Low earth orbit satellite mission state semantic closure scheduling method

Abstract

The present application relates to the technical field of low-orbit satellite network, and proposes a low-orbit satellite task state semantic closed scheduling method, which comprises the following steps: converting observation data into a unified state field containing state observation, state innovation and event set, performing hierarchical coding by a semantic package encapsulation module according to a basic state package, an event evidence package and an optional feature package, declaring inter-package preposition dependency, replacement group and fusion group relationship by a dependent semantic package graph construction module, generating package path scheduling decisions according to state utility and orbit closed feasibility by a PMEP online scheduling module in combination with time-slotted orbit contact graph and link resource constraints, preferentially transmitting information capable of supporting state updating and event explanation at the receiving end, verifying the satisfaction of dependency by the receiving end through a dependency closed checking module, executing state updating, and recording transmission and closed results by an audit log module. The present application significantly reduces the downlink transmission pressure while ensuring the consistency of state updating and the integrity of causal chain.

Description

Technical Field

[0001] This invention relates to the field of low-Earth orbit (LEO) satellite network technology, and more specifically, to a semantic closed-loop scheduling method for LEO satellite mission states. This invention can be deployed in LEO remote sensing satellites, integrated communication and remote sensing satellites, on-board edge computing payloads, inter-satellite link relay nodes, gateway ground stations, ground mission terminals, or combinations thereof; it can also be implemented as a semantic data plane and mission scheduling control plane module in a satellite network protocol stack. Background Technology

[0002] Low Earth Orbit (LEO) satellites, characterized by their low orbital altitude, relatively short revisit cycles, and flexible constellation deployment, are undertaking continuous monitoring tasks in sensitive areas such as vehicle target monitoring, wildfire risk tracking, critical facility inspection, port and airport situational awareness, and disaster emergency response. These tasks are typically not single-shot imaging missions, but rather require continuous maintenance of the states of interest to ground-based missions within multiple short monitoring windows. These states include, for example, the number of targets, target center location, spatial extent, direction of motion, risk level, event changes, state confidence, and uncertainty.

[0003] In existing technical solutions, observation satellites compress complete video frames, video windows, remote sensing images, or remote sensing raster data before transmitting them to a gateway, where ground-based systems perform target detection, wildfire identification, and status updates. This approach preserves complete pixel information but has a large downlink payload and is not suitable for short visible windows. Traditional video / image compression schemes use H.265, JPEG2000, remote sensing image compression, or deep joint source-channel coding to reduce the pixel stream size. These schemes improve the bit rate, but the core communication objective remains pixel restoration or visual quality, without explicitly expressing whether the receiver's status can be updated. Furthermore, on-board detection / feature extraction schemes run target detection, cloud filtering, fire detection, or remote sensing semantic segmentation models on the satellite and transmit detection boxes, heatmaps, or feature vectors. This approach reduces some redundancy but typically lacks a unified cross-task definition of status fields, event evidence hierarchy, and inter-packet dependencies. Ordinary packet scheduling schemes select transmission paths based on packet size, deadline, link capacity, path hop count, or information age. This scheme considers packet arrival as successful delivery, but lacks checks on the pre-existing relationships between basic state, event evidence, and optional characteristics. Dynamic LEO routing schemes utilize orbital contact plans, inter-satellite links, and gateway downlink windows for path selection. This scheme focuses on path reachability, capacity, and latency, but does not incorporate mission state semantics, state innovation, event interpretation, and dependency closure into a unified scheduling objective.

[0004] In summary, existing technologies suffer from the following common problems when handling continuous monitoring tasks in sensitive areas by low-Earth orbit satellites: Pixel-centric transmission methods do not prioritize receiver state updates, resulting in significant bandwidth consumption within short visibility windows for data that does not directly contribute to state maintenance. Onboard detection schemes lack unified state semantic definitions and inter-packet dependencies, making it difficult for the receiver to determine whether arriving data can support state updates. Conventional scheduling and routing schemes only focus on packet arrival and link capacity, failing to incorporate task state semantics, event evidence levels, and dependency closure into unified scheduling objectives, thus failing to prioritize state closure under dynamic link conditions. Summary of the Invention

[0005] In view of this, the present invention proposes a low-Earth orbit satellite mission state semantic closed-loop scheduling method, which enables low-Earth orbit satellite networks to prioritize the transmission of information that can support receiver state updates, event interpretation and dependency checks under conditions of short visibility windows, dynamic links and limited on-board resources.

