Unmanned aerial vehicle identity recognition method and system

Abstract

The invention provides an unmanned aerial vehicle identity recognition method and system, and the method comprises the steps: receiving a flight behavior data flow from an unmanned aerial vehicle; processing the flight behavior data flow by using time sequence analysis and a Bayesian network to obtain a multi-dimensional behavior trajectory model; comparing the multi-dimensional behavior track model with a preset identity behavior library through a graph matching algorithm so as to identify a behavior identifier matched with the multi-dimensional behavior track model; and determining the identity of the unmanned aerial vehicle by using a space-time consistency verification mechanism and a random forest algorithm according to the relevance between the cross validation behavior identifier and pre-stored unmanned aerial vehicle identity information. According to the invention, the accuracy and efficiency of unmanned aerial vehicle identity recognition are improved.

Description

Technical Field

[0001] The embodiments of the present application relate to the technical field of drone flight behavior data analysis and identity recognition, and in particular, to a drone identity recognition method and system. Background Art

[0002] With the widespread application of drones in logistics distribution, agricultural monitoring, security patrols, environmental monitoring and other fields, it has become crucial to ensure the safety and legality of drone operations. Drone identity recognition not only needs to process the position change information and flight attitude adjustment records in the drone flight behavior data stream, but also needs to combine multi-dimensional behavioral characteristics for comprehensive analysis to ensure accurate confirmation of the drone identity. In addition, in the face of complex and changing flight environments and dynamic flight modes, traditional methods based on single identifiers or static path matching can no longer meet the needs of modern drone management. Therefore, it is particularly necessary to develop a method that can effectively process large amounts of flight behavior data and use advanced data analysis technology to achieve high-precision identity recognition.

[0003] At present, drone identification mainly relies on the following methods: First, RFID or GPS tag-based authentication confirms the identity by sending a unique identifier through a device installed on the drone. This method is simple and direct, but it is easily affected by signal interference or equipment failure. Secondly, visual detection based on image recognition uses a camera to capture the appearance features of the drone and compares it with the images in the database. Although it is intuitive, it does not work well in complex environments such as low light and occlusion. Furthermore, traditional trajectory matching based on flight paths analyzes the historical flight paths of drones and compares them with preset standard paths to identify their identities. However, this method is difficult to cope with dynamically changing flight modes and complex flight environments. Finally, some studies have tried to use machine learning algorithms to classify drone behavior data, but they are often limited to specific types of flight behaviors and lack a comprehensive multidimensional behavior trajectory model construction and a systematic cross-validation mechanism.

[0004] Although existing solutions can meet certain identity recognition needs under specific conditions, they generally have the following defects: First, identity authentication methods based on RFID or GPS are easily affected by signal interference or equipment failure, resulting in recognition failure or misjudgment; second, image recognition-based methods perform poorly in complex environments such as low light and occlusion, limiting their scope of application; third, traditional trajectory matching methods only rely on historical path data, lack dynamic feature analysis of UAV flight behavior, and cannot effectively cope with changing flight modes and complex flight environments; finally, existing machine learning algorithms have difficulty achieving high-precision identity recognition when processing large-scale flight behavior data, especially in the presence of noise and partial data missing, the accuracy is greatly reduced. Summary of the invention

[0005] The embodiments of the present application provide a method and system for drone identity recognition, which are used to solve the problems of low accuracy and poor efficiency of drone identity recognition in the prior art.

[0006] In a first aspect, an embodiment of the present application provides a method for identifying a drone, including: Receiving a flight behavior data stream from the drone, the flight behavior data stream including position change information and flight attitude adjustment records; Using time series analysis and Bayesian networks, the flight behavior data stream is processed to obtain a multi-dimensional behavior trajectory model; By using a graph matching algorithm, the multidimensional behavior trajectory model is compared with a preset identity behavior library to identify a behavior identifier that matches the multidimensional behavior trajectory model; The identity of the drone is determined by cross-verifying the association between the behavior identifier and the pre-stored drone identity information using a spatiotemporal consistency check mechanism and a random forest algorithm.

[0007] Optionally, the multi-dimensional behavior trajectory model is compared with a preset identity behavior library by a graph matching algorithm to identify a behavior identifier that matches the multi-dimensional behavior trajectory model, including: The flight behavior data stream is subjected to dimensionality reduction processing by using a principal component analysis method and a random neighborhood embedding method, and anomaly detection is performed according to an autoencoder to obtain a flight behavior data set; Based on the flight behavior data set, a behavior map of the multi-dimensional behavior trajectory model is constructed, and a corresponding standard behavior map is constructed for each drone in the preset identity behavior library; Using a graph matching algorithm, the behavior graph is compared with the standard behavior graph to calculate a similarity score, based on the similarity score, candidate matching objects are preliminarily screened from the multidimensional behavior trajectory model and a preset identity behavior library, and the candidate matching objects are evaluated using a probabilistic graph model to obtain an optimized similarity score list; According to the set similarity threshold, combined with Bayesian decision theory, and using spatiotemporal context information, the behavior identifier is screened out from the optimized similarity score list.

[0008] Optionally, the graph matching algorithm is used to compare the behavior graph with the standard behavior graph to calculate a similarity score, and based on the similarity score, candidate matching objects are preliminarily screened from the multidimensional behavior trajectory model and a preset identity behavior library, and the candidate matching objects are evaluated using a probabilistic graph model to obtain an optimized similarity score list, including: Using a graph matching algorithm, the behavior graph is compared with the standard behavior graph to generate an initial similarity score; Using a machine learning algorithm to set a dynamic threshold, based on the dynamic threshold, screening the target behavior map from the initial similarity score, and using a clustering algorithm to determine the behavior identifier of the target behavior map, and using the behavior identifier as a candidate matching identifier; The probability of occurrence of the candidate matching identifiers under different conditions is evaluated by using a probabilistic graphical model, and the initial similarity scores of the candidate matching identifiers are adjusted by using a variational inference statistical method in combination with the time series characteristics of the UAV flight behavior and the influence of key environmental factors to generate an optimized similarity score; The optimized similarity scores are sorted by using a ranking learning algorithm to obtain an optimized similarity score list.

[0009] Optionally, the probabilistic graphical model is used to evaluate the probability of occurrence of the candidate matching identifier under different conditions, and the variational inference statistical method is used to adjust the initial similarity score of the candidate matching identifier in combination with the time series characteristics of the UAV flight behavior and the influence of key environmental factors to generate an optimized similarity score, including: Using a probabilistic graphical model, the occurrence probability of the candidate matching identifier under different conditions is evaluated to obtain an initial probability distribution; Based on the initial probability distribution, combined with time series analysis technology, the time evolution law of the candidate matching identifier is modeled to obtain an intermediate probability distribution; Collecting and analyzing data of key environmental factors, and adjusting the intermediate probability distribution based on the data of the key environmental factors to generate a target probability distribution, wherein the key environmental factors include wind speed, temperature, and humidity; A variational inference statistical method is applied to evaluate the probability distribution of the candidate matching identifiers based on the target probability distribution, and the initial similarity scores of the candidate matching identifiers are adjusted to generate optimized similarity scores.

