed by a specific dedicated satellite within the constellation; Task distribution: If the matching is successful, generate a collaborative observation task request and distribute it to the corresponding dedicated satellite; the collaborative observation task request includes the geographical location, type, and priority of the preliminarily screened target. The dedicated satellite will revisit and confirm when flying over the preliminarily screened target上空 later, and directly or via relay send the high-value results back to the ground; Rule 3: If C < T_low or the matching in Rule 2 is unsuccessful, it is considered that the preliminarily screened target is ambiguous or there is no on-board dedicated resource temporarily. Jump to S5 and send the preliminarily screened target or the target slice that cannot be distributed back to the ground.

[0012] Preferably, each preliminarily screened target inherits the semantic label of its clustering area as the type Type of the preliminary screening of this target. <00​Another aspect of the present invention provides a remote sensing constellation data integrated satellite-ground distributed processing system for implementing the remote sensing constellation data integrated satellite-ground distributed processing method described above, comprising: The onboard data acquisition module acquires raw remote sensing images of the Earth through various payloads, including optical, SAR, and hyperspectral sensors. The preprocessing module performs data preprocessing on the raw remote sensing images, including radiometric calibration, sensor correction, and geometric coarse correction, to eliminate physical distortions caused by the sensor itself and the observation geometry, and to form image data with accurate location and spatial information. The distributed processing module includes a fast region segmentation unit and a land cover classification and preliminary screening unit. The fast region segmentation unit accurately cuts out the region of interest from the image data based on preset geographic information and generates key area slices. The land cover classification and preliminary screening unit obtains a preliminary screening target list and its feature confidence. The on-board mission collaboration and distribution module includes an on-board target attribute and rule base that stores the attributes of different target types and their corresponding processing methods. It makes decisions on each initially screened target according to preset rules and determines the final flow of data. The on-board data encapsulation module encapsulates the slice data to be transmitted along with other additional attributes into a structured key data packet. The satellite-to-ground high-speed data transmission module transmits the encapsulated structured key data packets downlink. The ground data receiving module receives structured key data packets transmitted downlink; The ground data processing module receives and distributes structured key data packets, and based on their metadata tags, dispatches them to dedicated processing algorithms for in-depth analysis and final confirmation.

[0014] Preferably, the distributed processing module enters one of the following two processing modes, or executes them in a sequential combination, depending on whether there is explicit task input information: Task-oriented classification mode: This mode is designed for task inputs with clear semantic objectives or monitoring needs. In this mode, the land cover classification and initial screening units introduce task constraints and are executed only for task-related functional structures and land cover categories. Data-driven classification mode: For areas without specific task input, information is initially screened in a wide-area scene; in this processing mode, the land cover classification and initial screening unit performs general target mining on the remaining image area without depending on the specific task type.

[0015] Preferably, the rapid segmentation unit includes a lightweight geographic information database pre-stored on the satellite. This lightweight geographic information database contains peripheral boundary information of a macro-region, including but not limited to airports and ports. The lightweight geographic information database further includes geographic information about the internal functional structures of the macro-region, including but not limited to: The airport includes runways, taxiways, and aprons. Berths, wharf operation areas, and waterways within the port; Other infrastructure structures with clearly defined spatial functional semantics.

[0016] Compared with the prior art, the beneficial effects of the present invention are: This invention creatively designs a "coarse screening-splitting" space-ground collaborative processing flow and a multi-level processing architecture of "main satellite sub-guidance-star cluster system-ground fine analysis". By decomposing complex global problems into multiple simplified local data analysis problems, and by utilizing the division of labor and cooperation between lightweight models and dedicated models, it maximizes the on-orbit purification of data under the constraints of limited on-board resources, thereby accurately solving the contradiction between "data explosion" and "limited transmission / processing capacity". Attached Figure Description

