Unmanned aerial vehicle networking railway inspection method and system

Abstract

The application discloses a kind of unmanned aerial vehicle networking railway inspection method and system, the method includes: in the process of performing inspection task by the inspection cluster consisting of multiple unmanned aerial vehicles, based on the real-time evaluation result of communication link quality between unmanned aerial vehicles, trigger and execute network reconstruction decision to adjust the communication network topology and data forwarding path of inspection cluster;Based on the state information of unmanned aerial vehicle in inspection cluster, the load information of communication network and the progress information of inspection task, generate scheduling instruction through multi-objective optimization model, dynamically adjust unmanned aerial vehicle waypoint, task sequence and rotation strategy;Railway facility image data collected by unmanned aerial vehicle is executed end-edge-cloud collaborative processing, wherein, abnormal identification is carried out on image data by artificial intelligence model in edge computing node, and the transmission priority of data is determined according to the identification result;After inspection task is completed, control unmanned aerial vehicle to return to unmanned aerial vehicle nest deployed along railway, and generate inspection report based on identification result.

Description

Technical Field

[0001] This application relates to the field of unmanned aerial vehicle (UAV) application technology, and more specifically, to a UAV network railway inspection method and system. Background Technology

[0002] Traditional inspection methods mainly rely on manual ground patrols or large inspection vehicles, which have drawbacks such as low efficiency, high risk, and limited coverage. Drone technology is being gradually applied to railway inspection, enabling aerial observation of track facilities by carrying payloads such as visible light cameras.

[0003] However, when applying drones to large-scale, long-distance railway inspection scenarios, existing technological solutions still face a series of key technical challenges that urgently need to be addressed, restricting their level of fully automated and highly reliable application: 1. Insufficient communication reliability, making it difficult to adapt to complex and dynamic environments. Existing solutions mostly adopt static or pre-set networking strategies, with the communication network topology between UAVs fixed before the mission. In scenarios with complex terrain (such as tunnels and mountain obstructions) and variable electromagnetic environments along railway lines, fixed communication links are easily interrupted due to obstruction or interference, leading to data transmission failures or even UAV disconnection, seriously threatening the continuity and safety of inspection missions.

[0004] 2. The task scheduling strategy is simplistic, resulting in low overall system efficiency. Existing UAV inspection scheduling methods often focus on single-drone endurance or simple relay, failing to comprehensively consider multi-dimensional constraints such as communication load balancing, task urgency, and multi-drone collaboration. When a UAV runs out of power or encounters an emergency, the system cannot make globally optimal decisions and reallocate resources, leading to reduced inspection efficiency and insufficient resource utilization.

[0005] 3. Centralized data processing results in poor real-time response capabilities. Currently, the common practice is to transmit all the massive amounts of image and video data collected by drones back to a remote central server for processing and analysis. This centralized processing model places extremely high demands on communication bandwidth and introduces significant latency over long distances. For safety hazards requiring immediate response, such as damage to the overhead contact line or foreign objects on the track, it fails to meet the real-time requirement of "detecting during inspections and issuing warnings during the process," thus missing the optimal window for emergency response.

[0006] 4. The system lacks robustness and effective self-healing and fault-tolerance mechanisms. Existing systems typically lack rapid, automatic task reconfiguration and takeover mechanisms when faced with sudden malfunctions of a single drone, power depletion, or communication node failures. This often leads to the interruption of the entire inspection mission or necessitates manual intervention for replanning, making it difficult to achieve true unattended operation and routine automated operation.

[0007] In summary, existing UAV-based railway inspection technologies have significant shortcomings in terms of communication adaptability, intelligent scheduling, real-time processing, and system resilience. Therefore, there is an urgent need for a UAV-based networked inspection system and method capable of dynamic and reliable networking, multi-target intelligent collaboration, near real-time data processing, and high robustness to meet the fully automated and highly reliable operation and maintenance requirements of long-distance, complex railway lines. Summary of the Invention

[0008] To address the shortcomings of existing technologies, this invention provides a method and system for UAV-based railway inspection, which solves the problems of poor communication adaptability, intelligent scheduling, and real-time processing capabilities in current UAV-based railway inspection technologies. It enables UAV-based railway inspection with dynamic and reliable networking, multi-target intelligent collaboration, near real-time data processing, and high robustness.

[0009] To achieve the above objectives, according to a first aspect of the present invention, a method for unmanned aerial vehicle (UAV) network-based railway inspection is provided. The method includes: during the execution of an inspection task by an inspection cluster composed of multiple UAVs, triggering and executing a network reconfiguration decision based on real-time evaluation results of the communication link quality between the UAVs to adjust the communication network topology and data forwarding path of the inspection cluster; generating scheduling instructions through a multi-objective optimization model based on the UAV status information, the communication network load information, and the inspection task progress information, dynamically adjusting UAV waypoints, task sequences, and rotation strategies; performing end-edge-cloud collaborative processing on railway facility image data collected by the UAVs, wherein an artificial intelligence model is used at the edge computing node to identify anomalies in the image data and determine the data transmission priority based on the identification results; and controlling the UAVs to return to their deployment nests along the railway line after the inspection task is completed, and generating an inspection report based on the identification results.

[0010] In an exemplary embodiment, triggering and executing a network reconstruction decision to adjust the communication network topology and data forwarding path of the inspection cluster based on the real-time evaluation results of the inter-UAV communication link quality includes: monitoring the received signal strength indicator, signal-to-noise ratio, data packet delay, and data packet loss rate of the inter-UAV communication link, and obtaining a link quality index through weighted fusion calculation; when the link quality index is lower than a preset threshold, generating a network reconstruction scheme based on a reinforcement learning algorithm, wherein the network reconstruction decision includes at least one of the following: converting the inspection UAV into a relay UAV, updating the data route, or scheduling a standby UAV to take off and establish a temporary relay node. Specifically, this may involve instructing the inspection UAV to convert into a relay UAV, replanning the data backhaul route, or scheduling a standby UAV to take off and establish a temporary relay node.