[0006] To achieve the above objectives, the present invention proposes the following technical solution: A low-Earth orbit satellite mission state semantic closed-loop scheduling method, characterized by comprising: Acquire raw observation data of the sensitive area within the monitoring window; The raw observation data is processed through the on-board state semantic front-end to generate state observations, state innovations, and event sets; The state fields in the state observation, state innovation and event set are hierarchically encoded according to the basic state layer, event evidence layer and optional feature layer, and encapsulated into semantic packages. Package attributes of each semantic package are generated. The package attributes include at least the hierarchical identifier, payload, task utility, deadline slot, pre-dependency set, alternative dependency group and fusion dependency group. A dependency semantic package graph is constructed based on the dependency relationships between the semantic packages. The basic state package does not depend on other packages, the event evidence package depends on the basic state package in the same window, and the optional feature package depends on the basic state package or related event evidence package. Obtain a time-slotted orbit contact map, which includes dynamic resource constraints for observation links, inter-satellite links, and gateway downlinks; Read the queue of semantic packets to be scheduled, the current receiver closure set, the time-slotted track contact diagram and resource status, and select scheduling decisions from candidate packet-path pairs based on track closure feasibility and scheduling score; Transmit semantic packets according to the scheduling decision; The receiving end performs dependency closure checks on the received semantic packets. Semantic packets that meet the dependency closure conditions are used for state updates, while semantic packets that do not meet the dependency closure conditions are either put into the waiting cache or marked as invalid delivery.

[0007] Furthermore, the state observation includes at least the following fields: scalar state field, normalized center location, spatial expansion scale, direction of movement or diffusion, risk level, event set, confidence level, uncertainty, and state innovation score.

[0008] Furthermore, the semantic package includes: The basic state package carries the minimum state fields and has no prerequisites; the event evidence package carries event explanation information and depends on at least the basic state package of the same window; the optional feature package carries verification and accuracy improvement information and depends on the basic state package or related event evidence packages.

[0009] Furthermore, the construction of the dependency semantic package graph includes: Each semantic package is treated as a node; Establish a preceding dependency edge to indicate that one semantic packet depends on the arrival of another semantic packet; Establish an alternative group, meaning that the evidence condition is satisfied if at least one of the multiple external evidence packages arrives. Establishing a fusion group means that all of the multiple complementary evidence packages must arrive to form a complete interpretation; Ensure that the dependency graph within the same window is a directed acyclic graph.

[0010] Furthermore, the method for calculating the feasibility of orbit closure is as follows:

[0011] in, For the estimated arrival time, For the cutoff time slot, This indicates the probability of a link succeeding.

[0012] Furthermore, the method for calculating the scheduling score is as follows:

[0013] in, Contribute to the overall risk level, state innovation, deadline urgency, and dependency closure; For the package load; This represents the number of hops in the path. This is the cost of link or relay switching.

[0014] Furthermore, the dependency closure check at the receiving end includes: Once the basic state packet arrives, it is immediately added to the closed set and the minimum state is updated. Upon arrival of the event evidence package, check whether its preceding dependency set is already included in the closed set. If so, it is counted as a valid delivery and a supplementary explanation of the state change is provided. Otherwise, it is put into the waiting cache. Optional feature packets are only used for verification after their underlying state or relevant event evidence has been closed.

[0015] Furthermore, when the prerequisite dependencies of non-base layer semantic packages are missing, the scheduler adopts one of the following strategies: Block the packet and wait for the preceding packet to arrive before scheduling; group the packet with the missing preceding packet into a closed group and schedule them together; discard the packet and record it as an unclosed risk if the deadline is insufficient or the capacity is insufficient.

[0016] Furthermore, it also includes an exception handling mechanism: When the basic state packet is lost but the event evidence packet arrives, the event evidence packet is temporarily stored in the waiting cache and is not counted as a valid state update. It will automatically switch to the closed state after the basic state packet arrives. When the event evidence packet is lost but the basic status packet arrives normally, the receiving end initiates a retransmission request and retransmits the packets in order of risk level and innovation score. When optional feature packets are lost, they are marked as feature missing, which does not trigger retransmission and does not affect the main process of state update. When a transmission path is interrupted, the scheduler regenerates an alternative path based on the time-slotted track contact diagram. If all feasible paths fail to meet the deadline requirements, the track window is marked as missed. When the count changes abruptly, the scheduling priority of the corresponding event evidence packet is increased, and requests to transmit optional feature packets as verification evidence are allowed. When the credibility of the data source is insufficient, the status update must meet the conditions of the fusion dependency group. If any evidence package in the fusion group is missing, the high-risk related status will not be updated temporarily.

