Monitoring method and device and readable storage medium

Abstract

The invention provides a monitoring method and device and a readable storage medium. The method comprises the following steps: acquiring current flight data of an unmanned aerial vehicle; inputting the current flight data into a global AI model, performing feature extraction on the current flight data by using a global feature mapping function in the global AI model, and generating corresponding current observation trajectory features; calculating a trajectory difference degree between the current observation trajectory feature and a historical normal trajectory feature; and in response to the situation that the trajectory difference degree is greater than a preset abnormal threshold value, judging that the unmanned aerial vehicle has black flight. According to the method, the device and the readable storage medium, the problems that in an existing monitoring technology for black flight of the unmanned aerial vehicle, the precision of a radar monitoring technology is limited in a short-distance low-height scene, the monitoring error of a visual identification technology facing a high-speed unmanned aerial vehicle or complex weather is large, and comprehensive monitoring is difficult to achieve in an ADS-B technology due to the unmanned aerial vehicle configuration problem can be solved.

Description

Technical Field

[0001] The present invention relates to the technical field of unmanned aerial vehicles, and particularly to a monitoring method, device and readable storage medium. Background Art

[0002] With the wide application of unmanned aerial vehicles, the number of illegal "black flights" has gradually increased. In order to effectively monitor and prevent such behaviors, a variety of monitoring technologies for unmanned aerial vehicle black flights have emerged. Among them, the more common monitoring technologies include: ① Radar monitoring technology: By means of a radar system, radio frequency signals are emitted outward. These signals will reflect back to form echo signals after encountering the target (unmanned aerial vehicle). By deeply analyzing detailed information such as the time delay and frequency change of the echo signals, the radar can accurately determine the position, flight speed and forward direction of the unmanned aerial vehicle. ② Visual recognition technology: Mainly rely on a camera to collect images and use computer vision algorithms to achieve the recognition and tracking of unmanned aerial vehicles. This technology comprehensively analyzes the appearance shape, size, color characteristics and motion characteristics of the unmanned aerial vehicle, so that the visual recognition system can accurately distinguish the unmanned aerial vehicle from other objects, and then real-time track the position and flight trajectory of the unmanned aerial vehicle. ③ Automatic Dependent Surveillance - Broadcast (ADS-B) technology for aircraft: After the unmanned aerial vehicle is equipped with an ADS-B (Automatic Dependent Surveillance–Broadcast) system, it will actively send information such as its own position, speed, flight heading and identity identification. After the monitoring station receives this information, it can present the specific position and flight trajectory of the unmanned aerial vehicle on relevant equipment in real time.

[0003] However, the existing monitoring technologies for unmanned aerial vehicle black flights have some obvious deficiencies in practical applications:

[0004] 1) Radar monitoring technology: When in short-distance and low-altitude scenarios, the accuracy of radar monitoring is limited. Especially in the low-altitude area of 0 to 600 meters, due to the small size of the unmanned aerial vehicle and the dense surrounding buildings, it is difficult for the radar monitoring technology to accurately detect the unmanned aerial vehicle and cannot effectively meet the monitoring requirements in this scenario.

[0005] 2) Visual recognition technology: Visual recognition technology is only applicable to the close-range monitoring of unmanned aerial vehicles. Once faced with a high-speed flying unmanned aerial vehicle or in complex weather conditions, its monitoring error will increase significantly, resulting in a reduction in the reliability of the monitoring results.

[0006] 3) Automatic Dependent Surveillance - Broadcast (ADS-B) technology for aircraft: The unmanned aerial vehicle needs to be equipped with an ADS-B system, which increases the cost of the unmanned aerial vehicle.

[0007] 3) Automatic Dependent Surveillance - Broadcast (ADS-B) technology for aircraft: The unmanned aerial vehicle needs to be equipped with an ADS-B system, which increases the cost of the unmanned aerial vehicle.

[0008] 3) ADS-B technology: ADS-B technology is widely used in the field of civil aviation aircraft, but it has certain limitations in the aspect of unmanned aerial vehicles (UAVs). Not all UAVs can be equipped with ADS-B devices, and with the gradual popularization of low-altitude UAVs, it is expected that only important UAVs will be equipped with such devices, which makes it difficult to achieve comprehensive monitoring of UAVs with this technology. Summary of the Invention

[0009] The technical problem to be solved by the present invention is to provide a monitoring method, device and readable storage medium in view of the above deficiencies of the prior art, so as to solve the problems in the existing monitoring technologies for unlicensed flight of UAVs, such as the limited accuracy of radar monitoring technology in short-distance and low-altitude scenarios, the large monitoring errors of visual recognition technology for high-speed UAVs or in complex weather conditions, and the difficulty in achieving comprehensive monitoring due to the equipment problem of UAVs with ADS-B technology.

[0010] In the first aspect, the present invention provides a monitoring method, which is applied to any UAV takeoff and landing base node joined to the blockchain network. The method includes:

[0011] Obtain the current flight data of the UAV;

[0012] Input the current flight data into the global artificial intelligence (AI) model, and use the global feature mapping function in the global AI model to extract features from the current flight data to generate corresponding current observation trajectory features. Among them, the global AI model is obtained through distributed collaborative training based on federated learning, and the global feature mapping function is used to convert the current flight data into a discriminative feature representation;

[0013] Calculate the trajectory difference degree between the current observation trajectory features and the historical normal trajectory features;

[0014] In response to the trajectory difference degree being greater than a preset abnormal threshold, determine that the UAV has an unlicensed flight behavior.

[0015] Further, before obtaining the current flight data of the UAV, the method

[0016] further includes:

[0017] Package the basic information of the UAV takeoff and landing base node to generate initial message information, and broadcast it to other UAV takeoff and landing base nodes in the blockchain network for registration;

[0018] Among them, the basic information includes the basic attribute information of the UAV takeoff and landing base, the current number of UAVs in the base, the UAV list, and the detailed information of each UAV.

[0019] Further, before obtaining the current flight data of the UAV, the method further includes:

[0020] Receive the takeoff request message sent by the UAV, where the takeoff request message includes the identity identifier, estimated takeoff time, current position, current altitude, flight destination, flight time, and flight route of the UAV.

[0021] Further, before inputting the current flight data into the global artificial intelligence (AI) model and using the global feature mapping function in the global AI model to extract features from the current flight data to generate the corresponding current observation trajectory features, the method further includes:

[0022] Train a local AI model using local historical normal flight data to obtain local model parameters and a local feature mapping function;

[0023] Upload the local model parameters to the federated learning platform so that the federated learning platform aggregates the local model parameters received from all UAV takeoff and landing base nodes using a preset federated algorithm to form global AI model parameters;

[0024] Receive the global AI model parameters sent by the federated learning platform through the blockchain network;

[0025] Replace the local model parameters with the global AI model parameters and generate a global feature mapping function based on the global AI model parameters to form a global AI model.

[0026] Further, the current flight data includes the longitude and latitude, flight altitude, flight speed, heading angle, acceleration, and timestamp of multiple sampling points;

[0027] The global AI model includes a time series model and a convolutional neural network, or includes the time series model, convolutional neural network, and self-attention mechanism.