[0011] In one exemplary embodiment, the objective function of the multi-objective optimization model is... Specifically: ,in, For task coverage, For emergency response level, For network congestion, , , The weighting coefficients are dynamically adjusted based on the real-time task context; the generated scheduling instructions include waypoint instructions for controlling the UAV to perform task handover during flight, and flight waypoint instructions inserted to verify identified anomalies.

[0012] In an exemplary embodiment, the edge-cloud collaborative processing of the railway facility image data collected by the UAV includes: compressing and appending metadata to the collected image data at the UAV end to obtain preprocessed data, wherein the appending processing includes appending GNSS coordinates and timestamps; and using a pruned and quantized deep learning model at the edge computing node to perform target detection and defect identification on the preprocessed data, and outputting analysis results including target bounding boxes, category labels, and confidence scores.

[0013] In one exemplary embodiment, the method further includes: marking analysis results with confidence levels higher than a target threshold as high-priority anomalies and generating an alarm package containing an anomaly summary, key evidence image slices, and location information, and transmitting it to the cloud platform with priority; marking analysis results with confidence levels not higher than the target threshold as low-priority data and transmitting them to the cloud platform according to the normal process.

[0014] In one exemplary embodiment, the method further includes: recording network reconstruction events, scheduling instructions, and anomaly analysis results during task execution; and updating the parameters of the deep learning model, the multi-objective optimization model, and the lightweight artificial intelligence model based on the recorded data.

[0015] According to a second aspect of the present invention, a UAV-based networked railway inspection system is also provided, comprising: a dynamic self-organizing network module, integrated into each UAV, used to trigger and execute network reconstruction decisions to adjust the communication network topology and data forwarding path of the inspection cluster based on real-time evaluation results of the communication link quality between the UAVs during the execution of inspection tasks by an inspection cluster composed of multiple UAVs; and an intelligent scheduling module, deployed on a cloud platform or edge computing node, used to generate scheduling instructions through a multi-objective optimization model based on the status information of the UAVs in the inspection cluster, the load information of the communication network, and the progress information of the inspection tasks. The system dynamically adjusts UAV waypoints, mission sequences, and rotation strategies; a collaborative processing module performs edge-cloud collaborative processing on railway facility image data collected by the UAVs, wherein an artificial intelligence model is used at the edge computing node to identify anomalies in the image data and determine the data transmission priority based on the identification results; after the inspection task is completed, the UAVs are controlled to return to the UAV nests deployed along the railway line, and an inspection report is generated based on the identification results; the nests and UAV clusters are used, with the nests deployed along the railway line for the storage, take-off and landing, and energy replenishment of the UAVs; the UAV clusters are used to perform inspection flights and data collection.

[0016] In an exemplary embodiment, the coordination processing module includes: a preprocessing unit deployed on the UAV for compressing and appending metadata to the acquired raw image data; an analysis unit deployed on an edge computing node, which has a built-in lightweight deep learning model for defect detection and classification of the received preprocessed data; and a management unit deployed on a cloud platform for aggregating data, reviewing high-priority alarms, and generating inspection reports.

[0017] According to a third aspect of the present invention, a computer-readable storage medium is also provided, wherein a computer program is stored in the computer program, wherein the computer program is configured to execute the above-described UAV network railway inspection method when it is run.

[0018] According to a fourth aspect of the present invention, an electronic device is also provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the above-described unmanned aerial vehicle (UAV) network railway inspection method via the computer program.

[0019] In summary, compared with the prior art, the above-described technical solutions conceived by this invention can achieve the following beneficial effects: (1) This invention provides a UAV network railway inspection method. By introducing a comprehensive link health index and a real-time network reconstruction algorithm based on reinforcement learning, the UAV cluster can autonomously and dynamically adjust the communication topology and data routing according to the real-time channel quality, node location, and network load. This adaptive mechanism can effectively avoid communication interruptions caused by complex terrain (such as tunnels and mountains) or electromagnetic interference along the railway line, and can quickly self-heal when nodes or links fail. This significantly reduces the risk of UAV disconnection and loss of key inspection data, providing stable and reliable data transmission guarantee for continuous operation in long-distance and complex environments, and improving the overall robustness of the system.

[0020] (2) This invention constructs a multi-factor constrained intelligent scheduling model that integrates endurance continuity, mission urgency, and communication load balancing, breaking through the limitations of existing technologies that only focus on single-unit endurance or simple mission relay. The system can intelligently decide on the role switching (inspection / relay), waypoint adjustment, and precise rotation at dynamic handover points of drones based on the global real-time status (drone battery level, location, mission progress, network status, sudden anomalies, etc.). Thus, while ensuring seamless mission continuity and timely response to high-priority missions, it maximizes the utilization of the entire drone cluster resources (drones, communication bandwidth, time) and significantly improves the overall inspection efficiency.

[0021] (3) By adopting a three-level collaborative processing architecture of end-edge-cloud, an optimized lightweight artificial intelligence model is deployed at the network edge node (such as ground-based data centers or high-performance UAVs), which enables real-time analysis and anomaly screening of image data at or near the data acquisition source. This allows for the identification and early warning of critical equipment defects such as rail cracks, loose fasteners, and insulator bursts to achieve a response speed of minutes or even seconds.