[0017] Figure 1 This is an overall flowchart of the remote sensing constellation data integrated satellite-ground split-guide processing method in an embodiment of the present invention; Figure 2 This is a flowchart illustrating the rapid region segmentation, land cover classification, and initial screening in the remote sensing constellation data integrated satellite-ground split-processing method in this embodiment of the invention. Figure 3 This is a flowchart illustrating the task coordination and distribution decision-making process based on the confidence level C and type Type of each initially screened target in the remote sensing constellation data integrated satellite-ground distribution processing method in this embodiment of the invention. Figure 4 This is a structural block diagram of the remote sensing constellation data integrated satellite-ground distributed processing system in an embodiment of the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Figure 1 This is an overall flowchart of a remote sensing constellation data integrated satellite-ground distributed processing method according to an embodiment of the present invention. In the embodiments of the present invention, as... Figure 1 As shown, an embodiment of the present invention provides a satellite-ground integrated remote sensing constellation data processing method, comprising the following steps: A method for integrated satellite-ground processing of remote sensing constellation data includes the following steps: S1: Acquire raw remote sensing images in orbit; S2: Perform data preprocessing on the raw remote sensing data, including radiometric calibration, sensor correction, and geometric coarse correction; S3: Perform rapid region segmentation and land feature classification and initial screening; S4: Make task coordination and distribution decisions based on the confidence level C and type Type of each initial screening target; S5: Data encapsulation and downlink, encapsulating the data that needs to be transmitted back to the ground; S6: Ground system reception, receiving data packets transmitted up and down from the satellite; S7: The ground system distributes the data to the corresponding dedicated processing algorithms for final image interpretation and processing based on the data information.

[0020] The remote sensing constellation data integrated satellite-ground splitting processing method provided by the embodiments of the present invention creatively designs a "coarse screening-splitting" satellite-ground collaborative processing flow and a multi-level processing architecture of "main star splitting-constellation system-ground fine analysis". By decomposing complex global problems into multiple simplified local data analysis problems, and by utilizing the division of labor and cooperation between lightweight models and dedicated models, the on-orbit purification of data is maximized under the constraint of limited on-board resources, thereby accurately solving the contradiction between "data explosion" and "limited transmission / processing capacity".

[0021] Figure 2 This is a flowchart illustrating rapid region segmentation, land cover classification, and initial screening in a remote sensing constellation data integrated satellite-ground processing method according to an embodiment of the present invention. In one embodiment of the present invention, as... Figure 2 As shown, step S3 specifically includes: S31. Based on preset geographic information, macro-regional screening and functional structure-level fine segmentation are used to generate structural image slices with functional semantics. S32. Perform rapid target existence determination on structural-level image slices and remove functional structural regions with no targets or low value. S33. For the image areas that have been identified and preserved, perform land cover classification and target-level preliminary screening based on clustering algorithm, and generate a preliminary target list and its confidence level C.

[0022] Furthermore, in one embodiment of the present invention, step S31 specifically includes: S311, Coordinate Mapping: Extract macro-regional boundary vertices that are relevant to the current task or have potential high value from the pre-stored geographic information database, and use the linear transformation relationship between the corrected pixel coordinates and geographic coordinates to directly convert the latitude and longitude coordinates of all boundary vertices into the pixel coordinates of the current image. S312, Boundary Fitting: The least squares method is used to perform polygon fitting and registration on the transformed pixel vertices to generate the initial macroscopic region pixel boundaries; S313. Macro-region segmentation: Based on the fitted macro-region pixel boundaries, segment out the corresponding region of interest from the whole image.

[0023] During processing, the system first segments the entire imagery based on macroscopic region boundaries, identifying areas relevant to the mission or possessing potential high value. Then, within these macroscopic regions, the imagery is further segmented into refined sub-regions according to the aforementioned functional structure information, generating structure-level slices with clearly defined functional semantics. Through this multi-level segmentation from macroscopic regions to functional structures, the system can transform the entire imagery into structure-level data units with clear spatial semantics and functional attributes during the on-board phase, providing a structured foundation for subsequent target existence determination and data compression.