[0028] Further, calculating the trajectory difference degree between the current observation trajectory feature and the historical normal trajectory feature is specifically calculated according to the following formula:

[0029]

[0030] In the formula, D diff represents the trajectory difference degree, N represents the total number of sampling points, T obs (i) represents the current observation trajectory feature obtained from the i-th sampling point through the global feature mapping function, and T norm (i) represents the historical normal trajectory feature obtained from the i-th sampling point through the global feature mapping function.

[0031] Further, the method further includes at least one of the following:

[0032] If the UAV has illegal flight behavior, a warning measure is triggered;

[0033] Collect the latest normal flight data again according to a preset period, update the local historical normal flight data with the latest normal flight data, retrain the local AI model based on the updated local historical normal flight data, obtain new local model parameters and upload them to the federated learning platform, and receive the updated global AI model parameters issued by the federated learning platform, and update the local AI model with the updated global AI model parameters.

[0034] In a second aspect, the present invention provides a monitoring device, which is set at any UAV takeoff and landing base node joined to the blockchain network. The device includes:

[0035] A current flight data acquisition module, configured to acquire the current flight data of the UAV;

[0036] A flight data feature extraction module, connected to the current flight data acquisition module, configured to input the current flight data into a global artificial intelligence (AI) model, and use a global feature mapping function in the global AI model to extract features from the current flight data to generate corresponding current observation trajectory features; wherein, the global AI model is obtained through distributed collaborative training based on federated learning, and the global feature mapping function is used to convert the current flight data into a discriminative feature representation;

[0037] A trajectory difference degree calculation module, connected to the flight data feature extraction module, configured to calculate the trajectory difference degree between the current observation trajectory features and the historical normal trajectory features;

[0038] An illegal flight behavior judgment module, connected to the trajectory difference degree calculation module, configured to judge that the UAV has illegal flight behavior in response to the trajectory difference degree being greater than a preset abnormal threshold.

[0039] In a third aspect, the present invention provides a monitoring device, including a memory and a processor. A computer program is stored in the memory, and the processor is configured to run the computer program to implement the monitoring method described in the first aspect above.

[0040] In a fourth aspect, the present invention provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the monitoring method described in the first aspect above is implemented.

[0041] The monitoring method, device, and readable storage medium provided by the present invention. The drone takeoff and landing base node joined to the blockchain network first obtains the current flight data of the drone, and then inputs the current flight data into the global artificial intelligence (AI) model. The global feature mapping function in the global AI model is used to extract features from the current flight data to generate corresponding current observation trajectory features. Among them, the global AI model is obtained through distributed collaborative training based on federated learning, and the global feature mapping function is used to convert the current flight data into a discriminative feature representation. Then, the trajectory difference degree between the current observation trajectory feature and the historical normal trajectory feature is calculated. Finally, in response to the trajectory difference degree being greater than the preset anomaly threshold, it is determined that the drone has a black flight behavior. The present invention can ensure the immutability and decentralized storage of data by using blockchain technology. Through federated learning, each takeoff and landing base node only shares model parameters instead of raw data, effectively protecting the privacy of drone flight data. At the same time, the global AI model converts the current flight data (i.e., the original flight data) into a discriminative feature representation through feature mapping, which can effectively capture the tiny but crucial abnormal information in the flight trajectory, thereby achieving high-precision identification of black flight behaviors. In addition, compared with the radar monitoring technology whose accuracy is limited in short-distance and low-altitude scenarios, the present invention is not limited by the height and distance of specific scenarios and can accurately extract features from various flight data. Compared with the visual recognition technology, the present invention will not produce large monitoring errors due to the high-speed flight of the drone or complex weather, because it relies on the analysis of flight data rather than pure visual capture. In view of the dilemma that the ADS-B technology is difficult to achieve comprehensive monitoring due to the drone equipment problem, the present invention does not need to rely on the drone itself to be equipped with specific devices. Through the data collection of each takeoff and landing base node and the operation of the global AI model, it can achieve a wider and more comprehensive monitoring, greatly improving the accuracy, stability, and comprehensiveness of drone black flight monitoring, and effectively ensuring airspace safety. It solves the problems in the existing monitoring technologies for drone black flights, where the radar monitoring technology has limited accuracy in short-distance and low-altitude scenarios, the visual recognition technology has large monitoring errors in the face of high-speed drones or complex weather, and the ADS-B technology is difficult to achieve comprehensive monitoring due to the drone equipment problem. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 It is a flowchart of a monitoring method according to Embodiment 1 of the present invention;

[0043] Figure 2 It is a structural schematic diagram of a monitoring device according to Embodiment 2 of the present invention;

[0044] Figure 3 It is a structural schematic diagram of a monitoring device according to Embodiment 3 of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0045] To enable those skilled in the art to better understand the technical solution of the present invention, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings.

[0046] It can be understood that the specific embodiments and accompanying drawings described herein are only for explaining the present invention, rather than limiting the present invention.

[0047] It can be understood that, without conflict, the various embodiments in the present invention and the features in the embodiments can be combined with each other.

[0048] It can be understood that, for the convenience of description, only the parts related to the present invention are shown in the accompanying drawings of the present invention, and the parts unrelated to the present invention are not shown in the accompanying drawings.

[0049] It can be understood that each unit and module involved in the embodiments of the present invention may correspond to only one entity structure, or may be composed of multiple entity structures. Alternatively, multiple units and modules may also be integrated into one entity structure.

[0050] It can be understood that the terms "first", "second", etc. in the embodiments of the present invention are used to distinguish different objects, or to distinguish different processes for the same object, rather than to describe a specific order of the object.

[0051] It can be understood that, without conflict, the functions and steps marked in the flowcharts and block diagrams of the present invention may occur in a different order from that marked in the accompanying drawings.

[0052] It can be understood that in the flowcharts and block diagrams of the present invention, the possible system architectures, functions, and operations of the systems, devices, equipment, and methods according to the embodiments of the present invention are shown. Among them, each block in the flowchart or block diagram may represent a unit, module, program segment, or code, which contains executable instructions for implementing the specified function. Moreover, each block or combination of blocks in the block diagram and flowchart can be implemented by a hardware-based system for implementing the specified function, or can be implemented by a combination of hardware and computer instructions.

[0053] It can be understood that the units and modules involved in the embodiments of the present invention can be implemented in software or in hardware. For example, the units and modules can be located in the processor.

[0054] Embodiment 1:

[0055] This embodiment provides a monitoring method, which is applied to any drone takeoff and landing base node joined to the blockchain network. As Figure 1 shown, the method includes:

[0056] Step S101: Obtain the current flight data of the drone;

[0057] In this embodiment, a blockchain network is first constructed, and multiple UAV takeoff and landing bases are connected to the network. Each takeoff and landing base serves as a blockchain node (i.e., a UAV takeoff and landing base node). Each UAV takeoff and landing base node can obtain the current flight data of the UAVs connected thereto. These UAVs can be logistics UAVs, or urban patrol UAVs, or agricultural UAVs, etc.