[0010] Optionally, the using of the spatiotemporal consistency check mechanism and the random forest algorithm to determine the identity of the drone by cross-validating the association between the behavior identifier and pre-stored drone identity information includes: Using a spatiotemporal consistency verification mechanism, based on the time and space information extracted from the multidimensional behavior trajectory model, the behavior identifier is verified to obtain a candidate behavior identifier; Using a random forest algorithm, cross-validation is performed on the candidate behavior identifier and the pre-stored drone identity information to obtain a cross-validation result; Based on a set matching degree threshold, candidate behavior identifiers having a matching degree higher than the set matching degree threshold are screened out from the cross-validation results to generate a candidate list; In combination with the candidate behavior identifiers and the cross-validation result, a target behavior identifier is selected from the candidate list to confirm the identity of the drone.

[0011] Optionally, the cross-validation process is performed on the candidate behavior identifier and the pre-stored drone identity information by using a random forest algorithm to obtain a cross-validation result, including: Constructing a random forest algorithm model according to the candidate behavior identifiers; Pairing the candidate behavior identifier with pre-stored drone identity information to generate a data set; Using the random forest algorithm model, the data set is evaluated to calculate a matching probability value; The matching probability values ​​are summarized to obtain a cross-validation result.

[0012] Optionally, the flight behavior data stream is processed using time series analysis and Bayesian network to obtain a multidimensional behavior trajectory model, including: Using a time series analysis method, the time and space information in the flight behavior data stream is decomposed to extract the behavior patterns of the UAV in different time periods and generate a key time feature set; A multi-dimensional behavior trajectory model is generated by combining the key time feature set with a Bayesian network model.

[0013] In a second aspect, an embodiment of the present application provides a drone identification system, including: A receiving module, used to receive a flight behavior data stream from the UAV, wherein the flight behavior data stream includes position change information and flight attitude adjustment records; A processing module, used to process the flight behavior data stream using time series analysis and Bayesian network to obtain a multi-dimensional behavior trajectory model; A comparison module, used to compare the multi-dimensional behavior trajectory model with a preset identity behavior library through a graph matching algorithm to identify a behavior identifier that matches the multi-dimensional behavior trajectory model; The verification module is used to determine the identity of the drone by cross-verifying the association between the behavior identifier and the pre-stored drone identity information using a spatiotemporal consistency verification mechanism and a random forest algorithm.

[0014] In a third aspect, an embodiment of the present application provides a computing device, comprising a processor and a memory, wherein the memory stores a computer program, and the processor is configured to run the computer program to execute a drone identification method as described in any one of the first aspects.

[0015] In a fourth aspect, an embodiment of the present application provides a computer storage medium having computer program instructions stored thereon, wherein the computer program instructions, when executed by a processor, implement a drone identity recognition method as described in any one of the first aspects.

[0016] In an embodiment of the present application, a flight behavior data stream is received from a drone, wherein the flight behavior data stream includes position change information and flight attitude adjustment records; the flight behavior data stream is processed using time series analysis and a Bayesian network to obtain a multidimensional behavior trajectory model; the multidimensional behavior trajectory model is compared with a preset identity behavior library through a graph matching algorithm to identify a behavior identifier that matches the multidimensional behavior trajectory model; and the identity of the drone is determined based on cross-validation of the correlation between the behavior identifier and the pre-stored drone identity information using a spatiotemporal consistency check mechanism and a random forest algorithm.

[0017] The technical solution of this application has the following beneficial effects: This application uses time series analysis, Bayesian networks, graph matching algorithms, spatiotemporal consistency verification mechanisms, and random forest algorithms to accurately identify the behavior patterns of drones in complex environments and effectively screen out behavior identifiers that match the preset identity behavior library, thereby accurately confirming the identity of drones. Multi-level data processing and verification technologies, including spatiotemporal consistency verification and cross-validation of machine learning algorithms, are used to ensure the security and reliability of drone identity recognition. Even in the presence of interference or partial data loss, the system can still maintain a high degree of accuracy. This method is not only suitable for drone identity recognition in static environments, but also for dynamically changing flight environments. By conducting an in-depth analysis of the time series characteristics of the flight behavior data stream, the behavioral characteristics and evolution laws of drones in different time periods can be captured, thereby enhancing the adaptability of the system. Advanced machine learning algorithms (such as random forests) are used for fast and effective cross-validation, which significantly improves the speed and efficiency of drone identity recognition. Compared with the traditional one-by-one comparison method, this method can significantly shorten the recognition time and meet the needs of real-time monitoring. The method is flexible in design and easy to integrate new data analysis technologies and algorithm models, and can be adjusted and optimized according to the needs of specific application scenarios. In addition, as the identity behavior library continues to be enriched and improved, the system's recognition capabilities will be further enhanced and have good scalability.

[0018] Furthermore, the multidimensional behavior trajectory model is compared with the preset identity behavior library through a graph matching algorithm to identify matching behavior identifiers. First, the flight behavior data stream is reduced in dimension using the principal component analysis (PCA) and stochastic neighbor embedding (SNE) methods, and anomaly detection is performed based on the autoencoder to generate a clean flight behavior dataset. Then, a behavior graph of the multidimensional behavior trajectory model is constructed based on the dataset, and a corresponding standard behavior graph is constructed for each drone in the identity behavior library. Then, the graph matching algorithm is used to calculate the similarity score between the behavior graph and the standard behavior graph, and candidate matching objects are preliminarily screened out. These candidate objects are further evaluated using the probabilistic graph model to generate an optimized similarity score list. Finally, the final behavior identifier is screened out from the optimized similarity score list using the spatiotemporal context information, combined with the set similarity threshold and Bayesian decision theory.

[0019] This method achieves high-precision drone identification through a series of advanced data analysis techniques. First, dimensionality reduction processing is performed through PCA and SNE combined with anomaly detection of the autoencoder to ensure the quality of the input data and reduce the impact of noise and outliers. Secondly, a behavior graph is constructed and the similarity score is calculated through a graph matching algorithm, so that the system can accurately find the standard behavior graph that is closest to the target drone behavior pattern. In addition, the probabilistic graph model is used to evaluate the candidate matching objects to improve the reliability of the recognition results. Finally, the most suitable behavior identification is further screened out by combining Bayesian decision theory and spatiotemporal context information, which significantly improves the accuracy and robustness of identity recognition. This method not only enhances the adaptability and anti-interference ability of the system, but also can achieve efficient and accurate drone identity confirmation in complex environments.

[0020] These and other aspects of the present application will become more apparent from the description of the following embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0022] Figure 1 A flowchart of a drone identification method provided in an embodiment of the present application; Figure 2A schematic diagram of the structure of a drone identification system provided in an embodiment of the present application; Figure 3 A schematic diagram of the structure of a computing device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0023] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application.