[0024] In one embodiment of the present invention, step S32 specifically includes: S321. Detect whether there are abnormal pixels within the functional structure region that do not match the expected background spectral characteristics. Each functional structure region (such as runway, apron) has typical spectral characteristics in the absence of a target. Calculate the difference between the average spectrum of the current region and the pre-stored background spectral model using Mahalanobis distance. For all pixels within the functional structure region, calculate the average value for each band to obtain the average spectral vector of the current region. The Spectral Anomaly Index (SAI) is calculated as follows: (1); In equation (1), This represents the average spectral vector for the current region. This is a pre-stored mean vector of the background spectrum of this functional structural region; If the Spectral Anomaly Index (SAI) value is greater than the spectral index of the preset area, then the area is determined to be spectrally abnormal. S322. Detect whether the functional structural region has changed in spatial structure, and compare the structural similarity between the current image slice and the pre-stored targetless template. The structural similarity is compared from three dimensions: brightness, contrast, and structure. Compare the structural similarity (SSI) of the current image slice with a pre-stored targetless template: (2); In equation (2), The mean grayscale value of the current image slice (representing the average grayscale value of all pixels). The grayscale mean of the image without a target template is stored in advance. The standard deviation of grayscale values ​​for the current image slice (representing the degree of dispersion in grayscale distribution). The standard deviation of grayscale values ​​for a pre-stored template without a target. The gray-level covariance between the current image slice and the pre-stored targetless template (representing the degree of correlation between the gray-level changes of the two images). , It is a constant; The Structural Similarity Index (SSI) ranges from -1 to 1. A larger value indicates greater similarity. When the value is less than the preset similarity index, the structure is considered to have undergone significant changes. S323. The presence of a target object usually produces high contrast in a local area. By combining the standard deviation of the area gray level and the mean of the local gradient magnitude, determine whether there are areas with significantly increased local contrast within the functional structure area: Calculate the local contrast ratio (LCE) between the target and the background: (3); In equation (3), The standard deviation of gray levels in the region. The mean of the local gradient magnitude. and These are the weighting coefficients; S324. Calculate the confidence level of the existence of the target. : The spectral anomaly index (SAI), structural similarity index (SSI), and local contrast ratio (LCE) were normalized to the [0,1] interval, and a weighted voting mechanism was used to calculate the target existence confidence. : (4); In equation (4), , and These are the weighting coefficients, and , This is a normalization operation; If the confidence level of the existence of the target If the value is greater than the preset value, the target is determined to exist; The output of the fast target existence determination includes: Does the functional structure sub-region contain a target? Corresponding target existence confidence level ; A flag indicating whether to proceed to the next target-level processing flow.

[0025] After obtaining functional structure-level image slices, a rapid target presence determination is performed on each functional structure sub-region to determine whether there are any potential targets or abnormal changes within that region. If no pixel clusters matching preset target characteristics are detected within a functional structure sub-region, or if its target confidence level is below a minimum threshold, that sub-region is classified as an empty structure region. Its corresponding image data is not included in subsequent data encapsulation and downlink processes, thus achieving further compression of target-level slices and data volume during the on-board phase. Image data identified as empty structure regions is either discarded directly on-board or only retains statistical information, without participating in subsequent target-level initial screening and satellite-to-ground transmission processes.

[0026] In one embodiment of the present invention, step S33 specifically includes: S331. For the image areas that are retained after rapid target existence determination, the K-value clustering algorithm is used to classify the land cover categories: Suppose the image data has P pixels, and each pixel has D spectral band features. Divide the P pixels into K clusters { In the expression, the sum of squared distances J from all pixels to their respective cluster centers is minimized. (5); In equation (5), It is a pixel eigenvectors, It is clustering The centroid of the cluster, where i represents the cluster number; By iteratively updating the centroids and redistributing pixels, the final clustering result is obtained; After clustering is completed, based on each cluster center of mass The spectral features of the pixels are used to attach semantic labels, shape analysis is performed on the pixel groups, regions with regular geometric features are extracted, and they are initially screened as suspected man-made targets. Their minimum bounding rectangle is calculated to form a list of initial targets. For example, high reflectivity can be labeled as "clouds," and clusters with high reflectivity in the blue and green bands can be labeled as "sea areas," etc. The area of ​​the connected regions labeled as "clouds" in the classification map is calculated. If the area of ​​a region exceeds the set threshold Area_cloud, it is determined to be an invalid thick cloud area and is labeled and removed. In the "sea areas" or "bare land / buildings" categories, the shape of the pixel groups is analyzed to extract regions with regular geometric features (such as rectangles and circles), which are initially screened as suspected man-made targets (such as ships and vehicles). The minimum bounding rectangle is calculated to form a list of initially screened targets.