[0058] It should be noted that a UAV takeoff and landing base is a place that provides takeoff, landing, and related support services for UAVs. A UAV takeoff and landing base generally has a site suitable for UAV takeoff and landing, such as a runway, a helipad, etc. According to the type and size of the UAV, the length, width, and surface material of the runway may vary. In addition, facilities for parking, maintenance, and charging are also equipped to ensure that the UAVs can always be in good operating condition. The takeoff and landing base is provided with a control and command center. Staff plan, monitor, and control the flight tasks of UAVs through professional control equipment and software. The system can obtain the flight data of UAVs in real time, such as altitude, speed, position, etc., to ensure that the UAVs fly according to the predetermined route and take timely measures in case of abnormal situations. Since meteorological conditions have an important impact on the flight safety of UAVs, the takeoff and landing base usually equips meteorological monitoring equipment, such as anemometers, wind vanes, thermometers, hygrometers, etc., to monitor the meteorological data around the base in real time and provide accurate meteorological information for the takeoff, landing, and flight of UAVs. During the execution of tasks, UAVs may collect a large amount of data, such as images, videos, sensor data, etc. The data processing and transmission equipment of the takeoff and landing base can process and store these data in real time and transmit the data to relevant users or systems through wireless communication or other means. To ensure the safety of UAV takeoff, landing, and operation, the base will set up safety facilities such as safety fences and warning signs to prevent unauthorized personnel and vehicles from entering the takeoff and landing area. At the same time, fire-fighting equipment, first-aid equipment, etc. will also be equipped to deal with possible emergencies.

[0059] Optionally, before obtaining the current flight data of the UAV, the method further includes:

[0060] Pack the basic information of the UAV takeoff and landing base node to generate initial message information, and broadcast it to other UAV takeoff and landing base nodes in the blockchain network for registration;

[0061] Wherein, the basic information includes the basic attribute information of the UAV takeoff and landing base, the current number of UAVs in the base, the UAV list, and the detailed information of each UAV.

[0062] Specifically, in the blockchain network, when each UAV takeoff and landing base node registers, it will package its basic information (such as basic attribute information, the current number of UAVs at the base, the UAV list, and the detailed information of each UAV, etc.) into an initial message and broadcast it to the entire network. After receiving this information, other UAV takeoff and landing base nodes will record it in the local blockchain ledger. Therefore, each UAV takeoff and landing base node will ultimately store the initial registration information of all other nodes, forming a distributed database shared across the network. Among them, the basic attribute information of the UAV takeoff and landing base includes the location, altitude, temperature, UAV capacity, etc. of the base, and the detailed UAV information includes information such as the UAV's identity identifier, current location, current status, and current battery level. The UAV identity identifier includes relevant information such as the UAV's manufacturer, model, and serial number. Due to the uniqueness of the serial number, the identity identifier of each UAV is also unique.

[0063] Optionally, before obtaining the current flight data of the UAV, the method further includes:

[0064] Receiving a takeoff request message sent by the UAV, the takeoff request message including the UAV's identity identifier, estimated takeoff time, current location, current altitude, flight destination, flight time, and flight path.

[0065] In this embodiment, after the UAV receives the user's takeoff instruction, the UAV will first send a takeoff request message to the UAV takeoff and landing base where it is located. The request message contains information such as the UAV's identity identifier, estimated takeoff time, current location, current altitude, flight destination, flight time, and flight path.

[0066] Step S102: Input the current flight data into a global AI (Artificial Intelligence) model, and use the global feature mapping function in the global AI model to extract features from the current flight data to generate corresponding current observation trajectory features; wherein, the global AI model is obtained through distributed collaborative training based on federated learning, and the global feature mapping function is used to convert the current flight data into a discriminative feature representation.

[0067] In this embodiment, in order to effectively protect the privacy of UAV flight data, the distributed AI collaborative learning (federated learning) technology is adopted. The global AI model is jointly constructed by each UAV takeoff and landing base node. The global feature mapping function of the model is used to extract features from the current flight data, generating the current observation trajectory features with discriminative ability. Here, having discriminative ability means being able to distinguish normal and abnormal (i.e., unlicensed) flight modes, being able to capture subtle differences. Compared with directly using the original flight data (i.e., the current flight data), the feature representation with discriminative ability can capture key patterns and implicit information in the flight trajectory more sensitively.

[0068] Optionally, before inputting the current flight data into the global artificial intelligence AI model and using the global feature mapping function in the global AI model to extract features from the current flight data to generate the corresponding current observation trajectory features, the method further includes:

[0069] Training a local AI model using local historical normal flight data to obtain local model parameters and a local feature mapping function;

[0070] Uploading the local model parameters to the federated learning platform so that the federated learning platform uses a preset federated algorithm to aggregate the local model parameters received from all UAV takeoff and landing base nodes to form global AI model parameters;

[0071] Receiving the global AI model parameters sent by the federated learning platform through the blockchain network;

[0072] Replacing the local model parameters with the global AI model parameters and generating a global feature mapping function based on the global AI model parameters to form a global AI model.

[0073] In this embodiment, each UAV takeoff and landing base node uses local historical normal flight data to train a local AI model, extracts implicit features in the flight trajectory, and obtains local model parameters. The local model parameters record the flight features learned by this takeoff and landing base node. Each UAV takeoff and landing base node uploads the local model parameters to the federated learning platform, and the federated learning platform uses a preset federated algorithm to aggregate the local model parameters received from all UAV takeoff and landing base nodes to form global AI model parameters. Among them, the preset federated algorithm can be FedAvg (federated average algorithm), FedProx (federated proximal algorithm), FedAdagrad (federated adaptive gradient algorithm), or FedDyn (federated dynamic algorithm), etc. If FedAvg (federated average algorithm) is adopted, the calculation formula is:

[0074]

[0075] Among them, θ global is the global AI model parameter formed after aggregation, k represents the total number of UAV takeoff and landing base nodes, and |D i | represents the sample quantity of the local historical normal flight data of the i-th UAV takeoff and landing base node, θ i represents the local model parameter of the i-th UAV takeoff and landing base node, and is used to ensure that UAV takeoff and landing base nodes with a larger amount of data contribute more to the global AI model.

[0076] In this embodiment, the federated learning platform distributes the global AI model parameter to all UAV takeoff and landing base nodes through the blockchain network. Each UAV takeoff and landing base node updates the local AI model according to the received θ global The updated local AI model incorporates the global AI model parameter and is functionally consistent with the global AI model.

[0077] Optionally, the current flight data includes the longitude and latitude, flight altitude, flight speed, heading angle, acceleration, and timestamp of multiple sampling points;

[0078] The global AI model includes a time series model and a convolutional neural network, or includes the time series model, convolutional neural network, and self-attention mechanism.

[0079] In this embodiment, the current flight data is the flight data of the UAV at the current moment or within a short period of time, which is a time series data and includes the longitude and latitude, flight altitude, flight speed, heading angle, acceleration, and timestamp of different sampling points, etc.

