A signal identification and analysis system for a drone monitoring device

Abstract

The application relates to the technical field of unmanned aerial vehicle monitoring, and discloses a signal identification and analysis system for unmanned aerial vehicle monitoring equipment, which comprises a signal acquisition module, an environment fingerprint modeling module, a strategy generation module, a counter training module, a pre-switching module, an identification and analysis module and an updating module. The signal acquisition module is used for acquiring broadcast remote identification signals and detection quality indexes. The environment fingerprint modeling module is used for establishing environment fingerprint entries. The strategy generation module is used for generating an identification and analysis strategy package. The counter training module is used for training strategy parameters and writing the strategy parameters into the identification and analysis strategy package. The pre-switching module is used for preloading a corresponding identification and analysis strategy package. The identification and analysis module is used for carrying out protocol identification, message analysis, field checking and exception discrimination. The updating module is used for updating the environment fingerprint entries and the identification and analysis strategy package. The environment fingerprint entries are established according to spatial detection partitioning, and corresponding identification and analysis strategy packages are generated. Meanwhile, the identification and analysis strategy packages are written by using counter training, and strategy pre-switching is carried out in combination with the cross-zone movement of target unmanned aerial vehicles, so that the adaptive capacity of signal identification and analysis in a complex environment is improved.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) monitoring technology, specifically to a signal recognition and analysis system for UAV monitoring equipment. Background Technology

[0002] Unmanned aerial vehicle (UAV) monitoring equipment uses signal identification to receive, identify, and analyze UAV signals in the airspace. Typically, it receives wireless signals broadcast by UAVs during flight to acquire and process UAV identity information, location information, and flight status. As UAV application scenarios continue to expand, broadcast remote identification signals are gradually becoming an important source of information for UAV supervision. These signals are usually continuously transmitted in the airspace via wireless broadcast to characterize UAVs and their control information, thereby enabling UAV identification and tracking. Existing UAV monitoring equipment typically employs a unified signal identification and analysis strategy to perform candidate signal screening, protocol identification, and message parsing processing on the received broadcast signals.

[0003] However, in current technology, due to the significant differences in the wireless environment at different locations within the monitoring airspace, and the fact that the reception status and message characteristics of broadcast remote identification signals are prone to change under complex interference or abnormal scenarios, it is difficult to adaptively adjust to different environments using a unified signal identification and parsing strategy. As a result, there is a problem of insufficient adaptability of signal identification and parsing in complex environments. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a signal recognition and analysis system for UAV monitoring equipment, which solves the problem of insufficient adaptability of signal recognition and analysis in complex environments.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a signal identification and analysis system for unmanned aerial vehicle (UAV) monitoring equipment, comprising: The signal acquisition module collects broadcast-style remote identification signals and detection quality indicators within the monitored airspace; The environmental fingerprint modeling module establishes corresponding environmental fingerprint entries according to spatial detection zones; The strategy generation module generates an identification and parsing strategy package based on the environmental fingerprint entries; The adversarial training module generates physical layer interference scenarios and protocol layer attack scenarios through an interference generator, performs adversarial training through an identification and parsing module, obtains policy parameters, and writes them into the identification and parsing policy package. The pre-switching module loads the corresponding identification and parsing strategy package before the predicted target drone enters the target space detection zone; The identification and parsing module performs protocol identification, message parsing, field verification, and anomaly detection on candidate signals based on the currently loaded identification and parsing strategy package. The update module updates the environmental fingerprint entries and the identification and analysis strategy package based on the identification and analysis results.

[0006] Preferably, the signal acquisition module includes the following steps: The system receives wireless signals in the 2.4 GHz and 5.8 GHz bands within the monitored airspace to obtain the raw received signals. The original received signal is sampled synchronously by multiple antennas to obtain the corresponding multi-channel received data; Based on the multi-channel received data, the broadcast remote identification signal and the corresponding detection quality indicators are extracted. The detection quality indicators include received signal strength, signal-to-noise ratio, field integrity rate, and multi-antenna reception consistency.

[0007] Preferably, the environmental fingerprint modeling module includes the following steps: The monitored airspace is divided into multiple spatial detection zones, and each spatial detection zone is assigned a corresponding zone identifier. Based on the partition identifier and the preset statistical time window, the background wireless environment features, historical broadcast remote identification signal analysis results and abnormal message features collected in each spatial detection partition are spatiotemporally aligned, and the partition feature association matrix of the corresponding spatial detection partition is constructed according to the feature category and frequency of occurrence. The partition feature correlation matrix is ​​fused to generate environmental fingerprint entries for the corresponding spatial detection partition.

[0008] Preferably, the strategy generation module includes the following steps: Read the environmental fingerprint entries for the corresponding spatial detection partition; A set of identification and parsing rules is determined based on the environmental fingerprint entries. The set of identification and parsing rules includes candidate signal filtering rules, protocol matching rules, multi-frame splicing rules, field verification rules, and anomaly detection rules. Based on the set of recognition and parsing rules, a recognition and parsing strategy package for the corresponding spatial detection partition is generated.

[0009] Preferably, the adversarial training module includes the following steps: Construct a generative adversarial network, which includes a noise generator and a recognition parser; The interference generator generates physical layer interference scenarios and protocol layer attack scenarios to form adversarial training samples; The recognition parser performs recognition parsing training on the adversarial training samples to obtain policy parameters; The strategy parameters are written into the identification and parsing rule set and the identification and parsing strategy package.

[0010] Preferably, the generation of physical layer interference scenarios and protocol layer attack scenarios to constitute adversarial training samples includes the following steps: Based on historically collected broadcast remote identification signal samples, background wireless environment data, and abnormal message samples, a basic dataset for adversarial training is constructed. The interference generator perturbs and generates physical layer interference scenarios and protocol layer attack scenarios by perturbing the basic adversarial training dataset. The physical layer interference scenarios include weak signal superposition scenarios, co-frequency interference scenarios, and multipath reflection scenarios; the protocol layer attack scenarios include protocol field forgery scenarios, message timing disturbance scenarios, and mixed injection scenarios of legitimate messages and forged messages. The physical layer interference scenario and the protocol layer attack scenario are combined to form the adversarial training sample.

