Internet remote sensing satellite task autonomous planning method based on intelligent perception and feedback

Abstract

The application discloses an internet remote sensing satellite task autonomous planning method based on intelligent sensing and feedback, and relates to the field of intelligent satellite design.The method comprises the following steps: acquiring an observation task execution result and calculating a difference between a current target attribute and benchmark data; when the difference exceeds a limit, copying an initial task attribute, accumulating a priority according to an abnormal level, and modifying an identifier to a mandatory or collaborative state to generate a new task; calculating on-board processing energy consumption and storage consumption based on an observation range, locking a time window and reserving resources for the mandatory task; constructing a maximum benefit planning model satisfying resource constraints and solving the model to output a task set to be executed by the satellite; and identifying a collaborative task not selected and sending the collaborative task to an adjacent satellite through an inter-satellite link for planning. The technical solution provided by the application improves the efficiency of the on-orbit autonomous response of a remote sensing satellite to a sudden abnormal target.

Description

Technical Field

[0001] This application relates to the field of intelligent satellite design, and in particular to an autonomous planning method for Internet remote sensing satellite missions based on intelligent sensing and feedback. Background Technology

[0002] Currently, with the rapid development of aerospace technology, remote sensing satellites are playing an increasingly important role in fields such as Earth observation, disaster monitoring, and military reconnaissance. Point-to-point observation of specific targets tests satellite observation and planning capabilities. Therefore, scholars both domestically and internationally focus on research into satellite autonomous planning technology to achieve high revisit rates, high timeliness, and high-quality target acquisition. However, current research mainly concentrates on single-satellite closed-loop observation technologies relying on ground-triggered missions and on-board triggered missions, which rely on processing results. When dealing with highly dynamic and sudden abnormal targets, ground-triggered missions are limited by the distribution of telemetry and control windows and link transmission delays, resulting in lag in command uploading and making it difficult to capture fleeting target dynamics. On-board triggered missions, with their limited observation range and low timeliness of single-satellite data acquisition, struggle to meet the demands for efficient closed-loop response to sudden targets in complex scenarios. Summary of the Invention

[0003] This application provides an autonomous planning method for Internet remote sensing satellite missions based on intelligent sensing and feedback, which can be used to improve the efficiency of remote sensing satellites in autonomous on-orbit response to sudden abnormal targets.

[0004] The first aspect of this application provides an autonomous planning method for Internet remote sensing satellite missions based on intelligent sensing and feedback, the method comprising: The system acquires the execution results of observation tasks with planning and control attributes; calculates the attribute difference between the current target attribute data and the preset normal baseline data; if the attribute difference exceeds a preset judgment threshold, it copies the initial task attribute information contained in the initial observation task as the basic information for the new task; based on a preset level mapping strategy, it determines the priority increment value corresponding to the abnormal state level, adds the priority increment value to the task priority parameter in the basic information of the new task, and modifies the planning and control identifier parameter to indicate a valid state of mandatory execution or inter-satellite coordination, thus generating a new observation task; it calculates the amount of image data to be processed based on the observation range of the new observation task, and calculates the on-board processing of the image data to be processed based on a preset unit data volume resource consumption coefficient. The required energy consumption and storage space consumption values ​​are calculated. If the planning control flag parameter of the new observation task indicates that it is a mandatory execution state, the execution time window of the new observation task is locked, and the required energy and storage space are reserved. The new observation task is added to the set of tasks to be planned, and a task planning model with the goal of maximizing observation benefits is constructed. The task planning model is solved to generate the set of tasks to be executed on the local satellite, and the remaining tasks that have not been selected are identified from the set of tasks to be planned. The remaining task set is traversed, and the tasks whose planning control flag parameter indicates that they are in an effective state of inter-satellite cooperation are divided into a set of collaborative planning tasks. The set of tasks to be executed on the local satellite is output for execution by the local satellite. The set of collaborative planning tasks is sent to neighboring satellites for planning through the inter-satellite communication link.

[0005] In the above embodiments, the satellite can process images and analyze and detect anomalies in real time on orbit, and then automatically generate new high-priority observation tasks. These tasks are then incorporated into a planning model that considers both imaging and processing resource constraints. For mandatory tasks, resources are forcibly locked, and for collaborative tasks that cannot be executed, they are automatically transferred to neighboring satellites. This allows the satellite to transform sensing information into action plans without waiting for ground instructions. This not only ensures the feasibility of new tasks under resource-constrained conditions, but also breaks through the single-satellite capability bottleneck through inter-satellite collaboration, thereby eliminating the impact of ground-satellite communication delays. This improves the utilization rate of satellite observation resources, the frequency of target observation, and the timeliness of data acquisition, and enhances the efficiency of remote sensing satellites in responding autonomously to sudden anomalies on orbit.

[0006] In conjunction with some embodiments of the first aspect, in some embodiments, after sending the set of collaborative planning tasks to neighboring satellites for planning via inter-satellite communication links, the method further includes: In response to neighboring satellites receiving the collaborative planning task set, the system obtains the orbital parameters of the neighboring satellites, calculates the satellite observation range of the neighboring satellites based on the orbital parameters, and determines whether the satellite observation range covers the target observation area of ​​the target task in the collaborative planning task set. If so, the target task is marked as the highest priority fixed node, the imaging time window and corresponding energy storage resources of the neighboring satellites in the observation area are locked, and the task is added to the set of tasks to be planned.

[0007] In the above embodiments, when neighboring satellites confirm that their observation range covers the target observation area of ​​the target task in the collaboratively planned task set, they are marked as the highest priority fixed node and their imaging and energy resources are forcibly locked. This node is then added to the set of tasks to be planned and used for subsequent planning. Transforming cross-satellite task requests into irrevocable hard constraints ensures that high-value tasks discarded by the original satellite receive resource guarantees on neighboring satellites, preventing them from being rejected again due to resource competition from neighboring satellites. This achieves seamless relay at the constellation level and improves the collaborative capture rate and execution determinism for urgent and sudden tasks.

[0008] In conjunction with some embodiments of the first aspect, in some embodiments, after determining whether the satellite observation range covers the target observation area of ​​the target task in the collaborative planning task set, the method further includes: If not, the inter-satellite collaboration of the target mission is maintained in an effective state, and the target mission is added to the set of missions to be collaborated with by neighboring satellites. The target mission is then forwarded to the next collaborative node satellite that is connected to the neighboring satellite through the inter-satellite communication link, until a target execution satellite whose satellite observation range covers the target observation area is found.

[0009] In the above embodiments, when the observation range of adjacent satellites cannot cover the target observation area of ​​the target task in the collaborative planning task set, the task is continuously relayed to the next node using inter-satellite links until the target execution satellite is found. This expands the search range of task execution resources from local to the entire constellation network, ensuring that even when the current satellite is unreachable, an emergency task can eventually find a suitable execution node through multi-hop optimization. This improves the constellation system's wide-area response capability and execution success rate for time-sensitive tasks.

[0010] In conjunction with some embodiments of the first aspect, in some embodiments, after generating a new observation task based on a preset level mapping strategy, the method further includes: When the task type is target tracking task, obtain the target feature value from the previous time series observation results for the same target; calculate the feature difference degree between the current target attribute data and the target feature value; if the feature difference degree is greater than the preset tracking lock threshold, terminate the generation process of the new observation task and delete the corresponding task basic information; if the feature difference degree is less than or equal to the preset tracking lock threshold, obtain the latest target position coordinates in the current target attribute data; update the task position parameters of the new observation task with the latest target position coordinates.

[0011] In the above embodiments, a tracking verification and position correction mechanism based on temporal feature comparison is introduced. Before generating a new task, the target tracking status can be automatically identified based on the feature difference. On the one hand, the task generation process can be interrupted in time when the target is lost, preventing the waste of satellite resources caused by invalid observations. On the other hand, after confirming that the target is consistent, the task position parameters are dynamically corrected using the latest coordinates, eliminating the time accumulation error of the moving target, ensuring that subsequent planning can be based on the real-time position, and improving the satellite's continuous locking accuracy and resource utilization efficiency for highly dynamic moving targets.