[0024] In some of the processes described in the specification and claims of this application and the above-mentioned figures, multiple operations that appear in a specific order are included, but it should be clearly understood that these operations may not be executed in the order in which they appear in this article or executed in parallel. The serial numbers of the operations, such as 101, 102, etc., are only used to distinguish between different operations, and the serial numbers themselves do not represent any execution order. In addition, these processes may include more or fewer operations, and these operations may be executed in sequence or in parallel. It should be noted that the descriptions of "first", "second", etc. in this article are used to distinguish different messages, devices, modules, etc., do not represent the order of precedence, and do not limit the "first" and "second" to be different types.

[0025] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of this application.

[0026] Figure 1 A flowchart of a method for identifying a drone is provided for an embodiment of the present application, such as Figure 1 As shown, the method includes: Step 101: receiving a flight behavior data stream from a drone, wherein the flight behavior data stream includes position change information and flight attitude adjustment records; In this step, the flight behavior data stream is a series of continuous data points, including the drone's position change information (such as latitude and longitude, altitude, etc.) and flight attitude adjustment records (such as pitch angle, yaw angle, roll angle). These data are used to describe the drone's motion trajectory and attitude changes in a specific time period, and are the basis for subsequent analysis.

[0027] In actual operation, the system receives the flight behavior data sent by the drone in real time through the wireless communication module. The data is first stored in a buffer and then preprocessed to remove noise and outliers. The preprocessed data will be used to build a multidimensional behavior trajectory model.

[0028] For example, in a smart logistics distribution scenario, multiple drones are responsible for delivering packages from a warehouse to a customer's address. Each drone is equipped with a high-precision GPS module and an inertial measurement unit, which can send its position and attitude data to the central control system in real time. For example, when a drone takes off from a warehouse, it sends a set of data containing the current position coordinates (latitude, longitude, altitude) and attitude angles (pitch, yaw, roll) to the system every second. These data not only help the system monitor the flight status of the drone, but also provide basic data for subsequent behavior analysis.

[0029] Step 102: using time series analysis and Bayesian network to process the flight behavior data stream to obtain a multi-dimensional behavior trajectory model; In this step, time series analysis is a statistical method used to extract features in the time dimension, such as trends, seasonality, and random fluctuations. Bayesian network is a probabilistic graphical model used to represent the dependencies between variables. Combining these two methods, a multidimensional behavior trajectory model reflecting the behavior pattern of the drone can be extracted from the flight behavior data.

[0030] In actual operation, firstly, the flight behavior data stream is decomposed using time series analysis technology to identify trend, seasonality and random fluctuation components. Then, based on the extracted time feature set, Bayesian network modeling is used to learn the probability distribution of drone behavior patterns and their evolution over time, and finally a multi-dimensional behavior trajectory model is generated.

[0031] For example, continuing with the above-mentioned intelligent logistics distribution case, the central control system performs time series analysis on the received position and attitude data. The system first denoises and smoothes the data, and then uses the autoregressive integral moving average model to identify the behavior patterns of the drone in different time periods, such as takeoff, cruising, turning, hovering, etc. Next, the system uses Bayesian network modeling to train the model based on historical data to predict the future behavior trends of the drone. For example, during a package delivery process, the system found that a drone frequently turned and hovered in a certain area. This may be due to obstacles or traffic control in the area. The system adjusted the flight path of the drone accordingly to ensure that the task was completed safely and efficiently.

[0032] Step 103: Compare the multidimensional behavior trajectory model with a preset identity behavior library through a graph matching algorithm to identify a behavior identifier that matches the multidimensional behavior trajectory model; In this step, the graph matching algorithm is a technique used to compare the similarity between two graph structures. Here, it is used to compare the drone's behavior graph with the standard behavior graph in the preset identity behavior library. By calculating the similarity score, the standard behavior graph that matches the target drone's behavior pattern can be preliminarily screened out, thereby identifying the candidate behavior identity.

[0033] In actual operation, based on the multi-dimensional behavior trajectory model generated in the previous step, the system constructs a corresponding behavior graph and matches it with the standard behavior graph in the identity behavior library. The graph matching algorithm is used to calculate the similarity score between the two, and candidate matching objects are screened out according to the set threshold. This process helps to narrow the potential range of identities and improve the efficiency of subsequent verification.

[0034] For example, in a smart logistics distribution scenario, when a drone is found to have deviated from the preset path, the system will immediately trigger a graph matching algorithm to compare the drone's behavior graph with all standard behavior graphs in the identity behavior library. Assuming that the detected behavior pattern is highly similar to that of a drone responsible for delivery in a specific area, the system lists the drone as a suspected target. Specifically, the system extracts a series of behavior nodes (such as takeoff, cruising, turning, landing, etc.) of the drone in the past hour and matches them with the standard behavior graphs in the identity behavior library. If the similarity score exceeds the preset threshold, the drone is considered to be highly matched with a known drone, in preparation for further identity confirmation.

[0035] Step 104: Using the spatiotemporal consistency check mechanism and the random forest algorithm, the identity of the drone is determined by cross-validating the association between the behavior identifier and the pre-stored drone identity information.

[0036] In this step, the spatiotemporal consistency check mechanism is used to verify the continuity and rationality of the drone's behavior in time and space. The random forest algorithm is an ensemble learning method suitable for classification and regression tasks. The true identity of the drone can be determined by cross-validating the correlation between the behavior identification and the pre-stored drone identity information.

[0037] In actual operation, first, the temporal and spatial attributes of the candidate behavior identification are checked to see if they are reasonable using the spatiotemporal consistency check mechanism. Then, the random forest algorithm is applied to cross-validate these identifications, evaluate the strength of their association with the pre-stored identity information, and finally determine the identity of the drone. This method improves the accuracy and reliability of recognition.

[0038] For example, in the intelligent logistics distribution scenario, the suspected drones screened by the graph matching algorithm need to be further confirmed. The system verifies whether their behavior conforms to the normal operation logic through the spatiotemporal consistency verification mechanism. For example, the system checks whether the drone flies along the expected path within a specific time period and whether there is reasonable stop and turn behavior. Next, the system uses the random forest algorithm to cross-validate these behavior identifiers, and evaluates the degree of match between each behavior identifier and the known drone identity information through the trained random forest model. Assuming that the system finds that the behavior identifier of a drone is highly matched with the drone responsible for delivery in a specific area, and its behavior is completely consistent in time and space, the system finally confirms the identity of the drone and updates its mission status. This not only ensures the safe execution of the task, but also enhances the system's anti-interference ability.

[0039] Through the collaborative work of the above four steps, this method achieves efficient and accurate drone identity recognition. First, the flight behavior data stream of the drone is received and processed to ensure the integrity and accuracy of the input data; then, a multidimensional behavior trajectory model is constructed using time series analysis and Bayesian networks to lay the foundation for identity recognition; then, the graph matching algorithm is used to preliminarily screen out candidate behavior identifiers to narrow the identity range; finally, the spatiotemporal consistency verification mechanism and random forest algorithm are combined for final identity confirmation. The whole process not only improves the accuracy and robustness of recognition, but also shows good adaptability and anti-interference ability in complex environments, and is suitable for a variety of practical application scenarios. Especially in intelligent logistics distribution, this method effectively improves the safety and efficiency of drone operations and ensures the smooth execution of tasks.