[0022] (4) This invention deeply integrates dynamic adaptive networking, multi-objective intelligent scheduling, edge intelligent analysis and task closed-loop management. Compared with existing solutions that rely on segmented presets or a large amount of manual intervention, this invention achieves truly robust, unattended, fully automatic inspection, significantly reducing maintenance manpower costs and human error rate. At the same time, by recording operation logs to continuously optimize algorithm parameters, the system has the ability to self-evolve and adapt to complex environments in the long term, comprehensively improving the intelligence level, operational efficiency and reliability of railway infrastructure operation and maintenance. Attached Figure Description

[0023] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 A flowchart illustrating an optional UAV-based railway inspection method provided in this application embodiment; Figure 2 This is a schematic diagram of an optional electronic device provided in an embodiment of this application. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0026] The terms "first," "second," "third," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish different objects, not to describe a specific order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or apparatuses.

[0027] According to one aspect of the embodiments of this application, a method for unmanned aerial vehicle (UAV) network-based railway inspection is provided. The following is in conjunction with... Figure 1 This application describes a method for unmanned aerial vehicle (UAV) networked railway inspection provided in its embodiments.

[0028] Figure 1 This is a flowchart illustrating an optional UAV-based railway inspection method provided in an embodiment of this application. Figure 1 As shown, the process of this method may include the following steps: S102, during the inspection task performed by the inspection cluster composed of multiple drones, based on the real-time evaluation results of the communication link quality between the drones, a network reconstruction decision is triggered and executed to adjust the communication network topology and data forwarding path of the inspection cluster. S104, based on the status information of the UAVs in the inspection cluster, the load information of the communication network, and the progress information of the inspection task, a scheduling instruction is generated through a multi-objective optimization model to dynamically adjust the UAV waypoints, task sequences, and rotation strategies. S106, perform end-edge-cloud collaborative processing on the railway facility image data collected by the UAV, wherein an artificial intelligence model is used at the edge computing node to identify anomalies in the image data and determine the data transmission priority based on the identification results; S108: After completing the inspection task, control the drone to return to the drone nest deployed along the railway line, and generate an inspection report based on the identification results.

[0029] This application provides a UAV-based railway inspection method to achieve fully automated, highly reliable collaborative inspection and real-time status awareness of railway line facilities over large areas and long distances. It solves key technical challenges such as communication assurance, mission continuity maintenance, and real-time data processing for inspection UAV swarms in complex geographical environments. Optionally, this application can be applied to the daily inspection and emergency survey of the overhead contact system, tracks, roadbed, and surrounding environment of conventional railways, high-speed railways, and urban rail transit lines, significantly improving the intelligence level and safety assurance capabilities of railway infrastructure operation and maintenance.

[0030] Optionally, this inspection task can be initiated by the central control platform responding to a work order at a specific stage of the railway's entire lifecycle. The platform first reads the inspection task instruction submitted by the user or automatically generated by the system. This instruction clarifies the geographical range of the target railway section and the core objectives of this inspection (for example, during the construction phase, the integrity of slope protection and ballast laying needs to be monitored, while during the operation and maintenance phase, the focus is on rail smoothness and fastener tightness).

[0031] Subsequently, the platform will invoke the network of fixed drone nests deployed along the route, automatically calculating and locking the optimal nest set for this mission based on the start and end points of the mission segment. At the resource allocation level, the system dynamically allocates a cluster of multiple drones equipped with high-resolution visible light cameras from the target nests based on the total inspection mileage, mission urgency, and expected operational efficiency, designating one of them as the lead node for this mission.

[0032] Next, the central platform performs global planning for the entire cluster: based on the locked nest location, it assigns each UAV its own inspection sub-section and generates an initial flight path consisting of a series of precise geographic coordinates; at the same time, based on the linear topology characteristics of the railway line, it pre-calculates and sets an initial chain or tree-like communication network topology with the lead node as the root node, clarifying the initial role and data forwarding path of each UAV node in the network.

[0033] Finally, after all the planning was in place, the central platform issued a unified take-off command and mission configuration file to all participating drones and pods. Each drone took off autonomously from its respective pod in sequence according to the command. After reaching the predetermined cruising altitude, it quickly established an inter-drone communication link according to the initial network topology and entered a standby state to fly along the preset route, making full preparations for subsequent collaborative inspection operations.

[0034] Furthermore, after the drone swarm enters the patrol and inspection state, the system immediately initiates a full-link, multi-level real-time status monitoring and data acquisition process. It primarily performs two core tasks: firstly, continuously sensing the operational status of the drone swarm itself; and secondly, capturing visualized image data of the railway facilities.

[0035] In terms of status monitoring, each drone in the cluster acts as an independent intelligent node, continuously collecting its own status parameters through its onboard sensors. These key parameters include, but are not limited to: precise three-dimensional position and real-time flight speed obtained through the global navigation satellite system; remaining power and battery health monitored by the power management system; and wireless link quality (such as signal strength, signal-to-noise ratio, and packet loss rate) between the drone and neighboring drone nodes, relay drones, or ground-based drone nests, as assessed by the communication module. All this status data is encapsulated into lightweight telemetry data packets and transmitted hop-by-hop back to the central control platform at high frequency through the established inter-drone ad-hoc network.

[0036] Simultaneously, visual data collection of railway facilities is being carried out. All drones tasked with inspection are equipped with high-resolution visible light cameras as standard payloads. During flight, these cameras are automatically triggered according to preset shooting strategies (such as equal time intervals, equal distance intervals, or geographic tag-based triggering) to systematically capture images of the railway assets below.

[0037] It should be noted that the content collected by the visualized data acquisition is closely aligned with the management needs of the entire railway lifecycle: during the construction phase, the focus is on the integrity and compliance of earthwork work surfaces and temporary support structures; during the construction phase, attention is paid to the technological quality of sleeper laying, rail connection, and fastener installation; and during the operation and maintenance phase, the macroscopic and microscopic conditions of key facilities such as rail surfaces, track bed conditions, catenary supports, and insulators are precisely covered. Optionally, the acquired high-definition image or video stream data, after initial compression and preprocessing such as timestamps and location stamps, is temporarily stored in local storage and prepared for transmission to the backend via a dynamic self-organizing network link.

