Drone detection countermeasure method, system, device and medium

Abstract

The application discloses a kind of unmanned aerial vehicle detection countermeasure method, system, equipment and medium, unmanned aerial vehicle detection countermeasure method includes: real-time acquisition task instruction and the sensing signal of multiple sensing devices in monitoring area, task instruction and sensing signal are preprocessed, and standardization modal data are obtained;Task instruction and the standardization modal data corresponding to each sensing signal are extracted and high-level semantic extraction in parallel to feature extraction respectively, and the high-dimensional feature vector of each sensing signal and task context parameter are obtained;High-dimensional feature vector is fused based on task context parameter using spatiotemporal alignment multimodal fusion algorithm, and the multimodal fusion feature is obtained;Using neural network model, future time period is predicted to generate the flight trajectory of unmanned aerial vehicle according to historical trajectory data and multimodal fusion feature;Based on task context parameter and flight trajectory, threat level determination and behavior anomaly detection are carried out, alarm information is generated, and security countermeasure strategy based on alarm information is executed.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent perception and processing, and more particularly to a UAV detection countermeasure method, system, device and medium. BACKGROUND

[0002] With the rapid development of UAV technology, UAVs are increasingly widely used in civilian, commercial and military fields. However, since UAVs can fly at low altitudes and are small in size, flexible and highly maneuverable, if misused (such as illegal intrusion, privacy invasion or security threats, etc.), it will pose a certain security risk. Therefore, the counter-UAV system (C-UAS) has become a research hotspot in recent years, aiming to realize the detection, tracking, identification and interference of UAVs. However, the existing counter-UAV system mainly relies on a single sensor (such as radar or radio frequency detection) for detection, and has limited processing capability, and there are certain limitations in the detection capability of small UAVs, which is prone to false negatives or false positives, and it is difficult to cope with high dynamic UAV threats in complex environments. SUMMARY

[0003] The present application provides a UAV detection countermeasure method, system, device and medium to solve the technical problem that the existing counter-UAV system is prone to false negatives or false positives, and is difficult to cope with high dynamic UAV threats in complex environments.

[0004] In a first aspect, a UAV detection countermeasure method is provided, comprising:

[0005] Real-time acquisition of task instructions and sensing signals of multiple different sensing devices in a monitoring area, and preprocessing of the task instructions and the sensing signals to obtain standardized modal data corresponding to the task instructions and each sensing signal;

[0006] Parallel feature extraction and high-level semantic extraction are performed on the standardized modal data corresponding to the task instructions and each sensing signal, respectively, to obtain task context parameters and high-dimensional feature vectors corresponding to each sensing signal;

[0007] Using a spatio-temporal alignment multi-modal fusion algorithm, a confidence score strategy is dynamically adjusted based on the task context parameters, and the high-dimensional feature vectors corresponding to each sensing signal are fused to obtain multi-modal fusion features containing UAV state;

[0008] Using a pre-trained neural network model, the flight trajectory of the UAV in a future preset time period is predicted and generated according to the historical trajectory data of the UAV and the multi-modal fusion features;

[0009] Based on the task context parameters and the flight trajectory, threat level determination and behavior anomaly detection are performed to generate an alarm information, and a security countermeasure strategy is generated based on the alarm information and executed.

[0010] In a second aspect, a UAV detection countermeasure system is provided, comprising:

[0011] a data preprocessing unit configured to acquire task instructions and sensing signals of a plurality of different sensing devices in a monitoring area in real time, and preprocess the task instructions and the sensing signals to obtain standardized modal data corresponding to the task instructions and each sensing signal;

[0012] a feature extraction unit configured to perform feature extraction and high-level semantic extraction on the standardized modal data corresponding to the task instructions and each sensing signal in parallel to obtain a task context parameter and a high-dimensional feature vector corresponding to each sensing signal;

[0013] a feature fusion unit configured to dynamically adjust a confidence score strategy based on the task context parameter using a spatiotemporal alignment multi-modal fusion algorithm, and fuse the high-dimensional feature vector corresponding to each sensing signal to obtain multi-modal fusion features containing a UAV state;

[0014] a trajectory prediction unit configured to predict a flight trajectory of a UAV in a future preset time period based on historical trajectory data of the UAV and the multi-modal fusion features using a pre-trained neural network model;

[0015] a detection countermeasure unit configured to perform threat level determination and behavior anomaly detection based on the task context parameter and the flight trajectory, generate an alarm information, and generate a security countermeasure strategy based on the alarm information and execute the security countermeasure strategy.

