Unmanned aerial vehicle emergency inspection system integrating low-altitude communication and edge calculation

Abstract

The invention belongs to the technical field of intelligent unmanned systems and edge intelligent reasoning, and discloses an unmanned aerial vehicle emergency inspection system integrating low-altitude communication and edge calculation. The system is composed of a link state monitoring and scoring module, a reasoning calculation migration and model loading module, a task graph modeling and dynamic segmentation module, a multi-unmanned aerial vehicle task collaboration and trajectory continuation module, a flight state fusion and path adjustment module, a redundancy criterion driven fault-tolerant control module and a result consistency verification and data caching module. According to the method, the communication link scoring function fusing the packet loss rate, the bandwidth, the time delay and the signal-to-noise ratio is constructed, the comprehensive quality evaluation of different low-altitude communication links is realized, and the reasoning model switching and task migration strategy is driven by the scoring result, so that the communication adaptation capability of the system in a complex, dynamic or emergent scene is remarkably improved, and the system performance is improved. Compared with an existing communication mechanism depending on a fixed link or a preset priority.

Description

Technical Field

[0001] This invention belongs to the field of intelligent unmanned systems and edge intelligent reasoning technology, specifically a drone emergency inspection system that integrates low-altitude communication and edge computing. Background Technology

[0002] With the increasing demand for UAV technology in scenarios such as emergency management, urban power line inspection, and rapid response to natural disasters, low-altitude unmanned systems have been widely deployed in multi-source mission environments. However, in complex scenarios with imperfect communication infrastructure, uneven distribution of edge computing resources, and frequent fluctuations in flight status, existing UAV inspection systems generally suffer from problems such as delayed scheduling response, high mission interruption rate, and low resource adaptation efficiency, making it difficult to meet the comprehensive requirements of timeliness, stability, and fault tolerance in emergency missions.

[0003] Current research has attempted to introduce edge computing frameworks to improve the processing capabilities of UAVs. A search revealed that CN114815881A proposes an edge computing-supported inspection method that uses edge nodes to process image results and distribute tasks to UAV terminals. This approach alleviates data backhaul pressure to some extent, but its communication mechanism still relies on a pre-defined connection model and lacks real-time link quality assessment methods. Task allocation is primarily statically planned, failing to establish a dynamic task scheduling mechanism that can be split and reconstructed based on flight status, battery level, and task execution results. Furthermore, it lacks a task result consistency assessment strategy and an interruption recovery mechanism, which can easily lead to discrepancies in node inference results, loss of task status, and broken execution logic. Summary of the Invention

[0004] The purpose of this invention is to provide an emergency inspection system for unmanned aerial vehicles (UAVs) that integrates low-altitude communication and edge computing, so as to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: an emergency inspection system for unmanned aerial vehicles (UAVs) that integrates low-altitude communication and edge computing. The system consists of a link status monitoring and scoring module, an inference calculation migration and model loading module, a task graph modeling and dynamic segmentation module, a multi-UAV task collaboration and trajectory continuation module, a flight status fusion and path adjustment module, a redundancy criterion-driven fault-tolerant control module, and a result consistency verification and data caching module.

[0006] Link status monitoring and scoring module: used to access UHF communication module, 5G micro base station module and satellite communication module, collect packet loss rate, bandwidth, delay and signal-to-noise ratio parameters, and calculate the quality of each link through weighted communication link scoring function to obtain score S; when S is lower than the preset first threshold T1, output link switching command and inference migration request.

[0007] The inference computation migration and model loading module is used to receive the link score S and determine the current available bandwidth and resource conditions according to the model switching trigger function. It automatically loads one of the standard model, medium compression model, or summary inference model. When the link score S is lower than the first threshold T1, it triggers the inference task to migrate to the edge node for processing and outputs the model summary result to the verification module.

[0008] Task graph modeling and dynamic segmentation module: This module models the inspection area as a directed task graph G=(V,E) with a risk value W, where V is the target point set and E is the path edge set. When any condition in the graph segmentation trigger criterion function is met, including: the remaining battery power is lower than the second threshold T2, the current track deviation angle is greater than T3, or the number of mission failures exceeds N, the subgraph G′ is triggered and the task index table is updated synchronously.

[0009] Multi-UAV mission coordination and trajectory continuation module: used to receive subgraph G′ and broadcast it to UAV nodes within the communication range, and use a multi-dimensional cost function to comprehensively evaluate the path fitting distance and remaining processing capacity, and select the UAV node with the lowest cost value to execute the subgraph task; the continuation information includes subgraph index, starting node number and synchronization execution status code;

[0010] Flight status fusion and path adjustment module: used to collect the current UAV's flight attitude data, remaining power ratio and link score S, and calculate the status score P through the status fusion scoring function. The score P is used to dynamically control the adjustment of flight altitude, speed and inspection frequency, and the scoring result is fed back to the graph segmentation module to adjust the stability of the mission graph structure.