[0011] Preferably, the strategy parameters are written into the identification and parsing rule set and the identification and parsing strategy package, and are invoked by the identification and parsing module, including the following steps: The strategy parameters are respectively written into the candidate signal filtering rules, protocol matching rules, multi-frame splicing rules, field verification rules and anomaly discrimination rules in the recognition and parsing rule set; Based on the set of recognition and parsing rules written with the policy parameters, a corresponding recognition and parsing policy package is generated. The identification and parsing module calls the identification and parsing strategy package to perform protocol identification, message parsing, field verification, and abnormal message discrimination on the candidate signals.

[0012] Preferably, the pre-switching module includes the following steps: Obtain the target UAV's position, speed, and heading information obtained from continuous time-series analysis; The movement trend of the target UAV is determined based on the target UAV's position, speed, and heading information, and the target space detection zone that the target UAV will enter is predicted. Before the target UAV enters the target space detection zone, the identification and parsing strategy package corresponding to the target space detection zone is loaded.

[0013] Preferably, the identification and parsing module includes the following steps: The received signal is filtered for candidate signals according to the currently loaded identification and parsing strategy package to obtain candidate signals; The candidate signals are identified according to the protocol matching rules in the identification and parsing strategy package, and the identified broadcast remote identification signals are parsed. Based on the field validation rules and anomaly detection rules in the identification and parsing strategy package, the parsing results are validated and anomaly messages are detected.

[0014] Preferably, the update module includes the following steps: Obtain the identification and parsing results, abnormal message discrimination results, and parsing failure samples; The identification and parsing results, abnormal message discrimination results, and parsing failure samples are associated with the corresponding spatial detection partitions; The environmental fingerprint entries corresponding to the spatial detection partition are updated based on the association results. The update adopts a weighted fusion method, which combines the historical environmental fingerprint entries before the current spatial detection partition is updated with the current update information formed by the current identification and parsing results, abnormal message discrimination results and parsing failure samples. Among them, the weight coefficient corresponding to the historical environment fingerprint entries is used to control the degree of retention of historical basic features, and the weight coefficient corresponding to the current update information is used to control the degree of introduction of new data; Based on the updated environmental fingerprint entries, the identification and parsing rule set and the identification and parsing strategy package are modified.

[0015] This invention provides a signal identification and analysis system for unmanned aerial vehicle (UAV) monitoring equipment. It has the following advantages: 1. This invention establishes environmental fingerprint entries according to spatial detection partitions and generates corresponding identification and parsing strategy packages. At the same time, it uses adversarial training to obtain strategy parameters and writes them into the identification and parsing strategy packages, enabling the identification and parsing module to perform strategy pre-switching in combination with the cross-regional movement of the target UAV, thereby improving the adaptability of signal identification and parsing in complex environments.

[0016] 2. This invention divides the monitoring airspace into multiple spatial detection zones and associates background wireless environment characteristics, historical broadcast remote identification signal analysis results, and abnormal message characteristics to form environmental fingerprint entries corresponding to each spatial detection zone. Based on this, differentiated identification and analysis strategy packages are generated, so that the signal processing process in different areas can match the wireless environment status of the corresponding area, avoiding the use of uniform rules to process signals across the entire domain.

[0017] 3. This invention acquires the identification and parsing results, abnormal message discrimination results, and parsing failure samples, and associates them with the corresponding spatial detection partitions. It continuously updates the environmental fingerprint entries and identification and parsing strategy packages, enabling the system to dynamically correct the processing rules of each spatial detection partition as monitoring data accumulates. This gives the identification and parsing process continuous adjustment capabilities, making it easier to adapt to changes in the wireless environment and abnormal message characteristics within the monitored airspace. Attached Figure Description

[0018] Figure 1 This is an architecture diagram of a signal recognition and analysis system for drone monitoring equipment according to the present invention. Detailed Implementation

[0019] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Please see the appendix Figure 1 This invention provides a signal identification and analysis system for unmanned aerial vehicle (UAV) monitoring equipment, comprising: The signal acquisition module collects broadcast-style remote identification signals and detection quality indicators within the monitored airspace; Furthermore, the signal acquisition module includes the following steps: The system receives wireless signals in the 2.4 GHz and 5.8 GHz bands within the monitored airspace to obtain the raw received signals. The original received signal is sampled synchronously by multiple antennas to obtain the corresponding multi-channel received data; Based on the multi-channel received data, the broadcast remote identification signal and the corresponding detection quality indicators are extracted. The detection quality indicators include received strength, signal-to-noise ratio, field integrity rate and multi-antenna reception consistency.

[0021] Specifically, the monitoring equipment is deployed around the target airspace. It continuously listens to wireless signals within the monitored airspace via a receiving front-end. Considering that broadcast remote identification signals typically operate in the 2.4GHz and 5.8GHz frequency bands, the receiving front-end covers and receives these frequency bands, forming a raw received signal that includes the target signal, background wireless service signals, and interference signals. In some embodiments, the broadcast remote identification signal can be a Remote ID broadcast signal. After obtaining the original received signal, multiple receiving antennas are used to synchronously sample under a unified clock condition to form corresponding multi-channel received data. The purpose of multi-antenna synchronous sampling is to retain the receiving differences at different spatial locations at the same time, so that subsequent processing can not only obtain the broadcast remote identification signal itself, but also obtain auxiliary information that characterizes the current receiving state. Based on this, the multi-channel received data is preprocessed, and signal segments that meet the characteristics of broadcast remote identification are extracted. Simultaneously, detection quality indicators corresponding to the signal segments are statistically analyzed. These indicators include received signal strength (RSS), signal-to-noise ratio (SNR), field completeness rate, and multi-antenna reception consistency. RSS characterizes the received signal amplitude level; SNR characterizes the signal's distinguishability relative to background noise; field completeness rate characterizes the completeness of extractable fields in the signal; and multi-antenna reception consistency characterizes the consistency between the same signal received by different antennas. The broadcast remote identification signal and detection quality indicators obtained through this process serve as input for subsequent environmental fingerprint modeling and identification analysis strategy generation. For example, when a fixed monitoring device is installed at the perimeter of an airport, the device can continuously receive wireless signals in the 2.4 GHz and 5.8 GHz bands in the surrounding airspace, and extract the broadcast remote identification signal and the corresponding detection quality indicators broadcast by the UAV for subsequent processing; in some embodiments, the broadcast remote identification signal is the Remote ID broadcast signal.