[0040] In order to solve the problems of insufficient data dimension reduction and anomaly detection in the existing methods and further improve the accuracy and robustness of drone identity recognition, in some embodiments, the multi-dimensional behavior trajectory model is compared with a preset identity behavior library by a graph matching algorithm in step 103 to identify a behavior identifier that matches the multi-dimensional behavior trajectory model, including: The flight behavior data stream is subjected to dimensionality reduction processing by using a principal component analysis method and a random neighbor embedding method, and anomaly detection is performed according to an autoencoder to obtain a flight behavior data set; based on the flight behavior data set, a behavior map of the multidimensional behavior trajectory model is constructed, and a corresponding standard behavior map is constructed for each drone in a preset identity behavior library; a graph matching algorithm is used to compare the behavior map with the standard behavior map to calculate a similarity score, and based on the similarity score, candidate matching objects are preliminarily screened out from the multidimensional behavior trajectory model and the preset identity behavior library, and a probabilistic graph model is used to evaluate the candidate matching objects to obtain an optimized similarity score list; according to a set similarity threshold, in combination with Bayesian decision theory, and using spatiotemporal context information, behavior identifiers are screened out from the optimized similarity score list.

[0041] In this embodiment, principal component analysis is a statistical method used to reduce the dimension of data while retaining as much information as possible. Stochastic neighborhood embedding is a nonlinear dimensionality reduction technique that is particularly suitable for visualization and clustering of high-dimensional data. An autoencoder is a neural network structure that can automatically learn the feature representation of data and is used for anomaly detection. Through these methods, noise and outliers can be effectively removed to generate high-quality flight behavior data sets. The behavior maps and standard behavior maps constructed based on these data can more accurately reflect the real behavior patterns of drones.

[0042] In an embodiment of the present application, first, PCA and SNE are used to reduce the dimension of the flight behavior data stream, extract the main features and remove redundant information. Next, an autoencoder is used for anomaly detection to identify and filter out abnormal data points to ensure the quality of the data set. Then, a behavioral map of the multidimensional behavior trajectory model is constructed based on the processed data set, and a standard behavioral map is established for each drone in the identity behavior library. Next, a graph matching algorithm is used to calculate the similarity score between the two, and candidate matching objects are preliminarily screened out. Subsequently, these candidate objects are evaluated using a probabilistic graph model to generate an optimized similarity score list. Finally, combined with the set similarity threshold and Bayesian decision theory, the spatiotemporal context information is used to screen out the final behavior identifier from the optimized score list.

[0043] The following is a specific embodiment: In a smart city monitoring system, multiple drones are deployed for urban patrols and emergency response missions. To ensure the safety and legality of drone operations, the system needs to monitor the behavior of drones in real time and confirm their identities. First, the system receives flight behavior data streams from each drone, including position change information (such as latitude and longitude, altitude) and flight attitude adjustment records (such as pitch angle, yaw angle, roll angle). To improve data quality, the system uses PCA and SNE to reduce the dimensionality of these data, extract key features and remove redundant information. Next, an autoencoder is applied for anomaly detection to identify and filter out abnormal data points, such as sudden altitude changes or unreasonable attitude adjustments, to ensure the integrity and accuracy of the data set.

[0044] Based on the processed data set, the system constructs a behavior map of the multi-dimensional behavior trajectory model for each drone, and establishes a standard behavior map for each drone in the identity behavior library. During a routine patrol, a drone deviates from the preset path. The system immediately triggers the graph matching algorithm to compare the drone's behavior map with all the standard behavior maps in the identity behavior library and calculate the similarity score. Assume that the system finds that the drone's behavior pattern is highly similar to that of a drone responsible for patrolling a specific area, and lists it as a suspected target.

[0045] To further confirm the identity, the system uses a probabilistic graphical model to evaluate the association between the suspected target and other potential matches and generates an optimized similarity score list. Combining the set similarity threshold and Bayesian decision theory, the system uses spatiotemporal context information (such as the drone's historical flight path and current environmental conditions) to filter out the final behavior identification from the optimized score list. Assume that the system eventually confirms that the drone is indeed the one responsible for patrolling a specific area, and updates its mission status to ensure the safe execution of the mission. This method not only improves the accuracy and robustness of drone identity recognition, but also demonstrates good adaptability and anti-interference capabilities in complex environments.

[0046] Optionally, the step 103 uses a graph matching algorithm to compare the behavior graph with the standard behavior graph to calculate a similarity score, and based on the similarity score, preliminarily screens candidate matching objects from the multidimensional behavior trajectory model and a preset identity behavior library, and uses a probabilistic graph model to evaluate the candidate matching objects to obtain an optimized similarity score list, further comprising: The behavior map is compared with the standard behavior map by using a graph matching algorithm to generate an initial similarity score. A dynamic threshold is set by using a machine learning algorithm. Based on the dynamic threshold, a target behavior map is screened from the initial similarity score. The behavior identifier of the target behavior map is determined by using a clustering algorithm, and the behavior identifier is used as a candidate matching identifier. A probabilistic graph model is used to evaluate the probability of occurrence of the candidate matching identifier under different conditions. In combination with the time series characteristics of the UAV flight behavior and the influence of key environmental factors, a variational inference statistical method is used to adjust the initial similarity score of the candidate matching identifier to generate an optimized similarity score. A ranking learning algorithm is used to sort the optimized similarity scores to obtain an optimized similarity score list. Optionally, the method of using a probabilistic graphical model to evaluate the probability of occurrence of the candidate matching identifier under different conditions, and combining the time series characteristics of the drone flight behavior and the influence of key environmental factors, using a variational inference statistical method to adjust the initial similarity score of the candidate matching identifier to generate an optimized similarity score, including: using a probabilistic graphical model to evaluate the probability of occurrence of the candidate matching identifier under different conditions to obtain an initial probability distribution; based on the initial probability distribution, combining time series analysis technology, modeling the time evolution law of the candidate matching identifier to obtain an intermediate probability distribution; collecting and analyzing data on key environmental factors, adjusting the intermediate probability distribution based on the data on the key environmental factors to generate a target probability distribution, wherein the key environmental factors include wind speed, temperature and humidity; applying a variational inference statistical method to evaluate the probability distribution of the candidate matching identifier based on the target probability distribution, and adjusting the initial similarity score of the candidate matching identifier to generate an optimized similarity score.

[0047] In this embodiment, the graph matching algorithm is used to compare the similarity between two graph structures, which refers to comparing the behavior graph of the drone with the standard behavior graph in the preset identity behavior library. The machine learning algorithm is used to set a dynamic threshold and automatically adjust the threshold based on historical data to adapt to the changing environment. The clustering algorithm is used to identify behavioral identifiers with similar characteristics to form a set of candidate matching identifiers. The probabilistic graph model is used to evaluate the probability of occurrence of candidate matching identifiers under different conditions. Combining time series analysis technology and the influence of key environmental factors (such as wind speed, temperature, and humidity), the variational inference statistical method is used to adjust the initial similarity score and generate an optimized similarity score. The ranking learning algorithm is used to sort the optimized similarity scores to obtain the final score list.