[0011] Redundancy criterion-driven fault-tolerant control module: used to monitor the communication interruption duration Δt and the inference task accuracy E in parallel. Based on the dual-threshold fault trigger function, it determines whether the task abortion condition is met. If Δt>Δt1 and E>T5, the current task is aborted, and the unfinished task graph substructure G′ is forwarded to the backup drone or the operation and maintenance control center.

[0012] The result consistency verification and data caching module is used to compare the summary similarity error function between the edge inference output and the local inference result. When the result error is less than the set tolerance ε, it is determined to be consistent and the result is merged. When the communication link is abnormal, the module calls the sliding cache mechanism to retain the structured prediction summary within the last three minutes and completes the breakpoint resume transmission after the link is restored.

[0013] Preferably, the link status monitoring and scoring module includes:

[0014] (1) The UHF communication module, 5G micro base station module and satellite communication module are uniformly accessed and real-time communication parameters including packet loss rate, bandwidth, signal-to-noise ratio and communication delay are collected. The system calls the communication link scoring function to perform weighted calculation on the above indicators, constructs a unified scoring model, quantifies the current transmission quality of each link, and forms a status score S. The scoring result serves as an important basis for subsequent communication decisions and computation scheduling, ensuring that communication and computation behaviors have perception-driven capabilities.

[0015] The expression for the communication link scoring function is:

[0016]

[0017] In the formula, S: link score (range [0,100]), used to evaluate the quality of the current communication channel, p: packet loss rate (0~1), d: normalized communication delay (Latency, in ms, normalized), b: link bandwidth (in Mbps), SNR: signal-to-noise ratio (in dB), α, β, γ, δ: scoring weights, satisfying α+β+γ+δ=1;

[0018] Technical effect: Provides real-time computable metrics for communication state switching and computation migration;

[0019] (2) When the link status score S is lower than the preset first threshold T1, the system identifies it as a decline in communication quality and triggers the link switching mechanism to ensure the normal transmission of critical task data. At the same time, the system generates an inference migration request, starts the model unloading and edge node model loading process, and transfers the computing tasks originally executed locally to computing units with higher link quality to ensure that the tasks are not interrupted, the computing continuity is strong, and the energy consumption control is stable, so as to realize the integration and collaboration of low-altitude communication and edge computing.

[0020] Preferably, the inference computation transfer and model loading module includes:

[0021] (1) Receive the score value S from the link scoring module and, in conjunction with the model switching trigger function, determine the current available bandwidth and computing resource status in real time; if the score S meets the conditions of high bandwidth and sufficient computing power, the system automatically loads the standard inference model to perform complete feature extraction; if the system detects that the bandwidth is limited or the resources are restricted, it loads the compression model or the summary inference model to ensure that the computing task has stable feasibility in different link environments.

[0022] The expression for the model switching trigger function is:

[0023]

[0024] In the formula, Model: the currently loaded model structure ( As a standard model, It is a medium-precision compression model. (For summary model), b: communication link bandwidth (unit: Mbps);

[0025] Technical effect: Avoids delays or failures caused by insufficient bandwidth in large models;

[0026] (2) When the link score S is lower than the first threshold T1, the system determines that the communication quality cannot support the local model operation, automatically triggers the inference task migration process, schedules the current inference task to the edge node for execution, and unloads the local model to release resources; after the edge node completes the inference, it generates a structured model summary result and sends it back to the local node. This result serves as the basis for subsequent consistency verification and redundancy judgment, thereby ensuring the stability of cross-node inference and result synchronization.

[0027] Preferably, the task graph modeling and dynamic segmentation module includes:

[0028] (1) Construct a structured graph model for the inspection area, take the target to be inspected as the node set V of the graph, establish the possible paths between nodes as the edge set E, and combine the historical fault frequency, regional sensitivity level and target type to generate the node risk value W; by introducing this risk weight, the system can prioritize the allocation of resources to key task points, improve the inspection efficiency under limited energy conditions, and form a directed task graph G with priority scheduling capability;

[0029] (2) During the inspection process, when the system detects that the key operating status meets any abnormal condition in the graph segmentation trigger criterion function, when the remaining power is lower than the threshold T2, the current trajectory deviates more than the angle T3, or the number of consecutive mission failures exceeds the upper limit N, the task graph G segmentation mechanism is started; the system automatically divides the subgraph G′ from the original graph and synchronously corrects the current task index table to support subsequent multi-machine succession scheduling and task takeover processing, so as to realize a highly robust emergency collaborative inspection strategy;

[0030] The expression for the graph segmentation trigger criterion function is:

[0031]

[0032] In the formula, Triggering factor (triggers subgraph partitioning when ≥1), Remaining battery percentage (0~1), Angle between current heading and target path (unit: degrees). The number of consecutive failures of a target node. The threshold set by the system; : Indicator function, returns 1 if the condition is met, otherwise returns 0;

[0033] Technical effect: Ensures reasonable division of graph structure and improves multi-machine collaboration efficiency.