[0022] The environmental fingerprint modeling module establishes corresponding environmental fingerprint entries according to spatial detection zones; Furthermore, the environmental fingerprint modeling module includes the following steps: The monitored airspace is divided into multiple spatial detection zones, and each spatial detection zone is assigned a corresponding zone identifier. Based on the partition identifier and the preset statistical time window, the background wireless environment features, historical broadcast remote identification signal analysis results and abnormal message features collected in each spatial detection partition are spatiotemporally aligned, and the partition feature association matrix of the corresponding spatial detection partition is constructed according to the feature category and frequency of occurrence. The partition feature correlation matrix is ​​fused to generate environmental fingerprint entries for the corresponding spatial detection partition.

[0023] Specifically, the environmental fingerprint modeling module performs a structured representation of the wireless environment status of different spatial detection zones within the monitored airspace, enabling the subsequent policy generation module to generate corresponding identification and parsing policy packages based on the environmental status of different spatial detection zones. This module receives broadcast remote identification signal related data output by the signal acquisition module and forms environmental fingerprint entries corresponding to each spatial detection zone through partitioning, spatiotemporal alignment, and feature fusion processing. In practice, the monitoring airspace is first divided into multiple spatial detection zones based on the installation location, coverage area, and spatial distribution characteristics of the monitoring equipment. Each spatial detection zone is then assigned a corresponding zone identifier. These spatial detection zones can be divided according to a fixed grid or by combining key flight paths, building distribution, obstruction areas, or historical signal distribution. Their purpose is to ensure that the wireless environment status within the same spatial detection zone is relatively consistent. The zone identifier is used to locate and assign various types of data subsequently collected, enabling data from different spatial detection zones to be processed separately. After partitioning, based on the partition identifier and preset statistical time window, the background wireless environment features, historical broadcast remote identification signal analysis results, and abnormal message features collected in each spatial detection partition are spatiotemporally aligned. A partition feature association matrix is ​​then constructed according to feature category and frequency of occurrence for the corresponding spatial detection partition. The preset statistical time window is used to limit the time range of the data participating in the modeling, ensuring that data within the same statistical period and belonging to the same spatial detection partition can be processed at a unified time scale. Spatiotemporal alignment refers to uniformly matching the background wireless environment features, historical analysis results, and abnormal message features within the same spatial detection partition and the same preset statistical time window according to time and spatial affiliation. After spatiotemporal alignment, the data is then classified and organized according to feature category, and a feature association matrix is ​​constructed based on the frequency of occurrence of each type of feature within the preset statistical time window. This matrix reflects the distribution of environmental features, historical analysis status, and abnormal situation distribution of the corresponding spatial detection partition within the statistical period. Based on this, the partition feature correlation matrix is ​​fused to generate environmental fingerprint entries for the corresponding spatial detection partitions. In this embodiment, the environmental fingerprint entry corresponding to the k-th spatial detection partition can be represented as: ; in, Indicates the first Environmental fingerprint entries for each spatial detection partition Indicates the first The partition feature correlation matrix corresponding to each spatial detection partition The function represents the fusion processing function, which is used to perform fusion processing on the partition feature association matrix to obtain the corresponding environmental fingerprint entries. The fusion processing can be implemented by feature splicing, weighted combination, normalization processing or mapping encoding, and the data representing the environmental state of the spatial detection partition in the partition feature association matrix is ​​uniformly converted into environmental fingerprint entries that can be called later. For example, when deploying fixed monitoring equipment around an airport, the runway extension direction, the area around the terminal building, and the outer edge of the parking lot can be divided into different spatial detection zones, and each zone can be assigned a zone identifier. Within a certain preset statistical time window, if there is a high frequency of background wireless interference, a large number of historical parsing anomalies, and multiple abnormal message records in the spatial detection zone corresponding to the area around the terminal building, the above data can be spatiotemporally aligned based on the zone identifier of the spatial detection zone and the preset statistical time window. The zone feature association matrix of the spatial detection zone can be constructed according to the feature category and the frequency of occurrence. Then, the matrix is ​​fused to generate the corresponding environmental fingerprint entry. Thus, the subsequent strategy generation module can generate an identification and parsing strategy package adapted to the environmental state of the area based on the environmental fingerprint entry. Through the above processing, the environmental fingerprint modeling module realizes a continuous process from spatial detection partitioning, spatiotemporal alignment of partition data, construction of partition feature correlation matrix to generation of environmental fingerprint entries, enabling the environmental status of different areas in the monitored airspace to be represented in a structured manner.

[0024] The strategy generation module generates an identification and parsing strategy package based on the environmental fingerprint entries; Furthermore, the strategy generation module includes the following steps: Read the environmental fingerprint entries for the corresponding spatial detection partition; The set of identification and parsing rules is determined based on the environmental fingerprint entries. The set of identification and parsing rules includes candidate signal screening rules, protocol matching rules, multi-frame splicing rules, field verification rules, and anomaly detection rules. The identification and parsing strategy package for the corresponding spatial detection partition is generated based on the set of identification and parsing rules.