[0048] In an embodiment of the present application, first, a graph matching algorithm is used to compare the behavior map with the standard behavior map to generate an initial similarity score. Next, a dynamic threshold is set using a machine learning algorithm, and the target behavior map is screened out from the initial similarity score based on the threshold, and the behavior identifiers of these maps are determined by a clustering algorithm as candidate matching identifiers. Then, a probabilistic graph model is used to evaluate the probability of occurrence of candidate matching identifiers under different conditions, and the time series analysis technology is combined to model the law of their time evolution, and data on key environmental factors are collected to adjust the intermediate probability distribution to generate a target probability distribution. The variational inference statistical method is applied to evaluate the probability distribution of candidate matching identifiers based on the target probability distribution, adjust the initial similarity score, and generate an optimized similarity score. Finally, the optimized similarity score is sorted using a ranking learning algorithm to obtain an optimized similarity score list.

[0049] The following is a specific embodiment: In an intelligent agricultural monitoring system, multiple drones are deployed for farmland monitoring and crop spraying tasks. To ensure the safety and legality of drone operations, the system needs to monitor the behavior of drones in real time and confirm their identities.

[0050] First, the system receives flight behavior data streams from each drone, including position change information (such as latitude and longitude, altitude) and flight attitude adjustment records (such as pitch angle, yaw angle, and roll angle). The system uses a graph matching algorithm to compare the behavior graph of each drone with the standard behavior graph in the identity behavior library to generate an initial similarity score. For example, during a routine monitoring mission, a drone deviated from the preset path. The system triggered the graph matching algorithm and found that its behavior pattern was highly similar to that of a drone responsible for monitoring a specific area, generating a higher initial similarity score.

[0051] Next, the system uses a machine learning algorithm to set a dynamic threshold, based on which it selects the target behavior graphs from the initial similarity score, and uses a clustering algorithm to determine the behavior identifiers of these graphs as candidate matching identifiers. Assume that the system selects two suspected drones and lists them as candidate matching identifiers.

[0052] To further confirm the identity, the system uses a probabilistic graphical model to evaluate the probability of occurrence of candidate matching logos under different conditions. The system first establishes an initial probability distribution based on historical data and the time series characteristics of the current flight behavior. Next, the system collects data on key environmental factors such as wind speed, temperature, and humidity, and adjusts the intermediate probability distribution based on this data to generate a target probability distribution. For example, assuming that the current wind speed is high, the system predicts that this may affect the flight trajectory of the drone, and thus adjusts its probability distribution.

[0053] The variational inference statistical method is applied to evaluate the probability distribution of candidate matching identifiers based on the target probability distribution, and adjust their initial similarity scores to generate optimized similarity scores. For example, the system finds that the probability distribution of one of the candidate drones is significantly higher than that of the other, indicating that it is more likely to be the target drone. Finally, the system uses a ranking learning algorithm to sort the optimized similarity scores and obtain an optimized similarity score list. Assume that the system eventually confirms that the drone is indeed the drone responsible for monitoring a specific area, and updates its mission status to ensure the safe execution of the mission.

[0054] In order to further improve the accuracy and reliability of drone identification, in some embodiments, the use of the spatiotemporal consistency check mechanism and the random forest algorithm in step 104 to determine the drone's identity by cross-validating the association between the behavior identifier and the pre-stored drone identity information includes: Using a spatiotemporal consistency verification mechanism, based on the time and space information extracted from the multidimensional behavior trajectory model, the behavior identifier is verified to obtain a candidate behavior identifier; using a random forest algorithm, the candidate behavior identifier is cross-validated with the pre-stored drone identity information to obtain a cross-validation result; based on a set matching threshold, the candidate behavior identifiers with a matching degree higher than the set threshold are screened out from the cross-validation result to generate a candidate list; combining the candidate behavior identifier and the cross-validation result, a target behavior identifier is selected from the candidate list to confirm the identity of the drone. Optionally, using a random forest algorithm to cross-validate the candidate behavior identifier with the pre-stored drone identity information to obtain a cross-validation result includes: constructing a random forest algorithm model based on the candidate behavior identifier; pairing the candidate behavior identifier with the pre-stored drone identity information to generate a data set; using the random forest algorithm model to evaluate the data set to calculate a matching probability value; summarizing the matching probability values ​​to obtain a cross-validation result.

[0055] In this embodiment, the spatiotemporal consistency verification mechanism is a method for verifying the consistency of data in time and space dimensions. It verifies the authenticity and legitimacy by analyzing the time series and geographic location information in the behavior identification. The random forest algorithm is an ensemble learning method that makes predictions by building multiple decision trees and aggregating their results to obtain more accurate and reliable classification or regression results. In this scenario, the random forest algorithm is used to evaluate the matching probability value between the candidate behavior identification and the pre-stored drone identity information.

[0056] In an embodiment of the present application, first, a spatiotemporal consistency verification mechanism is used to verify the authenticity of the behavior identifier based on the time and space information extracted from the multidimensional behavior trajectory model, and to screen out qualified candidate behavior identifiers. Next, a random forest algorithm is used to build a model, and the candidate behavior identifiers are paired with the pre-stored drone identity information to form a data set. Then, the random forest algorithm is used to evaluate this data set, and the matching probability value between each candidate behavior identifier and the drone identity information is calculated. These matching probability values ​​are summarized to form a cross-validation result. Based on a pre-set matching threshold, candidate behavior identifiers with a high degree of matching are screened out from these results to generate a candidate list. Finally, the candidate behavior identifiers and their corresponding cross-validation results are comprehensively considered, and the most likely target behavior identifier is selected to confirm the identity of the drone.

[0057] The following is a specific embodiment: In an urban logistics distribution system, drones are widely used to transport express parcels. To ensure the safety and legality of each drone's mission, the system needs to monitor the drone's behavior in real time and confirm its identity.

[0058] The system first collects various data during the flight of the drone, including the flight path (spatial information) and flight time (temporal information). Through the spatiotemporal consistency verification mechanism, the system verifies these behavior identifiers, such as checking whether a flight conforms to the expected route and schedule, thereby obtaining a set of candidate behavior identifiers.

[0059] Next, the system uses the random forest algorithm to train a model based on historical data to evaluate the correlation between candidate behavior identifiers and pre-stored drone identity information. Assuming a group of drones have participated in multiple delivery missions, the system pairs the historical behavior data of these drones with the current mission behavior identifiers to create a dataset with rich features. Using the random forest algorithm, the system evaluates this dataset and calculates the match probability value between each candidate behavior identifier and a specific drone identity information.

[0060] Based on the set matching threshold, the system screens out candidate behavior identifiers whose matching probability values ​​exceed the threshold and forms a candidate list. Considering the complexity of the actual operating environment, the system not only relies on the matching probability value, but also combines other factors (such as weather conditions, flight area restrictions, etc.) to select the final target behavior identifier from the candidate list to confirm the identity of the drone.