[0034] Preferably, the multi-UAV mission coordination and trajectory continuation module includes:

[0035] (1) When the mission graph is segmented, the system will broadcast the generated subgraph G′ to other UAV nodes within the current communication range through the low-altitude communication module; each receiving node will comprehensively evaluate the fitting distance between its own trajectory and the subgraph path, as well as the remaining computing resources and power processing capabilities through the built-in multidimensional cost function, and complete the succession optimization judgment; this evaluation mechanism ensures the spatial continuity and resource adaptability of mission relay.

[0036] The expression for the multidimensional cost function is:

[0037]

[0038] In the formula, :No. The collaborative value of drones The Hausdorff distance (in meters) between its track and the subgraph path. Its remaining task capacity percentage The weighting coefficients set empirically satisfy... ;

[0039] Technical benefits: Selecting the optimal connection node reduces handover costs;

[0040] (2) After the cost assessment is completed, the system selects the UAV node with the lowest comprehensive cost to execute the subgraph task and pushes the task continuation instruction package to it. The instruction package contains subgraph index information, continuation starting point number and synchronous execution status code, which are used for unified identification and scheduling control of task status. After the selected node loads the task, it automatically updates its local index table and sends back the continuation success signal to the original node to achieve seamless task transition.

[0041] Preferably, the flight status fusion and path adjustment module includes:

[0042] (1) Real-time acquisition of the current UAV's flight attitude data, remaining battery percentage and communication link score S, and call the state fusion scoring function to perform weighted fusion of the above multi-source state indicators to generate a single score result P; the score result comprehensively reflects the current flight stability, endurance and communication support level, and provides a quantitative basis for subsequent control commands and mission adjustments;

[0043] The state fusion scoring function expression is:

[0044]

[0045] In the formula, P: flight status score (0~1), e: remaining battery power (proportion), φ: attitude stability factor (e.g., based on IMU angular velocity or Euler angle change rate), S: communication score (derived from formula F1), μ1, μ2, μ3: state fusion weights, μ1+μ2+μ3=1;

[0046] Technical effect: Enables dynamic adjustment of mission and path reconstruction during flight;

[0047] (2) The state score P serves as the core input signal for the execution control strategy. It is used to dynamically adjust the flight altitude, speed and inspection frequency of the UAV to achieve a balance between flight performance and mission efficiency. At the same time, the system feeds back the score results to the mission graph segmentation module and adjusts the stability weight of the mission graph structure according to the state change trend to ensure that the graph model has an adaptive response capability to changes in flight risk and improve the robustness of mission scheduling.

[0048] Preferably, the redundancy criterion-driven fault-tolerant control module includes:

[0049] (1) Parallel monitoring of the current communication interruption duration Δt and the inference task execution accuracy E to construct a state detection channel oriented towards task reliability; the system sets the communication interruption time threshold Δt1 and the accuracy threshold T5 based on the dual-threshold fault trigger function to jointly determine the current state; this criterion avoids false triggering due to fluctuations in a single indicator and enhances the accuracy and robustness of fault detection.

[0050] The expression for the dual-threshold fault trigger function is:

[0051]

[0052] In the formula, : Whether to trigger fault tolerance (triggered when value is 1) Duration of communication interruption (in seconds) , , Indicator functions;

[0053] Technical effect: Prevents accidental triggering of game content, improving system stability;

[0054] (2) When the system detects that the composite conditions of Δt being greater than Δt1 and E being higher than T5 are met simultaneously, the current task termination logic is immediately triggered to prevent the spread of data errors due to communication interruption or model inaccuracy. The system automatically extracts the unfinished substructure G′ in the current task graph and, based on the current communication status, prioritizes forwarding it to the backup UAV node; when all air nodes lose contact, it forwards it to the ground operation and maintenance control center to realize a multi-level fault tolerance and task continuation strategy.