[0027] S332. By calculating the difference in spectral and texture features between the initial screening target area and the surrounding background area, and combining this with the shape regularity, the feature confidence level C of each initial screening target is calculated comprehensively. Spectral Dissimilarity (SD): Calculated by the average Mahalanobis distance between the initial screening target area and the surrounding background area, reflecting the spectral distinction between the target and the background. The calculation formula is as follows: (6); In equation (6), This represents the average spectral vector of the target region in the initial screening. This represents the average spectral vector of the 3×3 pixel wide background region surrounding the target in the initial screening. The background region's spectral covariance matrix; Texture Difference (TD): Based on the grayscale matrix, the difference in texture contrast between the target and the background is calculated, reflecting the distinguishability of texture features. The calculation formula is as follows: (7); In equation (7), The grayscale matrix contrast of the target region. The grayscale matrix contrast of the background area; Shape regularity (SR): Combining the minimum bounding rectangle fill rate and aspect ratio, this quantifies the regular geometric features of man-made targets. The calculation formula is as follows: (8); In equation (8), For the target pixel area, The goal is to find the area of ​​the minimum bounding rectangle. The aspect ratio of the minimum bounding rectangle. The typical length-to-width ratio is taken as 8:1 for ships and 2:1 for vehicles. The maximum aspect ratio threshold; Normalize the three features to the [0,1] interval, and use a weighted summation to obtain the feature confidence C: (9); In equation (9), , , These are the weighting coefficients, and .

[0028] Figure 3 This is a flowchart illustrating the task coordination and distribution decision-making process based on the confidence level C and type Type of each initially screened target in the remote sensing constellation data integrated satellite-ground distributed processing method of this invention. Figure 3 As shown, in one embodiment of the present invention, step S4 specifically includes: Based on the characteristic confidence C and type Type of the preliminary screening target, make a judgment according to the preset rules: Rule 1: If C ≥ T_high, where T_high represents the highest confidence threshold, then jump to S5 and transmit the target slice back to the ground; Rule 2: If T_low ≤ C < T_high, where T_low represents the lowest confidence threshold, then query the on-board target attributes and the rule library; Matching judgment: If there is a record in the on-board target attributes and the rule library, it indicates that the preliminary screening target of this Type can be processed by a specific dedicated satellite within the constellation; Task distribution: If the match is successful, generate a collaborative observation task request and distribute it to the corresponding dedicated satellite; the collaborative observation task request includes the geographical location, type, and priority of the preliminary screening target, and the dedicated satellite will re-visit and confirm when flying over the preliminary screening target上空 later, and directly or via relay transmit the high-value results back to the ground; Rule 3: If C < T_low or the match in Rule 2 is unsuccessful, it is considered that the preliminary screening target is ambiguous or there is no dedicated on-board resource for the time, jump to S5 and transmit the preliminary screening target or the target slice that cannot be distributed back to the ground.

[0029] In an embodiment of the present invention, each preliminary screening target inherits the semantic label of its所在 clustering area as the type Type for preliminary screening this target. For example, a preliminary screening target located in the "sea area" clustering area is marked with Type as "ship"; a preliminary screening target located in the "bare land / building / airport" clustering area is marked with Type as "vehicle / building / aircraft".

[0030] Figure 4 This is the structural block diagram of the space-ground integrated and guided processing system for remote sensing constellation data in the embodiment of the present invention. The present invention also provides a space-ground integrated and guided processing system for remote sensing constellation data, which is used to implement the space-ground integrated and guided processing method for remote sensing constellation data as described above, as Figure 4 shown, including: On-board data acquisition module, which acquires the original remote sensing images of earth observation through multiple payloads including optical, SAR, and hyperspectral; Pretreatment module, which performs data preprocessing on the original remote sensing images including radiometric calibration, sensor correction, and geometric rough correction, eliminates the physical distortion caused by the sensor itself and the observation geometry, and forms image data with accurate position and spatial information; The onboard data acquisition module and preprocessing module are responsible for acquiring raw remote sensing images of Earth observation through various payloads such as optics, SAR, and hyperspectral imaging. They also perform preprocessing operations such as radiometric calibration, sensor correction, and system-level geometric coarse correction on the raw images to eliminate physical distortions introduced by sensor characteristics and observation geometry during the imaging process, and generate preprocessed image data with accurate spatial positional relationships, providing basic data for subsequent onboard intelligent sub-guidance processing. The distributed processing module includes a fast region segmentation unit and a land cover classification and preliminary screening unit. The fast region segmentation unit accurately cuts out the region of interest from the image data based on preset geographic information and generates key area slices. The land cover classification and preliminary screening unit obtains a preliminary screening target list and its feature confidence. The on-board mission collaboration and distribution module includes an on-board target attribute and rule base that stores the attributes of different target types and their corresponding processing methods. It makes decisions on each initially screened target according to preset rules and determines the final flow of data. The on-board data encapsulation module integrates the slice data to be transmitted (including slices of key areas, slices of high-confidence targets, and slices of low-confidence suspicious targets) with other additional attributes (including but not limited to payload type, imaging time, geographic coordinates, initial screening classification confidence, image parameters, etc.) to form a structured key data package. The satellite-to-ground high-speed data transmission module transmits the encapsulated structured key data packets downlink. The ground data receiving module receives structured key data packets transmitted downlink; The ground data processing module integrates powerful computing resources and advanced algorithms. It receives and distributes structured key data packets and, based on their metadata tags, dispatches them to dedicated processing algorithms for in-depth analysis and final confirmation.