[0025] Specifically, the environmental fingerprint entries corresponding to the target spatial detection partition are retrieved from the environmental fingerprint entry storage area. Since the aforementioned environmental fingerprint modeling module has completed the structured expression of the monitoring spatial partition and the environmental characteristics of each partition, after receiving the partition identifier of a certain spatial detection partition, the environmental fingerprint entries corresponding to the partition identifier can be directly read as the input data for the current policy generation. This reading process enables the policy generation module to organize subsequent rules based on the environmental state of the specific spatial detection partition, rather than relying on a unified global default parameter. After obtaining the corresponding environmental fingerprint entry, a set of identification and parsing rules is determined based on the environmental fingerprint entry. This set of rules describes the combination state of various processing rules in the subsequent signal identification and parsing process, including candidate signal filtering rules, protocol matching rules, multi-frame splicing rules, field verification rules, and anomaly detection rules. Specifically, candidate signal filtering rules determine the conditions for subsequent identification processes of the received signal; protocol matching rules determine the protocol type or protocol characteristics corresponding to the broadcast remote identification signal; multi-frame splicing rules associate multiple frames of data for the same target when a single frame field is incomplete; field verification rules perform consistency checks on the parsed field content; and anomaly detection rules identify parsed abnormal messages, forged messages, or conflicting messages. In this embodiment, the set of identification and parsing rules corresponding to the k-th spatial detection partition can be represented as: ; in, This represents the set of recognition and parsing rules corresponding to the k-th spatial detection partition. This represents the environmental fingerprint entry corresponding to the k-th spatial detection partition. This represents a rule set generation function, which is used to determine the set of identification and parsing rules suitable for the spatial detection partition based on the background wireless environment characteristics, detection quality indicators, historical broadcast remote identification signal parsing results, and abnormal message characteristics in the environmental fingerprint entry. The rule set generation function can be implemented by rule mapping, table lookup matching, weighted selection, or parameter configuration, as long as it can output the corresponding set of identification and parsing rules based on the environmental fingerprint entry. After the set of identification and parsing rules is determined, it is organized into an identification and parsing strategy package for the corresponding spatial detection partition. This identification and parsing strategy package can be understood as a rule carrier directly invoked by the subsequent identification and parsing module during signal processing. It contains the set of identification and parsing rules corresponding to the current spatial detection partition and can further record rule version information, partition identification information, and invocation priority information as needed. In this embodiment, the identification and parsing strategy package corresponding to the k-th spatial detection partition can be represented as: ; in, This represents the identification and parsing strategy packet corresponding to the k-th spatial detection partition. This represents the set of recognition and parsing rules corresponding to the k-th spatial detection partition. Indicates the strategy parameters, This indicates the strategy package generation function, used to generate a callable identification and parsing strategy package based on the set of identification and parsing rules. At this stage, if strategy parameters have not yet been introduced, then... The parameter set can be empty or the default parameter set. After the strategy parameters are output by the subsequent adversarial training module, the strategy parameters can be written into the identification and parsing strategy package. This makes the strategy package more adaptable to complex interference scenarios and protocol attack scenarios on the basis of the original rules. By organizing the set of identification and parsing rules into a strategy package, the subsequent pre-switching module and identification and parsing module can directly call the corresponding strategy according to the spatial detection partition without having to repeat the rule generation process during online processing. For example, in scenarios where fixed monitoring equipment is deployed around an airport, for spatial detection zones near the terminal building, due to the dense background wireless signals, high probability of signal collisions, and high frequency of abnormal packets during historical parsing, the set of identification and parsing rules generated from the environmental fingerprint entries of this spatial detection zone can be configured with stricter candidate signal filtering conditions, higher priority field verification rules, and more sensitive anomaly discrimination rules. For spatial detection zones in the runway extension direction, due to the relatively open background environment and high historical parsing success rate, a set of rules that are more inclined towards fast matching and direct parsing can be configured. The different identification and parsing strategy packages generated are then associated and stored with the corresponding spatial detection zones for subsequent module calls. Through the above processing, the policy generation module realizes the transformation from environmental fingerprint entries to a set of identification and parsing rules, and then to an identification and parsing policy package. This enables the partitioned environmental representation results output by the environmental fingerprint modeling module to be further transformed into policy data that can be directly executed in subsequent signal identification and parsing processing. This provides a foundation for the subsequent adversarial training module to write policy parameters, the pre-switching module to load the target partition policy package, and the identification and parsing module to perform protocol identification and message parsing according to the policy package.

[0026] The adversarial training module generates physical layer interference scenarios and protocol layer attack scenarios through an interference generator, performs adversarial training through an identification and parser, obtains policy parameters, and writes them into the identification and parsing policy package. Furthermore, the adversarial training module includes the following steps: Construct a generative adversarial network, which includes a noise generator and a recognition parser; The interference generator generates physical layer interference scenarios and protocol layer attack scenarios to form adversarial training samples; The recognition parser performs recognition parsing training on adversarial training samples to obtain policy parameters; Write the strategy parameters into the recognition and parsing rule set and the recognition and parsing strategy package.