[0012] In conjunction with some embodiments of the first aspect, in some embodiments, when the attribute difference exceeds a preset judgment threshold, the initial task attribute information contained in the initial observation task is copied as the basic information of the new task, specifically including: The system identifies the task type of the initial observation task. If the task type is target detection, it calculates the coefficient of variation between the number of detected targets and the preset number of normal targets. If the coefficient of variation is greater than a preset detection threshold, it is considered an anomaly and a new task is triggered. If the task type is target identification and the coefficient of variation is less than or equal to the preset detection threshold, it further determines whether the identified target attributes contain preset important target attributes. If so, it is considered an anomaly and a new task is triggered. If the task type is target tracking, it calculates the target feature difference between the current target feature value and the target feature value to be tracked. If the target feature difference is less than a preset tracking threshold, tracking is considered successful and a new task is triggered. If the task type is change detection, it calculates the proportion of the changed area to the total observation area. If the proportion is greater than a preset area change threshold, it is considered an anomaly and a new task is triggered. In response to triggering a new task, it copies the task priority parameters, planning control flag parameters, and latest execution time parameters of the initial observation task, and extends the latest execution time parameter by one task planning cycle as the basic information for the new task.

[0013] In the above embodiments, multi-dimensional anomaly judgment logic based on statistical variation, attribute matching, feature comparison, and region proportion was constructed for different task types such as target detection, recognition, tracking, and change detection. This logic can identify key information in various scenarios and quickly generate new tasks with longer timeliness based on the original task parameters. This avoids missed detections or false alarms caused by a single judgment standard and also achieves a seamless connection from anomaly detection to continuous monitoring, improving the satellite's intelligent perception and continuous monitoring capabilities in complex and ever-changing environments.

[0014] In conjunction with some embodiments of the first aspect, in some embodiments, a task planning model is constructed with the objective of maximizing observational gains, specifically including: Obtain the list of regular tasks that have been pre-queued in the current set of tasks to be planned, identify and remove target regular tasks in the list of regular tasks that overlap with or conflict with the execution time windows of the mandatory tasks; mark the mandatory tasks as fixed nodes and insert them into the time axis of task planning, lock the imaging time window and the total amount of resources required; construct a task planning model with the goal of maximizing observation benefits.

[0015] In the above embodiments, when a new task is marked as mandatory, conflicting regular tasks are identified and removed. The new task is then marked as a fixed node, and its imaging time window and required total resources are forcibly locked. This constrains the solution space to include the node, transforming high-value emergency tasks from competitive planning to deterministic hard constraints. This ensures absolute execution rights even under extreme resource conflicts, preventing them from being squeezed out by regular tasks and enhancing the satellite's ability to support critical emergencies and its execution certainty.

[0016] In conjunction with some embodiments of the first aspect, in some embodiments, after outputting the set of tasks to be executed by the local satellite for execution by the local satellite, the method further includes: Determine whether the planning control identifier parameter of the new observation mission indicates a periodic execution status; if so, copy the mission basic information of the new observation mission to construct a derivative observation mission; extend the deadline parameter of the derivative observation mission by one mission planning cycle; inject the derivative observation mission into the satellite's long-term mission repository, and add the derivative observation mission to the set of missions to be planned when the next planning cycle starts.

[0017] In the above embodiments, an automatic task propagation and cross-cycle pre-injection mechanism was established. When a periodic demand is identified, a derivative task can be automatically constructed based on the current task and its effective time limit can be extended, seamlessly transferring it to the next planning cycle. This breaks the time slice limitation of a single planning cycle, enabling the satellite to achieve one-time triggering and continuous execution without relying on frequent ground injection commands. This avoids the interruption of long-cycle monitoring tasks and improves the satellite's autonomous continuous monitoring capability of dynamically evolving targets and the level of intelligence in on-orbit operation.

[0018] Secondly, embodiments of this application provide an autonomous planning system for Internet remote sensing satellite missions. The autonomous planning system for Internet remote sensing satellite missions includes: one or more processors and a memory; the memory is coupled to the one or more processors, and the memory is used to store computer program code, the computer program code including computer instructions, and the one or more processors call the computer instructions to cause the autonomous planning system for Internet remote sensing satellite missions to perform the methods described in the first aspect and any possible implementation thereof.

[0019] Thirdly, embodiments of this application provide a computer program product containing instructions that, when the computer program product is run on an Internet remote sensing satellite mission autonomous planning system, cause the Internet remote sensing satellite mission autonomous planning system to execute the method described in the first aspect and any possible implementation thereof.

[0020] Fourthly, embodiments of this application provide a computer-readable storage medium including instructions that, when executed on an Internet remote sensing satellite mission autonomous planning system, cause the Internet remote sensing satellite mission autonomous planning system to perform the method described in the first aspect and any possible implementation thereof.

[0021] It is understood that the Internet remote sensing satellite mission autonomous planning system provided in the second aspect, the computer program product provided in the third aspect, and the computer storage medium provided in the fourth aspect are all used to execute the Internet remote sensing satellite mission autonomous planning method based on intelligent sensing and feedback provided in the embodiments of this application. Therefore, the beneficial effects it can achieve can be referred to the beneficial effects in the corresponding methods, and will not be repeated here.

[0022] One or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages: 1. The satellite in this application is capable of processing images and analyzing them in real time on orbit to detect anomalies. It then automatically generates new high-priority observation tasks and incorporates them into a planning model that considers both imaging and processing resource constraints. For mandatory tasks, resources are forcibly locked, and for collaborative tasks that cannot be executed, they are automatically transferred to neighboring satellites. This allows the satellite to transform sensing information into action plans without waiting for ground instructions. This not only ensures the feasibility of new tasks under resource-constrained conditions but also breaks through the single-satellite capability bottleneck through inter-satellite collaboration. This eliminates the impact of ground-satellite communication delays, improves the utilization rate of satellite observation resources, the frequency of target observation, and the timeliness of data acquisition, and enhances the efficiency of remote sensing satellites in responding autonomously to sudden anomalies on orbit.

[0023] 2. In this application, when neighboring satellites confirm that their observation range covers the target observation area of ​​the target task in the collaboratively planned task set, the task is marked as a high-priority fixed node, and the resources required for imaging and processing are forcibly locked. This node is then added to the set of tasks to be planned and used for subsequent planning. This transforms cross-satellite task requests into irrevocable hard constraints, ensuring that high-value tasks discarded by the original satellite receive resource guarantees on neighboring satellites. This avoids being rejected again due to resource competition from neighboring satellites, thus achieving seamless relay at the constellation level and improving the collaborative capture rate and execution certainty for urgent and unexpected tasks.

[0024] 3. This application extends the search range of mission execution resources from local to the entire constellation network by using inter-satellite links to continuously relay the mission to the next node when the observation area cannot be covered by adjacent satellites, until the target execution satellite is found. This ensures that even when the current satellite is unreachable, emergency missions can eventually find a suitable execution node through multi-hop optimization, thereby improving the constellation system's wide-area response capability and execution success rate for time-sensitive missions. Attached Figure Description

[0025] Figure 1 This is a flowchart illustrating an autonomous planning method for Internet remote sensing satellite missions based on intelligent perception and feedback, as described in this application. Figure 2 This is another flowchart illustrating the autonomous planning method for Internet remote sensing satellite missions based on intelligent perception and feedback in the embodiments of this application; Figure 3 This is an exemplary hardware structure diagram of an autonomous planning system for Internet remote sensing satellite missions in this application embodiment. Detailed Implementation

[0026] The terminology used in the following embodiments of this application is for the purpose of describing particular embodiments only and is not intended to be limiting of this application. As used in the specification and appended claims of this application, the singular expressions “a,” “an,” “the,” “the,” “the,” and “this” are intended to include the plural expressions as well, unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in this application refers to and includes any or all possible combinations of one or more of the listed items.

[0027] Hereinafter, the terms "first" and "second" are used for descriptive purposes only and should not be construed as implying or suggesting relative importance or implicitly indicating the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature, and in the description of the embodiments of this application, unless otherwise stated, "multiple" means two or more.

[0028] In related technologies, responding to emergencies mainly employs ground-triggered missions or on-board autonomous triggering modes. Ground-triggered missions are limited by the distribution of telemetry and control windows and link transmission delays, resulting in lag in command uploading and making it difficult to capture fleeting target dynamics. On-board triggering missions, with their single-satellite closed-loop observation technology, suffer from limited satellite observation range and low timeliness of single-satellite data acquisition, making it difficult to meet the requirements for efficient closed-loop response to sudden targets in complex scenarios.