[0080] In this embodiment, both the global AI model and the local AI model adopt deep learning technology, and their core components include:

[0081] (1) Time series model: such as LSTM (Long Short-Term Memory) / GRU (Gated Recurrent Unit), which is used to learn the time series features of the UAV trajectory and identify trends such as speed, acceleration, and altitude changes.

[0082] (2) Convolutional neural network (Convolution Neural Network, CNN): which is used to extract the spatial features of the flight trajectory and enhance the model's ability to identify different flight patterns.

[0083] (3) Self-attention mechanism (Transformer) (optional): which is used to improve the understanding of long-term trajectory patterns and enhance the anomaly detection ability.

[0084] Step S103: Calculate the trajectory difference degree between the current observed trajectory feature and the historical normal trajectory feature.

[0085] In this embodiment, the trajectory difference degree is used to represent the deviation degree between the current observed trajectory feature and the historical normal trajectory feature. The larger the data, the more obvious the deviation degree.

[0086] Optionally, to calculate the trajectory difference degree between the current observed trajectory feature and the historical normal trajectory feature, it is specifically calculated according to the following formula:

[0087]

[0088] In the formula, D diff represents the trajectory difference degree, N represents the total number of sampling points, T obs (i) represents the current observed trajectory feature obtained by the global feature mapping function for the i-th sampling point, and T norm (i) represents the historical normal trajectory feature obtained by the global feature mapping function for the i-th sampling point.

[0089] Step S104: In response to the trajectory difference degree being greater than a preset abnormal threshold, determine that the drone has black flight behavior.

[0090] In this embodiment, the abnormal threshold δ is determined by historical data and expert experience and is used to determine the sensitivity of the system to the deviation of the flight trajectory. When the drone has deviations such as deviation from the flight route, abnormal time, abnormal speed or altitude, illegal entry into the no-fly zone, or excessive stay time, etc., it will leave a pattern trace in the current observed trajectory feature, resulting in the trajectory difference degree D diff > δ.

[0091] Optionally, the method further includes at least one of the following:

[0092] If the drone has black flight behavior, trigger a warning measure;

[0093] Collect the latest normal flight data again according to a preset period, update the local historical normal flight data with the latest normal flight data, retrain the local AI model based on the updated local historical normal flight data, obtain new local model parameters and upload them to the federated learning platform, and, receive the updated global AI model parameters issued by the federated learning platform and update the local AI model with the updated global AI model parameters.

[0094] In this embodiment, if the drone has black flight behavior, a warning measure is automatically triggered for operations such as black flight warning and interception; if the trajectory difference degree D diff ≤ δ, it is considered that the flight behavior of the drone is normal, and regular monitoring continues.

[0095] In this embodiment, each UAV takeoff and landing base node collects the latest normal flight data again at a preset period (such as every few hours or every day), updates the local historical normal flight data with the latest normal flight data, retrains the local AI model based on the updated local historical normal flight data, obtains new local model parameters and uploads them to the federated learning platform. The federated learning platform recalculates the global AI model parameters through the federated algorithm, and distributes the updated global AI model parameters to all UAV takeoff and landing base nodes through the blockchain network. After receiving the updated global AI model parameters distributed by the federated learning platform, each UAV takeoff and landing base node updates the local AI model with the updated global AI model parameters, so that the local AI model is consistent with the global AI model and always reflects the latest flight mode.

[0096] It should be noted that the monitoring method provided by the present invention adopts the distributed AI collaborative learning (federated learning) technology. The global AI model is jointly constructed and dynamically updated by each UAV takeoff and landing base node. The original flight data is mapped to a high-dimensional feature space by using this model, and then the current trajectory is compared with the historical normal trajectory in the feature space, so as to quantify the degree of abnormality and trigger an early warning in time when an abnormality is detected. The advantages of the present invention are mainly reflected in the following aspects:

[0097] (1) The blockchain technology is used to achieve the tamper-proof and decentralized storage of data. At the same time, through federated learning, each UAV takeoff and landing base node only shares the model parameters rather than the original data, thus effectively protecting the privacy of UAV flight data.

[0098] (2) The global AI model converts the original flight data into a high-dimensional feature representation through feature mapping, enabling the system to capture tiny but critical abnormal information in the flight trajectory, so as to achieve high-precision identification of illegal flight behaviors.

[0099] (3) By periodically iteratively updating the global AI model, the detection standard can be continuously optimized with the changes of new data and flight modes, so as to adapt to new illegal flight means in time, quickly trigger early warnings and response measures. Each UAV takeoff and landing base node independently collects and preprocesses data and collaboratively participates in model training. The entire system adopts a distributed architecture, which is suitable for large-scale deployment and reduces the risk of single-point failures.

[0100] In a specific embodiment, the monitoring method may include the following steps:

[0101] Step 1: Data collection and local preprocessing

[0102] 1. Build a blockchain network and connect multiple drone takeoff and landing bases to the blockchain. Each takeoff and landing base serves as a blockchain node, that is, each node represents an independent drone takeoff and landing base. During the registration process, the drone takeoff and landing base needs to provide some basic information, such as the location, altitude, temperature, drone capacity and other basic attribute information of the base, as well as the current number of drones at the base, the drone list and the detailed information of each drone. Among them, the detailed drone information includes the identity identifier of the drone, the current location, the current status, the current battery level and other information. The drone identity identifier includes relevant information such as the manufacturer, model, and serial number of the drone. Due to the uniqueness of the serial number, the identity identifier of each drone is also unique. It should be noted that during the registration phase, each blockchain node will package the above basic information to generate the initial message information and broadcast it to all other nodes on the blockchain node, and each node will update the basic information by itself.

[0103] 2. After the drone receives the takeoff instruction from the user, the drone will first send a takeoff request message to the drone takeoff and landing base where it is located. The request message contains information such as the identity identifier of the drone, the estimated takeoff time, the current location, the current altitude, the flight destination, the flight time, and the flight path. During the flight of the drone, each drone takeoff and landing base will collect the original flight data of the drone and record it as X raw . After filtering and normalization processing, a local dataset D i (that is, the local historical normal flight data) is generated for subsequent model training.

[0104] Step 2. Model construction and training

[0105] Each node uses the local data to train the local AI model and extract the implicit features in the flight trajectory. Specifically, each drone takeoff and landing base i uses a predefined neural network structure and uses the local dataset D i to train the local AI model and obtain the local model parameters θ i (the local model parameters of the drone takeoff and landing base node i, which record the flight features learned by this takeoff and landing base node) and the local feature mapping function f θi (), which maps the original flight data X raw to a discriminative feature representation T.

[0106] Among them, the flight trajectory refers to the time series data generated during the flight of the drone in the air, including its spatial position information and flight state parameters at consecutive moments. These trajectory data are collected by the drone itself and collected and processed through the takeoff and landing base nodes it accesses.