[0027] Specifically, the adversarial training module is used to generate complex interference and attack samples during the offline training phase, and to train the identification and parsing module accordingly to obtain policy parameters applicable to actual monitoring equipment. It takes the identification and parsing rule set and identification and parsing policy package output by the policy generation module, trains and enhances them, so that the subsequent identification and parsing module can still perform signal identification and parsing according to the trained policy package when facing complex electromagnetic environments, abnormal messages or attack scenarios. Since the generative adversarial network itself can be implemented using the existing adversarial training framework, it can use the interference generator to construct training samples suitable for broadcast remote identification signal scenarios, and write the trained policy parameters into the identification and parsing rule set and identification and parsing policy package. In practice, a generative adversarial network (GAN) is first constructed. The GAN includes an interference generator and a recognition parser. The interference generator is used to generate training scenarios, and the recognition parser is used to perform signal recognition and message parsing training in the training scenarios. The GAN can be deployed on the offline training platform corresponding to the monitoring equipment and establish data connections with the aforementioned environmental fingerprint entries, recognition and parsing rule sets, and historical collection samples to ensure that the training process is consistent with the characteristics of broadcast remote recognition signals in the actual monitoring airspace. After the generative adversarial network is constructed, the interference generator generates physical layer interference scenarios and protocol layer attack scenarios to form adversarial training samples. The physical layer interference scenarios are used to characterize the changes in the reception state of broadcast remote identification signals in complex wireless environments, while the protocol layer attack scenarios are used to characterize abnormal disturbances to the content or organization of broadcast remote identification messages. The constructed adversarial training samples are no longer limited to conventional samples collected in history, but can cover complex scenarios that rarely occur in actual monitoring but may affect the identification and resolution process, thereby providing more sufficient sample input for the training of the subsequent identification and resolution device. After obtaining the adversarial training samples, the recognition parser performs recognition and parsing training under the adversarial training samples to obtain policy parameters. The recognition and parsing training takes the recognition and parsing task of broadcast remote recognition signals as the goal and trains and adjusts the processing processes such as candidate signal screening, protocol matching, multi-frame stitching, field verification and anomaly detection. This enables the recognition parser to gradually form parameter configuration results that are more suitable for the current monitoring needs when facing complex scenarios generated by the interference generator. Among them, the policy parameters can be the parameterized correction results of various rules in the aforementioned recognition and parsing rule set, or they can be a set of parameters used to adjust the rule calling conditions, threshold range or matching priority. After training is completed, the policy parameters are written into the recognition and parsing rule set and the recognition and parsing policy package. The policy parameters can be associated with the recognition and parsing rule set of the corresponding spatial detection partition and written into the corresponding recognition and parsing policy package. This allows the subsequent recognition and parsing module to directly use the trained parameters to perform the recognition and parsing of the broadcast remote recognition signal when calling the recognition and parsing policy package, without having to retrain in the online processing stage. Thus, the training results output by the adversarial training module are transformed into deployable and callable policy data and are connected with the aforementioned policy generation module. For example, in airport perimeter monitoring scenarios, it is rare for multiple abnormal signals to overlap simultaneously in conventionally collected samples. By using the interference generator in the offline training platform, training samples can be constructed where broadcast remote identification signals and background interference signals coexist and the message fields exhibit abnormal disturbances. After the identification parser completes training under such samples, it can obtain policy parameters applicable to such complex scenarios and write them into the identification and parsing policy package of the corresponding spatial detection partition for subsequent online identification and parsing. Through the above processing, the adversarial training module realizes the process from generating complex training samples, identifying and parsing training to writing policy parameters. This enables the aforementioned identification and parsing policy package to not only reflect the environmental state of the corresponding spatial detection partition, but also to include the parameter configuration results trained for complex interference and attack scenarios, providing a foundation for subsequent policy parameter writing methods.

[0028] Furthermore, the interference generator generates physical layer interference scenarios and protocol layer attack scenarios to constitute adversarial training samples, including the following steps: Based on historically collected broadcast remote identification signal samples, background wireless environment data, and abnormal message samples, a basic dataset for adversarial training is constructed. The physical layer interference scenario and the protocol layer attack scenario are obtained by perturbing the basic dataset of adversarial training using the interference generator. Among them, physical layer interference scenarios include weak signal superposition scenarios, co-frequency interference scenarios, and multipath reflection scenarios; protocol layer attack scenarios include protocol field forgery scenarios, message timing disturbance scenarios, and mixed injection scenarios of legitimate messages and forged messages. Physical layer interference scenarios and protocol layer attack scenarios are combined to form adversarial training samples.

[0029] Specifically, the generation of adversarial training samples is based on historical data and extended by combining the interference and anomaly features of broadcast remote identification signals, so that the training samples can both retain the data features in the actual monitoring scenario and cover complex scenarios that are not easy to obtain in the conventional acquisition process. First, based on historically collected broadcast remote identification signal samples, background wireless environment data, and abnormal message samples, a basic dataset for adversarial training is constructed. Among them, broadcast remote identification signal samples are used to characterize the characteristics of normal target signals, background wireless environment data are used to characterize the background interference in the monitoring airspace, and abnormal message samples are used to characterize the characteristics of abnormal messages that occurred during historical monitoring. After obtaining the basic dataset for adversarial training, the interference generator perturbs and generates physical layer interference scenarios and protocol layer attack scenarios. The perturbation generation combines the reception characteristics and message characteristics of broadcast remote identification signals to perturb the signal strength, signal overlap relationship, propagation path characteristics, message field content and message timing relationship. Among them, physical layer interference scenarios include weak signal superposition scenarios, co-frequency interference scenarios, and multipath reflection scenarios; protocol layer attack scenarios include protocol field forgery scenarios, message timing disturbance scenarios, and mixed injection scenarios of legitimate and forged messages. By generating the above scenarios, the training samples can cover complex disturbances at both the physical layer and the protocol layer. After generating physical layer interference scenarios and protocol layer attack scenarios, the two are combined to form adversarial training samples, which are then input into the recognition parser for training. For example, in airport perimeter monitoring scenarios, weak signal superposition scenarios, co-channel interference scenarios, and protocol field forgery scenarios can be generated based on historically collected broadcast remote identification signal samples, background wireless environment data, and abnormal message samples. These can be combined to form adversarial training samples for use in training the identification parser.

[0030] Furthermore, the strategy parameters are written into the recognition and parsing rule set and the recognition and parsing strategy package, and then called by the recognition and parsing module, including the following steps: The strategy parameters are respectively written into the candidate signal filtering rules, protocol matching rules, multi-frame splicing rules, field verification rules and anomaly discrimination rules in the recognition and parsing rule set; Based on the set of recognition and parsing rules written with the policy parameters, a corresponding recognition and parsing policy package is generated. The identification and parsing module calls the identification and parsing strategy package to perform protocol identification, message parsing, field verification, and abnormal message discrimination on candidate signals.