[0061] In order to further improve the analysis accuracy of the UAV flight behavior data and ensure that the multi-dimensional behavior trajectory model can accurately reflect the behavior pattern of the UAV, in some embodiments, the flight behavior data stream is processed by using time series analysis and Bayesian network in step 102 to obtain a multi-dimensional behavior trajectory model, including: The time and space information in the flight behavior data stream is decomposed using a time series analysis method to extract the behavior patterns of the UAV in different time periods and generate a key time feature set. A Bayesian network model is combined with the key time feature set to generate a multidimensional behavior trajectory model.

[0062] In this embodiment, time series analysis is a statistical technique used to identify patterns, trends, and periodicities in time series data. It can help understand the behavior patterns of drones in different time periods, such as takeoff, cruising, turning, landing, etc. The key time feature set includes useful information extracted from the original data, such as trend components, seasonal components, and random fluctuation components. The Bayesian network is a probabilistic graphical model used to represent the dependencies between variables. By combining the key time feature set, the Bayesian network can learn the probability distribution of drone behavior patterns and their evolution over time, thereby generating a multidimensional behavior trajectory model that comprehensively describes the drone behavior.

[0063] In an embodiment of the present application, first, a time series analysis method is used to decompose and process the time and space information in the flight behavior data stream. Specifically, the system decomposes the raw data into different components, such as long-term trends, seasonal changes, and short-term fluctuations. These components together constitute a key time feature set that reflects the behavior patterns of drones in different time periods. Next, using a Bayesian network model combined with these key time feature sets, the system can learn the probability distribution of drone behavior patterns and capture their changes over time. Ultimately, a multidimensional behavior trajectory model is generated based on this information, which can comprehensively describe the behavioral characteristics of the drone and provide a basis for subsequent identity recognition.

[0064] The following is a specific embodiment: In an intelligent agricultural monitoring system, multiple drones are deployed for farmland monitoring and crop spraying tasks. To ensure the safety and legality of drone operations, the system needs to monitor the behavior of drones in real time and confirm their identities.

[0065] First, the system receives a flown data stream of flight behavior from each drone, including position change information (such as latitude and longitude, altitude) and flight attitude adjustment records (such as pitch angle, yaw angle, and roll angle). The system uses time series analysis methods to decompose and process these data. For example, in a daily monitoring mission, a drone performed multiple round trips to cover the entire farmland area. The system first decomposes the flight path data into long-term trends (such as overall flight direction), seasonal changes (such as changes in daily flight hours), and short-term fluctuations (such as small path deviations caused by changes in wind speed), and extracts a set of key time features.

[0066] Next, the system uses the Bayesian network model to combine these key time feature sets to generate a multi-dimensional behavior trajectory model. Suppose the system finds that the drone exhibits obvious cruising and hovering behavior patterns in a specific time period, which may be due to the presence of obstacles in the area or areas that need to be monitored. The Bayesian network establishes the probability distribution of the drone's behavior patterns by learning historical data and predicts its future behavior trends. For example, the system predicts that the drone may exhibit similar behavior patterns again in the same area during the next flight.

[0067] Based on the generated multi-dimensional behavior trajectory model, the system can not only better understand the current behavior pattern of the drone, but also predict its future behavior trend, helping to optimize scheduling and path planning. For example, in a crop spraying mission, the system predicted that a certain field might need additional spraying operations based on the multi-dimensional behavior trajectory model, and arranged the drone to go to the area in advance, improving the efficiency and accuracy of the operation.

[0068] This application takes into account the lack of accuracy and robustness of drone behavior recognition in complex environments, especially in dynamically changing flight modes and complex flight environments, where traditional methods are difficult to provide high-precision identity recognition. Therefore, a new optional solution is proposed, which includes: The graph matching algorithm is used to compare the behavior graph with the standard behavior graph to calculate a similarity score, and based on the similarity score, candidate matching objects are preliminarily screened from the multidimensional behavior trajectory model and the preset identity behavior library, and the candidate matching objects are evaluated using a probabilistic graph model to obtain an optimized similarity score list, including: Using a graph matching algorithm, the behavior graph is compared with the standard behavior graph to generate an initial similarity score, wherein the initial similarity score The calculation formula is as follows: ; in, represents the initial similarity score, represents the time window length, is the improved time decay factor, and Respectively at time points The node set and edge set of the largest common subgraph on , and The nodes are and edge At the point in time The importance weight of and The nodes are and edge At the point in time The attribute matching degree of and is a nonlinear adjustment term, and Respectively at time points The node collection of the behavior graph and the standard behavior graph, and Respectively at the time point The edge set of the behavior graph and the standard behavior graph; The following is a detailed explanation of each parameter: Represents the initial similarity score, which is used to measure the similarity between the drone behavior map and the standard behavior map.

[0069] Indicates the length of time, indicating the time period to be considered. Usually set according to the application scenario, such as every second, every minute, etc.

[0070] represents the time decay factor, reflecting the impact of behavioral characteristics at different time points on the final score. ) is defined, where is the parameter that controls the decay rate.

[0071] and Respectively at the time point The set of nodes and edges of the largest common subgraph on . The elements in these sets are the common parts between the two graphs and are obtained through the graph matching algorithm.

[0072] and The nodes are and edge At the point in time The weight reflects the importance of a node or edge to the overall behavior pattern and can be pre-set based on domain knowledge or automatically adjusted through a learning algorithm.

[0073] and The nodes are and edge At the point in time The matching degree measures the similarity of the attributes of the corresponding nodes or edges in the two graphs, and the value range is usually It is obtained by comparing the attributes of nodes or edges.

[0074] and is a nonlinear adjustment item used to enhance or weaken the influence of attribute matching. The specific nonlinear function can be the imperial function (such as ),in and It is a parameter that controls the strength of the matching effect.

[0075] and Respectively at the time point A collection of nodes for behavior graphs and standard behavior graphs.

[0076] and Respectively at the time point The edge set of the behavior graph and the standard behavior graph.

[0077] The following is an introduction to the design reasons of each sub-item: This part calculates the sum of the weighted attribute matching of nodes and edges in the largest common subgraph. Through weighting and nonlinear adjustment, it ensures that important nodes and edges have a greater contribution to the score, while avoiding unimportant nodes and edges from having too much influence on the score. The reason for weighted summation is to comprehensively consider the importance of each node and edge and their matching degree, so as to make the score more comprehensive and accurate.

[0078] This section calculates the total weights of nodes and edges in the behavior graph and the standard behavior graph, and selects the larger one as the denominator. The purpose of this is to standardize the score to prevent it from being too high or too low. The reason for using the maximum value is that considering that the behavior graph and the standard behavior graph may be asymmetric, choosing a larger total weight can ensure the stability of the score.