[0017] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention enables low-Earth orbit satellite networks to prioritize the transmission of critical information supporting receiver state updates, event interpretation, and dependency checks under conditions of short visibility windows, dynamic links, and limited onboard resources. This significantly reduces downlink transmission pressure while ensuring consistency of state updates and integrity of causal chains. Specifically, this invention abstracts raw observation data into lightweight semantic state fields, effectively reducing satellite-to-ground transmission bandwidth consumption and improving adaptability to short windows. By dividing semantic packets into basic state packets, event evidence packets, and optional feature packets, and establishing multi-dimensional dependency relationships such as pre-dependencies, alternative groups, and fusion groups, it ensures that the receiver only updates its state after dependency closure, avoiding erroneous alarms or state jumps due to missing critical information. The PMEP online scheduling module makes optimal packet-path matching decisions based on orbit closure feasibility and comprehensive scheduling scores, under multiple constraints such as link capacity, deadline slots, and handover costs. This ensures the priority and reliable delivery of basic state and high-risk event evidence, while effectively avoiding invalid transmissions through dependency-aware blocking, group scheduling, or dropping strategies. Attached Figure Description

[0018] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. In the drawings: Figure 1 This is a schematic diagram of the overall framework of the semantic closed-loop scheduling method for continuous monitoring of sensitive areas of low-orbit satellites in this invention. Figure 2 This is a schematic diagram of the task state semantic representation and dependency semantic package graph in an embodiment of the present invention; Figure 3 This is a schematic diagram of the PMEP online scheduling and counting closed loop in an embodiment of the present invention. Detailed Implementation

[0019] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the disclosure to those skilled in the art. It should be noted that, unless otherwise specified, the embodiments and features described herein can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0020] This embodiment proposes a semantic closure scheduling method for low-Earth orbit satellite mission states, including a sensitive area observation source, observation satellite, on-board state semantic front-end, semantic packet encapsulation module, dependency-based semantic packet graph construction module, PMEP online scheduling module, orbit contact graph and link resource management module, cooperating / relay satellite, gateway ground station, ground mission terminal, receiver dependency closure check module, and audit log module. All of these modules can be deployed in the same satellite and ground terminal combined system, or they can be distributed across constellations, gateways, and mission centers.

[0021] Specifically, the on-board state semantic front-end is used to convert observation data into maintainable state fields at the receiver. This module takes the original video window or remote sensing raster sequence and the state of the previous window as input, and outputs the state observation. State innovation event collection ; The semantic packet encapsulation module is used to encode state fields hierarchically according to minimum state, event evidence, and optional features. This module takes state observations as input. State innovation and link context Output B / E / F semantic packets and packet attributes. ; The SDPG building module is used to declare whether non-base layer packages depend on base state or event evidence. This module takes B / E / F semantic packages, event types, and state sources as inputs, and outputs semantic package nodes, preceding dependency edges, alternative groups, and fusion groups. The OCSG resource module provides dynamic link constraints. It takes as input ephemeris, link visibility, capacity, buffer, success probability, and remaining window, and outputs a set of candidate paths. With path attributes; The PMEP scheduling module selects transmission objects based on state utility and track closure feasibility. This module takes semantic packet queues, SDPG, OCSG, and task weights as inputs and outputs packet-path scheduling decisions. ; The receiver closure check module uses packets whose dependencies are satisfied for state updates. This module takes arrived semantic packets, the dependency graph, and historical states as input, and outputs a set of valid packets. Status after update ; The audit log module is used to check the results of counting, scheduling, and status updates. This module takes as input status source, packet ID, path, arrival, closure, and error information, and outputs traceable records and performance indicators.

[0022] 2. Unified task state semantic generation like Figure 2 As shown, for any sensitive region a and monitoring window w, the original observation is denoted as... In satellite video missions These are continuous video clips from the wildfire mission. This example uses a state generation function to predict raster or remote sensing product sequences for discrete remote sensing. State observations obtained: .in The task status maintained or predicted by the receiving end after the previous window.

[0023] A state object must include at least the following fields:

[0024] Where m is a scalar state field, which is the number of targets or density integral in the vehicle task and the total predicted response in the wildfire task; As a normalization center; For spatial expansion scale; Indicates the direction of movement or diffusion; Risk level; For a collection of events; Confidence level; For uncertainty; Create a score for the state.

[0025] 3. Vehicle Target Count Details In this embodiment, vehicle target counting is a continuous process of "tile-level response counting—window-level deduplication counting—state-level change counting—event-level evidence counting—effective delivery counting." This design aims to avoid inconsistencies in counting caused by small targets, overlapping tiles, cross-frame repetition, and low-confidence detection.

[0026] (1) Tile division. For the input video window... Each frame or representative frame is divided into P×P pixel tiles. During project implementation, the tile size can be 512×512, the density map size can be 64×64, the number of historical windows h=3, and each window can retain K=12 high-response tiles; the above parameters can be adjusted according to the on-board computing power and camera resolution.