[0029] In this embodiment, the satellite can process images in real time on orbit and compare them with baseline data to detect anomalies, thereby automatically generating new observation tasks based on the degree of anomaly. More importantly, when constructing the mission planning model, not only is the energy consumption of the imaging operation itself considered, but the energy consumption and storage space consumption required for on-board data processing and information extraction are also accurately calculated based on the amount of data to be processed. These dynamic computational loads are then incorporated into the mission planning resource constraints for unified solution. This mechanism allows the satellite to transform perceived information into action plans without waiting for ground commands, and ensures the strict executability of new tasks under conditions of limited computing and storage resources. This eliminates the impact of ground-to-satellite communication delays and improves the on-orbit autonomous response efficiency of remote sensing satellites to sudden anomalies.

[0030] Figure 1 This is a flowchart illustrating the autonomous planning method for Internet remote sensing satellite missions based on intelligent sensing and feedback, as described in this application, including the following steps: S101. Obtain the execution results of the observation task with planning control attributes.

[0031] Among them, the observation task with planning and control attributes refers to the observation task that the satellite is currently performing. The attributes include parameters that guide the subsequent processing logic (such as task type, priority, etc.). The execution result refers to the data product generated after the physical execution of the task, including various types of image products and information product data obtained after camera imaging and on-board processing. Image product data refers to standard data products with geographic coordinate information and physical meaning generated after preprocessing such as radiometric correction and geometric correction of the original image (usually corresponding to L1 or L2 level products defined in the field of remote sensing). Information product data is a data product containing target location, related attributes, etc., generated after information extraction and processing of image products. Target attributes include, but are not limited to, the type and level of the target.

[0032] Specifically, this step is usually divided into two stages: data processing and information extraction.

[0033] First, the satellite payload generates a raw binary data stream at level L0 when performing its mission.

[0034] Subsequently, the onboard processing unit invokes preset algorithms to process the data, including radiometric correction to eliminate sensor distortion, geometric correction to eliminate distortion and assign geographic coordinates, generating standard information product data in a two-dimensional raster format.

[0035] Next, information is extracted from the processed image data. Computer vision or artificial intelligence algorithms are used to analyze the image to identify content of interest to specific industries. During this process, through feature extraction (such as edge and contour recognition) and subsequent analysis, product data containing information such as target quantity and location is ultimately output.

[0036] This process is completed entirely autonomously on the satellite, realizing the conversion from image data to decision-making information.

[0037] In some embodiments, the above steps can be implemented in multiple ways: Optionally, a deep learning-based target detection workflow can be adopted: First, the initial remote sensing image data is input into the on-board GPU-accelerated preprocessing module to perform radiometric and geometric corrections to generate orthophotos; second, the orthophoto slices are input into the lightweight YOLO series target detection network; finally, the network outputs the bounding box coordinates and confidence scores of the targets, which are then post-processed to obtain information product data.

[0038] It is understandable that this can also be achieved using a processing method based on multimodal data fusion, which is not limited here.

[0039] S102. Calculate the attribute difference between the current target attribute data and the preset normal baseline data.

[0040] Among them, the preset normal baseline data refers to the characteristic indicators of the target area under normal conditions, which are pre-set based on historical observation data statistics, ground intelligence compilation or domain expert experience. This data is usually uploaded to the satellite in the form of a configuration file. The attribute difference is used to represent the degree of deviation between the current observation results and the normal state. It can be the absolute difference in numerical value, the relative proportion or the coefficient of variation in statistics. The preset judgment threshold is the critical value used to distinguish between normal fluctuations and abnormal situations. This threshold is usually derived by those skilled in the art through a large number of sample tests or simulations based on mission requirements and false alarm rate requirements.

[0041] Specifically, the satellite compares the current target attribute data extracted from S101 with preset normal baseline data stored in the onboard database. The comparison method depends on the type of attribute data: for numerical attributes, the difference or ratio between the current value and the baseline value is calculated; for character attributes (such as type, model), the character values ​​are compared to see if they match.

[0042] S103. If the attribute difference exceeds the preset judgment threshold, copy the initial task attribute information contained in the initial observation task as the basic information of the new task.

[0043] Among them, the preset judgment threshold refers to the numerical boundary used to define the abnormal and normal values. This threshold is preset based on the distribution pattern of historical data and the sensitivity of the task. The initial task attribute information refers to the metadata describing the execution requirements and control logic of the initial observation task, including but not limited to the task ID, observation area coordinates, sensor type, task priority parameters, and planning control identifier parameters. The new task basic information is used to represent the initial state data of the newly generated observation task and is constructed based on a copy of the initial task attribute information.

[0044] Specifically, the attribute difference calculated in S102 is compared with a preset judgment threshold. If the attribute difference is less than or equal to the threshold, the current scene is considered normal and no additional operation is required; if the attribute difference exceeds the threshold, it is determined that an abnormal target has been found, triggering a new task generation process.

[0045] To ensure that the new task can inherit the basic context information of the original task (such as who the observation target is and what camera is used to take the picture), the attribute list of the initial observation task is directly read, and the key parameters (such as priority, control identifier, observation geometric constraints, etc.) are completely copied as the draft or basic information of the new task.

[0046] S104. Based on the preset level mapping strategy, determine the priority increment value corresponding to the abnormal state level, add the priority increment value to the task priority parameter in the new task basic information, and modify the planning control identifier parameter to indicate the effective state of mandatory execution or inter-satellite coordination, and generate a new observation task.

[0047] Among them, the preset level mapping strategy refers to a predefined rule table or function relationship used to map the severity of anomalies to the increase in task priority. This strategy is usually formulated by task planning experts based on the urgency of the business. The anomaly status level is the degree of difference between the current target attribute data and the preset normal baseline data, used to quantify the severity of the current anomaly. The priority increment value refers to the value that needs to be added on the basis of the original task priority. The mandatory execution status in the planning control identifier parameters indicates that the task must be selected during planning. The inter-satellite coordination status indicates that the task is allowed to be sent to other satellites for execution.

[0048] Specifically, the first step is to calculate the level of the abnormal state, for example, to calculate how many times the current target number is greater than the normal value.

[0049] Then, the preset level mapping strategy is queried. Based on the lookup result, the priority increment value is determined and added to the original priority of the new task, giving the new task a higher execution weight.

[0050] Meanwhile, to ensure that this important task triggered by the anomaly can be effectively executed, the planning control flags were modified to mark them as mandatory to prevent them from being taken over by regular tasks, or marked as effective for inter-satellite coordination so that neighboring satellites can be sought to help when single-satellite resources are insufficient.

[0051] After these parameter adjustments are completed, a complete new observation task object is finally generated.

[0052] S105. Calculate the amount of image data to be processed based on the observation range of the new observation mission, and calculate the energy consumption and storage space consumption required for on-board processing of the image data to be processed based on the preset unit data volume resource consumption coefficient.

[0053] Among them, the observation range refers to the geographical area or scan strip length to be covered by the new mission plan; the amount of image data to be processed is used to indicate the size of the raw data expected to be generated when the observation mission is completed; the preset unit data volume resource consumption coefficient refers to the average power consumption and storage space occupied by processing a unit size of image data, and this coefficient is an empirical value derived from ground test data or on-orbit historical operation statistics; on-board processing specifically refers to the entire process of data processing and information extraction in S101; the energy consumption value refers to the additional power required for the on-board processing unit to run the algorithm; and the storage space consumption value refers to the capacity required to store the processing results and intermediate data.

[0054] Specifically, the first step is to estimate the total number of pixels and corresponding data volume that the imaging will generate, based on the observation range of the new mission (such as latitude and longitude) and camera resolution. Unlike traditional planning that only considers the energy consumption of the satellite imaging stage, this step pays special attention to the energy consumption of the onboard intelligent processing stage.

[0055] By using a preset resource consumption coefficient (e.g., X joules of electrical energy are required for CPU / GPU computation and Y GB of storage space for intermediate results and final products per 1 GB of data processed), the amount of image data to be processed is multiplied by this coefficient to calculate the energy consumption and storage space consumption required for on-board processing of the task. This step transforms the soft algorithmic processing into hard resource constraints, providing data support for subsequent planning.

[0056] In some embodiments, the above steps can be implemented in multiple ways: Optionally, a linear model-based calculation method is used: First, based on the observation area S and ground resolution, the data volume is calculated as: observation area / ground resolution² × bit depth; second, the preset unit energy consumption coefficient and unit storage coefficient are read; finally, the energy consumption is calculated as: energy volume × unit energy consumption coefficient, and storage consumption is calculated as: data volume × unit storage coefficient.