[0055] Preferably, the result consistency verification and data caching module includes:

[0056] (1) The result summary returned by the edge reasoning node is compared with the local reasoning output in a structured manner. The difference between the results is calculated by calling the summary similarity error function. When the comparison result is lower than the set tolerance threshold ε, the system recognizes that the two have logical consistency and automatically completes the merging and synchronous update of the reasoning results to ensure that the judgment results under the heterogeneous reasoning path have fusion capability and reliability.

[0057] The expression for the abstract similarity error function is:

[0058]

[0059] In the formula, : Consistency error (relative value), H (edge): Hash vector of edge node inference summary, H (local): Hash vector of UAV local inference summary, ‖·‖: Vector magnitude (e.g., L2 norm);

[0060] Technical benefits: Ensures consistent results and reliable data transmission;

[0061] (2) When the system detects a communication link interruption or network anomaly, it immediately starts the sliding cache mechanism to temporarily store the structured prediction summaries generated within the last three minutes locally; the cached content is organized using timestamps and content indexes, supporting sequential tracking and duplicate verification; after the link is restored, the system automatically restores the cached state and performs breakpoint resume transmission on the summary data that has not yet been transmitted, so as to ensure data continuity and control information integrity.

[0062] The beneficial effects of this invention are as follows:

[0063] 1. This invention constructs a communication link scoring function that integrates packet loss rate, bandwidth, latency, and signal-to-noise ratio to achieve comprehensive quality assessment of different low-altitude communication links (UHF, 5G, satellite). The scoring results drive inference model switching and task migration strategies, thereby significantly improving the system's communication adaptability in complex, dynamic, or sudden scenarios. Compared with existing communication mechanisms that rely on fixed links or preset priorities, this invention can switch the optimal communication path in real time according to changes in link quality, ensuring that the inspection task is not interrupted and the inference process is not lost, thus enhancing the stability and continuity of task execution.

[0064] 2. This invention introduces a task graph segmentation mechanism based on flight status, remaining battery power, and execution deviation criteria. This mechanism supports dynamic segmentation of the original task graph during inspection and evaluates the path cost and resource availability of each candidate node through a multi-dimensional cost function. This allows for the selection of the optimal UAV to take over the task. This mechanism can quickly trigger subgraph continuation in situations such as nodes about to lose contact, mission failure, or severe yaw, effectively reducing the risk of global mission interruption caused by single node failure. Compared with existing fixed path planning schemes, this invention has stronger task scheduling flexibility and multi-UAV collaborative capabilities.

[0065] 3. This invention constructs a consistency comparison mechanism between edge and local inference results based on a summary similarity error function, realizing intelligent verification and result fusion judgment among multiple source nodes. This avoids data conflicts caused by differences in model accuracy or environmental disturbances. At the same time, combined with a sliding caching mechanism, it automatically retains the recent inference summary when communication is interrupted and automatically triggers breakpoint resumption after the link is restored, ensuring the continuity of task execution and data recoverability. Compared with existing solutions that rely only on single-end inference or lack consistency evaluation, this invention significantly improves the system's intelligent discrimination and data integrity assurance capabilities under abnormal conditions. Attached Figure Description

[0066] Figure 1 This is a flowchart of the UAV emergency inspection system that integrates low-altitude communication and edge computing, as described in this invention. Detailed Implementation

[0067] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0068] like Figure 1 As shown, this embodiment of the invention provides an emergency inspection system for unmanned aerial vehicles (UAVs) that integrates low-altitude communication and edge computing. The system consists of a link status monitoring and scoring module, an inference computing migration and model loading module, a task graph modeling and dynamic segmentation module, a multi-UAV task collaboration and trajectory continuation module, a flight status fusion and path adjustment module, a redundancy criterion-driven fault-tolerant control module, and a result consistency verification and data caching module.

[0069] Link status monitoring and scoring module: used to access UHF communication module, 5G micro base station module and satellite communication module, collect packet loss rate, bandwidth, delay and signal-to-noise ratio parameters, and calculate the quality of each link through weighted communication link scoring function to obtain score S; when S is lower than the preset first threshold T1, output link switching command and inference migration request.

[0070] The inference computation migration and model loading module is used to receive the link score S and determine the current available bandwidth and resource conditions according to the model switching trigger function. It automatically loads one of the standard model, medium compression model, or summary inference model. When the link score S is lower than the first threshold T1, it triggers the inference task to migrate to the edge node for processing and outputs the model summary result to the verification module.

[0071] Task graph modeling and dynamic segmentation module: This module models the inspection area as a directed task graph G=(V,E) with a risk value W, where V is the target point set and E is the path edge set. When any condition in the graph segmentation trigger criterion function is met, including: the remaining battery power is lower than the second threshold T2, the current track deviation angle is greater than T3, or the number of mission failures exceeds N, the subgraph G′ is triggered and the task index table is updated synchronously.