[0079] Using a machine learning algorithm to set a dynamic threshold, based on the dynamic threshold, screening the target behavior map from the initial similarity score, and using a clustering algorithm to determine the behavior identifier of the target behavior map, and using the behavior identifier as a candidate matching identifier; The probability of occurrence of the candidate matching identifier under different conditions is evaluated by using a probabilistic graphical model, and the initial similarity score of the candidate matching identifier is adjusted by using a variational inference statistical method in combination with the time series characteristics of the UAV flight behavior and the influence of key environmental factors to generate an optimized similarity score. The calculation formula is as follows: ; in, represents the optimized similarity score, represents the sigmoid function, is the adaptive learning rate, represents the initial similarity score, and is the exponential adjustment parameter, represents the conditional probability, It is a penalty item.

[0080] The following is a detailed explanation of each parameter: Represents the optimized similarity score, which is used to measure the matching degree between the adjusted drone behavior map and the standard behavior map.

[0081] Represents the sigmoid function, which is used to map any real value to The specific form is This function helps to constrain the range of scores and provide a smooth probability interpretation.

[0082] Represents the adaptive learning rate, which is used to control the speed of score update. High values ​​make the score more responsive to new information, while low values ​​are more conservative and make the score change more gradually. Usually adjusted automatically by a machine learning algorithm or set based on experience.

[0083] Represents the initial similarity score, generated by the graph matching algorithm, which indicates the preliminary matching degree between the behavior graph and the standard behavior graph.

[0084] and Represents an index adjustment parameter used to enhance or weaken the initial similarity score and conditional probability The impact on the final score. These parameters can be adjusted according to specific application scenarios to optimize model performance.

[0085] Represents conditional probability, indicating that given environmental factors Next, candidate matching identifier Probability of occurrence. This probability is usually estimated through a Bayesian network or other probability model, reflecting the likelihood of occurrence of candidate matching identifiers under different conditions.

[0086] Represents a penalty term, which is used to reduce the impact of certain unfavorable features on the score. The specific penalty term can be about the candidate matching identifier A function of the frequency of occurrence or any negative influence factors associated with environmental factors.

[0087] The following is an introduction to the design reasons of each sub-item: This part is the initial similarity score Perform index adjustment. By introducing the index adjustment parameter , can flexibly enhance or weaken the impact of the initial similarity score. When the larger The value will be further amplified; when When the smaller The value will be relatively reduced. This design makes the score more sensitive to changes in the initial similarity.

[0088] Conditional Probability Reflects the situation under certain environmental conditions. By introducing this item, the score can be adjusted in accordance with the actual environmental factors to improve the accuracy and robustness of recognition.

[0089] The penalty term is used to reduce the impact of certain unfavorable features on the score. For example, if a candidate match has exhibited abnormal behavior many times in the past, its score can be lowered by increasing its penalty term. This helps filter out those behavior patterns that do not meet expectations and improves the reliability of the final recognition result.

[0090] This part comprehensively considers the impact of the initial similarity score, conditional probability and penalty term, and uses an adaptive learning rate and exponential adjustment parameters Further adjust the score. Index adjustment parameters Similar to , used to increase or decrease the influence of the entire expression.

[0091] By multiplying the sub-terms together, and A multiplication operation is used between them. This is because both reflect the degree of match between the behavior map and the standard behavior map, but with different emphases. The focus is on the similarity of graph structure, while The multiplication operation can effectively combine the information from these two aspects, making the score more comprehensive and accurate. From This is to reduce the impact of factors that are not conducive to scoring. In this way, those unreasonable candidate matching identifiers can be effectively filtered out and the reliability of the final score can be improved.

[0092] Here is a specific example: In an intelligent logistics distribution system, multiple drones are deployed for package delivery tasks. Assume that there are two drones A and B, which perform different delivery tasks. In order to ensure the safety and legality of drone operations, the system needs to monitor the behavior of drones in real time and confirm their identities.

[0093] Generating an initial similarity score Assume that after comparing the behavior graph of drone A with the standard behavior graph in the identity behavior library, the following parameters are obtained: Time window length , time decay factor , the importance weights of nodes and edges and The attribute matching degree is 0.8 and 0.6 respectively. and 0.9 and 0.7 respectively, nonlinear adjustment term and ; According to the formula: ; Calculated .

[0094] Set dynamic thresholds and filter target behavior maps. The dynamic threshold is set to 0.8 using a machine learning algorithm. If the value is greater than 0.8, the behavior graph of drone A is identified as the target behavior graph. The clustering algorithm is used to further determine its behavior identifier as a candidate matching identifier.

[0095] Generate optimized similarity score , Assume that the initial similarity score of drone A is , conditional probability , penalty term , adaptive learning rate , exponential adjustment parameter and According to the formula: ; in, is the sigmoid function. Calculated: ; From the calculation results, we can see that the final optimized similarity score is 0.61. This means that based on the existing data and conditions, the behavior pattern of drone A is highly consistent with the standard behavior map, but considering the influence of some uncertainties and environmental factors, its final score is slightly lower than the initial score. This shows that drone A is likely to belong to the expected identity, but further verification is still needed to ensure accuracy. This method not only improves the accuracy of recognition, but also enhances the robustness and adaptability of the system, which is particularly suitable for application scenarios such as intelligent logistics distribution that require efficiency and safety.

[0096] Figure 2 A schematic diagram of a drone identification system is provided for an embodiment of the present application. Figure 2 As shown, the system includes: The receiving module 21 is used to receive the flight behavior data stream from the UAV, wherein the flight behavior data stream includes position change information and flight attitude adjustment records; A processing module 22, for processing the flight behavior data stream using time series analysis and Bayesian network to obtain a multi-dimensional behavior trajectory model; A comparison module 23, configured to compare the multi-dimensional behavior trajectory model with a preset identity behavior library through a graph matching algorithm to identify a behavior identifier that matches the multi-dimensional behavior trajectory model; The verification module 24 is used to determine the identity of the drone by cross-verifying the association between the behavior identifier and the pre-stored drone identity information using a spatiotemporal consistency verification mechanism and a random forest algorithm.

[0097] Figure 2 The drone identification system can be implemented Figure 1 The implementation principle and technical effect of the drone identification method described in the embodiment are not described in detail. The specific way in which each module and unit performs operations in the drone identification system in the above embodiment has been described in detail in the embodiment of the method, and will not be described in detail here.

[0098] In one possible design, Figure 2 A drone identification system of the illustrated embodiment may be implemented as a computing device, such as Figure 3 As shown, the computing device may include a storage component 31 and a processing component 32; The storage component 31 stores one or more computer instructions, wherein the one or more computer instructions are called and executed by the processing component 32 .

[0099] The processing component 32 is used for the above Figure 1 The embodiment provides a method for identifying a drone.

[0100] The processing component 32 may include one or more processors to execute computer instructions to complete all or part of the steps in the above method. Of course, the processing component may also be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors or other electronic components to perform the above method.

[0101] The storage component 31 is configured to store various types of data to support operations at the terminal. The storage component can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0102] Of course, the computing device may also include other components, such as input / output interfaces, display components, communication components, etc.

[0103] The input / output interface provides an interface between the processing component and the peripheral interface module, which may be an output device, an input device, etc.