[0031] Specifically, the strategy parameters are not stored independently and called separately, but are written into the recognition and parsing rule set, and further generated into a recognition and parsing strategy package that can be directly called by online processing. In this way, the parameter configuration results obtained in the adversarial training stage can be directly converted into the strategy content that the recognition and parsing module can execute, so that the offline training results can be used in the online recognition and parsing process. First, the strategy parameters are written into the candidate signal filtering rules, protocol matching rules, multi-frame splicing rules, field verification rules and anomaly discrimination rules in the recognition and parsing rule set respectively. That is, according to the category and target of the strategy parameters, the parameter items in each type of rule are adjusted so that the recognition and parsing rule set changes from the initial rule state to the rule state after training and correction. After the strategy parameters are written, a corresponding identification and parsing strategy package is generated based on the set of identification and parsing rules written with the strategy parameters. The identification and parsing strategy package is used to organize and encapsulate the set of identification and parsing rules in a unified manner so that they can be stored, loaded and called in subsequent spatial detection partitions. During the online identification and parsing process, the identification and parsing module calls the identification and parsing strategy package to perform protocol identification, message parsing, field verification and abnormal message discrimination on the candidate signal. Therefore, when the identification and parsing module processes broadcast remote identification signals, it no longer relies on the initial rules, but on the identification and parsing strategy package containing strategy parameters. For example, in an airport perimeter monitoring scenario, if the strategy parameters obtained from adversarial training indicate that the sensitivity of abnormal message discrimination rules should be increased and the correlation conditions in multi-frame stitching rules should be adjusted within a certain spatial detection zone, then the corresponding parameters can be written into the recognition and parsing rule set, and an updated recognition and parsing strategy package can be generated for the recognition and parsing module to call within that spatial detection zone. Through the above processing, the training results can be directly applied to the actual recognition and parsing process.

[0032] The pre-switching module loads the corresponding identification and parsing strategy package before the predicted target drone enters the target space detection zone; Furthermore, the pre-switching module includes the following steps: Obtain the target UAV's position, speed, and heading information obtained from continuous time-series analysis; The movement trend of the target UAV is determined based on its position, speed, and heading information, and the target space detection zone that the target UAV will enter is predicted. Before the target drone enters the target space detection zone, the corresponding identification and analysis strategy for the target space detection zone is loaded.

[0033] Specifically, the pre-switching module is used to load the corresponding identification and analysis strategy package in advance before the target UAV enters the next spatial detection zone. This allows the identification and analysis module to directly perform signal processing according to the strategy adapted to the environmental conditions of the area after the target enters the spatial detection zone, avoiding the processing delay caused by temporarily switching strategies after the target crosses the zone. In practice, the current motion state of the target UAV is first determined based on the position, speed and heading information of the target UAV obtained at continuous time intervals. The position, speed and heading information of the target UAV can be obtained from the analysis results of the broadcast remote identification signal in the previous identification and analysis process, and stored and updated in chronological order to form continuous motion data of the target UAV. Through the motion data at continuous time intervals, the flight direction and displacement changes of the target UAV in the current monitoring period can be reflected, providing a basis for subsequent motion trend judgment. After obtaining the target UAV's position, speed, and heading information at continuous intervals, the movement trend of the target UAV is determined based on the information, and the target space detection zone it will enter is predicted. In this embodiment, the target UAV at each interval... The predicted location can be represented as: ; in, Indicates the target drone at time The predicted location, Indicates the target drone at time Current location Indicates the target drone at time The velocity vector, This represents the prediction time interval. When heading information is given in the form of heading angles, the velocity vector can be expressed as: ; in, Indicates the target drone at time speed magnitude, Indicates the target drone at time The heading angle, based on the correspondence between the predicted position and the boundaries of each spatial detection zone, can determine the target spatial detection zone that the target UAV will enter. Through this process, the area that the target is about to enter can be determined in advance before the target crosses the zone. After determining the target space detection partition, the corresponding identification and parsing strategy package is loaded before the target UAV enters the detection partition. Based on the partition identifier of the target space detection partition, the identification and parsing strategy package corresponding to the partition can be retrieved from the strategy storage area and set to a pending state or a currently valid state. This allows the identification and parsing module to directly perform candidate signal screening, protocol identification, message parsing, field verification, and abnormal message discrimination according to the identification and parsing strategy package after the target UAV enters the target space detection partition. For example, in an airport perimeter monitoring scenario, when the position and heading information of a target UAV at continuous intervals indicates that the target is moving from the runway extension direction towards the area surrounding the terminal, it can be predicted that the target will soon enter the corresponding spatial detection zone around the terminal based on its current position, speed, and heading. Before the target enters, the identification and analysis strategy package corresponding to that spatial detection zone is preloaded. Since the background wireless environment around the terminal is usually quite complex, after preloading the corresponding strategy package, the identification and analysis module can directly call the rules adapted to that area for processing when the target enters the area. Thus, the pre-switching module realizes a continuous process from target motion state acquisition and motion trend prediction to strategy preloading, making the switching of identification and analysis strategies coordinated with the cross-area movement process of the target UAV.

[0034] The identification and parsing module performs protocol identification, message parsing, field verification, and anomaly detection on candidate signals based on the currently loaded identification and parsing strategy package. Furthermore, the identification and parsing module includes the following steps: The received signal is filtered for candidate signals according to the currently loaded identification and parsing strategy package to obtain candidate signals; The candidate signals are identified according to the protocol matching rules in the identification and parsing strategy package, and the identified broadcast remote identification signals are parsed. Based on the field validation rules and anomaly detection rules in the identification and parsing strategy package, the parsing results are validated and anomaly messages are identified.

[0035] Specifically, under the action of the currently loaded identification and parsing strategy package, the identification and parsing module performs candidate signal filtering, protocol identification, message parsing, field verification and abnormal message discrimination on the received signal, and outputs the corresponding identification and parsing results. It is the execution link of the online processing of broadcast remote identification signals. Its processing basis is no longer a unified fixed rule, but the identification and parsing strategy package corresponding to the current spatial detection partition, so that the identification and parsing process matches the environmental status of the target area. Based on the currently loaded identification and parsing strategy package, candidate signals are filtered for the received signal to obtain candidate signals. The candidate signal filtering process can judge the frequency band characteristics, time domain characteristics, received strength, signal-to-noise ratio, field integrity rate and multi-antenna reception consistency of the received signal according to the candidate signal filtering rules in the identification and parsing strategy package. From this, signal segments that meet the subsequent identification conditions are selected as candidate signals. The target-related signals in the original received signal can be initially distinguished from the background signals that obviously do not meet the identification conditions. After obtaining candidate signals, protocol identification is performed on the candidate signals according to the protocol matching rules in the identification and parsing strategy package, and message parsing is performed on the identified broadcast remote identification signals. Protocol identification is used to determine whether the candidate signals conform to the protocol characteristics corresponding to broadcast remote identification signals, and message parsing is used to extract the field content carried in the signal, including UAV identification, location information, altitude information, speed information, and pilot position information, thereby transforming the candidate signals obtained in the previous step into structured parsing results.