[0038] In an exemplary embodiment, the step of triggering and executing a network reconfiguration decision to adjust the communication network topology and data forwarding path of the inspection cluster based on the real-time evaluation results of the inter-UAV communication link quality includes: Monitor the received signal strength indication, signal-to-noise ratio, data packet delay, and data packet loss rate of the communication link between UAVs, and obtain the link quality index through weighted fusion calculation; When the link quality index is lower than a preset threshold, a network reconstruction scheme is generated based on a reinforcement learning algorithm. The network reconstruction decision includes at least one of the following: converting the inspection drone into a relay drone, updating the data route, or scheduling a standby drone to take off and establish a temporary relay node.

[0039] The command can be used to convert the inspection drone into a relay drone, re-plan the data backhaul route, or dispatch a backup drone to take off and establish a temporary relay node.

[0040] In this embodiment, the stability of the communication link is a fundamental prerequisite for ensuring reliable data transmission and effective command issuance during the flight of the UAV swarm along a preset route. The core objective of this step is to address the dynamic fluctuations in the communication channel caused by the long railway lines, geographical factors, and continuous changes in the relative position of the UAVs, thereby achieving adaptive reconstruction and optimization of the communication network. This process begins with a continuous quantitative assessment of the communication link quality. Each UAV integrates a link status awareness module, which monitors multiple key performance indicators of the communication link with other UAV nodes and ground-based UAV base stations at a high frequency (e.g., several times per second). These indicators mainly include received signal strength indication, signal-to-noise ratio, data packet round-trip time, and bandwidth utilization.

[0041] The aforementioned raw metrics are input into a predefined link health calculation model, which outputs a comprehensive, dimensionless link health score through a weighted fusion algorithm:

[0042] here, The calculated link quality index, This is the normalized received signal strength indication. This represents the normalized signal-to-noise ratio. This represents the normalized average latency of data packets. This represents the packet loss rate (percentage).

[0043] , , , The corresponding weight coefficients, and satisfying The weighting coefficients are determined using the analytic hierarchy process (AHP). First, the relative importance of each indicator to link stability is assessed based on historical communication data, constructing a judgment matrix. Then, the eigenvectors of this matrix are calculated and normalized to obtain the weight values ​​for each indicator. This method combines expert experience with mathematical calculations, ensuring the rationality and objectivity of the weight allocation.

[0044] For example, when any node in the system detects that the health score of one or more of its links is lower than a preset stability threshold, a network reconstruction decision mechanism is triggered. This triggering event is considered a network state anomaly, and the relevant node immediately generates a network alarm information packet containing its own identity, the abnormal link identifier, and the current health score, and sends it to the central control platform via the currently available paths. Upon receiving the alarm, the central platform does not perform centralized path calculation, but instead activates a distributed network optimization algorithm. This algorithm is trained based on a reinforcement learning framework. Its state space is defined as the current location information, remaining energy, and health of all inter-drone links of all UAV nodes; its action space is defined as the set of all possible UAV role (inspection node or relay node) allocation combinations and data packet forwarding paths; its reward function is set to maximize the overall network throughput and minimize end-to-end communication latency as the primary objectives, while also considering the balance of network energy consumption.

[0045]

[0046] Where R is the immediate reward value of the reinforcement learning model, and T i Let D be the throughput of the i-th data stream. i Let be the end-to-end latency of the i-th data stream. The standard deviation of the remaining energy of all drones in the cluster is used to measure energy balance.

[0047] , , These are weighting coefficients used to balance the priorities of different optimization objectives. The weights are calibrated through multi-objective optimization experiments in a simulation environment. The system runs under a large number of random network topologies; by adjusting the weight combinations and observing network performance (such as total throughput and average latency), the optimal weight combination is ultimately selected that enables the system to achieve its best performance in most scenarios.

[0048] For example, after the algorithm runs, it will output a near-optimal reconstruction scheme for the current network conditions. This scheme may include one or more of the following instructions: instructing an inspection drone in a better geographical location to temporarily switch its role to a pure relay drone to bypass communication obstacle areas; recalculating the data return route to avoid using links with poor health; or, in areas where communication quality is severely degraded, instructing a standby drone in a ready state to take off from the nearest nest and go to a specific location to act as a temporary communication relay node.

[0049] Understandably, all these reconfiguration instructions will be encrypted in the form of task frames and broadcast to the relevant UAVs. Upon receiving the instructions, the UAVs will perform rapid parameter reconfiguration at the network and data link layers of their communication protocol stacks, including updating routing tables, switching communication frequencies, or adjusting transmission power, and will complete the network topology switch at a negotiated synchronization time.

[0050] This embodiment ensures that the data transmission path is always in the optimal or suboptimal state under the current conditions, significantly reducing the risk of data loss or drone disconnection due to communication interruption, and providing a stable and resilient data carrying network for the entire system.

[0051] In an exemplary embodiment, the objective function of the multi-objective optimization model is specifically: , among which, among which, For task coverage, For emergency response level, For network congestion, , , These are weighting coefficients that are dynamically adjusted based on the real-time task context. The generated scheduling instructions include waypoint instructions for controlling the UAV to hand over tasks during flight, and flight waypoint instructions inserted to verify identified anomalies.

[0052] In this embodiment, while maintaining a stable communication network, the goal is to maximize overall inspection efficiency and task reliability under multiple constraints by optimizing resource allocation and dynamically assigning tasks to multiple UAVs in long-term inspection missions. The intelligent scheduling model, as the core of this step, acts as a continuously running decision engine. Its input comes from the real-time outputs of steps two and three, forming a multi-source heterogeneous data stream, including but not limited to: real-time position vectors, remaining battery estimates, flight speed vectors, current task execution progress (such as the percentage of completed inspection mileage), and preliminary identification of suspected fault points, their geographical coordinates, and preliminary classification information reported from edge computing nodes.