[0031] Furthermore, in one embodiment of the present invention, the decentralization processing module enters one of the following two processing modes, or executes them in a sequential combination, depending on whether there is explicit task input information: Task-oriented classification mode: This mode is designed for task inputs with clear semantic objectives or monitoring needs. In this mode, the land cover classification and initial screening units introduce task constraints and are executed only for task-related functional structures and land cover categories. Data-driven classification mode: For areas without specific task input, information is initially screened in a wide-area scene; in this processing mode, the land cover classification and initial screening unit performs general target mining on the remaining image area without depending on the specific task type.

[0032] In one embodiment of the present invention, the rapid segmentation unit includes a lightweight geographic information database pre-stored on a satellite. The lightweight geographic information database contains peripheral boundary information of a macro-region, including but not limited to airports and ports. The lightweight geographic information database further contains geographic information of the internal functional structures of the macro-region, including but not limited to: The airport includes runways, taxiways, and aprons. Berths, wharf operation areas, and waterways within the port; Other infrastructure structures with clearly defined spatial functional semantics The present invention provides an innovative satellite-ground integrated remote sensing constellation data processing system that constructs a multi-level processing architecture of "main satellite sub-guidance - constellation system - ground fine analysis", thereby systematically solving the core contradiction between the explosive growth of on-board data and the limited bandwidth of satellite-ground data transmission, as well as the immaturity of on-orbit processing technology.

[0033] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A remote sensing constellation data satellite-ground integrated sub-directive processing method, characterized in that, Includes the following steps: S1: Acquire raw remote sensing images in orbit; S2: Perform data preprocessing on the raw remote sensing data, including radiometric calibration, sensor correction, and geometric coarse correction; S3: Perform rapid region segmentation and land feature classification and initial screening; S4: Make task coordination and distribution decisions based on the confidence level C and type Type of each initial screening target; S5: Data encapsulation and downlink, encapsulating the data that needs to be transmitted back to the ground; S6: Ground system reception, receiving data packets transmitted up and down from the satellite; S7: The ground system distributes the data to the corresponding dedicated processing algorithms for final image interpretation and processing based on the data information.

2. The remote sensing constellation data integrated satellite-ground processing method according to claim 1, characterized in that, Step S3 specifically includes: S31. Based on preset geographic information, macro-regional screening and functional structure-level fine segmentation are used to generate structural image slices with functional semantics. S32. Perform rapid target existence determination on structural-level image slices and remove functional structural regions with no targets or low value. S33. For the image areas that have been identified and preserved, perform land cover classification and target-level preliminary screening based on clustering algorithm, and generate a preliminary target list and its confidence level C.

3. The remote sensing constellation data integrated satellite-ground processing method according to claim 2, characterized in that, Step S31 specifically includes: S311, Coordinate Mapping: Extract macro-regional boundary vertices that are relevant to the current task or have potential high value from the pre-stored geographic information database, and use the linear transformation relationship between the corrected pixel coordinates and geographic coordinates to directly convert the latitude and longitude coordinates of all boundary vertices into the pixel coordinates of the current image. S312, Boundary Fitting: The least squares method is used to perform polygon fitting and registration on the transformed pixel vertices to generate the initial macroscopic region pixel boundaries; S313. Macro-region segmentation: Based on the fitted macro-region pixel boundaries, segment out the corresponding region of interest from the whole image.