[0072] Multi-UAV mission coordination and trajectory continuation module: It is used to receive subgraph G′ and broadcast it to UAV nodes within the communication range. It uses a multi-dimensional cost function to comprehensively evaluate the path fitting distance and remaining processing capacity, and selects the UAV node with the lowest cost value to execute the subgraph task. The continuation information includes subgraph index, starting node number and synchronous execution status code.

[0073] Flight status fusion and path adjustment module: It is used to collect the current flight attitude data, remaining power ratio and link score S of the UAV, and calculate the status score P through the status fusion scoring function. The score P is used to dynamically control the adjustment of flight altitude, speed and inspection frequency, and feeds the scoring results back to the graph segmentation module to adjust the stability of the mission graph structure.

[0074] Redundancy criterion-driven fault-tolerant control module: used to monitor the communication interruption duration Δt and the inference task accuracy E in parallel. Based on the dual-threshold fault trigger function, it determines whether the task abortion condition is met. If Δt>Δt1 and E>T5, the current task is aborted, and the unfinished task graph substructure G′ is forwarded to the backup drone or the operation and maintenance control center.

[0075] The result consistency verification and data caching module is used to compare the summary similarity error function between the edge inference output and the local inference result. When the result error is less than the set tolerance ε, it is determined to be consistent and the result is merged. When the communication link is abnormal, the module calls the sliding cache mechanism to retain the structured prediction summary within the last three minutes and completes the breakpoint resume transmission after the link is restored.

[0076] Example 1: Link-based scoring-driven model switching and edge migration

[0077] During an emergency inspection mission following an earthquake in a mountainous area, the UAV system simultaneously accessed UHF, 5G micro base stations, and satellite communication links, collecting real-time data on packet loss rate, bandwidth, latency, and signal-to-noise ratio. The system calculated a score S using a weighted communication link scoring function. When the score S fell below a first threshold T1, a link switch was automatically triggered, prioritizing the link with the highest score. If the communication quality was insufficient to support the operation of the complete inference model, the system further evaluated the available bandwidth and the idle computing power of the edge nodes. It then automatically loaded an inference model with a compression level appropriate (such as a medium compression model or a summary inference model) through a model switching trigger function and migrated the current inference task to the edge node for execution. After the inference was completed, the summary result was sent back to the result verification module, achieving highly robust information processing capabilities and effectively ensuring uninterrupted mission operation in low-bandwidth scenarios.

[0078] Example 2: Mission Graph Segmentation and Multi-UAV Collaborative Succession Mechanism

[0079] In a joint inspection mission at a large petrochemical park, the system first models the inspection target area as a directed graph G=(V,E) with risk value weights, where nodes V are the set of inspection points and edges E are the path structure. As the mission progresses, the remaining battery power of a certain UAV gradually decreases to the second threshold T2, and at the same time, the system detects that the trajectory deviation angle exceeds the set threshold T3. The system determines that the current mission state has met the graph segmentation trigger criterion, so it automatically divides the subgraph G′ and updates the mission index table in real time. Subsequently, the system broadcasts the subgraph continuation request to all cooperative UAV nodes within the communication range. Each node calculates the multidimensional cost value based on the path fitting distance and the local remaining processing capacity. The system selects the node with the lowest cost value to continue executing the sub-task graph. During the continuation execution, the UAV continuously collects flight attitude, battery ratio, and link score data, and generates a score P through the state fusion scoring function. The system dynamically adjusts the flight speed, altitude, and inspection frequency based on the score P, and further feeds it back to the graph segmentation module to optimize the stability of the mission structure.

[0080] Example 3: Fault Tolerance and Data Consistency Guarantee Mechanism

[0081] During coastal power transmission inspections in areas with frequent high-wind-speed interference, the system detected that the duration of the current communication link interruption Δt exceeded the preset upper limit Δt1, while the inference task accuracy E was still higher than the acceptable threshold T5. Based on this, the system determined that the task was at risk of going out of control and automatically executed the task abort command through a dual-threshold fault trigger function. The system also forwarded the unfinished task subgraph G′ to the backup drone or the operation and maintenance command center to ensure that the task chain was not lost. At the same time, after the edge node completed the inference summary output, the system compared the result with the drone's local inference summary result using a similarity error function to determine whether the result was within the tolerance ε, ensuring data consistency. If the communication was interrupted, the system would call a sliding cache mechanism to temporarily save the structured summary result for nearly three minutes. After the link was restored, the system would automatically resume the transmission from the breakpoint, thereby achieving lossless data delivery and continuous task progress.