[0053] In other words, the aforementioned scheduling model is essentially a complex, constrained multi-objective optimization problem. Its decision space encompasses waypoint sequence adjustment, UAV role assignment, task command issuance, and nest start / stop control. The optimization process requires balancing the following key objectives simultaneously: First, endurance continuity, ensuring that any UAV can safely return to the nest for charging or rotation before its battery runs out. This requires a high-precision remaining endurance prediction model to dynamically calculate and designate dynamic handover points between UAVs, achieving seamless relay of inspection tasks. Second, task urgency, when edge analysis identifies high-priority anomalies (such as structural deformation during construction or suspected rail cracks during maintenance), the model must respond immediately, reassess cluster resources, potentially interrupting a UAV's routine inspection task, and planning an optimal path for it to the anomaly point for detailed verification and multi-angle photography. Third, communication load balancing, scheduling decisions need to consider the current dynamic network load status to avoid a large influx of data streams onto a single critical relay link, causing network congestion. Therefore, it may be necessary to balance network traffic by adjusting the geographical location of UAVs or data backhaul strategies.

[0054] Here, S represents the overall utility score of the scheduling scheme. Task coverage rate is the ratio of completed inspection mileage to total mileage. Emergency response speed is positively correlated with the speed of processing high-priority anomalies. The network congestion level is the variance of the load on each relay link. , , These are dynamic weighting coefficients. The system adjusts them based on the real-time task context: in the early stages of the task, (Coverage) has a higher weight; when a high-priority anomaly alarm is received, The (emergency response) weight is temporarily and significantly increased; when uneven network traffic is detected, it is appropriately increased. (Congestion level) weights are used to balance the load.

[0055] Optionally, the model solution relies on heuristic algorithms from operations research, such as genetic algorithms or particle swarm optimization, to obtain a feasible optimized scheduling scheme within a reasonable timeframe. The solver's output is a specific, executable set of scheduling instructions. This set of instructions may include: sending a command to a drone with battery power below a threshold, instructing it to end its current inspection task, fly to a pre-calculated dynamic handover point, complete the task data handover with a newly launched backup drone from the pod, and then autonomously return to land; simultaneously, sending a task update command to another drone closest to the newly identified suspected fault location and capable of operation, reconstructing its subsequent waypoint sequence, inserting waypoints to the fault location, and improving its camera's shooting resolution and frequency. All instructions are distributed through the stable network maintained in step three. The drone's flight control system and mission management system receive and parse these instructions, ultimately driving the drone platform to execute corresponding flight maneuvers and mission operations.

[0056] Through this embodiment, the intelligent scheduling process realizes the transformation from passive and fixed task execution to proactive and flexible resource scheduling, ensuring the continuity, efficiency and high responsiveness of task execution under limited resources.

[0057] In one exemplary embodiment, the edge-cloud collaborative processing of the railway facility image data collected by the UAV includes: The collected image data is compressed and metadata is appended on the UAV to obtain preprocessed data, wherein the appending process includes adding GNSS coordinates and timestamps. At the edge computing node, a pruned and quantized deep learning model is used to perform target detection and defect identification on the preprocessed data, and output analysis results including target bounding boxes, category labels and confidence scores.

[0058] In this embodiment, edge-cloud collaborative data processing and anomaly identification are employed to address the challenges of real-time processing, intelligent analysis, and efficient transmission of massive amounts of inspection image data. The core of this approach is the construction of a clearly defined, collaborative three-tiered edge-cloud data processing architecture to overcome the bandwidth pressure and processing latency associated with transmitting all raw data back to the central processing center. This architecture rationally allocates tasks across the terminal, edge, and cloud based on the timeliness requirements and computational complexity of the data processing.

[0059] The terminal refers to the drone itself. At this level, the main tasks are front-end data preprocessing and caching. The embedded computing unit on the drone performs preliminary processing on the raw high-definition images or video streams captured by the visible light camera. This processing includes, but is not limited to: lossy or lossless compression to reduce data size, such as using JPEG 2000 or HEVC encoding formats; attaching rich metadata tags, such as precise GNSS coordinates, drone attitude angles, timestamps, and camera parameters; and data filtering based on simple rules, such as removing invalid frames caused by cloud cover or momentary motion blur. The preprocessed data is temporarily stored in onboard solid-state storage and ready for transmission to the next level. This step reduces data redundancy at the source, conserving valuable wireless communication resources.

[0060] Edge computing nodes are crucial for achieving real-time intelligent response in this solution. They are typically deployed in high-performance computing units within ground-based drone nests or on designated drone nodes with strong computing capabilities within a cluster. The edge layer undertakes the core tasks of near-real-time intelligent analysis and data classification. Preprocessed image data transmitted from the drone terminal is pushed to the edge nodes. An optimized, lightweight artificial intelligence model is pre-loaded and runs on these nodes. This AI model is a visual detection model based on a deep convolutional network, with its architecture selectable from lightweight versions of single-stage or two-stage object detection models such as YOLO, SSD, or Faster R-CNN, after pruning and quantization. The core task of this model is to perform efficient multi-object detection and classification. Trained on massive amounts of railway facility image data (covering normal and abnormal state samples from all stages of construction, maintenance, and operation), it can accurately identify specific target components in the images (such as sleepers, fasteners, insulators, and overhead contact line supports), and further determine whether they possess pre-defined defects (e.g., sleeper cracks, loose fastener bolts, burst insulators, surface contamination, etc.). The model performs fast inference on the input image and outputs analysis results including target bounding boxes, category labels, and corresponding confidence scores.

[0061] In one exemplary embodiment, the method further includes: Analysis results with confidence levels higher than the target threshold are marked as high-priority anomalies, and alarm packages containing anomaly summaries, key evidence image slices, and location information are generated and transmitted to the cloud platform with priority. Analysis results with confidence levels not exceeding the target threshold are marked as low-priority data and transmitted to the cloud platform according to the standard procedure.