[0057] Understandably, estimation methods based on historical task analogies can also be used, and this is not limited here.

[0058] S106. If the planning control flag parameter of the new observation task indicates that it is a mandatory execution state, then lock the execution time window of the new observation task and reserve the required energy and storage space.

[0059] Among them, "mandatory execution status" means that the task has the highest necessity for execution and cannot be abandoned during the planning process; "locked execution time window" means that the start and end time of the task is forcibly defined on the timeline, and other tasks are prohibited from occupying it; "reserved energy and storage space" means that the amount of resources required for the task (including imaging and processing consumption) is deducted from the satellite's current available resource pool to ensure that there are sufficient resources available.

[0060] Specifically, check the planning control indicators for new observation tasks. If the indicator is mandatory, initiate the resource preemption process.

[0061] First, the start and end time window of the task is calculated based on the orbital parameters and marked as a fixed node, exclusively locking this time period on the timeline. If this time period is already occupied by a regular task, the conflicting regular task is directly removed.

[0062] Secondly, based on the total resource consumption (imaging + processing) calculated by S105, the corresponding values ​​are pre-deducted from the satellite's remaining energy and storage space. By adopting a strategy of locking in before planning, it is ensured that the mandatory tasks become unshakable hard constraints when constructing the planning model, avoiding their abandonment by the optimization algorithm due to resource competition.

[0063] In some embodiments, the above steps can be implemented in multiple ways: An optional implementation based on timeline conflict resolution is as follows: First, traverse the currently scheduled tasks and find the task whose time window overlaps with the new task; second, remove the overlapping task from the queue and release its resources; finally, insert the new task into the queue and mark it as "locked".

[0064] It is understandable that a forced insertion method based on priority weights can also be used to achieve this, and no limitation is made here.

[0065] S107. Add new observation tasks to the set of tasks to be planned, and construct a task planning model with the goal of maximizing observation benefits.

[0066] The set of tasks to be planned refers to the list of all observation tasks currently awaiting execution; the task planning model is a mathematical model used to describe the task scheduling problem, which usually includes decision variables, objective function and constraints. Constraints include that the satellite's remaining energy is greater than the sum of imaging energy consumption and on-board processing energy consumption, and that the satellite's remaining storage space is greater than the sum of imaging data volume and on-board processing data volume; maximizing observation benefits means that the goal of planning is to maximize the total value of the selected tasks; the satellite's remaining energy and remaining storage space represent the current available power of the satellite's batteries and the remaining capacity of its memory, respectively.

[0067] Specifically, the new observation task generated by S104 is first inserted into the current set of tasks to be planned.

[0068] Next, a mathematical programming model (such as an integer programming model or a constraint satisfaction model) is constructed. In this model, the objective function is set to maximize the total revenue of all selected tasks. The key lies in the construction of the constraints, which not only consider conventional imaging resource consumption but also forcibly incorporate the onboard processing energy consumption and storage space consumption calculated by S105. That is, the constraints explicitly require that the satellite's remaining energy must be sufficient to cover the sum of the energy consumption of both imaging and AI computation operations; similarly, the remaining storage space must also be able to accommodate the sum of imaging data and processing data. This dual constraint ensures the physical feasibility of the planning results and avoids task execution interruptions due to neglecting processing energy consumption.

[0069] In some embodiments, the above steps can be implemented in multiple ways: An optional modeling approach based on graph theory is as follows: First, treat each task as a node in a graph, and the connections between tasks as edges; second, assign weights (revenue) and costs (energy consumption + storage) to each node; finally, under the premise of satisfying the total cost constraint, find the largest independent set with the largest sum of weights.

[0070] It is understandable that this can also be achieved using a modeling approach based on constraint satisfaction problems (CSP), which is not limited here.

[0071] S108. Solve the task planning model to generate a set of tasks to be executed on a single planet, and identify the set of remaining tasks that have not been selected from the set of tasks to be planned.

[0072] Among them, solving refers to the process of using a specific algorithm to find the decision variable values ​​that satisfy all constraints and make the objective function optimal; the set of tasks to be executed by the satellite refers to the optimal solution output by the planning model, that is, the list of tasks to be executed by the satellite; the set of remaining tasks that were not selected refers to the list of tasks that were discarded by the planning algorithm in the set of tasks to be planned due to insufficient resources, time conflicts, or low priority.

[0073] Specifically, the onboard algorithms (such as heuristic algorithms and dynamic programming algorithms) are invoked to solve the mission planning model constructed by S107. During the search of the solution space, the algorithm continuously optimizes different combinations of tasks and checks whether the dual constraints of energy and storage are met.

[0074] Ultimately, the algorithm outputs the optimal task combination, which constitutes the set of tasks to be executed on this satellite. Simultaneously, a difference operation is performed between the original set of tasks to be planned and the set of tasks to be executed on this satellite, filtering out those tasks that failed to enter the execution sequence, forming a set of unselected remaining tasks.

[0075] In some embodiments, the above steps can be implemented in multiple ways: Optionally, a greedy algorithm can be used to solve the problem: First, sort all tasks in descending order of priority; second, try adding tasks to the execution set in turn, adding them if the constraints are met, otherwise assigning them to the remaining set; finally, output two sets.

[0076] Optionally, a genetic algorithm can be used to solve the problem: First, find the individual with the highest fitness (i.e., the optimal execution sequence) through iterative evolution; second, decode the individual to obtain the set of tasks to be executed on this planet; finally, traverse the original task list and mark the tasks that are not in the optimal sequence as the remaining task set.

[0077] It is understandable that ensemble learning algorithms or deep learning algorithms can also be used to achieve this, and no specific restrictions are imposed here.

[0078] S109. Traverse the remaining task set and divide the tasks whose planning control identifier parameters indicate that the inter-satellite cooperation is in an effective state into a collaborative planning task set.

[0079] The remaining task set comes from the unselected tasks in S108; the inter-satellite collaboration effective status means that the planning control flag in the task attribute is set to effective (usually set in S104 due to abnormal triggering), indicating that although the task cannot be executed by this satellite, it has high timeliness or value, allowing and encouraging the seeking of assistance from other satellites through the inter-satellite network; the collaborative planning task set refers to the final list of tasks that need to be sent to neighboring satellites.

[0080] Specifically, this step aims to salvage high-value emergency missions from those abandoned by the planet.

[0081] The remaining task set generated by S108 is iterated through one by one, and the planning control flag parameters of each task are checked. If a task is found to be marked as having effective inter-satellite coordination, it means that the system determines that the task needs to be given priority (such as sudden abnormal observations) and cannot be discarded directly. These tasks are extracted and added to the coordinated planning task set. For ordinary low-priority tasks that are neither selected nor have a coordination flag, they are officially confirmed to be discarded or postponed in this step.

[0082] S110. Output the set of tasks to be executed by this satellite for execution by this satellite.

[0083] After the set of tasks to be executed by this satellite is output to the satellite management system, the sequence of execution actions is obtained through instruction arrangement, and then the actions of each subsystem are controlled to complete the corresponding actions.

[0084] S111, The collaborative planning task set is sent to adjacent satellites for planning via inter-satellite communication links.

[0085] Inter-satellite communication links refer to communication channels established between satellites for data transmission, such as laser links or microwave links; the collaborative planning task set consists of high-value unselected tasks selected from S109; neighboring satellites refer to other satellite nodes that have established stable communication connections with this satellite at the current moment and are in the same constellation network.

[0086] Specifically, the collaborative planning task set obtained from S109 is encapsulated into a standard data packet (containing metadata such as task ID, observation area, deadline, and priority). Then, the inter-satellite link communication module is invoked to identify currently visible neighboring satellite nodes. This data packet is then sent to the neighboring satellite via the inter-satellite link. Upon receiving the task packet, the neighboring satellite incorporates it into its own planning process (i.e., triggers the neighboring satellite to execute a similar planning method), thereby utilizing the neighboring satellite's orbital coverage advantage or remaining resources to complete urgent tasks that the local satellite cannot accomplish, achieving a constellation-level collaborative closed loop.

[0087] In some embodiments, the above steps can be implemented in multiple ways: Optionally, a topology-aware directional transmission method is used: First, the dynamic topology table of the constellation network maintained on the satellite is read to determine the neighboring satellite with the best current link quality; second, a point-to-point connection handshake is established with the neighboring satellite; finally, the cooperative task packet is reliably transmitted to the neighboring satellite through the directional link.