[0027] (2) Tile target response. A local density response map is generated for each tile. and confidence level If the front-end outputs a detection bounding box, first map the center of the detection bounding box to the density map using a Gaussian kernel; if the front-end directly outputs the density map, then use density map integration. The initial tile count is... .

[0028] (3) High-response tile selection. Calculate the tile scheduling score.

[0029] Take the TopK tiles into the timing encoder, and at the same time record the low-confidence background count estimate of those not included in the TopK tiles.

[0030] (4) Duplicate Target Removal. For candidate targets in adjacent or overlapping tiles, if the center distance is less than 1 / 3, the duplicate targets are removed. Or the detection box IoU is greater than If the density map is empty, it is considered to be the same target, and the one with the highest confidence level is retained; if there is only a density map without a bounding box, non-maximum suppression is applied to the density peak, and the radius can be set to a value that is not specified. =2 to 5 density map grids. The deduplicated target set is denoted as .

[0031] (5) Window-level counting. Number of vehicles at the window level. ,in As the target confidence weight, For low response background count compensation term, This is the compensation coefficient. If the system requires conservative alarms, ceil can be used; if reducing false alarms is required, round can be used and a minimum confidence threshold can be set. .

[0032] (6) Stable counting across frames and across windows. The trajectory IDs of the same target in consecutive frames are associated, and the same trajectory ID is counted only once within the same window; trajectory matching is only used when calculating "new target event" across windows, and the current state count is still based on the deduplicated set of the current window. For reference only. New target number. Number of vanished targets .

[0033] (7) Events involving changes in quantity. Let ,in Predict the quantity from the previous state. If If so, the "target quantity change" event is triggered; if If this happens, the "new target appears" event will be triggered.

[0034] 4. Wildfire remote sensing response counting The counts in the wildfire mission are not a direct measurement of the actual burned area, but rather a state-based statistical analysis of high-response cells, total response, response zone proportion, and event trigger counts within the predicted response grid. These statistics are used for communication and status maintenance and should not be equated with fire line boundaries or actual heat release power.

[0035] Wildfire prediction grid First, normalize to obtain the response graph. Total response Number of response units Response area percentage Possible threshold Alternatively, it can be adaptively calibrated based on region, season, sensor, and false alarm rate.

[0036] Wildfire Response Center Response space expansion scale ,in To be according to Weighted coordinate covariance matrix. Response changes include... , , and If a high-response unit changes from 0 to non-zero or the area of ​​a new connected region exceeds a threshold, a "new fire point" event is triggered; if the change in the proportion of the response region exceeds a threshold, a "response region change" event is triggered; if the growth rate of the expansion scale exceeds a threshold, a "diffusion acceleration" event is triggered; if the risk level changes, a "risk level change" event is triggered.

[0037] 5. Semantic Packet Layering and Encoding Details This embodiment encapsulates the state field into three semantic packages: a basic state package (B), an event evidence package (E), and an optional feature package (F). Each semantic package has the following attributes: .

[0038] in As a hierarchy, For load, For task utility, For the cutoff time slot, For the set of preceding dependencies, To replace dependency groups, To merge dependency groups, This is the identifier for the source of the status.

[0039] 6. Dependency-based semantic package graph construction Construct a dependency semantic package graph for region a and window w.

[0040] in For semantic packet nodes, For preceding dependency edges, As an alternative group, For the fusion group.

[0041] The dependency rules include: (1) The basic state package does not depend on any other package; (2) The event evidence package depends on at least the basic state package in the same window; (3) The optional feature package depends on the basic state package or the related event evidence package; (4) Multiple external evidence packages can form a substitution group, and as long as one of them arrives, a certain evidence condition can be satisfied; (5) Multiple complementary evidence packages can form a fusion group, and all of them must arrive to form a complete interpretation; (6) The dependency graph in the same window should be a directed acyclic graph, and self-dependency and circular dependency are not allowed.

[0042] The receiving end maintains a closed set. For any package p, if =B, then ;like ,but The valid delivery determination of package p is as follows: ,in This indicates that packet p arrived before time t.

[0043] 7. PMEP Dependency Closed-Loop Online Scheduling The PMEP scheduling module reads the sendable semantic packets, the dependency semantic packet graph, and the time-slotted track contact graph in each time slot to generate candidate packet-path pairs. The scores are determined based on track closure feasibility, packet priority, load, path hop count, and switching cost.