[0062] The analysis results obtained above are divided into two categories: one category is normal or without obvious defects, which is marked as low priority and queued for return according to the normal process; the other category is identified as suspected anomalies, and its anomaly type, confidence level, and location bounding box in the image are automatically labeled. Once a high-confidence key anomaly (such as missing bolts or structural deformation) is detected, the edge node will immediately trigger an alarm mechanism, generating a lightweight alarm information packet containing an anomaly summary, key evidence image slices (not the full-size original image), and their precise location information. This alarm packet is given the highest transmission priority and is immediately reported through the dynamic network. The edge AI model's confidence level for the identification results:

[0063] Where C is the confidence level of the model output, and its value ranges from (0, 1); F is the feature vector extracted from the input image by a convolutional neural network; w is a weight vector corresponding to a specific anomaly category; b is the bias term; e is a natural constant.

[0064] The weight vector w and bias term b were ultimately determined by iterative optimization through model training using the backpropagation algorithm on a massive dataset of labeled railway component images, by minimizing the cross-entropy loss function between the predicted values ​​and the true labels.

[0065] The cloud serves as the central control platform, responsible for data aggregation, in-depth analysis, and archiving. The cloud receives all data from edge nodes: first, routine low-priority image data streams for complete data archiving, subsequent offline in-depth mining, and AI model retraining; second, high-priority alarm information packets. For alarm information, the cloud can perform secondary verification, conducting confirmatory analysis on evidence image slices uploaded from the edge, and immediately pushing confirmed alarms to the operations and maintenance management interface to initiate emergency procedures. Simultaneously, the cloud integrates and analyzes all historical inspection data, generating reports to assess the changing trends of railway facility conditions, and iteratively updates and optimizes the lightweight model on the edge side using global data before distributing it to each edge node via the network.

[0066] This embodiment effectively balances the conflict between processing speed, bandwidth consumption, and analysis depth, enabling rapid transformation from data to decision.

[0067] In one exemplary embodiment, the method further includes: Record network reconfiguration events, scheduling instructions, and anomaly analysis results during task execution; The parameters of the deep learning model, the multi-objective optimization model, and the lightweight artificial intelligence model are updated based on the recorded data.

[0068] In this embodiment, in order to maintain the long-term operational resilience of the entire system in the face of emergencies, the task closed-loop management and system resilience maintenance links cover the entire process from task termination and resource recovery to the formation of a management closed loop based on data analysis results.

[0069] First, there is the sequential termination of the mission and resource recovery. When the last UAV performing the inspection mission completes its assigned inspection sub-section, or receives a global mission termination command from the central platform, its behavior logic transitions to the mission completion phase. The UAV will autonomously fly to its designated ground nest according to the optimal return path pre-calculated by the intelligent scheduling model. Upon arrival above the nest, it communicates with the nest's guidance system to execute a precise landing procedure, including alignment, a slow descent, and a safe landing on the charging platform. After landing, the UAV rapidly exports any remaining inspection data from its onboard memory to the nest's local storage unit or directly uploads it to the central cloud platform via a wired connection, ensuring data integrity. Simultaneously, the nest's automated support system activates, charging or hot-swapping the UAV's battery and performing basic health checks on the UAV platform to prepare for the next mission. This process achieves the recovery and regeneration of UAV resources, ensuring the system's sustainable operation.

[0070] Secondly, there is the final processing and report generation of the inspection data. In the cloud, once it is confirmed that all participating drones have returned safely and that the data transmission is complete, the system initiates the data post-processing pipeline. This includes temporal and spatial stitching and alignment of all sequentially collected image data to form a visualized digital orthophoto or panoramic image of the railway line. Based on this, and combined with all anomaly records identified at the edge and in the cloud in step five, the system automatically generates a structured inspection report. This report details the scope and time of the inspection, the various defects found, their precise geographical locations, severity levels, and visual evidence images. It can also perform comparative analysis based on historical data to indicate trends in condition changes. The report is output in a standardized format and can be integrated into the railway asset management system.

[0071] Finally, the formation of a closed-loop operation and maintenance system and the demonstration of system resilience are key aspects. The generated inspection report and its anomaly list are not the end of the process. The system automatically converts confirmed anomalies into standardized maintenance work orders and pushes them to the workflow system of the railway facility maintenance management department via an application programming interface, assigning them to the appropriate responsible personnel and teams.

[0072] Furthermore, the system continuously learns throughout its entire operational cycle. It records every network reconfiguration event and its triggers, every intelligent scheduling decision and its effect, and the accuracy of every anomaly identification. These operational logs are used for offline analysis to optimize the algorithm parameters of dynamic networking, improve the weight allocation of the scheduling model, and enhance the recognition accuracy of the edge AI model. This self-optimization capability based on historical data and operational experience enables the system to continuously adapt to complex and ever-changing external environments, demonstrating strong long-term operational resilience and value.

[0073] Through steps S102 to S108, during the inspection task performed by an inspection cluster composed of multiple UAVs, network reconstruction decisions are triggered and executed based on the real-time evaluation results of the communication link quality between UAVs to adjust the communication network topology and data forwarding path of the inspection cluster. Based on the status information of the UAVs in the inspection cluster, the load information of the communication network, and the progress information of the inspection task, scheduling instructions are generated through a multi-objective optimization model to dynamically adjust the UAV waypoints, task sequences, and rotation strategies. Edge-cloud collaborative processing is performed on the railway facility image data collected by the UAVs. At the edge computing node, an artificial intelligence model is used to identify anomalies in the image data, and the data transmission priority is determined based on the identification results. After the inspection task is completed, the UAVs are controlled to return to their nests deployed along the railway line, and an inspection report is generated based on the identification results. This achieves dynamic and reliable networking, multi-objective intelligent collaboration, near real-time data processing, and highly robust UAV network inspection.