[0082] The link status monitoring and scoring module refers to the unified access of the UHF communication module, 5G micro base station module, and satellite communication module, collecting real-time communication parameters including packet loss rate, bandwidth, signal-to-noise ratio, and communication latency. The system calls the communication link scoring function to perform weighted calculations on the above indicators, constructs a unified scoring model, quantifies the current transmission quality of each link, and forms a status score S. This scoring result serves as an important basis for subsequent communication decisions and computation scheduling, ensuring that communication and computation behaviors have perception-driven capabilities. When the link status score S is lower than the preset first threshold T1, the system identifies it as a decline in communication quality and triggers a link switching mechanism to ensure the normal transmission of critical task data. At the same time, the system generates an inference migration request, initiates the model unloading and edge node model loading process, and transfers the computation tasks originally executed locally to the computation unit with higher link quality, ensuring uninterrupted tasks, strong computational continuity, and stable energy consumption control, realizing the integration and collaboration of low-altitude communication and edge computing.

[0083] The inference computation migration and model loading module receives the score value S from the link scoring module and, in conjunction with the model switching trigger function, determines the current available bandwidth and computing resource status in real time. When the score S meets the requirements of high bandwidth and sufficient computing power, the system automatically loads a standard inference model to perform complete feature extraction. If the system detects insufficient bandwidth or limited resources, it loads a compressed model or a summary inference model to ensure the computational task is stable and feasible under different link environments. When the link score S is lower than the first threshold T1, the system determines that the communication quality cannot support the local model's operation, automatically triggers the inference task migration process, schedules the current inference task to an edge node for execution, and unloads the local model to release resources. After the edge node completes inference, it generates a structured model summary result and sends it back to the local node. This result serves as the basis for subsequent consistency verification and redundancy judgment, ensuring the stability of cross-node inference and result synchronization.

[0084] The task graph modeling and dynamic segmentation module involves constructing a structured graph model of the inspection area, treating the target to be inspected as the node set V of the graph, establishing the possible paths between nodes as the edge set E, and comprehensively generating node risk values ​​W by combining historical failure frequency, regional sensitivity level, and target type. By introducing this risk weight, the system can prioritize the allocation of resources to critical task points, improve inspection efficiency under limited energy conditions, and form a directed task graph G with priority scheduling capabilities. During the inspection process, when the system detects that a critical operating state meets any abnormal condition in the graph segmentation trigger criterion function, such as when the remaining power is lower than the threshold T2, the current trajectory deviation exceeds the angle T3, or the number of consecutive task failures exceeds the upper limit N, the segmentation mechanism of the task graph G is activated. The system automatically divides the original graph into a subgraph G′ and simultaneously corrects the current task index table to support subsequent multi-aircraft succession scheduling and task takeover processing, realizing a highly robust emergency collaborative inspection strategy.

[0085] The multi-UAV mission coordination and trajectory continuation module refers to the system broadcasting the generated sub-graph G′ to other UAV nodes within the current communication range via a low-altitude communication module after the mission graph is segmented. Each receiving node comprehensively evaluates the fitting distance between its own trajectory and the sub-graph path, as well as its remaining computing resources and power processing capabilities, through a built-in multi-dimensional cost function to complete the continuation selection decision. This evaluation mechanism ensures the spatial continuity and resource adaptability of mission relay. After the cost evaluation is completed, the system selects the UAV node with the lowest comprehensive cost value to execute the sub-graph task and pushes a mission continuation instruction package to it. The instruction package contains sub-graph index information, continuation start point number, and synchronous execution status code for unified identification and scheduling control of mission status. After the selected node loads the task, it automatically updates its local index table and sends back a continuation success signal to the original node, achieving seamless mission transition.

[0086] The flight state fusion and path adjustment module involves real-time acquisition of the UAV's flight attitude data, remaining battery percentage, and communication link score S. It then calls a state fusion scoring function to weightedly fuse these multi-source state indicators, generating a single score P. This score comprehensively reflects the current flight stability, endurance, and communication reliability, providing a quantitative basis for subsequent control commands and mission adjustments. The state score P serves as the core input signal for executing control strategies, dynamically adjusting the UAV's flight altitude, speed, and inspection frequency to achieve a balance between flight performance and mission efficiency. Simultaneously, the system feeds the score back to the mission graph segmentation module, adjusting the stability weights of the mission graph structure based on state change trends. This ensures the graph model has adaptive response capabilities to changes in flight risks, improving the robustness of mission scheduling.