[0044] Time-slotted track contact diagram This includes observation links, inter-satellite links, and gateway downlinks. Each edge e has a capacity in time slot t. Remaining capacity Service delay Probability of success and remaining visible windows Candidate paths The following conditions must be met: link visibility, sufficient capacity, estimated arrival time no later than the deadline time slot, and path reliability no less than the threshold.

[0045] Orbit closure feasibility Scheduling score .in, Contribute to the overall risk level, state innovation, deadline urgency, and dependency closure; For the package load; This represents the number of hops in the path. This is the cost of link or relay switching.

[0046] When multiple packets compete for the same link resource, PMEP prioritizes scheduling packets that can close the basic state and key event evidence. If the prerequisites of a non-basic layer packet are missing, the scheduler can adopt three strategies: (1) block the packet; (2) group the packet with the missing prerequisite packet into a closure group and schedule them together; (3) discard the packet and record it as an unclosed risk when the deadline is insufficient or the capacity is insufficient.

[0047] 8. Receiver status update, retransmission, and auditing After receiving the packet, the receiving end first verifies the CRC and task number, and then... , and Write to the receive buffer. Once the basic status packet arrives, proceed directly to the next step. It updates the minimum state; after the event evidence packet arrives, it checks whether its Par set is closed. If closed, it is used to supplement the explanation of state changes; if not closed, it enters the waiting queue; optional feature packets are only used for verification after their basic state or related event evidence is closed.

[0048] If a packet does not arrive before the deadline, the receiver can trigger a retransmission via ACK / NACK or a status digest. Retransmission priority is ordered as follows: "Basic State Packet > High-Risk Event Evidence Packet > Ordinary Event Evidence Packet > Optional Feature Packet". If the basic state packet misses its deadline, non-basic layer packets that depend on it will generally not be retransmitted unless the task requests historical verification.

[0049] The audit log should record at least the following: status source, region ID, window ID, packet ID, packet level, event type, number of field units, payload length, list of prerequisites, scheduling slot, path node, link type, expected arrival, actual arrival, closure result, cause of error, and receiver status version number. This log is used to calculate the number of delivered packets, the number of valid packets, OER, DAUC, status mismatch rate, and track window miss rate.

[0050] Example 1 This embodiment presents the specific process of generating and encapsulating the semantic representation of the continuous monitoring status of satellite video vehicles in the above method, including: Step 1: The observation satellite acquires video clips of the sensitive region a within window w. A window can consist of 3, 5, or more consecutive / interval frames. The onboard front end records the observation timestamp, region ID, source satellite ID, and sensor parameters.

[0051] Step 2: Perform radiometric normalization, geometric alignment, and motion compensation on the video frames. If the satellite platform attitude varies significantly, georeferencing or feature point matching is used for correction; if the observation area is fixed, a region template is used for direct cropping.

[0052] Step 3: Divide the video frame into tiles. Calculate the local target response, density map, confidence score, and temporal difference for each tile. Optional parameters are: 512×512 pixels for the tile, 64×64 for the density map, history window h=3, TopK=12 tiles per window, and token dimension 106.

[0053] Step 4: Count the targets for each tile. If using a density map, then... If a bounding box is used, the statistical confidence level is higher than [previous level]. Furthermore, the number of detection frames retained after NMS; if a trajectory is used, the same trajectory ID is counted only once within the same window.

[0054] Step 5: Perform cross-tile deduplication. For candidate targets at the boundary of adjacent tiles, merge duplicate targets based on center distance, IoU, or trajectory ID. The number of deduplicated targets in the window is denoted as . .

[0055] Step 6: Calculate the center of the vehicle group Computational space expansion Direction of movement of the computing center .

[0056] Step 7: Calculate state innovation. , Innovation score It can be calculated by weighted sum: .

[0057] Step 8: Generate an event set. If Then a new target event is generated; if Then a quantity change event is generated; if Then a center offset event is generated; if Changes generate risk change events; if the trajectory set is significantly updated, trajectory update events are generated.

[0058] Step 9: Generate the B base state package. Fields include... and This package has no prerequisites, and its deadline is set to the current visible window or the window of the nearest gateway.

[0059] Step 10: Generate an E-event evidence package for each event. A new target event package may include the number of new targets, target cluster centers, new target confidence levels, and associated B-package IDs; a quantity change event package may include... The change ratio, trigger threshold ID, and event confidence level.

[0060] Step 11: Determine whether to generate an optional F-package based on link capacity. The F-package may include a TopK tile summary, a low-dimensional projection of the local density map, a context token, a description representing the target cluster, and uncertainty supplementation.

[0061] Step 12: Construct the SDPG. Package B serves as the parent node, package E depends on package B, and package F depends on package B or a related package E. If multiple pieces of evidence can prove the same quantity change, they can be added to the substitution group; if local density and trajectory evidence must be combined to explain the event, they can be added to the fusion group.