[0107] Among them, the AI model mainly uses deep learning technology, and its core components include:

[0108] 1. Neural network architecture:

[0109] (1) Temporal model (such as LSTM / GRU): Used to learn the time series features of the drone trajectory and identify trends such as speed, acceleration, and altitude changes.

[0110] (2) Convolutional neural network (CNN): Used to extract the spatial features of the flight trajectory and enhance the model's ability to recognize different flight patterns.

[0111] (3) Self-attention mechanism (Transformer) (optional): Used to improve the understanding of long-term trajectory patterns and enhance the anomaly detection ability.

[0112] 2. Input data:

[0113] GPS (Global Positioning System) coordinates (latitude and longitude), flight altitude, flight speed, heading angle, acceleration, and timestamp as auxiliary features.

[0114] 3. Output data:

[0115] Trajectory feature vector T obs 、T norm 。

[0116] In this step, the AI model is not independently trained by a single node, but through federated learning for distributed collaborative training to ensure that each node (drone takeoff and landing base) shares model knowledge while ensuring data privacy.

[0117] In summary, the AI in this solution refers to a drone trajectory anomaly detection model based on deep learning. This solution realizes black flight detection through neural network feature extraction + trajectory anomaly analysis + federated learning optimization, ensuring that the system can accurately and efficiently identify abnormal flight patterns, and continuously optimize the detection criteria through continuous learning to improve the low-altitude safety supervision ability.

[0118] Step 3. Upload of local model parameters and global aggregation

[0119] Each drone takeoff and landing base node i trains a local AI model with local data and uploads its local model parameters θ i to the federated learning platform. The federated learning platform uses the federated average algorithm to calculate the global AI model parameters θ global for the parameters uploaded by all drone takeoff and landing base nodes using the FedAvg algorithm. The calculation formula is:

[0120]

[0121] where, To ensure that nodes with a large amount of data contribute more to the global AI model, k represents the total number of drone takeoff and landing base nodes, and θ global is the global AI model parameter formed after aggregation and is used as the global AI model θ AI , |D i | represents the number of samples in the dataset D i . The flight data of each drone can be used as a training sample.

[0122] It should be noted that in traditional machine learning, it is usually necessary to collect data scattered in different places on a central server for training. However, in many scenarios, due to reasons such as data privacy, security regulations, and data ownership, the data cannot be directly centralized. Federated learning allows each participating party (such as different institutions, organizations, or devices) to retain data locally and only exchange model parameters or intermediate calculation results with other participating parties or the central server (i.e., the federated learning platform) through encryption to jointly train a global AI model. Its advantage is that the data of each participating party does not need to leave the local area, avoiding the risk of privacy leakage during data transmission and storage. It enables different institutions or organizations to integrate data resources from all parties without sharing the original data, achieving "usable but invisible" data and breaking data silos. Since flight data has a certain degree of sensitivity, it is suitable for using the federated learning model.

[0123] The main types of federated learning include horizontal federated learning, vertical federated learning, and federated transfer learning. Among them, horizontal federated learning is also called federated learning with feature alignment. It is applicable when the feature spaces of the samples in the datasets of multiple participating parties overlap more, but the sample ID spaces are different. Since the drone flight data collected by all takeoff and landing bases in the present invention includes the same fields (such as longitude and latitude, speed, time, etc.), the feature dimensions are consistent. The data of each node belongs to the drones in its own jurisdiction, the sample sets do not overlap, and the data belongs to different entities. Therefore, the present invention mainly adopts horizontal federated learning, and each node can independently train the model on its own local trajectory data without accessing the original data of other nodes.

[0124] In federated learning, participants usually need to locally deploy some components and algorithms related to federated learning. The participants need to process their own data locally and execute the model training process (such as gradient calculation, etc.). Taking the horizontal federated learning scenario as an example, when jointly conducting AI model training at multiple UAV takeoff and landing base nodes, each UAV takeoff and landing base node needs to locally deploy the relevant code and environment that can train the model based on the UAV flight data it collects, and complete the local model update calculation. These components and algorithms locally deployed by the participants enable them to interact with the central server (i.e., the federated learning platform) or other UAV takeoff and landing bases according to the protocols and instructions of federated learning, upload the model parameters generated by local training (such as gradients, weight update values, etc.), and receive update information from other parties to further update the local model.

[0125] Central server: The main role of the central server is to coordinate operations such as communication and model aggregation among various participants. In some federated learning architectures, the central server stores the global AI model and is responsible for receiving the local model updates uploaded by each participant, then aggregating these updates into the global AI model, and then distributing the updated global AI model to each participant. In this case, the central server needs to deploy relevant components that can execute model aggregation algorithms, such as the federated averaging algorithm, etc. However, the central server does not necessarily need to deploy a complete federated learning model for training on local data. Because the central server usually does not directly participate in the data training process (the data remains local to the participants), its main function is to manage and coordinate the training process of the model and parameter transfer. But in some more complex federated learning scenarios, the central server may also undertake some additional computing tasks, and at this time, more model and algorithm components related to federated learning may need to be deployed.

[0126] Step 4: Global AI model distribution and feature extraction

[0127] The federated learning platform distributes the global AI model parameters θ global to all UAV takeoff and landing base nodes i through the blockchain network. Each UAV takeoff and landing base node i updates the local model θ global with the received θ AI , that is, let θ AI＝ θ global .

[0128] Each UAV takeoff and landing base node i uses the global feature mapping function f θglob () to extract features from the newly collected flight data, obtaining the feature representation T (i.e., the trajectory feature T obs of the currently observed UAV), where:

[0129] T obs(i) = f θglobal (X obs (i)) represents the feature vector of the current UAV flight sampling point data (i.e., the current observed trajectory feature). The meaning of this formula is: using the AI model f θglobal to map the current trajectory raw data X obs (i) into a trajectory feature T obs (i) for subsequent comparison with the historical normal trajectory feature to determine anomalies. Among them,

[0130] X obs (i): represents the raw flight data of the i-th sampling point in the current UAV flight trajectory (such as longitude, latitude, altitude, speed, etc.);

[0131] T obs (i): represents the representation of the i-th sampling point of the current trajectory in the feature space, which is the trajectory expression "understood" by the model.

[0132] Suppose a UAV collected a data point at 09:00:02:

[0133] X obs (3) = [longitude, latitude, altitude, speed, heading angle] = [113.32, 23.11.

[0134] 100, 12.5, 270]

[0135] Then: through the global AI model f θglob , an output feature vector T obs (3) is obtained. This vector is no longer the original physical quantity, but the "behavior semantic" representation of this trajectory point learned by the AI model during training. Directly comparing the raw flight data (longitude, latitude, altitude, etc.) can only capture the gap between trajectory points, but in the feature space learned by the AI model, it can more accurately distinguish between "normal turns" and "abnormal deviations", etc., which is more intelligent.

[0136] T norm (i) = f θglobal (X norm (i)) represents the feature vector of the corresponding sampling point data of the historical normal trajectory. Here, the correspondence means the mapping established between the i-th sampling point of the current trajectory and the point at the same progress position of the trajectory in the historical normal trajectory.