[0036] After obtaining the parsing results, the parsing results are validated and abnormal messages are identified according to the field validation rules and anomaly discrimination rules in the parsing strategy package. Field validation is used to determine whether each field in the parsing results meets the preset format requirements, consistency requirements and correlation requirements. Abnormal message discrimination is used to identify situations such as missing fields, field conflicts, timing anomalies, forgery injection, or the mixed occurrence of legitimate messages and abnormal messages. The validity and abnormality of the parsing results can be checked before outputting the parsing results.

[0037] For example, in an airport perimeter monitoring scenario, when a target UAV enters the spatial detection zone corresponding to the terminal building, the identification and parsing module calls the identification and parsing strategy package corresponding to that spatial detection zone. First, it filters out candidate signals that meet the conditions from the received signals, then performs protocol identification and message parsing on the candidate signals to obtain the identification information and location information of the target UAV. Finally, it combines field verification rules and anomaly discrimination rules to determine whether the message has field anomalies or suspected forgery, and outputs the corresponding identification and parsing results. This realizes a continuous processing process from candidate signal screening to protocol identification, message parsing, field verification, and anomaly message discrimination.

[0038] The update module updates the environmental fingerprint entries and the identification and analysis strategy package based on the identification and analysis results.

[0039] Furthermore, the update module includes the following steps: Obtain the identification and parsing results, abnormal message discrimination results, and parsing failure samples; The identification and parsing results, abnormal message discrimination results, and parsing failure samples are associated with the corresponding spatial detection partitions; The environmental fingerprint entries corresponding to the spatial detection partition are updated based on the association results. The update adopts a weighted fusion method, which combines the historical environmental fingerprint entries before the current spatial detection partition is updated with the current update information formed by the current identification and parsing results, abnormal message discrimination results and parsing failure samples. Among them, the weight coefficient corresponding to the historical environment fingerprint entries is used to control the degree of retention of historical basic features, and the weight coefficient corresponding to the current update information is used to control the degree of introduction of new data; Based on the updated environmental fingerprint entries, the identification and parsing rule set and the identification and parsing strategy package are modified.

[0040] Specifically, the update module corrects the environmental status representation of the corresponding spatial detection partition based on the identification and parsing results, abnormal message discrimination results and parsing failure samples generated during the online identification and parsing process, and adjusts the identification and parsing strategy package on this basis, so that the system can dynamically update the processing strategy of different spatial detection partitions as the monitoring data continues to accumulate. In practice, the identification and parsing results, abnormal message discrimination results, and parsing failure samples are first obtained. The identification and parsing results are used to characterize the parsing status of the broadcast remote identification signal in the current monitoring period. The abnormal message discrimination results are used to characterize whether there are field abnormalities, timing abnormalities, or suspected counterfeit messages in the current monitoring period. The parsing failure samples are used to characterize signal segments that have not completed effective parsing but have candidate signal characteristics. By collecting the data, an updated input reflecting the current monitoring status can be formed. After obtaining the update input, the identification and parsing results, abnormal message discrimination results and parsing failure samples are associated with the corresponding spatial detection partitions. Based on the partition identifier, spatial location or policy call record corresponding to the target signal, the above data can be assigned to the corresponding spatial detection partition, so that subsequent update processing is always carried out on a specific spatial detection partition basis, rather than mixed updates in the entire monitoring airspace. After the association is completed, the environmental fingerprint entries corresponding to the spatial detection partition are updated according to the association results. In this embodiment, the first... The spatial detection partition is in the first The updated environment fingerprint entry can be represented as follows: ; in, Indicates the first The first spatial detection zone Environment fingerprint entries prior to the last update Indicates the first The first spatial detection zone The updated environmental fingerprint entries This represents the update information formed by the current identification and parsing results, the abnormal message discrimination results, and the parsing failure samples. This indicates updating the weight coefficients, and This update method allows for the introduction of new data from the current monitoring period while preserving the basic characteristics of historical environmental fingerprint entries. After the environmental fingerprint entries are updated, the identification and parsing rule set and identification and parsing strategy package are modified based on the updated environmental fingerprint entries. In this embodiment, the updated identification and parsing strategy package can be represented as: ; in, Indicates the first The updated identification and parsing strategy package for each spatial detection partition. This indicates the updated environmental fingerprint entry. This represents the function that generates the rule set. This indicates the strategy package generation function. This represents the policy parameters. Through this process, the updated environmental fingerprint entries can be remapped to the corresponding set of identification and parsing rules, and further a revised identification and parsing policy package can be generated for subsequent online identification and parsing calls. For example, in the airport perimeter monitoring scenario, if the spatial detection zone around the terminal building shows a continuous decrease in field integrity, an increase in the number of abnormal messages, and failure to parse some candidate signals during recent monitoring, the corresponding identification and parsing results, abnormal message discrimination results, and parsing failure samples can be associated with the spatial detection zone, and the environmental fingerprint entries of the zone can be updated accordingly. Then, a new identification and parsing strategy package is generated so that the corrected processing rules are used when performing identification and parsing in the spatial detection zone later. Through the above processing, the update module realizes the updating of environmental fingerprint entries and the correction of identification and parsing strategy packages at the spatial detection partition level, enabling the system to dynamically adjust the identification and parsing strategy corresponding to each spatial detection partition based on the continuously accumulated monitoring results.

[0041] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A signal recognition and analysis system for unmanned aerial vehicle (UAV) monitoring equipment, characterized in that, include: The signal acquisition module collects broadcast-style remote identification signals and detection quality indicators within the monitored airspace; The environmental fingerprint modeling module establishes corresponding environmental fingerprint entries according to spatial detection zones; The strategy generation module generates an identification and parsing strategy package based on the environmental fingerprint entries; The adversarial training module generates physical layer interference scenarios and protocol layer attack scenarios through an interference generator, performs adversarial training through an identification and parsing module, obtains policy parameters, and writes them into the identification and parsing policy package. The pre-switching module loads the corresponding identification and parsing strategy package before the predicted target drone enters the target space detection zone; The identification and parsing module performs protocol identification, message parsing, field verification, and anomaly detection on candidate signals based on the currently loaded identification and parsing strategy package. The update module updates the environmental fingerprint entries and the identification and analysis strategy package based on the identification and analysis results.