[0088] It is understandable that a forwarding method based on a smart routing protocol can also be used to achieve this, and this is not limited here.

[0089] In the above embodiments, the satellite can process images and analyze and detect anomalies in real time on orbit, and then automatically generate new high-priority observation tasks. These tasks are then incorporated into a planning model that considers both imaging and processing resource constraints. For mandatory tasks, resources are forcibly locked, and for collaborative tasks that cannot be executed, they are automatically transferred to neighboring satellites. This allows the satellite to transform sensing information into action plans without waiting for ground instructions. This not only ensures the feasibility of new tasks under resource-constrained conditions, but also breaks through the single-satellite capability bottleneck through inter-satellite collaboration, thereby eliminating the impact of ground-satellite communication delays. This improves the utilization rate of satellite observation resources, the frequency of target observation, and the timeliness of data acquisition, and enhances the efficiency of remote sensing satellites in responding autonomously to sudden anomalies on orbit.

[0090] In other embodiments of this application, when faced with complex and ever-changing target characteristics and limited satellite resources, anomalies may be missed or high-value emergency missions may be abandoned due to conflicts. The autonomous planning method for Internet remote sensing satellite missions based on intelligent perception and feedback provided in this application can achieve accurate perception and ensure execution through multi-dimensional anomaly judgment logic and a mandatory resource locking mechanism for required missions.

[0091] like Figure 2 The diagram shown is another flowchart illustrating the autonomous planning method for Internet remote sensing satellite missions based on intelligent sensing and feedback provided in this application, including the following steps: S201. Obtain the execution results of observation tasks with planning and control attributes; the execution results include various types of information product data obtained after camera imaging and on-board processing.

[0092] S202. Calculate the attribute difference between the current target attribute data and the preset normal baseline data.

[0093] Steps S201-S202 and Figure 1 Steps S101-S102 in the illustrated embodiment are similar and can be found in the descriptions of steps S101-S102, which will not be repeated here.

[0094] S203. Identify the task type of the initial observation task.

[0095] The initial observation task refers to the observation activities that the satellite is currently performing, which serve as the basis for anomaly determination; the task type refers to the specific business scenario classification that the task targets, including but not limited to target detection, target recognition, target tracking and change detection, etc., and this type identifier is usually included in the task's attribute field.

[0096] Specifically, since different types of tasks have different definitions of anomalies (e.g., detecting fluctuations in the number of observations, tracking the matching of observed features), it is essential to first determine which type of task is being processed. The attribute fields of the initial observation task are read, the task type code is parsed, and the control flow is directed to the corresponding decision logic branch based on the parsing result.

[0097] In some embodiments, the above steps can be implemented in multiple ways: Optionally, a task ID prefix-based identification method is used: First, read the task ID string; second, extract the prefix characters of the ID (e.g., "DET" represents detection, "TRK" represents tracking); finally, determine the task type according to the prefix mapping table.

[0098] It is understandable that this can also be achieved using a method based on instruction code parsing, and this is not a limitation here.

[0099] S204. When the task type is target detection task, calculate the coefficient of variation between the number of detected targets and the preset number of normal targets.

[0100] Among them, the target detection task refers to the task with the main purpose of counting the number and location of targets in a specific area; the number of detected targets is the real-time statistical value output by the on-board processing in S201; the preset normal target number refers to the baseline value set in advance based on the statistical average value of the historical long-term observation data of the area, which is derived from the analysis of historical big data; the coefficient of variation is used to represent the degree of fluctuation of the current observation number relative to the historical baseline, and is usually expressed as the ratio of the standard deviation to the mean or the relative error.

[0101] Specifically, when S203 identifies the mission type as target detection, it enters the quantity fluctuation analysis logic. The core indicator the satellite focuses on is whether the number of targets has changed significantly. It reads the total number of targets detected in the current image and retrieves the preset normal target number for the observation area from the parameter file. Then, it uses a mathematical formula to calculate the degree of difference between the two, i.e., the coefficient of variation. The coefficient of variation can eliminate the influence of orders-of-magnitude differences (for example, adding 5 targets has different implications when the base is 10 versus 100), thus reflecting the severity of the fluctuation.

[0102] In some embodiments, the above steps can be implemented in multiple ways: Optionally, a calculation method based on relative error is as follows: First, obtain the current quantity Ncurr and the baseline quantity Nbase; second, calculate the absolute difference |Ncurr−Nbase|; finally, calculate the coefficient of variation = |Ncurr−Nbase| / Nbase.

[0103] It is understandable that a normalized calculation method based on the historical extreme value range can also be used to achieve this, and no limitation is made here.

[0104] S205. If the coefficient of variation is greater than the preset detection threshold, it is determined to be abnormal and a new task is triggered.

[0105] Among them, the preset detection threshold is a critical value used to define whether the quantity fluctuation is within the normal range. This threshold is usually preset according to the business tolerance and false alarm rate requirements; triggering new task creation means generating a signal to start the subsequent task generation process.

[0106] Specifically, the coefficient of variation calculated by S204 is compared with a preset detection threshold. If the coefficient of variation is less than or equal to the threshold, it indicates that the fluctuation in the number of targets is within the normal range, and it is judged as normal, so no action is taken.

[0107] If the coefficient of variation is greater than the preset detection threshold, it indicates that the number of targets has experienced an abnormal and drastic increase or decrease, and the area is determined to be abnormal. At this time, the abnormality flag is set, and a new task creation process is triggered to revisit the area for confirmation or continuous monitoring.

[0108] In some embodiments, the above steps can be implemented in multiple ways: Optionally, a threshold-based triggering method is used: First, multiple threshold levels are set (mild, moderate, severe); second, it is determined which interval the coefficient of variation falls into; finally, if the minimum threshold is exceeded, creation is triggered, and different priority parameters are set according to the interval level.

[0109] It is understandable that a dynamic determination method based on adaptive thresholds can also be used to achieve this, and this is not limited here.

[0110] S206. If the task type is target recognition task and the coefficient of variation is less than or equal to the preset detection threshold, further determine whether the identified target attributes contain preset important target attributes.

[0111] Among them, the target identification task refers to not only focusing on the target's location, but also confirming the target's specific model or category; the preset important target attributes refer to a predefined list of high-value target features, which is preset and stored on the satellite.

[0112] Specifically, when the task type is target recognition, the criteria for anomaly detection are not limited to quantity. Even if the coefficient of variation calculated by S204 is within the acceptable range (i.e., the number of targets is normal), it cannot be directly determined that there are no anomalies. Further in-depth examination of the specific attributes (such as category and model) of each identified target is required. The attribute labels of all targets in the current image are traversed and matched against a pre-defined list of important target attributes. This step aims to discover targets that are few in number but of extremely high value, preventing important targets from being obscured by sheer quantity.

[0113] In some embodiments, the above steps can be implemented in multiple ways: Optionally, a whitelist-based matching method can be used: First, load a list of important target models; second, iterate through all currently identified target models; and finally, check if the current model exists in the list.

[0114] Optionally, an attribute feature vector matching method is used: First, obtain the feature vector template of the preset important target; second, calculate the similarity between the feature vector of the current target and the template; finally, if the similarity is higher than the threshold, it is considered to contain important attributes.

[0115] It is understandable that this can also be achieved using attribute reasoning based on a rule engine, and this is not a limitation here.

[0116] S207. If it contains preset important target attributes, it is judged as an anomaly and a new task is created.

[0117] Among them, the inclusion of preset important target attributes indicates that high-value or sensitive targets have been discovered in the current observation field.

[0118] Specifically, if any target's attribute matches a preset important target attribute during the S206 check, the current scenario will be immediately determined to be abnormal. At this point, regardless of whether the quantity fluctuates, the new task creation process will be triggered directly.

[0119] This ensures immediate response to critical targets, enabling satellites to generate follow-up tracking or detailed investigation missions for these high-value targets.

[0120] S208. If the task type is a target tracking task, calculate the target feature difference between the current target feature value and the target feature value to be tracked.

[0121] Among them, the target tracking task refers to the task of continuously locking and monitoring a specific moving target; the current target feature value is the target feature extracted from the current image (such as category, model, position, length, width, speed, direction, etc.); the target feature value to be tracked is the target template feature locked in the previous time series or the initial time; the target feature difference is used to quantify the similarity or distance between the current target and the tracked target.