[0074] According to another aspect of the embodiments of this application, a UAV network railway inspection system is also provided, which performs the above-described UAV network railway inspection method, characterized in that it includes: The dynamic self-organizing network module, integrated into each UAV, is used to trigger and execute network reconstruction decisions to adjust the communication network topology and data forwarding path of the inspection cluster during the inspection task performed by an inspection cluster composed of multiple UAVs, based on the real-time evaluation results of the communication link quality between the UAVs. The intelligent scheduling module, deployed on a cloud platform or edge computing node, is used to generate scheduling instructions based on the status information of UAVs in the inspection cluster, the load information of the communication network, and the progress information of the inspection task, and dynamically adjust the UAV waypoints, task sequences, and rotation strategies through a multi-objective optimization model. The collaborative processing module is used to perform end-edge-cloud collaborative processing on the railway facility image data collected by the UAV. In this module, an artificial intelligence model is used at the edge computing node to identify anomalies in the image data and to determine the data transmission priority based on the identification results. After the inspection task is completed, the drone is controlled to return to the drone nest deployed along the railway line, and an inspection report is generated based on the identification results; It also includes drone nests and drone swarms. The drone nests are deployed along the railway line for storing, taking off and landing, and refueling the drones. The drone swarms are used to perform inspection flights and data collection.

[0075] In one exemplary embodiment, the coordination processing module includes: The preprocessing unit deployed on the drone is used to compress the acquired raw image data and append metadata. The analysis unit, deployed on the edge computing node, has a built-in lightweight deep learning model for defect detection and classification of the received preprocessed data; The management unit, deployed on a cloud platform, is used to aggregate data, review high-priority alarms, and generate inspection reports.

[0076] This embodiment enables dynamic and reliable networking, multi-target intelligent collaboration, near real-time data processing, and highly robust UAV network inspection.

[0077] According to another aspect of the embodiments of this application, a storage medium is also provided. Optionally, in this embodiment, the storage medium can be used to execute the program code of any of the above-described UAV network railway inspection methods in the embodiments of this application.

[0078] Optionally, in this embodiment, the storage medium is configured to store program code for performing the following steps: S1, during the inspection task performed by the inspection cluster composed of multiple drones, based on the real-time evaluation results of the communication link quality between the drones, a network reconstruction decision is triggered and executed to adjust the communication network topology and data forwarding path of the inspection cluster. S2, based on the status information of the UAVs in the inspection cluster, the load information of the communication network, and the progress information of the inspection task, a scheduling instruction is generated through a multi-objective optimization model to dynamically adjust the UAV waypoints, task sequences, and rotation strategies. S3, perform end-edge-cloud collaborative processing on the railway facility image data collected by the UAV, wherein an artificial intelligence model is used at the edge computing node to identify anomalies in the image data and determine the data transmission priority based on the identification results; S4. After the inspection task is completed, control the drone to return to the drone nest deployed along the railway line, and generate an inspection report based on the identification results.

[0079] Optionally, specific examples in this embodiment can refer to the examples described in the above embodiments, and will not be repeated in this embodiment.

[0080] The computer-readable storage medium may include, but is not limited to, any type of disk, including floppy disks, optical disks, DVDs, CD-ROMs, microdrives, as well as magneto-optical disks, ROMs, RAMs, EPROMs, EEPROMs, DRAMs, VRAMs, flash memory devices, magnetic cards or optical cards, nanosystems (including molecular memory ICs), or any type of medium or device suitable for storing instructions and / or data.

[0081] According to another aspect of the embodiments of this application, an electronic device for implementing the above-described UAV network railway inspection method is also provided. The electronic device may be a server, a terminal, or a combination thereof.

[0082] Figure 2 This is a schematic diagram of the structure of an optional electronic device according to an embodiment of this application, such as... Figure 2 As shown, it includes a processor 202, a communication interface 204, a memory 206, and a communication bus 208. The processor 202, communication interface 204, and memory 206 communicate with each other via the communication bus 208. Memory 206 is used to store computer programs; When processor 202 executes a computer program stored in memory 206, it performs the following steps: S1, during the inspection task performed by the inspection cluster composed of multiple drones, based on the real-time evaluation results of the communication link quality between the drones, a network reconstruction decision is triggered and executed to adjust the communication network topology and data forwarding path of the inspection cluster. S2, based on the status information of the UAVs in the inspection cluster, the load information of the communication network, and the progress information of the inspection task, a scheduling instruction is generated through a multi-objective optimization model to dynamically adjust the UAV waypoints, task sequences, and rotation strategies. S3, perform end-edge-cloud collaborative processing on the railway facility image data collected by the UAV, wherein an artificial intelligence model is used at the edge computing node to identify anomalies in the image data and determine the data transmission priority based on the identification results; S4. After the inspection task is completed, control the drone to return to the drone nest deployed along the railway line, and generate an inspection report based on the identification results.

[0083] Optionally, the communication bus can be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. This communication bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 2 The symbol is represented by a single thick line, but this does not indicate that there is only one bus or one type of bus. The communication interface is used for communication between the aforementioned electronic device and other devices.

[0084] Memory may include RAM or non-volatile memory. Volatile memory, for example, at least one disk storage device. Alternatively, the memory may also be at least one storage device located remotely from the aforementioned processor.

[0085] The processors mentioned above can be general-purpose processors, including but not limited to: CPU (Central Processing Unit), NP (Network Processor), etc.; they can also be DSP (Digital Signal Processor), ASIC (Application Specific Integrated Circuit), FPGA (Field-Programmable Gate Array), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0086] Optionally, specific examples in this embodiment can refer to the examples described in the above embodiments, and will not be repeated here.

[0087] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

[0088] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0089] In the several embodiments provided in this application, it should be understood that the disclosed apparatus can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some service interface; the indirect coupling or communication connection between devices or units may be electrical or other forms.