4. The remote sensing constellation data integrated satellite-ground processing method according to claim 2, characterized in that, Step S32 specifically includes: S321. Detect whether there are abnormal pixels within the functional structure region that do not match the expected background spectral characteristics: For all pixels within the functional structure region, calculate the average value for each band to obtain the average spectral vector of the current region. The Spectral Anomaly Index (SAI) is calculated as follows: （1）； In equation (1), This represents the average spectral vector for the current region. This is a pre-stored mean vector of the background spectrum of this functional structural region; If the Spectral Anomaly Index (SAI) value is greater than the spectral index of the preset area, then the area is determined to be spectrally abnormal. S322. Detect whether the spatial structure of the functional structure area has changed: Compare the structural similarity (SSI) of the current image slice with a pre-stored targetless template: （2）； In equation (2), The average grayscale value of the current image slice. The grayscale mean of the image without a target template is stored in advance. The standard deviation of the grayscale of the current image slice. The standard deviation of grayscale values ​​for a pre-stored template without a target. The grayscale covariance between the current image slice and the pre-stored targetless template. , It is a constant; The Structural Similarity Index (SSI) ranges from -1 to 1. A larger value indicates greater similarity. When the value is less than the preset similarity index, the structure is considered to have undergone significant changes. S323. Determine whether there are areas with significantly increased local contrast within the functional structural region: Calculate the local contrast ratio (LCE) between the target and the background: （3）； In equation (3), The standard deviation of gray levels in the region. The mean of the local gradient magnitude. and These are the weighting coefficients; S324. Calculate the confidence level of the existence of the target. : The spectral anomaly index (SAI), structural similarity index (SSI), and local contrast ratio (LCE) were normalized to the [0,1] interval, and a weighted voting mechanism was used to calculate the target existence confidence. : （4）； In equation (4), , and These are the weighting coefficients, and , This is a normalization operation; If the confidence level of the existence of the target If the value is greater than the preset value, the target is determined to exist; The output of the fast target existence determination includes: Does the functional structure sub-region contain a target? Corresponding target existence confidence level ; A flag indicating whether to proceed to the next target-level processing flow.

5. The remote sensing constellation data integrated satellite-ground processing method according to claim 4, characterized in that, Step S33 specifically includes: S331. For the image areas that are retained after rapid target existence determination, the K-value clustering algorithm is used to classify the land cover categories: Suppose the image data has P pixels, and each pixel has D spectral band features. Divide the P pixels into K clusters { In the expression, the sum of squared distances J from all pixels to their respective cluster centers is minimized. （5）； In equation (5), It is a pixel eigenvectors, It is clustering The centroid of the cluster, where i represents the cluster number; By iteratively updating the centroids and redistributing pixels, the final clustering result is obtained; After clustering is completed, based on each cluster center of mass The spectral features of the pixels are used to attach semantic labels, shape analysis is performed on the pixel groups, regions with regular geometric features are extracted, and they are initially screened as suspected man-made targets. Their minimum bounding rectangle is calculated to form a list of initial targets. S332. By calculating the difference in spectral and texture features between the initial screening target area and the surrounding background area, and combining this with the shape regularity, the feature confidence level C of each initial screening target is calculated comprehensively. Spectral Difference Degree SD: Calculate the average spectral Mahalanobis distance between the initially screened target area and the surrounding background area, which reflects the spectral discrimination between the target and the background. The calculation formula is as follows: （6）； In equation (6), This represents the average spectral vector of the target region in the initial screening. This represents the average spectral vector of the 3×3 pixel wide background region surrounding the target in the initial screening. The background region's spectral covariance matrix; Texture Difference Degree TD: Calculate the difference in texture contrast between the target and the background based on the gray matrix, which reflects the discrimination of texture features. The calculation formula is as follows: （7）； In equation (7), The grayscale matrix contrast of the target region. The grayscale matrix contrast of the background area; Shape Regularity SR: Combine the filling rate of the minimum bounding rectangle and the aspect ratio to quantify the regular geometric features of artificial targets. The calculation formula is as follows: （8）； In equation (8), For the target pixel area, The goal is to find the area of ​​the minimum bounding rectangle. The aspect ratio of the minimum bounding rectangle. For the target typical aspect ratio, The maximum aspect ratio threshold; Normalize the above three features to the interval [0, 1] respectively, and use weighted summation to obtain the feature confidence C: （9）； In equation (9), , , These are the weighting coefficients, and .