[0122] Specifically, when the task type is target tracking, the core logic lies in confirming whether what is currently seen is the target to be tracked. This involves obtaining the previous time-series observation results for the same target (i.e., the template data of the target to be tracked). Next, the feature difference between the current target attribute data and the template feature values ​​is calculated. This calculation process typically involves calculating the Euclidean distance or cosine similarity between two feature vectors, or calculating the weighted coefficient of variation of shape and motion attributes.

[0123] In some embodiments, the above steps can be implemented in multiple ways: An optional method for calculating cosine similarity is as follows: First, calculate the dot product of the two vectors; second, divide by the product of the magnitudes of the two vectors to obtain the cosine value; finally, convert the cosine value into a difference (e.g., 1−cosθ).

[0124] It is understandable that a similarity scoring method based on Siamese networks can also be used, and this is not limited here.

[0125] S209. If the difference in target features is less than the preset tracking threshold, the tracking is determined to be successful and a new task is created.

[0126] The preset tracking threshold is a critical similarity value used to determine whether two targets are the same object, which is obtained based on a large number of samples trained. It is usually set in advance by those skilled in the art based on the performance curve of the AI ​​model. Successful tracking means confirming that the current target is the target to be tracked. The purpose of triggering the creation of a new task here is to maintain the continuous lock on the target and prevent loss of tracking.

[0127] Specifically, the feature differences calculated by S208 are compared with a preset tracking threshold. If the difference is less than the preset threshold, it indicates that the target's identity has been confirmed and tracking is successful. At this point, although the scene itself may not be abnormal (i.e., the target is still there), a new task needs to be created to achieve continuous tracking. This new task will be generated based on the target's latest position, thereby ensuring that the satellite can continue to observe the moving target in the next moment.

[0128] S210. When the task type is change detection task, calculate the proportion of the changed area to the total observed area.

[0129] Among them, change detection is to detect changes in targets in images before and after the transition; change area refers to the area in images before and after the transition; area ratio is the proportion of the change area to the total observation area.

[0130] Specifically, when the task type is change detection, the focus is on changes in the target. First, an on-board target feature extraction algorithm (such as a deep learning model) is used to obtain the target area's range and outline. Then, the changes are compared with the previous feature extraction results uploaded to the satellite to output change patches. Next, the area (number of pixels) of the change patches is counted, and their percentage of the total observed area is calculated.

[0131] In some embodiments, the above steps can be implemented in multiple ways: Optionally, change detection based on previous and subsequent images: First, upload previous images of the area to be detected to the satellite, then send the previous images and the current images together into the change detection model to output change patches; second, calculate the sum of the geometric areas of all change patches; finally, divide by the total geometric area of ​​the observed area.

[0132] S211. If the area ratio is greater than the preset area change threshold, it is judged as abnormal and a new task is triggered.

[0133] The preset area change threshold is the boundary used to define whether the change is significant. This threshold is set according to the sensitivity of the monitored object; a judgment of an anomaly indicates that a significant change event has occurred.

[0134] Specifically, the region proportion calculated by S210 is compared with a preset threshold. If the proportion is greater than the threshold, it indicates a significant change and is judged as an anomaly. At this point, a new task is triggered to perform a more frequent or higher-resolution detailed investigation of this significantly changed region.

[0135] In some embodiments, the above steps can be implemented in multiple ways: Optionally, a ratio-based triggering method is used: First, load the threshold parameter; second, compare the current ratio with the threshold; finally, if the limit is exceeded, generate an exception event report and trigger the task.

[0136] It is understandable that this can also be achieved by using a triggering method based on the level of semantic change, and this is not limited here.

[0137] S212. In response to triggering the creation of a new task, copy the task priority parameters, planning control identifier parameters, and latest execution time parameters of the initial observation task, and extend the latest execution time parameter by one task planning cycle as the basic information of the new task.

[0138] Among them, responding to triggering new task creation refers to the action performed after any of the above-mentioned decision logics (S205, S207, S209, S211) issues a trigger signal; the latest execution time parameter refers to the deadline for the task to be executed; the task planning cycle duration refers to the time interval for the satellite to perform a complete task planning.

[0139] Specifically, once a new task creation is triggered, the basic information for the new task needs to be constructed. To ensure task continuity, attribute inheritance and extension operations are performed: the task priority, planning control flag, and latest execution time parameter of the initial observation task are directly copied. In particular, the latest execution time parameter is extended by one task planning cycle. This is done to give the new task validity in the future time period, preventing it from expiring due to using the old deadline, thus solving the continuity problem of sudden tasks in the time dimension and preventing the high-value target tracking chain from breaking due to the planning cycle deadline.

[0140] In some embodiments, the above steps can be implemented in multiple ways: Optionally, an implementation based on object property operations is as follows: First, obtain the original task object Task_Old; second, create Task_New and assign it Task_New.Priority = Task_Old.Priority; finally, Task_New.Deadline = Task_Old.Deadline + CycleTime.

[0141] S213. Based on the preset level mapping strategy, determine the priority increment value corresponding to the abnormal state level, add the priority increment value to the task priority parameter in the new task basic information, and modify the planning control identifier parameter to indicate the effective state of mandatory execution or inter-satellite coordination, and generate a new observation task.

[0142] Step S213 and Figure 1 Step S104 in the illustrated embodiment is similar and can be found in the description of step S104, which will not be repeated here.

[0143] In some embodiments, after generating a new observation task based on a preset level mapping strategy, for target tracking tasks, further validity verification and dynamic position correction steps based on temporal feature comparison can be performed to ensure the continuity and accuracy of the tracking task.

[0144] Specifically, when the task type of the new observation task generated in S213 is identified as target tracking, the deep verification process begins.

[0145] First, the previous time-series observation results for the same target (i.e., template data of the target to be tracked) are obtained, and the target feature values ​​are acquired. These feature values ​​typically include the target's category features, geometric features (such as aspect ratio and area), and motion features (such as velocity vector). Next, the feature difference between the current target attribute data and the target feature values ​​from the previous time series is calculated. This calculation process usually involves calculating the Euclidean distance or cosine similarity between two feature vectors, or calculating the weighted coefficient of variation of shape and motion attributes. The preset tracking lock threshold is a critical similarity value used to determine whether two targets are the same object, obtained based on a large number of samples trained on it, and is usually preset by those skilled in the art based on the performance curve of the AI ​​model.

[0146] Subsequently, a binary branch logic is executed based on the feature difference degree: In the first case, if the feature difference degree is greater than the preset tracking and locking threshold, it indicates that the currently detected target does not match the features of the target to be tracked, and it is determined that the target has been lost or lost. At this time, in order to avoid wasting on-board resources due to invalid observations, a circuit breaker operation is immediately executed to terminate the subsequent generation process of the current new observation task and completely delete the basic information of the created task to prevent the invalid task from entering the subsequent planning queue.

[0147] In the second scenario, if the feature difference is less than or equal to the preset tracking and locking threshold, it indicates that the target's identity has been confirmed and tracking is successful. At this point, since the target is in motion and its position has changed, the latest target position coordinates are extracted from the current target attribute data. Subsequently, these latest coordinates are used to update the task position parameters in the new observation task. The advantage of this approach is that it corrects for positional deviations caused by target movement, ensuring more accurate prediction of the target's current position during subsequent planning.

[0148] The above technical steps introduce a feature comparison-circuit / correction mechanism, which on the one hand, stops losses in time when tracking fails, avoiding the waste of satellite energy and storage space by invalid tasks; on the other hand, dynamically corrects position parameters when tracking is successful, ensuring that the satellite can continuously lock onto and observe moving targets, thereby improving the success rate of mission execution and resource utilization efficiency in highly dynamic scenarios.

[0149] S214. Calculate the amount of image data to be processed based on the observation range of the new observation mission, and calculate the energy consumption and storage space consumption required for on-board processing of the image data to be processed based on the preset unit data volume resource consumption coefficient.

[0150] S215. If the planning control flag parameter of the new observation task indicates that it is a mandatory execution state, then lock the execution time window of the new observation task and reserve the required energy and storage space.

[0151] S216. Add new observation tasks to the set of tasks to be planned.

[0152] Steps S214-S216 and Figure 1 Steps S105-S107 in the illustrated embodiment are similar and can be found in the descriptions of steps S105-S107, which will not be repeated here.