[0062] Step 13: The PMEP scheduler generates candidate paths based on the OCSG and calculates... Prioritize sending B packets and high-risk E packets. If an E packet arrives at the receiver before the B packet is closed, the E packet enters the waiting buffer and is not counted in the valid state update.

[0063] Step 14: The receiving end updates the status. After packet B is closed, the number of vehicles, the center, and the risk level are updated; after packet E is closed, an explanation is provided for "why the number has changed or the risk has increased"; after packet F is closed, a delayed review can be performed.

[0064] Example 2 This embodiment provides the specific process of the wildfire remote sensing temporal state semantic generation and counting encapsulation method in the above method, including: Step 1: Observe satellites or ground preprocessing systems to generate wildfire prediction grids by pressing window w. Alternatively, it can read external fire points, hot spots, and smoke products and rasterize them to the same coordinate system.

[0065] Step 2: Normalization to response diagram Masking is applied to outliers, missing values, cloud-obscured areas, or sensor edge areas to ensure that counting is performed only on valid grid cells.

[0066] Step 3: Calculate the total response Calculate the number of high-response units Calculate the proportion of the response area . It can be taken as 0.7, or it can be determined adaptively using historical quantiles or false alarm rate.

[0067] Step 4: Perform connected component labeling on the high-response units to obtain the number of connected components in the response. The maximum area of ​​a connected component, the center of each connected component, and its bounding box. A new fire event is generated when a newly added connected component cannot be matched with the previous window.

[0068] Step 5: Calculate the response center and spatial expansion The centroid of the response is obtained by weighting the response values, and the spatial extension is obtained by taking the square root of the trace of the weighted covariance matrix of the response values.

[0069] Step 6: Calculate the state changes. .like If the threshold is exceeded, a diffusion acceleration event is triggered.

[0070] Step 7: Risk Level Mapping. This can be... , , , and Input rules Outputs low / medium / high risk. If the risk level changes, a risk level change event is generated.

[0071] Step 8: Generate package B. Package B includes... , , , Risk level, confidence level, and uncertainty. If there is no high-response area, an empty state B packet can also be generated to indicate that the status of this window has been updated to no obvious response.

[0072] Step 9: Generate E-packets. E-packets include evidence of events such as new fire points, changes in response zones, accelerated spread, and changes in risk. Each event may include a trigger threshold, the magnitude of the change, a connected component number, and an associated B-packet ID.

[0073] Step 10: Generate the F-package. The F-package may contain a low-resolution summary of the response graph, key connected component outlines, an index of external product evidence, a smoke polygon summary, or meteorological auxiliary features.

[0074] Step 11: Receiver dependency closure. After packet B arrives, the total wildfire response, center, and risk level are updated; packet E depends on packet B, and closure of the packet explains new fire spots or spread; packet F is used to supplement evidence.

[0075] Example 3: like Figure 3 As shown, this embodiment provides the specific process of PMEP online scheduling and receiver closure in the above method, including: At any time slot t, PMEP reads the queue of packets to be scheduled. Current closed set Track contact diagram and resource status The scheduler first marks packets that can be sent: B packets can directly enter the candidate set; E / F packets only enter the candidate set if their prerequisites are closed, or if they can form a closed group with the necessary prerequisite packets.

[0076] Then, PMEP generates a maximum of [number] packets per candidate package. There are 10 candidate paths. A path can be a direct downlink from "observation satellite to gateway ground station," or a relay path from "observation satellite to cooperating satellite to gateway ground station," and may also include a store-and-carry-forward window. Paths must meet constraints on capacity, buffering, latency, deadline, and reliability.

[0077] For candidate package-path pairs PMEP calculation and ,according to Select from high to low. After selecting a packet-path pair, deduct the remaining capacity of each link on the path and the relay node buffer; if resources are insufficient after deduction, skip the candidate pair. If a packet misses its deadline slot, it is marked as a missed track window.

[0078] Upon receiving a packet, the receiving end performs a closure check. If p is a B packet, it is directly added. If p is an E / F packet, then check... Is it included in If the conditions are met, a valid delivery is formed and the state is updated; if not, it is considered an unclosed dependency arrival, and the delivery awaits the preceding package or is discarded after the deadline.

[0079] Example 4 This embodiment provides examples of semantic packet fields and byte counts in the above method, specifically including: Taking a vehicle monitoring window as an example, suppose a window generates one B packet, three E packets, and two F packets. The payload of the B packet includes the number of vehicles, center, expansion, direction, risk, confidence level, uncertainty, and innovation score, approximately 20 to 40 bytes; each E packet is approximately 20 to 60 bytes; and each F packet, if carrying 12 tile digests and a 96-dimensional int8 token, can reach approximately 1,200 to 1,800 bytes. With the packet header and dependency index added, the average size of the complete state semantic packet graph can be in the KB range.