[0104] The communication component is configured to facilitate, among other things, wired or wireless communications between the computing device and other devices.

[0105] Among them, the computing device can be a physical device or an elastic computing host provided by a cloud computing platform, etc. In this case, the computing device can refer to a cloud server, and the above-mentioned processing components, storage components, etc. can be basic server resources rented or purchased from the cloud computing platform.

[0106] The present application also provides a computer storage medium storing a computer program, wherein the computer program can achieve the above-mentioned Figure 1 A method for identifying a drone in the illustrated embodiment.

[0107] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0108] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0109] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0110] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for identifying a drone, characterized in that: include: Receiving a flight behavior data stream from the drone, the flight behavior data stream including position change information and flight attitude adjustment records; Using time series analysis and Bayesian networks, the flight behavior data stream is processed to obtain a multi-dimensional behavior trajectory model; By using a graph matching algorithm, the multidimensional behavior trajectory model is compared with a preset identity behavior library to identify a behavior identifier that matches the multidimensional behavior trajectory model; The identity of the drone is determined by cross-verifying the association between the behavior identifier and the pre-stored drone identity information using a spatiotemporal consistency check mechanism and a random forest algorithm.

2. The method according to claim 1, characterized in that The multi-dimensional behavior trajectory model is compared with a preset identity behavior library by using a graph matching algorithm to identify a behavior identifier that matches the multi-dimensional behavior trajectory model, including: The flight behavior data stream is subjected to dimensionality reduction processing by using a principal component analysis method and a random neighborhood embedding method, and anomaly detection is performed according to an autoencoder to obtain a flight behavior data set; Based on the flight behavior data set, a behavior map of the multi-dimensional behavior trajectory model is constructed, and a corresponding standard behavior map is constructed for each drone in the preset identity behavior library; Using a graph matching algorithm, the behavior graph is compared with the standard behavior graph to calculate a similarity score, based on the similarity score, candidate matching objects are preliminarily screened from the multidimensional behavior trajectory model and a preset identity behavior library, and the candidate matching objects are evaluated using a probabilistic graph model to obtain an optimized similarity score list; According to the set similarity threshold, combined with Bayesian decision theory, and using spatiotemporal context information, the behavior identifier is screened out from the optimized similarity score list.

3. The method according to claim 2, characterized in that The graph matching algorithm is used to compare the behavior graph with the standard behavior graph to calculate a similarity score, and based on the similarity score, candidate matching objects are preliminarily screened from the multidimensional behavior trajectory model and the preset identity behavior library, and the candidate matching objects are evaluated using a probabilistic graph model to obtain an optimized similarity score list, including: Using a graph matching algorithm, the behavior graph is compared with the standard behavior graph to generate an initial similarity score; Using a machine learning algorithm to set a dynamic threshold, based on the dynamic threshold, screening the target behavior map from the initial similarity score, and using a clustering algorithm to determine the behavior identifier of the target behavior map, and using the behavior identifier as a candidate matching identifier; The probability of occurrence of the candidate matching identifiers under different conditions is evaluated by using a probabilistic graphical model, and the initial similarity scores of the candidate matching identifiers are adjusted by using a variational inference statistical method in combination with the time series characteristics of the UAV flight behavior and the influence of key environmental factors to generate an optimized similarity score; The optimized similarity scores are sorted by using a ranking learning algorithm to obtain an optimized similarity score list.

4. The method according to claim 3, characterized in that The probability graph model is used to evaluate the occurrence probability of the candidate matching identifier under different conditions, and the variational inference statistical method is used to adjust the initial similarity score of the candidate matching identifier in combination with the time series characteristics of the UAV flight behavior and the influence of key environmental factors to generate an optimized similarity score, including: Using a probabilistic graphical model, the occurrence probability of the candidate matching identifier under different conditions is evaluated to obtain an initial probability distribution; Based on the initial probability distribution, combined with time series analysis technology, the time evolution law of the candidate matching identifier is modeled to obtain an intermediate probability distribution; Collecting and analyzing data of key environmental factors, and adjusting the intermediate probability distribution based on the data of the key environmental factors to generate a target probability distribution, wherein the key environmental factors include wind speed, temperature, and humidity; A variational inference statistical method is applied to evaluate the probability distribution of the candidate matching identifiers based on the target probability distribution, and the initial similarity scores of the candidate matching identifiers are adjusted to generate optimized similarity scores.

5. The method according to claim 1, characterized in that The method of using the spatiotemporal consistency check mechanism and the random forest algorithm to determine the identity of the drone by cross-validating the association between the behavior identifier and the pre-stored drone identity information includes: Using a spatiotemporal consistency verification mechanism, based on the time and space information extracted from the multidimensional behavior trajectory model, the behavior identifier is verified to obtain a candidate behavior identifier; Using a random forest algorithm, cross-validation is performed on the candidate behavior identifier and the pre-stored drone identity information to obtain a cross-validation result; Based on a set matching degree threshold, candidate behavior identifiers having a matching degree higher than the set matching degree threshold are screened out from the cross-validation results to generate a candidate list; In combination with the candidate behavior identifiers and the cross-validation result, a target behavior identifier is selected from the candidate list to confirm the identity of the drone.

6. The method according to claim 5, characterized in that The random forest algorithm is used to cross-validate the candidate behavior identifier and the pre-stored drone identity information to obtain a cross-validation result, including: Constructing a random forest algorithm model according to the candidate behavior identifiers; Pairing the candidate behavior identifier with pre-stored drone identity information to generate a data set; Using the random forest algorithm model, the data set is evaluated to calculate a matching probability value; The matching probability values ​​are summarized to obtain a cross-validation result.

7. The method according to claim 1, characterized in that The flight behavior data stream is processed by using time series analysis and Bayesian network to obtain a multi-dimensional behavior trajectory model, including: Using a time series analysis method, the time and space information in the flight behavior data stream is decomposed to extract the behavior patterns of the UAV in different time periods and generate a key time feature set; A multi-dimensional behavior trajectory model is generated by combining the key time feature set with a Bayesian network model.

8. A drone identification system, characterized in that: include: A receiving module, used to receive a flight behavior data stream from the UAV, wherein the flight behavior data stream includes position change information and flight attitude adjustment records; A processing module, used to process the flight behavior data stream using time series analysis and Bayesian network to obtain a multi-dimensional behavior trajectory model; A comparison module, used to compare the multi-dimensional behavior trajectory model with a preset identity behavior library through a graph matching algorithm to identify a behavior identifier that matches the multi-dimensional behavior trajectory model; The verification module is used to determine the identity of the drone by cross-verifying the association between the behavior identifier and the pre-stored drone identity information using a spatiotemporal consistency verification mechanism and a random forest algorithm.

9. A computing device, characterized in that It comprises a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are used to be called and executed by the processing component to implement a drone identification method as described in any one of claims 1 to 7.

10. A computer storage medium, characterized in that: A computer program is stored, and when the computer program is executed by a computer, a method for identifying a drone as described in any one of claims 1 to 7 is implemented.

Images