[0153] S217. Obtain the list of regular tasks that have been pre-queued in the current set of tasks to be planned, identify and remove target regular tasks in the list of regular tasks that have overlapping or conflicting execution time windows with the new observation task.

[0154] Among them, the regular task list refers to ordinary observation tasks with low priority and not required in the set to be planned; the execution time window refers to the start and end time period occupied by the task plan; overlap and conflict refer to the time windows of two tasks intersecting on the time axis, which makes it physically impossible to execute them at the same time.

[0155] Specifically, when a new observation task is marked as mandatory, the first step is to retrieve the list of pre-queued regular tasks from the current set of tasks to be planned. Then, the execution time windows of these regular tasks are checked one by one to see if they overlap with the window of the new observation task. Since a satellite can only perform one observation action at a time (or cannot perform them consecutively due to maneuver time constraints), any temporal overlap is an unacceptable physical conflict. Therefore, all target regular tasks with temporal overlap conflicts are identified and removed directly from the list to free up time slots for the mandatory task.

[0156] In some embodiments, the above steps can be implemented in multiple ways: An optional implementation based on a conflict graph involves: first, constructing a task conflict graph; second, identifying all regular task nodes connected to the new task node; and finally, deleting these nodes.

[0157] It is understandable that a bitmap detection method based on time axis occupancy can also be used to achieve this, and no limitation is made here.

[0158] S218. Mark the new observation task as a fixed node and insert it into the timeline of the task planning, locking the imaging time window and the total amount of resources required.

[0159] Among them, fixed nodes refer to task nodes whose position and status cannot be changed during the planning process; locking means marking resources as occupied and prohibiting other tasks from preempting them; the total amount of resources required includes imaging energy consumption, on-board processing energy consumption, imaging data volume, and on-board processing data volume.

[0160] Specifically, after clearing time conflicts, the new observation mission is formally inserted into the mission planning timeline and its status is marked as fixed. This means that in subsequent optimization algorithms, this mission will always be selected and will not be eliminated. Simultaneously, based on the resource consumption values ​​calculated by S215, the corresponding energy and storage space are pre-deducted from the satellite's available resource pool. This operation is equivalent to allocating a dedicated area for this mission, ensuring that regardless of subsequent planning adjustments, the resources required for the mission are physically guaranteed.

[0161] In some embodiments, the above steps can be implemented in multiple ways: Optionally, a constraint injection-based implementation is as follows: First, generate a hard constraint rule "Task_New must be executed in Time_T"; second, inject this rule into the planning engine; and finally, reserve resource quotas.

[0162] It is understandable that a priority-based locking approach can also be used, and this is not a limitation here.

[0163] S219. Construct a task planning model with the goal of maximizing observation gains.

[0164] Specifically, a task planning model (such as an integer programming model) is constructed. In this model, the objective function aims to maximize the reward. The key is that, since the new task is already marked as a fixed node in S218, the solution space is strongly constrained when constructing the model. That is, when searching for the optimal solution, the algorithm no longer decides whether to execute the new task (because it is already fixed), but instead, around this fixed node, decides which regular tasks can be inserted into the remaining resources and time gaps. This approach narrows the search range and mathematically guarantees the absolute execution right of the mandatory task.

[0165] In some embodiments, the above steps can be implemented in multiple ways: An optional implementation based on segmented programming is as follows: First, the time axis is divided into several segments with fixed nodes as boundaries; second, a sub-programming model is constructed in each idle time segment; finally, the solution of the sub-model is concatenated with the fixed node.

[0166] It is understandable that a soft constraint to hard constraint conversion based on a penalty function can also be used to achieve this, and this is not limited here.

[0167] S220. Solve the task planning model to generate a set of tasks to be executed in the current system, and identify the set of remaining tasks that have not been selected from the set of tasks to be planned.

[0168] S221. Traverse the remaining task set and divide the tasks whose planning control identifier parameters indicate that the inter-satellite cooperation is in an effective state into a collaborative planning task set.

[0169] S222. Output the set of tasks to be executed by this satellite for execution by this satellite.

[0170] Steps S220-S222 and Figure 1 Steps S108-S110 in the illustrated embodiment are similar and can be found in the descriptions of steps S108-S110, which will not be repeated here.

[0171] In some embodiments, after completing the task planning solution for the current cycle, tasks in the periodic execution state can be executed based on cross-cycle task propagation and pre-injection operations, thereby achieving continuous autonomous monitoring of long-cycle dynamic targets.

[0172] Specifically, first, the system checks whether the planning and control flag parameters of the new observation task indicate a periodic execution state. If so, it means that the task needs to be executed regularly over a long period of time, rather than a one-time observation. The system reads all the basic information of the current new observation task (such as target location, task type, priority, etc.) and constructs the derived observation task object in memory.

[0173] Next, to ensure the derivative task remains valid in future planning cycles, a time parameter extension operation is performed: the deadline parameter for the derivative observation task is extended by one mission planning cycle. The mission planning cycle duration refers to the time interval between two adjacent planning calculations performed by the satellite mission planning system (e.g., depending on the satellite's orbital period of 90 minutes or a fixed duration set by the system). This parameter is pre-set after comprehensively considering the solution efficiency and quality of the mission planning algorithm, as well as the satellite's response speed to unexpected tasks. This is done to give the derivative task validity in future time periods and prevent it from expiring and being cleared due to using the old deadline.

[0174] Finally, this configured derivative observation task is injected into the satellite's long-term mission repository. When the satellite enters the start time of the next planning cycle, it automatically scans the repository, extracts these tasks, and adds them to the set of tasks to be planned, thereby automatically triggering the next round of observation planning without ground intervention.

[0175] The above technical steps solve the problem of continuity of periodic tasks in the time dimension by triggering the current task, reversing the future task, and pre-injecting the next cycle through an automated chain. This avoids the task chain being broken due to the end of a single planning, and improves the satellite's ability to autonomously track and continuously acquire intelligence on long-cycle evolution events.

[0176] S223. The collaborative planning task set is sent to neighboring satellites for planning via inter-satellite communication links.

[0177] Step S223 and Figure 1Step S111 in the illustrated embodiment is similar and can be found in the description of step S111, which will not be repeated here.

[0178] In some embodiments, when neighboring satellites receive a set of collaboratively planned tasks, they can also execute autonomous decision-making and multi-hop relay forwarding processes based on orbital coverage capabilities, thereby achieving resource coordination and mission closure at the constellation level.

[0179] Specifically, after neighboring satellites receive the collaborative planning task set via inter-satellite links, they first parse the geographical location information of the target tasks contained within. Then, based on their own orbital parameters, they calculate the satellite's position and observation range for the next planning cycle. A geometric intersection operation is performed to determine whether the satellite's observation range covers the target observation area of ​​the target task. Based on the determination result, a differentiated response is executed. If the judgment result is yes, it indicates that the adjacent satellite meets the execution conditions. To ensure the absolute priority of this emergency mission, the target mission is marked as a high-priority fixed node, and the corresponding imaging time window is forcibly locked on the timeline. At the same time, energy and storage resources required for imaging and processing are reserved from its own resource pool. Subsequently, planning is carried out with this fixed node as the core, optimizing the utilization of remaining resources while ensuring the execution of this collaborative mission.

[0180] If the judgment result is negative, it means that although the adjacent satellite received the request, it is physically unable to execute it (e.g., the satellite's observation range does not cover the target area). In this case, the task is not discarded; instead, the status of effective inter-satellite cooperation is maintained, and it is temporarily stored in the set of tasks to be cooperated with. Then, using the routing function of the inter-satellite network, the target task is encapsulated and forwarded to the next cooperating node satellite that is currently communicating with the satellite. This process continues in the constellation network, forming a multi-hop relay search, until a target execution satellite whose observation range covers the target observation area is found.

[0181] The aforementioned technical steps overcome the limitations of a single satellite in terms of physical coverage and revisit time by constructing an inter-satellite cooperation mechanism of coverage decision-resource locking-relay relay. When a single satellite cannot respond, the constellation network can automatically find nodes with the necessary execution conditions, integrating the capabilities of isolated satellites into a dynamic response network, thereby improving the success rate of executing sudden, large-scale, or time-sensitive tasks.