[0080] Example 5 In actual satellite-to-ground transmission, the arrival of semantic packets may be abnormal due to reasons such as link interruption, on-board buffer overflow, inter-satellite handover failure, or ground station receiving conflicts. To ensure the robustness and traceability of receiver state updates, this embodiment further specifies the following anomaly handling and security verification mechanisms: (i) Scenario where the basic state packet is lost but the event evidence packet arrives. In this case, although the event evidence packet has been transmitted to the receiving end, the receiving end cannot independently interpret the event evidence because its prerequisite basic state packet is missing. Therefore, the receiving end temporarily stores the event evidence packet in the waiting cache and does not count it as a valid state update. If the basic state packet finally arrives within its deadline slot, the event evidence packet automatically enters the closed state and is used for supplementary interpretation; if the basic state packet has not arrived before the deadline slot, the arrived event evidence packet is marked as "dependency not closed upon arrival," is not counted in the consumable state, and is recorded as an invalid delivery for subsequent delivery efficiency statistics.

[0081] (ii) Scenario where the event evidence packet is lost but the basic status packet arrives normally. The receiver can update the minimum task status based on the basic status packet, but the lack of event explanation information makes the causal chain of state changes incomplete. In this case, the receiver can initiate a retransmission request to the onboard scheduler. The retransmission priority is sorted according to the risk level and innovation score of the event evidence packet. If the link conditions permit, the scheduler will prioritize retransmitting high-risk or high-innovation event evidence packets; if the link conditions are tight, it is allowed to maintain only the basic status update, and the event explanation will be supplemented in a subsequent window or idle period.

[0082] (III) Situation of Optional Feature Packet Loss. Since optional feature packets do not constitute a necessary prerequisite for basic state updates or event interpretation, their loss does not affect the receiver's maintenance of minimum state and event evidence. The receiver only uses optional feature packets for verification, accuracy enhancement, or delay analysis, provided that the basic state packet or relevant event evidence packet has been closed. If the optional feature packet does not arrive within the deadline, it is directly marked as "feature missing," without triggering retransmission or affecting the main state update process.

[0083] (iv) Transmission Path Interruption Situation. When an inter-satellite link or satellite-to-ground downlink among the candidate paths selected by the PMEP scheduler is interrupted during transmission due to satellite movement, link obstruction, or node failure, the scheduler will detect that the path is unavailable and immediately regenerate an alternative path based on the time-slotted orbit contact diagram. If the alternative path can complete transmission before the deadline of the semantic packet, transmission will continue; if all feasible paths fail to meet the deadline requirement, the semantic packet will be marked as "orbit window missed," the receiver will maintain the previous state, and the audit log will record the reason for the miss.

[0084] (v) Situations of Abnormal Sudden Changes in Counts. When the number of vehicles, the total number of wildfire responses, or the risk level changes significantly between adjacent windows, it may be due to observation noise, sensor anomalies, or real events. This invention quantifies such changes through a state innovation score. If the innovation score exceeds a preset threshold, the scheduling priority of the corresponding event evidence packet is automatically increased to ensure that the change information is transmitted preferentially. At the same time, after the basic state is updated, if the receiving end finds that the event explanation is insufficient to support the magnitude of the change, it can request the transmission of optional feature packets as verification evidence to reduce the risk of false alarms.

[0085] (vi) Situations where the reliability of data sources is insufficient. When the confidence or uncertainty field output by the on-board state semantic front-end indicates that the current observation quality is low, the receiver, according to a preset security policy, requires that the state update of this window must meet the fusion dependency group condition. That is, event evidence packets from multiple independent evidence sources must be received simultaneously to form a complete interpretation and update the state. If any evidence packet in the fusion dependency group is missing or not closed, the receiver will not update the high-risk related state and will record it as "fusion condition not met" to avoid misjudgment due to low-reliability data.

[0086] Through the above-mentioned anomaly handling and security verification mechanisms, this embodiment can still ensure the correctness, interpretability, and auditability of the receiver's status update under the conditions of dynamic low-Earth orbit links and limited on-board resources, effectively avoiding erroneous alarms or status jumps caused by partial packet loss or delay.