[0090] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0091] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0092] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage device (CMD). Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a memory and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned memory includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0093] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, which may include: a flash drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, etc.

[0094] The foregoing description is merely an exemplary embodiment of this disclosure and should not be construed as limiting the scope of this disclosure. Any equivalent changes and modifications made in accordance with the teachings of this disclosure shall still fall within the scope of this disclosure. Those skilled in the art will readily conceive of embodiments of this disclosure upon considering the specification and practicing the disclosure herein. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not described herein. The specification and embodiments are to be considered exemplary only, and the scope and spirit of this disclosure are defined by the claims.

[0095] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0096] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for unmanned aerial vehicle (UAV) network-based railway inspection, characterized in that, include: During the inspection mission performed by an inspection cluster composed of multiple drones, a network reconstruction decision is triggered and executed based on the real-time evaluation results of the communication link quality between the drones to adjust the communication network topology and data forwarding path of the inspection cluster. Based on the status information of the UAVs in the inspection cluster, the load information of the communication network, and the progress information of the inspection task, a multi-objective optimization model is used to generate scheduling instructions and dynamically adjust the UAV waypoints, task sequences, and rotation strategies. The railway facility image data collected by the UAV is processed by end-edge-cloud collaborative processing. At the edge computing node, an artificial intelligence model is used to identify anomalies in the image data and the data transmission priority is determined based on the identification results. After the inspection task is completed, the drone is controlled to return to the drone nest deployed along the railway line, and an inspection report is generated based on the identification results.

2. The UAV-based railway inspection method as described in claim 1, characterized in that, The step of triggering and executing network reconfiguration decisions to adjust the communication network topology and data forwarding paths of the inspection cluster based on real-time evaluation results of the inter-UAV communication link quality includes: Monitor the received signal strength indication, signal-to-noise ratio, data packet delay, and data packet loss rate of the communication link between UAVs, and obtain the link quality index through weighted fusion calculation; When the link quality index is lower than a preset threshold, a network reconstruction scheme is generated based on a reinforcement learning algorithm. The network reconstruction decision includes at least one of the following: converting the inspection drone into a relay drone, updating the data route, or scheduling a standby drone to take off and establish a temporary relay node. The command can be used to convert the inspection drone into a relay drone, re-plan the data backhaul route, or dispatch a backup drone to take off and establish a temporary relay node.

3. The UAV network railway inspection method as described in claim 1, characterized in that, The objective function of the multi-objective optimization model Specifically: in, For task coverage, For emergency response level, For network congestion, , , These are weighting coefficients that are dynamically adjusted based on the real-time task context. The generated scheduling instructions include waypoint instructions for controlling the UAV to hand over tasks during flight, and flight waypoint instructions inserted to verify identified anomalies.

4. The UAV-based railway inspection method as described in claim 1, characterized in that, The edge-cloud collaborative processing of the railway facility image data collected by the UAV includes: The collected image data is compressed and metadata is appended on the UAV to obtain preprocessed data, wherein the appending process includes adding GNSS coordinates and timestamps. At the edge computing node, a pruned and quantized deep learning model is used to perform target detection and defect identification on the preprocessed data, and output analysis results including target bounding boxes, category labels and confidence scores.

5. The UAV-based railway inspection method as described in claim 4, characterized in that, The method further includes: Analysis results with confidence levels higher than the target threshold are marked as high-priority anomalies, and alarm packages containing anomaly summaries, key evidence image slices, and location information are generated and transmitted to the cloud platform with priority. Analysis results with confidence levels not exceeding the target threshold are marked as low-priority data and transmitted to the cloud platform according to the standard procedure.

6. The UAV-based railway inspection method as described in claim 1, characterized in that, The method further includes: Record network reconfiguration events, scheduling instructions, and anomaly analysis results during task execution; The parameters of the deep learning model, the multi-objective optimization model, and the lightweight artificial intelligence model are updated based on the recorded data.

7. A networked railway inspection system for unmanned aerial vehicles (UAVs), characterized in that, include: The dynamic self-organizing network module, integrated into each UAV, is used to trigger and execute network reconstruction decisions to adjust the communication network topology and data forwarding path of the inspection cluster during the inspection task performed by an inspection cluster composed of multiple UAVs, based on the real-time evaluation results of the communication link quality between the UAVs. The intelligent scheduling module, deployed on a cloud platform or edge computing node, is used to generate scheduling instructions based on the status information of UAVs in the inspection cluster, the load information of the communication network, and the progress information of the inspection task, and dynamically adjust the UAV waypoints, task sequences, and rotation strategies through a multi-objective optimization model. The collaborative processing module is used to perform end-edge-cloud collaborative processing on the railway facility image data collected by the UAV. In this module, an artificial intelligence model is used at the edge computing node to identify anomalies in the image data and to determine the data transmission priority based on the identification results. After the inspection task is completed, the drone is controlled to return to the drone nest deployed along the railway line, and an inspection report is generated based on the identification results; It also includes drone nests and drone swarms. The drone nests are deployed along the railway line for storing, taking off and landing, and refueling the drones. The drone swarms are used to perform inspection flights and data collection.

8. The UAV-based networked railway inspection system as described in claim 7, characterized in that, The coordination processing module includes: The preprocessing unit deployed on the drone is used to compress the acquired raw image data and append metadata. The analysis unit, deployed on the edge computing node, has a built-in lightweight deep learning model for defect detection and classification of the received preprocessed data; The management unit, deployed on a cloud platform, is used to aggregate data, review high-priority alarms, and generate inspection reports.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored program, wherein the program, when executed, performs the method of any one of claims 1 to 6.

10. An electronic device comprising a memory and a processor, characterized in that, The memory stores a computer program, and the processor is configured to execute the method of any one of claims 1 to 6 through the computer program.

Images