[0087] The redundancy criterion-driven fault-tolerant control module refers to the parallel monitoring of the current communication interruption duration Δt and the inference task execution accuracy E, constructing a state detection channel oriented towards task reliability. The system uses a dual-threshold fault trigger function, setting a communication interruption time threshold Δt1 and an accuracy threshold T5 to jointly determine the current state. This criterion avoids false triggering due to fluctuations in a single indicator, enhancing the accuracy and robustness of fault detection. When the system detects that the combined conditions of Δt greater than Δt1 and E higher than T5 are simultaneously met, it immediately triggers the current task termination logic to prevent the spread of data errors due to communication interruption or model inaccuracy. The system automatically extracts the incomplete substructure G′ in the current task graph and, based on the current communication status, prioritizes forwarding it to the backup UAV node. When all air nodes lose contact, it forwards it to the ground operation and maintenance control center, realizing a multi-level fault tolerance and task continuation strategy.

[0088] The result consistency verification and data caching module refers to the structured comparison between the result summary returned by the edge inference node and the local inference output. By calling the summary similarity error function, the difference between the results is calculated. When the comparison result is lower than the set tolerance threshold ε, the system determines that the two have logical consistency and automatically completes the merging and synchronous update of the inference results, ensuring that the judgment results under heterogeneous inference paths have fusion capability and reliability. When the system detects a communication link interruption or network anomaly, it immediately starts a sliding cache mechanism to temporarily store the structured prediction summaries generated within the past three minutes locally. The cached content is organized using timestamps and content indexes, supporting sequential tracking and duplicate verification. After the link is restored, the system automatically restores the cache state and performs breakpoint resume transmission for the summary data that has not yet been transmitted, realizing data continuity assurance and information integrity control.

[0089] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0090] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An emergency inspection system for unmanned aerial vehicles (UAVs) integrating low-altitude communication and edge computing, characterized in that: The system includes: Link status monitoring and scoring module: Access UHF, 5G micro base station and satellite communication links, collect packet loss rate, bandwidth, latency and signal-to-noise ratio, output link score S through weighted communication link scoring function and trigger handover command; Inference computation migration and model loading module: Based on the link score S and the current bandwidth and computing power conditions, load the standard model, medium compression model or summary model. If the score S is lower than the threshold, migrate the inference task and output the results. Task graph modeling and dynamic segmentation module: Constructs the inspection area into a directed graph with risk weights, and determines whether to segment it into sub-task graphs based on battery level, track deviation and number of mission failures and updates the task index information; Multi-UAV mission collaboration and trajectory continuation module: Receives task subgraph information, calculates path distance and the cost of remaining processing resources, selects the UAV node with the lowest cost to execute the task and synchronizes task status data; Flight status fusion and path adjustment module: Collects flight attitude, battery percentage and link score, calculates score P using the status fusion scoring function, and adjusts flight altitude, speed and inspection frequency parameters accordingly; Redundancy criterion-driven fault-tolerant control module: Real-time monitoring of communication interruption duration and inference accuracy, determining whether to abort the task based on dual threshold functions, and sending the incomplete subgraph to the backup drone or control node; The result consistency verification and data caching module compares the edge and local inference digest errors to determine whether the results are consistent. In the event of a link interruption, it starts caching to save inference data and resumes transmission after the link is restored.

2. The UAV emergency inspection system integrating low-altitude communication and edge computing according to claim 1, characterized in that: The link status monitoring and scoring module includes: (1) The UHF communication module, 5G micro base station module and satellite communication module are uniformly accessed and real-time communication parameters including packet loss rate, bandwidth, signal-to-noise ratio and communication delay are collected. The system calls the communication link scoring function to perform weighted calculation on the above indicators, constructs a unified scoring model, quantifies the current transmission quality of each link, and forms a status score S. (2) When the link status score S is lower than the preset first threshold T1, the system identifies it as a decline in communication quality and triggers the link switching mechanism to ensure the normal transmission of critical task data. At the same time, the system generates an inference migration request, starts the model unloading and edge node model loading process, and transfers the computing tasks originally executed locally to computing units with higher link quality to ensure that the tasks are not interrupted, the computing continuity is strong, and the energy consumption control is stable, so as to realize the integration and collaboration of low-altitude communication and edge computing.

3. The UAV emergency inspection system integrating low-altitude communication and edge computing according to claim 1, characterized in that: The inference computation migration and model loading module includes: (1) Receive the score value S from the link scoring module and, in conjunction with the model switching trigger function, determine the current available bandwidth and computing resource status in real time; if the score S meets the conditions of high bandwidth and sufficient computing power, the system automatically loads the standard inference model to perform complete feature extraction; if the system detects that the bandwidth is limited or the resources are restricted, it loads the compression model or the summary inference model to ensure that the computing task has stable feasibility in different link environments. (2) When the link score S is lower than the first threshold T1, the system determines that the communication quality cannot support the local model operation, automatically triggers the inference task migration process, schedules the current inference task to the edge node for execution, and unloads the local model to release resources; after the edge node completes the inference, it generates a structured model summary result and sends it back to the local machine.