[0087] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A semantic closed-loop scheduling method for low-Earth orbit satellite mission states, characterized in that, include: Acquire raw observation data of the sensitive area within the monitoring window; The raw observation data is processed through the on-board state semantic front-end to generate state observations, state innovations, and event sets; The state fields in the state observation, state innovation, and event set are hierarchically encoded according to the basic state layer, event evidence layer, and optional feature layer, encapsulated into semantic packages, and package attributes of each semantic package are generated. Construct a dependency semantic package graph based on the dependencies between semantic packages; Obtain a time-slotted orbit contact map, which includes dynamic resource constraints for observation links, inter-satellite links, and gateway downlinks; Read the queue of semantic packets to be scheduled, the current receiver closure set, the time-slotted track contact diagram and resource status, and select scheduling decisions from candidate packet-path pairs based on track closure feasibility and scheduling score; Transmit semantic packets according to the scheduling decision; The receiving end performs dependency closure checks on the received semantic packets. Semantic packets that meet the dependency closure conditions are used for state updates, while semantic packets that do not meet the dependency closure conditions are either put into the waiting cache or marked as invalid delivery.

2. The low-Earth orbit satellite mission state semantic closed-loop scheduling method according to claim 1, characterized in that, The state observation includes at least the following fields: scalar state field, normalized center location, spatial expansion scale, direction of movement or diffusion, risk level, event set, confidence level, uncertainty, and state innovation score.

3. The low-Earth orbit satellite mission state semantic closed-loop scheduling method according to claim 1, characterized in that, The semantic package includes: The basic state package carries the minimum state fields and has no prerequisites; the event evidence package carries event explanation information and depends on at least the basic state package of the same window; the optional feature package carries verification and accuracy improvement information and depends on the basic state package or related event evidence packages.

4. The low-Earth orbit satellite mission state semantic closed-loop scheduling method according to claim 1, characterized in that, The construction of the dependency semantic package graph includes: Each semantic package is treated as a node; Establish a preceding dependency edge to indicate that one semantic packet depends on the arrival of another semantic packet; Establish an alternative group, meaning that the evidence condition is satisfied if at least one of the multiple external evidence packages arrives. Establishing a fusion group means that all of the multiple complementary evidence packages must arrive to form a complete interpretation; Ensure that the dependency graph within the same window is a directed acyclic graph.

5. The low-Earth orbit satellite mission state semantic closed-loop scheduling method according to claim 1, characterized in that, The method for calculating the feasibility of orbit closure is as follows: ， in, For the estimated arrival time, For the cutoff time slot, This indicates the probability of a link succeeding.

6. The low-Earth orbit satellite mission state semantic closed-loop scheduling method according to claim 1, characterized in that, The scheduling score is calculated as follows: ， in, Contribute to the overall risk level, state innovation, deadline urgency, and dependency closure; For the package load; This represents the number of hops in the path. This is the cost of link or relay switching.

7. The low-Earth orbit satellite mission state semantic closed-loop scheduling method according to claim 1, characterized in that, The dependency closure check at the receiving end includes: Once the basic state packet arrives, it is immediately added to the closed set and the minimum state is updated. Upon arrival of the event evidence package, check whether its preceding dependency set is already included in the closed set. If so, it is counted as a valid delivery and a supplementary explanation of the state change is provided. Otherwise, it is put into the waiting cache. Optional feature packets are only used for verification after their underlying state or relevant event evidence has been closed.

8. The low-Earth orbit satellite mission state semantic closed-loop scheduling method according to claim 1, characterized in that, When the prerequisite dependencies of non-base layer semantic packages are missing, the scheduler adopts one of the following strategies: Block the packet and wait for the preceding packet to arrive before scheduling; group the packet with the missing preceding packet into a closed group and schedule them together; discard the packet and record it as an unclosed risk if the deadline is insufficient or the capacity is insufficient.

9. The low-Earth orbit satellite mission state semantic closed-loop scheduling method according to claim 1, characterized in that, It also includes an exception handling mechanism: When the basic state packet is lost but the event evidence packet arrives, the event evidence packet is temporarily stored in the waiting cache and is not counted as a valid state update. It will automatically switch to the closed state after the basic state packet arrives. When the event evidence packet is lost but the basic status packet arrives normally, the receiving end initiates a retransmission request and retransmits the packets in order of risk level and innovation score. When optional feature packets are lost, they are marked as feature missing, which does not trigger retransmission and does not affect the main process of state update. When a transmission path is interrupted, the scheduler regenerates an alternative path based on the time-slotted track contact diagram. If all feasible paths fail to meet the deadline requirements, the track window is marked as missed. When the count changes abruptly, the scheduling priority of the corresponding event evidence packet is increased, and requests to transmit optional feature packets as verification evidence are allowed. When the credibility of the data source is insufficient, the status update must meet the conditions of the fusion dependency group. If any evidence package in the fusion group is missing, the high-risk related status will not be updated temporarily.

Images