[0137] It should be noted that the "current observed trajectory feature" refers to the flight data of the UAV at the current moment or within a short period of time passing through the global AI model f θglobalThe obtained feature vectors after processing, these features reflect information such as the current flight state, position, speed, and trajectory changes; while the "historical normal trajectory features" refer to the feature vectors obtained by processing the flight data that was previously verified as normal through the same global AI model f θglobal The obtained feature vectors after processing serve as the benchmark for the normal flight mode. The comparison of these two types of features in the feature space can help the system quantify whether there is an anomaly in the current flight, thus achieving precise anomaly detection.

[0138] Specifically, each UAV takeoff and landing base node i first trains a model locally to obtain local model parameters; then, these local model parameters are aggregated into global AI model parameters θ global , and then distributed to each base as θ AI .

[0139] Step 5, Abnormal trajectory detection and quantification

[0140] In the feature space extracted by each UAV takeoff and landing base, compare the current observed trajectory feature T obs with the historical normal trajectory feature T norm , and calculate the trajectory difference degree D diff . The larger the value, the more obvious the deviation degree of the current observed trajectory feature from the historical normal trajectory feature.

[0141]

[0142] Among them, N represents the total number of sampling points of the flight data.

[0143] T obs (i) and T norm (i): respectively represent the feature vectors obtained by the i-th sampling point through the global mapping function, which can reflect the flight behavior better than directly using the original flight data.

[0144] Step 6, Intelligent determination and early warning response

[0145] Compare the calculated D diff with the preset abnormal threshold δ to determine whether there is a risk of unlicensed flight. If the system determines that D diff >δ, the monitoring system determines that the current flight trajectory is abnormal and automatically triggers early warning measures for unlicensed flight warning, interception, etc.; if D diff ≤δ, the system considers the flight behavior normal and continues with routine monitoring.

[0146] δ is the anomaly threshold, determined by historical data and expert experience, which determines the sensitivity of the system to deviations in the flight trajectory. In this embodiment, "current observed trajectory features" represent the flight state of the UAV at the current moment or in the recent past, while "historical normal trajectory features" represent flight patterns that have been verified in the past and are considered normal. Both have been processed by the global AI model f θglobal for extraction and converted into high-dimensional feature vectors that can capture key patterns and implicit information in the flight trajectory. Therefore, when the system compares these two feature vectors, if the degree of difference between them (e.g., D diff ) calculated by the mean square error exceeds the preset threshold δ, it indicates that there is a significant deviation between the current flight state and the normal mode, and thus it can be determined that the UAV may be engaged in illegal flight behavior. This method utilizes the subtle differences in the feature space and is more sensitive and accurate than directly comparing the original flight data, helping to detect anomalies such as illegal flights in a timely manner.

[0147] Step 7: Dynamic iterative update

[0148] Each UAV takeoff and landing base node re-collects the latest normal flight data at a preset period (such as every few hours or every day) and updates the local dataset D i . Based on the updated D i , each takeoff and landing base node retrains the local AI model to obtain new local parameters θ i ; subsequently, each node uploads the new θ i to the federated learning platform. The federated learning platform uses the latest data to recalculate the global AI model parameters θ global through the federated averaging algorithm and distributes the updated θ global to all nodes to ensure that the global AI model θ AI always reflects the latest flight pattern. Decide how often the model is updated to adapt to environmental changes. The dynamic update of θ AI improves the long-term robustness and anomaly detection accuracy of the model.

[0149] In one example, the UAV is a logistics UAV, and the normal flight pattern of the logistics UAV refers to a flight trajectory that follows established rules, a predetermined route, or conforms to historical patterns, which are considered safe and compliant flight behaviors.

[0150] Suppose a certain logistics UAV takes off from distribution center A at a fixed time every day (such as 9:00 - 10:00 am), flies along a predetermined route to the receiving point B, maintains a stable speed and altitude during the flight, avoids no-fly zones, and completes the delivery task within 10 minutes. This trajectory has been stored in the system as a "historical normal trajectory", and no anomalies have occurred in this route during multiple past flight missions.

[0151] If one day, this logistics drone:

[0152] a) Deviates from the route: Suddenly flies towards area C (an unregistered flight destination);

[0153] Input reflection method: The longitude and latitude coordinates continuously deviate from the normal flight path;

[0154] Feature expression: The spatial path of the trajectory is significantly different, and the position change does not conform to the historical pattern;

[0155] Result: The T extracted by the model obs and the historical normal trajectory T norm have a large difference D diff very large.

[0156] b) Time anomaly: Takes off at 2:00 am instead of the normal 9:00 - 10:00 time period;

[0157] Input reflection method: The time deviates from the common time;

[0158] Feature expression: For example, there are very few trajectories in the "early morning time period" that the model has seen. At this time, "the same path + different time" will also be considered to deviate from the normal pattern;

[0159] Result: The model will regard "the behavior occurs in a time period when it should not" as an anomaly, D diff > δ.

[0160] c) Speed or altitude anomaly: It should fly at a constant speed originally, but suddenly accelerates or the altitude fluctuates violently;

[0161] Input reflection method: The flight speed and altitude are the input fields;

[0162] Feature expression: It is manifested as violent fluctuations and high-frequency changes in the trajectory, and neural networks (especially RNN / CNN) can effectively capture them;

[0163] Result: The trajectory feature changes greatly, and the difference from the normal stable trajectory is obvious, D diff > δ.

[0164] d) Illegally enters the no-fly zone: Flies into sensitive areas, such as above airports and relevant institutions;

[0165] Input reflection method: The GPS coordinates directly fall within the coordinate range of the no-fly zone;

[0166] Feature expression: The model may not have learned such a trajectory point distribution, and such positions deviate from the normal distribution in the feature space;

[0167] Result: Even if the speed and altitude are normal, the trajectory route is different from the normal state, D diff > δ.

[0168] e) Excessive residence time: The flight should have been completed in 10 minutes, but it circled or hovered in mid-air for 30 minutes.

[0169] Input reflection method: It can be seen from the speed field that "close to 0 m / s for a long time"; or there is no change in longitude and latitude for a long time (position points accumulate);

[0170] Feature expression: The trajectory stays in a specific area for a long time and has weak spatial motion characteristics;

[0171] Result: The model can perceive that this behavior is very different from the typical trajectory, D diff > δ.

[0172] It should be noted that although the corresponding features of time anomaly, illegal entry into the no-fly zone, and excessive residence time are not directly encoded as a tag field, they will all leave pattern traces in the trajectory time-series features of the current input (that is, "naturally appear" as anomalies in the feature space, which in turn leads to D diff > δ).

[0173] In the above situation, the system compares the current flight trajectory feature T obs with the historical normal trajectory feature T norm to calculate the difference degree D diff . If the difference is large (such as the deviation from the flight route is too far or the flight mode changes violently), it will be determined as an abnormal illegal flight and a warning will be triggered.

[0174] This method is not only applicable to logistics UAVs, but also can be applied to urban patrol UAVs, agricultural UAVs, etc., to ensure flight safety and effectively identify illegal flights.