[0182] In the above embodiments, multi-dimensional anomaly judgment logic based on statistical variation, attribute matching, feature comparison, and regional proportion was constructed for different task types such as target detection, identification, tracking, and change detection. This logic can identify abnormal states and important targets in various scenarios and automatically generate new tasks with longer timeliness based on the original task parameters. Furthermore, when a new task is marked as mandatory, a strategy of locking the time window and reserving resources ensures that high-value, sudden tasks still have absolute execution rights even under extreme conditions of resource conflict. This avoids missed detections or false alarms caused by a single judgment standard and achieves seamless transition from anomaly detection to continuous monitoring, enhancing the satellite's intelligent perception and continuous monitoring capabilities in complex and ever-changing scenarios.

[0183] The following describes an exemplary autonomous planning system 300 for Internet remote sensing satellite missions provided in an embodiment of this application. Figure 3 This is an exemplary hardware structure diagram of the Internet remote sensing satellite mission autonomous planning system 300 provided in this application embodiment.

[0184] In some embodiments, the Internet remote sensing satellite mission autonomous planning system 300 is a computer device or includes a computer device in the system. The computer device includes a processor, memory, and a network interface connected via a system bus. The processor of the computer device provides computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. When the computer program is executed by the processor, it implements the methods described in the embodiments of this application.

[0185] Those skilled in the art will understand that Figure 3 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0186] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit it. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

[0187] As used in the above embodiments, depending on the context, the term "when..." can be interpreted as "if...", "after...", "in response to determining...", or "in response to detecting...". Similarly, depending on the context, the phrase "when determining..." or "if (the stated condition or event) is interpreted as "if determining...", "in response to determining...", "when (the stated condition or event) is detected", or "in response to detecting (the stated condition or event)".

[0188] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer is a dedicated computer. The computer instructions can be stored in a computer-readable storage medium or transferred from one computer-readable storage medium to another. The computer-readable storage medium can be any available medium accessible to a computer or a data storage device integrating one or more available media. The available medium can be a magnetic medium (e.g., a hard disk drive) or a semiconductor medium (e.g., a solid-state drive), etc.

[0189] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. This program can be stored in a computer-readable storage medium, and when executed, it can include the processes described in the above method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as ROM or RAM.

Claims

1. A method for autonomous planning of Internet remote sensing satellite missions based on intelligent sensing and feedback, characterized in that, include: Obtain the execution results of the initial observation task with planning and control attributes; the execution results include various types of information product data obtained after camera imaging and on-board processing; The information product data includes current target attribute data; Calculate the attribute difference between the current target attribute data and the preset normal baseline data; If the attribute difference exceeds a preset judgment threshold, the initial task attribute information contained in the initial observation task is copied as the basic information of the new task; the initial task attribute information includes task priority parameters and planning control identifier parameters; Based on a preset level mapping strategy, a priority increment value corresponding to the abnormal state level is determined, the priority increment value is accumulated into the task priority parameter in the new task basic information, and the planning control identifier parameter is modified to indicate the effective state of mandatory execution or inter-satellite coordination, thereby generating a new observation task. The abnormal state level is the degree of difference between the current target attribute data and the preset normal baseline data; The amount of image data to be processed is calculated based on the observation range of the new observation mission, and the energy consumption and storage space consumption required for on-board processing of the image data to be processed are calculated based on the preset unit data volume resource consumption coefficient. If the planning control flag parameter of the new observation task indicates that it is a mandatory execution state, then the execution time window of the new observation task is locked, and the required energy and storage space are reserved. The new observation task is added to the set of tasks to be planned, and a task planning model is constructed with the goal of maximizing observation benefits. The constraints of the task planning model include that the satellite's remaining energy is greater than the sum of the imaging energy consumption and the on-board processing energy consumption, and that the satellite's remaining storage space is greater than the sum of the imaging data volume and the on-board processing data volume. Solve the task planning model to generate a set of tasks to be executed on a given day, and identify the set of remaining tasks that have not been selected from the set of tasks to be planned; Traverse the remaining task set and divide the tasks that are indicated by the planning control identifier parameter as having an effective inter-satellite coordination state into a collaborative planning task set; Output the set of tasks to be executed by this satellite for execution by this satellite; The collaborative planning task set is sent to neighboring satellites via inter-satellite communication links for planning. In response to the neighboring satellite receiving the collaborative planning task set, the orbital parameters of the neighboring satellite are obtained, the satellite observation range of the neighboring satellite is calculated based on the orbital parameters, and it is determined whether the satellite observation range covers the target observation area of ​​the target task in the collaborative planning task set; If so, the target task is marked as the highest priority fixed node, the imaging time window and corresponding energy storage resources of the adjacent satellite in the observation area are locked, and it is added to the set of tasks to be planned; If not, maintain the effective inter-satellite cooperation status of the target mission and add the target mission to the set of tasks to be cooperated with by the neighboring satellite; The target mission is forwarded to the next cooperating node satellite that is connected to the neighboring satellite via the inter-satellite communication link, until a target execution satellite whose observation range covers the target observation area is found.

2. The method according to claim 1, characterized in that, After generating a new observation task based on the preset level mapping strategy, the method further includes: If the task type of the new observation task is a target tracking task, obtain the target feature value from the previous time series observation results for the same target; Calculate the feature difference degree between the current target attribute data and the target feature value; If the feature difference is greater than a preset tracking lock threshold, the generation process of the new observation task is terminated and the corresponding task basic information is deleted. If the feature difference is less than or equal to the preset tracking and locking threshold, obtain the latest target position coordinates in the current target attribute data; The mission position parameters of the new observation mission are updated using the latest target position coordinates.

3. The method according to claim 1, characterized in that, When the attribute difference exceeds a preset threshold, copying the initial task attribute information contained in the initial observation task as the basic information for the new task specifically includes: Identify the task type of the initial observation task; When the task type is a target detection task, calculate the coefficient of variation between the number of detected targets and the preset number of normal targets; If the coefficient of variation is greater than the preset detection threshold, it is determined to be abnormal and a new task is triggered. If the task type is a target recognition task and the coefficient of variation is less than or equal to the preset detection threshold, it is further determined whether the identified target attributes contain preset important target attributes. If so, it is judged as an exception and a new task is created; If the task type is a target tracking task, then calculate the target feature difference between the current target feature value and the target feature value to be tracked; If the difference in the target features is less than a preset tracking threshold, the tracking is deemed successful and a new task is triggered. When the task type is a change detection task, calculate the proportion of the changed area to the total observation area. If the area ratio is greater than the preset area change threshold, it is determined to be abnormal and a new task is created. In response to the triggering of new task creation, the task priority parameters, planning control identifier parameters, and latest execution time parameters of the initial observation task are copied, and the latest execution time parameter is extended by one task planning cycle as the basic information of the new task.

4. The method according to claim 3, wherein constructing a task planning model with the objective of maximizing observational gains specifically includes: Obtain the list of regular tasks that have been pre-queued in the current set of tasks to be planned, identify and remove target regular tasks in the list of regular tasks that have overlapping or conflicting execution time windows with the mandatory execution tasks; Mark the required tasks as fixed nodes and insert them into the timeline of the task planning to lock the imaging time window and the total amount of resources required; A task planning model is constructed with the goal of maximizing observation gains; the solution space of the task planning model is constrained to include the fixed nodes.

5. The method according to claim 1, characterized in that, After outputting the set of tasks to be executed by the local satellite for execution, the method further includes: Determine whether the planning and control flag parameters of the new observation task indicate a periodic execution state; If so, then copy the basic task information of the new observation task to construct a derived observation task; Extend the deadline parameter of the derived observation task by one task planning cycle. The derived observation tasks are injected into the satellite's long-term mission repository, and then added to the set of planned tasks at the start of the next planning cycle.

6. An autonomous planning system for internet remote sensing satellite missions, characterized in that, The Internet remote sensing satellite mission autonomous planning system includes: one or more processors and a memory; the memory is coupled to the one or more processors, the memory is used to store computer program code, the computer program code includes computer instructions, and the one or more processors call the computer instructions to cause the Internet remote sensing satellite mission autonomous planning system to perform the method as described in any one of claims 1-5.

7. A computer program product containing instructions, characterized in that, When the computer program product is run on the Internet remote sensing satellite mission autonomous planning system, the Internet remote sensing satellite mission autonomous planning system performs the method as described in any one of claims 1-5.

8. A computer-readable storage medium comprising instructions, characterized in that, When the instructions are executed on the Internet remote sensing satellite mission autonomous planning system, the Internet remote sensing satellite mission autonomous planning system performs the method as described in any one of claims 1-5.

Images