4. The UAV emergency inspection system integrating low-altitude communication and edge computing according to claim 1, characterized in that: The task graph modeling and dynamic segmentation module includes: (1) Construct a structured graph model of the inspection area, take the target to be inspected as the set of nodes V of the graph, establish the possible paths between nodes as the set of edges E, and combine the historical failure frequency, regional sensitivity level and target type to generate the node risk value W. (2) During the inspection process, when the system detects that the key operating status meets any abnormal condition in the graph segmentation trigger criterion function, when the remaining power is lower than the threshold T2, the current track deviates more than the angle T3, or the number of consecutive mission failures exceeds the upper limit N, the task graph G segmentation mechanism is started; the system automatically divides the subgraph G′ from the original graph and synchronously corrects the current task index table.

5. The UAV emergency inspection system integrating low-altitude communication and edge computing according to claim 1, characterized in that: The multi-UAV mission coordination and trajectory continuation module includes: (1) When the task map is segmented, the system will broadcast the generated sub-map G′ to other UAV nodes within the current communication range through the low-altitude communication module; each receiving node will comprehensively evaluate the fitting distance between its own trajectory and the sub-map path, as well as the remaining computing resources and power processing capabilities through the built-in multi-dimensional cost function, and complete the succession optimization judgment. (2) After the cost assessment is completed, the system selects the UAV node with the lowest comprehensive cost to execute the subgraph task and pushes the task continuation instruction package to it. The instruction package contains subgraph index information, continuation starting point number and synchronous execution status code, which are used for unified identification and scheduling control of task status. After the selected node loads the task, it automatically updates its local index table and sends back the continuation success signal to the original node to achieve seamless task transition.

6. The UAV emergency inspection system integrating low-altitude communication and edge computing according to claim 1, characterized in that: The flight status fusion and path adjustment module includes: (1) Real-time acquisition of the current UAV's flight attitude data, remaining battery percentage and communication link score S, and call the state fusion scoring function to perform weighted fusion of the above multi-source state indicators to generate a single score result P; the score result comprehensively reflects the current flight stability, endurance and communication support level, and provides a quantitative basis for subsequent control commands and mission adjustments; (2) The state score P serves as the core input signal for the execution control strategy. It is used to dynamically adjust the flight altitude, speed and inspection frequency of the UAV to achieve a balance between flight performance and mission efficiency. At the same time, the system feeds back the score results to the mission graph segmentation module and adjusts the stability weight of the mission graph structure according to the state change trend.

7. The UAV emergency inspection system integrating low-altitude communication and edge computing according to claim 1, characterized in that: The redundancy criterion-driven fault-tolerant control module includes: (1) Parallel monitoring of the current communication interruption duration Δt and the inference task execution accuracy E, constructing a state detection channel oriented towards task reliability; the system sets the communication interruption time threshold Δt1 and the accuracy threshold T5 based on the dual-threshold fault trigger function, and jointly judges the current state; (2) When the system detects that the composite conditions of Δt greater than Δt1 and E higher than T5 are met simultaneously, the current task termination logic is immediately triggered. The system automatically extracts the unfinished substructure G′ in the current task graph and forwards it to the backup UAV node according to the current communication status. When all air nodes lose contact, it forwards to the ground operation and maintenance control center to realize multi-level fault tolerance and task continuation strategy.

8. The UAV emergency inspection system integrating low-altitude communication and edge computing according to claim 1, characterized in that: The result consistency verification and data caching module includes: (1) The result summary returned by the edge reasoning node is compared with the local reasoning output in a structured manner. The difference between the results is calculated by calling the summary similarity error function. When the comparison result is lower than the set tolerance threshold ε, the system recognizes that the two have logical consistency and automatically completes the merging and synchronous update of the reasoning results to ensure that the judgment results under the heterogeneous reasoning path have fusion capability and reliability. (2) When the system detects a communication link interruption or network anomaly, it immediately starts the sliding cache mechanism to temporarily store the structured prediction summaries generated within the last three minutes locally; the cached content is organized using timestamps and content indexes, supporting sequential tracking and duplicate verification; after the link is restored, the system automatically restores the cached state and performs breakpoint resume transmission on the summary data that has not yet been transmitted, so as to ensure data continuity and control information integrity.

Images