[0175] The monitoring method provided by the embodiment of the present invention is as follows: The drone takeoff and landing base node joined to the blockchain network first obtains the current flight data of the drone, and then inputs the current flight data into the global artificial intelligence (AI) model. The global feature mapping function in the global AI model is used to extract features from the current flight data to generate corresponding current observation trajectory features. Among them, the global AI model is obtained through distributed collaborative training based on federated learning. The global feature mapping function is used to convert the current flight data into a discriminative feature representation. Then, the trajectory difference degree between the current observation trajectory feature and the historical normal trajectory feature is calculated. Finally, in response to the trajectory difference degree being greater than the preset abnormal threshold, it is determined that the drone has a black flight behavior. The present invention utilizes blockchain technology to ensure the immutability and decentralized storage of data. Through federated learning, each takeoff and landing base node only shares model parameters instead of raw data, effectively protecting the privacy of drone flight data. At the same time, the global AI model converts the current flight data (i.e., the original flight data) into a discriminative feature representation through feature mapping, which can effectively capture the minute but crucial abnormal information in the flight trajectory, thereby achieving high-precision identification of black flight behavior. In addition, compared with the radar monitoring technology whose accuracy is limited in short-distance and low-altitude scenarios, the present invention is not limited by the height and distance of specific scenarios and can accurately extract features from various flight data. Compared with the visual recognition technology, the present invention will not generate large monitoring errors due to the high-speed flight of the drone or complex weather, because it relies on the analysis of flight data rather than simple visual capture. In view of the dilemma that it is difficult to achieve comprehensive monitoring for the ADS-B technology due to the drone equipment problem, the present invention does not need to rely on the drone itself to be equipped with specific devices. Through the data collection of each takeoff and landing base node and the operation of the global AI model, it can achieve a wider and more comprehensive monitoring, greatly improving the accuracy, stability and comprehensiveness of drone black flight monitoring, and effectively ensuring airspace safety. It solves the problems in the existing monitoring technologies for drone black flight, including that the radar monitoring technology has limited accuracy in short-distance and low-altitude scenarios, the visual recognition technology has large monitoring errors in the face of high-speed drones or complex weather, and the ADS-B technology is difficult to achieve comprehensive monitoring due to the drone equipment problem.

[0176] Embodiment 2:

[0177] As Figure 2 shown, this embodiment provides a monitoring device, which is set in any drone takeoff and landing base node joined to the blockchain network and is used to execute the above monitoring method, including:

[0178] A current flight data acquisition module 11, which is used to acquire the current flight data of the drone;

[0179] The flight data feature extraction module 12, connected to the current flight data acquisition module 11, is configured to input the current flight data into the global artificial intelligence (AI) model, and use the global feature mapping function in the global AI model to extract features from the current flight data, generating corresponding current observation trajectory features; wherein, the global AI model is obtained through distributed collaborative training based on federated learning, and the global feature mapping function is used to convert the current flight data into a discriminative feature representation;

[0180] The trajectory difference degree calculation module 13, connected to the flight data feature extraction module 12, is configured to calculate the trajectory difference degree between the current observation trajectory features and the historical normal trajectory features;

[0181] The black flight behavior judgment module 14, connected to the trajectory difference degree calculation module 13, is configured to determine that the drone has black flight behavior in response to the trajectory difference degree being greater than a preset abnormal threshold.

[0182] Optionally, the device further includes:

[0183] The node registration module is configured to package the basic information of the drone takeoff and landing base node to generate initial message information, and broadcast it to other drone takeoff and landing base nodes in the blockchain network for registration;

[0184] Wherein, the basic information includes the basic attribute information of the drone takeoff and landing base, the current number of drones at the base, the drone list, and the detailed information of each drone.

[0185] Optionally, the device further includes:

[0186] The takeoff request message receiving module is configured to receive the takeoff request message sent by the drone, and the takeoff request message includes the identity identifier of the drone, the expected takeoff time, the current position, the current altitude, the flight destination, the flight time, and the flight route.

[0187] Optionally, the device further includes:

[0188] The local training module is configured to train a local AI model using local historical normal flight data to obtain local model parameters and a local feature mapping function;

[0189] The parameter upload module is configured to upload the local model parameters to the federated learning platform, so that the federated learning platform aggregates the local model parameters received from all drone takeoff and landing base nodes using a preset federated algorithm to form global AI model parameters;

[0190] A global AI model parameter receiving module, configured to receive global AI model parameters sent by the federated learning platform through the blockchain network;

[0191] A local update module, configured to replace the local model parameters with the global AI model parameters, and generate a global feature mapping function based on the global AI model parameters to form a global AI model.

[0192] Optionally, the current flight data includes the latitude and longitude, flight altitude, flight speed, heading angle, acceleration, and timestamp of multiple sampling points;

[0193] The global AI model includes a time series model and a convolutional neural network, or includes the time series model, convolutional neural network, and self-attention mechanism.

[0194] Optionally, the trajectory difference degree calculation module 13 specifically calculates the trajectory difference degree between the current observed trajectory feature and the historical normal trajectory feature according to the following formula:

[0195]

[0196] In the formula, D diff represents the trajectory difference degree, N represents the total number of sampling points, T obs (i) represents the current observed trajectory feature obtained by the i-th sampling point through the global feature mapping function, and T norm (i) represents the historical normal trajectory feature obtained by the i-th sampling point through the global feature mapping function.

[0197] Optionally, the device further includes at least one of the following:

[0198] An early warning module, configured to trigger an early warning measure if the drone has illegal flight behavior;

[0199] An iterative update module, configured to re-collect the latest normal flight data at a preset period, update the local historical normal flight data with the latest normal flight data, re-train the local AI model based on the updated local historical normal flight data, obtain new local model parameters and upload them to the federated learning platform, and receive the updated global AI model parameters sent by the federated learning platform to update the local AI model with the updated global AI model parameters.

[0200] Embodiment 3:

[0201] Refer to Figure 3 , this embodiment provides a monitoring device, including a memory 21 and a processor 22. A computer program is stored in the memory 21, and the processor 22 is configured to run the computer program to execute the monitoring method in Embodiment 1.

[0202] Among them, the memory 21 is connected to the processor 22. The memory 21 can be a flash memory, a read-only memory, or other memories, and the processor 22 can be a central processing unit or a single-chip microcomputer.

[0203] Embodiment 4:

[0204] This embodiment provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the monitoring method in Embodiment 1 above is implemented.

[0205] The computer-readable storage medium includes volatile or non-volatile, removable or non-removable media implemented in any method or technology for storing information such as computer-readable instructions, data structures, computer program modules, or other data. The computer-readable storage medium includes, but is not limited to, RAM (Random Access Memory), ROM (Read-Only Memory), EEPROM (Electrically Erasable Programmable Read Only Memory), flash memory or other memory technologies, CD-ROM (Compact Disc Read-Only Memory), digital versatile disc (DVD) or other optical disc storage, magnetic cassette, tape, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to store the desired information and can be accessed by a computer.