2. The signal recognition and analysis system for UAV monitoring equipment according to claim 1, characterized in that, The signal acquisition module includes the following steps: The system receives wireless signals in the 2.4 GHz and 5.8 GHz bands within the monitored airspace to obtain the raw received signals. The original received signal is sampled synchronously by multiple antennas to obtain the corresponding multi-channel received data; Based on the multi-channel received data, the broadcast remote identification signal and the corresponding detection quality indicators are extracted. The detection quality indicators include received signal strength, signal-to-noise ratio, field integrity rate, and multi-antenna reception consistency.

3. The signal recognition and analysis system for unmanned aerial vehicle (UAV) monitoring equipment according to claim 1, characterized in that, The environmental fingerprint modeling module includes the following steps: The monitored airspace is divided into multiple spatial detection zones, and each spatial detection zone is assigned a corresponding zone identifier. Based on the partition identifier and the preset statistical time window, the background wireless environment features, historical broadcast remote identification signal analysis results and abnormal message features collected in each spatial detection partition are spatiotemporally aligned, and the partition feature association matrix of the corresponding spatial detection partition is constructed according to the feature category and frequency of occurrence. The partition feature correlation matrix is ​​fused to generate environmental fingerprint entries for the corresponding spatial detection partition.

4. The signal recognition and analysis system for UAV monitoring equipment according to claim 1, characterized in that, The strategy generation module includes the following steps: Read the environmental fingerprint entries for the corresponding spatial detection partition; A set of identification and parsing rules is determined based on the environmental fingerprint entries. The set of identification and parsing rules includes candidate signal filtering rules, protocol matching rules, multi-frame splicing rules, field verification rules, and anomaly detection rules. Based on the set of recognition and parsing rules, a recognition and parsing strategy package for the corresponding spatial detection partition is generated.

5. The signal identification and analysis system for unmanned aerial vehicle (UAV) monitoring equipment according to claim 1, characterized in that, The adversarial training module includes the following steps: Construct a generative adversarial network, which includes a noise generator and a recognition parser; The interference generator generates physical layer interference scenarios and protocol layer attack scenarios to form adversarial training samples; The recognition parser performs recognition parsing training on the adversarial training samples to obtain policy parameters; The strategy parameters are written into the identification and parsing rule set and the identification and parsing strategy package.

6. The signal recognition and analysis system for UAV monitoring equipment according to claim 5, characterized in that, The process of generating physical layer interference scenarios and protocol layer attack scenarios to form adversarial training samples includes the following steps: Based on historically collected broadcast remote identification signal samples, background wireless environment data, and abnormal message samples, a basic dataset for adversarial training is constructed. The interference generator perturbs and generates physical layer interference scenarios and protocol layer attack scenarios by perturbing the basic adversarial training dataset. The physical layer interference scenarios include weak signal superposition scenarios, co-frequency interference scenarios, and multipath reflection scenarios; the protocol layer attack scenarios include protocol field forgery scenarios, message timing disturbance scenarios, and mixed injection scenarios of legitimate messages and forged messages. The physical layer interference scenario and the protocol layer attack scenario are combined to form the adversarial training sample.

7. The signal recognition and analysis system for unmanned aerial vehicle (UAV) monitoring equipment according to claim 5, characterized in that, The strategy parameters are written into the identification and parsing rule set and the identification and parsing strategy package, and are invoked by the identification and parsing module, including the following steps: The strategy parameters are respectively written into the candidate signal filtering rules, protocol matching rules, multi-frame splicing rules, field verification rules and anomaly discrimination rules in the recognition and parsing rule set; Based on the set of recognition and parsing rules written with the policy parameters, a corresponding recognition and parsing policy package is generated. The identification and parsing module calls the identification and parsing strategy package to perform protocol identification, message parsing, field verification, and abnormal message discrimination on the candidate signals.

8. The signal identification and analysis system for unmanned aerial vehicle (UAV) monitoring equipment according to claim 1, characterized in that, The pre-switching module includes the following steps: Obtain the target UAV's position, speed, and heading information obtained from continuous time-series analysis; The movement trend of the target UAV is determined based on the target UAV's position, speed, and heading information, and the target space detection zone that the target UAV will enter is predicted. Before the target UAV enters the target space detection zone, the identification and parsing strategy package corresponding to the target space detection zone is loaded.

9. The signal identification and analysis system for unmanned aerial vehicle (UAV) monitoring equipment according to claim 1, characterized in that, The identification and parsing module includes the following steps: The received signal is filtered for candidate signals according to the currently loaded identification and parsing strategy package to obtain candidate signals; The candidate signals are identified according to the protocol matching rules in the identification and parsing strategy package, and the identified broadcast remote identification signals are parsed. Based on the field validation rules and anomaly detection rules in the identification and parsing strategy package, the parsing results are validated and anomaly messages are detected.

10. The signal recognition and analysis system for unmanned aerial vehicle (UAV) monitoring equipment according to claim 1, characterized in that, The update module includes the following steps: Obtain the identification and parsing results, abnormal message discrimination results, and parsing failure samples; The identification and parsing results, abnormal message discrimination results, and parsing failure samples are associated with the corresponding spatial detection partitions; The environmental fingerprint entries corresponding to the spatial detection partition are updated based on the association results. The update adopts a weighted fusion method, which combines the historical environmental fingerprint entries before the current spatial detection partition is updated with the current update information formed by the current identification and parsing results, abnormal message discrimination results and parsing failure samples. Among them, the weight coefficient corresponding to the historical environment fingerprint entries is used to control the degree of retention of historical basic features, and the weight coefficient corresponding to the current update information is used to control the degree of introduction of new data; Based on the updated environmental fingerprint entries, the identification and parsing rule set and the identification and parsing strategy package are modified.

Images