6. The remote sensing constellation data integrated satellite-ground processing method according to claim 5, characterized in that, Step S4 specifically includes: According to the feature confidence C and type Type of the initially screened target, make a judgment based on the preset rules: Rule 1: If C ≥ T_high, where T_high represents the highest threshold of confidence, then jump to S5 and send the target slice back to the ground; Rule 2: If T_low ≤ C < T_high, where T_low represents the lowest threshold of confidence, then query the on-board target attributes and the rule library; Matching judgment: If there are records in the on-board target attributes and the rule library, it indicates that the initially screened target of this Type can be processed by specific dedicated satellites within the constellation; Task distribution: If the matching is successful, generate a collaborative observation task request and distribute it to the corresponding dedicated satellite; the collaborative observation task request includes the geographical location, type, and priority of the initially screened target, and the dedicated satellite revisits and confirms when flying over the initially screened target上空 later, and directly or through relay transmits the high-value results back to the ground; Rule 3: If C < T_low or the matching in Rule 2 is unsuccessful, it is considered that the initially screened target is ambiguous or there are no on-board dedicated resources temporarily, then jump to S5 and send the initially screened target or the target slice that cannot be distributed back to the ground.

7. The remote sensing constellation data integrated satellite-ground processing method according to claim 6, characterized in that, Each initially screened target inherits the semantic label of its所在聚类区域 as the type Type of initially screening this target.

8. A remote sensing constellation data integrated satellite-ground distributed processing system, used to implement the remote sensing constellation data integrated satellite-ground distributed processing method as described in any one of claims 1-7, characterized in that, It includes: On-board data acquisition module, which acquires the original remote sensing images of对地观测 through multiple payloads including optical, SAR, and hyperspectral; Preprocessing module, which performs data preprocessing on the original remote sensing images including radiometric calibration, sensor correction, and geometric rough correction, eliminates the physical distortions caused by the sensor itself and the observation geometry, and forms image data with accurate position and spatial information; Guided processing module, which includes a fast region segmentation unit and a ground object classification and initial screening unit. The fast region segmentation unit accurately cuts out the area of interest from the image data based on the preset geographical information and generates key area slices. The ground object classification and initial screening unit obtains the list of initially screened targets and their feature confidence; On-board task collaboration and distribution module, which includes an on-board target attribute and rule library storing the attributes of different target types and their corresponding processing methods, makes decisions on each initially screened target according to the preset rules, and determines the final flow direction of the data; On-board data encapsulation module, which integrally encapsulates the slice data to be downlinked and other additional attributes to form a structured key data packet; Space-ground high-speed data transmission module, which downlinks the encapsulated structured key data packet; Ground data reception module, which receives the downlinked structured key data packet; The ground data processing module receives and distributes structured key data packets, and based on their metadata tags, dispatches them to dedicated processing algorithms for in-depth analysis and final confirmation.

9. The remote sensing constellation data integrated satellite-ground processing method according to claim 8, characterized in that, The distributed processing module, depending on whether there is explicit task input information, enters one of the following two processing modes, or executes them in a sequential combination: Task-oriented classification mode: This mode is designed for task inputs with clear semantic objectives or monitoring needs. In this mode, the land cover classification and initial screening units introduce task constraints and are executed only for task-related functional structures and land cover categories. Data-driven classification mode: For areas without specific task input, information is initially screened in a wide-area scene; in this processing mode, the land cover classification and initial screening unit performs general target mining on the remaining image area without depending on the specific task type.

10. The remote sensing constellation data integrated satellite-ground processing method according to claim 9, characterized in that, The rapid segmentation unit includes a lightweight geographic information database pre-stored on the satellite. This lightweight geographic information database contains peripheral boundary information of a macro-region, including but not limited to airports and ports. The lightweight geographic information database further contains geographic information about the internal functional structures of the macro-region, including but not limited to: The airport includes runways, taxiways, and aprons. Berths, wharf operation areas, and waterways within the port; Other infrastructure structures with clearly defined spatial functional semantics.

Images