[0206] In summary, for the monitoring method, device, and readable storage medium provided by the embodiments of the present invention, the drone takeoff and landing base node joined to the blockchain network first obtains the current flight data of the drone, and then inputs the current flight data into the global artificial intelligence (AI) model. The global feature mapping function in the global AI model is used to extract features from the current flight data to generate corresponding current observation trajectory features. Among them, the global AI model is obtained through distributed collaborative training based on federated learning, and the global feature mapping function is used to convert the current flight data into a discriminative feature representation. Then, the trajectory difference degree between the current observation trajectory features and the historical normal trajectory features is calculated. Finally, in response to the trajectory difference degree being greater than a preset abnormal threshold, it is determined that the drone has black flight behavior. The present invention uses blockchain technology to ensure the immutability and decentralized storage of data. Through federated learning, each takeoff and landing base node only shares model parameters instead of raw data, effectively protecting the privacy of drone flight data. At the same time, the global AI model converts the current flight data (i.e., the original flight data) into a discriminative feature representation through feature mapping, which can effectively capture small but key abnormal information in the flight trajectory, thereby achieving high-precision identification of black flight behavior. In addition, compared with the radar monitoring technology, whose accuracy is limited in short-distance and low-altitude scenarios, the present invention is not limited by the height and distance of specific scenarios and can accurately extract features from various flight data. Compared with visual recognition technology, the present invention will not produce large monitoring errors due to the high-speed flight of drones or complex weather, because it relies on the analysis of flight data rather than simple visual capture. For the dilemma that the ADS-B technology is difficult to achieve comprehensive monitoring due to the equipment problem of drones, the present invention does not need to rely on specific equipment equipped on the drones themselves. Through data collection at each takeoff and landing base node and the operation of the global AI model, it can achieve a wider and more comprehensive monitoring, greatly improving the accuracy, stability, and comprehensiveness of drone black flight monitoring, and effectively ensuring airspace safety. It solves the problems in the existing monitoring technologies for drone black flight, where the radar monitoring technology has limited accuracy in short-distance and low-altitude scenarios, the visual recognition technology has large monitoring errors in the face of high-speed drones or complex weather, and the ADS-B technology is difficult to achieve comprehensive monitoring due to the equipment problem of drones.

[0207] It can be understood that the above embodiments are merely exemplary embodiments adopted to illustrate the principles of the present invention. However, the present invention is not limited thereto. For those of ordinary skill in the art, various modifications and improvements can be made without departing from the spirit and essence of the present invention, and these modifications and improvements are also regarded as the protection scope of the present invention.

Claims

1. A monitoring method, characterized in that, Applied to any drone takeoff and landing base node joining the blockchain network, the method includes: Obtain the current flight data of the drone; Input the current flight data into the global artificial intelligence (AI) model, and use the global feature mapping function in the global AI model to extract features from the current flight data, generating corresponding current observation trajectory features; wherein, the global AI model is obtained through distributed collaborative training based on federated learning, and the global feature mapping function is used to convert the current flight data into a discriminative feature representation; Calculate the trajectory difference degree between the current observation trajectory features and the historical normal trajectory features; In response to the trajectory difference degree being greater than a preset abnormal threshold, determine that the drone has illegal flight behavior.

2. The method according to claim 1, wherein Before obtaining the current flight data of the drone, the method further includes: Package the basic information of the drone takeoff and landing base node to generate initial message information, and broadcast it to other drone takeoff and landing base nodes in the blockchain network for registration; Wherein, the basic information includes the basic attribute information of the drone takeoff and landing base, the current number of drones at the base, the drone list, and the detailed information of each drone.

3. The method according to claim 1, characterized in that, Before obtaining the current flight data of the drone, the method further includes: Receive the takeoff request message sent by the drone, and the takeoff request message includes the identity identifier of the drone, the expected takeoff time, the current position, the current altitude, the flight destination, the flight time, and the flight route.

4. The method according to claim 1, wherein Before inputting the current flight data into the global artificial intelligence (AI) model and using the global feature mapping function in the global AI model to extract features from the current flight data to generate corresponding current observation trajectory features, the method further includes: Train a local AI model using local historical normal flight data to obtain local model parameters and a local feature mapping function; Upload the local model parameters to the federated learning platform, so that the federated learning platform aggregates the local model parameters received from all drone takeoff and landing base nodes using a preset federated algorithm to form global AI model parameters; Receive the global AI model parameters sent by the federated learning platform through the blockchain network; Replace the local model parameters with the global AI model parameters, and generate a global feature mapping function based on the global AI model parameters to form a global AI model.

5. The method according to claim 1, characterized in that, The current flight data includes the longitude and latitude, flight altitude, flight speed, heading angle, acceleration, and timestamp of multiple sampling points; The global AI model includes a time series model and a convolutional neural network, or includes the time series model, convolutional neural network, and self-attention mechanism.

6. The method according to claim 5, wherein The calculation of the trajectory difference degree between the current observation trajectory features and the historical normal trajectory features is specifically calculated according to the following formula: Where D diff represents the trajectory difference degree, N represents the total number of sampling points, and T obs (i) represents the current observed trajectory feature obtained by the global feature mapping function for the i-th sampling point, and T norm (i) represents the historical normal trajectory feature obtained by the global feature mapping function for the i-th sampling point.

7. The method according to claim 4, characterized in that The method further includes at least one of the following: If the drone has illegal flight behavior, trigger a warning measure; Collect the latest normal flight data again according to a preset period, update the local historical normal flight data with the latest normal flight data, retrain the local AI model based on the updated local historical normal flight data, obtain new local model parameters and upload them to the federated learning platform, and, receive the updated global AI model parameters sent by the federated learning platform, and update the local AI model with the updated global AI model parameters.

8. A monitoring device, characterized in that, It is set at any UAV takeoff and landing base node that joins the blockchain network. The device includes: A current flight data acquisition module, configured to acquire the current flight data of the UAV; A flight data feature extraction module, connected to the current flight data acquisition module, configured to input the current flight data into a global artificial intelligence (AI) model, and use a global feature mapping function in the global AI model to extract features from the current flight data, generating corresponding current observation trajectory features; wherein, the global AI model is obtained through distributed collaborative training based on federated learning, and the global feature mapping function is used to convert the current flight data into a discriminative feature representation; A trajectory difference degree calculation module, connected to the flight data feature extraction module, configured to calculate the trajectory difference degree between the current observation trajectory features and the historical normal trajectory features; A black flight behavior judgment module, connected to the trajectory difference degree calculation module, configured to, in response to the trajectory difference degree being greater than a preset abnormal threshold, judge that the UAV has black flight behavior.

9. A monitoring device, characterized in that, It includes a memory and a processor. A computer program is stored in the memory, and the processor is configured to run the computer program to implement the monitoring method according to any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, A computer program is stored on the computer-readable storage medium, and when the computer program is executed by the processor, it implements the monitoring method according to any one of claims 1-7.

Images