[0016] The application also provides a UAV detection countermeasure system, comprising a server, a plurality of different types of sensing devices arranged in a monitoring area, a plurality of different types of interference devices arranged in the monitoring area, and a terminal, wherein

[0017] a plurality of sensing devices configured to collect information of a UAV invading the monitoring area from different dimensions in real time to generate sensing signals;

[0018] The server is configured to acquire task instructions and sensing signals of a plurality of sensing devices in a monitoring area in real time, pre-process the task instructions and the sensing signals to obtain standardized modal data corresponding to the task instructions and each sensing signal, perform feature extraction and high-level semantic extraction on the standardized modal data corresponding to the task instructions and each sensing signal in parallel to obtain a task context parameter and a high-dimensional feature vector corresponding to each sensing signal, perform fusion on the high-dimensional feature vector corresponding to each sensing signal by using a spatiotemporal alignment multi-modal fusion algorithm and dynamically adjusting a confidence score strategy based on the task context parameter to obtain multi-modal fusion features containing a state of the unmanned aerial vehicle, and predict a flight trajectory of the unmanned aerial vehicle in a preset time period in the future by using a pre-trained neural network model according to historical trajectory data of the unmanned aerial vehicle and the multi-modal fusion features. Threat level determination and behavior anomaly detection are performed based on the task context parameter and the flight trajectory, an alarm information is generated, and a security countermeasure strategy is generated based on the alarm information.

[0019] The terminal is configured to send the task instructions and display the alarm information.

[0020] A plurality of interference devices are connected to the server through an MQTT or HTTP interface, and are configured to execute the security countermeasure strategy pushed by the server.

[0021] In a third aspect, a computer device is provided, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the steps of the unmanned aerial vehicle detection and countermeasure method when executing the computer program.

[0022] In a fourth aspect, a computer readable storage medium is provided, which stores a computer program, and the computer program implements the steps of the unmanned aerial vehicle detection and countermeasure method when executed by a processor.

[0023] Compared with the prior art, the unmanned aerial vehicle detection countermeasure method provided by the application can obtain standardized modal data by uniformly preprocessing the real-time task instructions and the sensing signals of multiple different sensing devices in the monitoring area, can obtain task context parameters and high-dimensional feature vectors corresponding to each sensing signal by performing feature extraction and high-level semantic extraction on the standardized modal data of different modalities in parallel, and can obtain multi-modal fusion features containing the state of the unmanned aerial vehicle by using a spatiotemporal alignment multi-modal fusion algorithm, dynamically adjusting a confidence score strategy based on the task context parameters, and fusing the high-dimensional feature vectors corresponding to each sensing signal. Then, the flight trajectory of the unmanned aerial vehicle in a future preset time period is predicted by using a pre-trained neural network model according to the historical trajectory data of the unmanned aerial vehicle and the multi-modal fusion features. Finally, threat level determination and behavior anomaly detection are performed based on the task context parameters and the flight trajectory, an alarm information is generated, a safety countermeasure strategy is generated based on the alarm information and is executed. It can be seen that the unmanned aerial vehicle detection countermeasure method can detect intruding unmanned aerial vehicles according to multi-source data, can improve the adaptability of detection in complex environments, can avoid false negatives or false positives, and can improve the data transmission efficiency by performing parallel feature extraction of multi-source data. In the detection process, the task instructions are used as control parameters for unmanned aerial vehicle detection to guide feature fusion and decision-making of different sensing signals, forming a dynamic, intelligent and task-oriented perception-decision system, which can improve the accuracy and robustness of unmanned aerial vehicle detection in complex environments. BRIEF DESCRIPTION OF DRAWINGS

[0024] Figure 1 A flowchart of the unmanned aerial vehicle detection countermeasure method provided by the first embodiment of the application is shown in the figure.

[0025] Figure 2 A flowchart of a specific implementation of step S130 in the first embodiment is shown in the figure. Figure 1

[0026] Figure 3 A flowchart of a specific implementation of step S140 in the first embodiment is shown in the figure. Figure 1

[0027] Figure 4 A flowchart of a specific implementation of step S160 in the first embodiment is shown in the figure. Figure 1

[0028] Figure 5 A flowchart of the unmanned aerial vehicle detection countermeasure method provided by the second embodiment of the application is shown in the figure.

[0029] Figure 6 A flowchart of the unmanned aerial vehicle detection countermeasure method provided by the third embodiment of the application is shown in the figure.

[0030] Figure 7 A structural diagram of the unmanned aerial vehicle detection countermeasure system provided by the first embodiment of the application is shown in the figure.​​​

[0031] Figure 8 is a structural schematic diagram of the unmanned aerial vehicle detection countermeasure system provided by the second embodiment of the present application;

[0032] Figure 9 is Figure 8 is a structural schematic diagram of the internal architecture of the server in the unmanned aerial vehicle detection countermeasure system shown in

[0033] Figure 10 is Figure 8 is an interface framework schematic diagram of the database in the server in the unmanned aerial vehicle detection countermeasure system shown in

[0034] Figure 11 is a structural schematic diagram of the computer device in the embodiment provided by the present application. DETAILED DESCRIPTION

[0035] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are some of the embodiments of the present application, but not all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the scope of protection of the present application.

[0036] It should also be understood that the term "and / or" used in the specification and the appended claims of the present application means any combination of one or more of the associated listed items and all possible combinations, and includes these combinations.

[0037] Please refer to Figure 1 , Figure 1 is a schematic flowchart of the unmanned aerial vehicle detection countermeasure method provided by the first embodiment of the present application. In the embodiment shown in the drawing, the unmanned aerial vehicle detection countermeasure method comprises the following steps S110-160:

[0038] S110, real-time acquisition of task instructions and sensing signals of a plurality of different sensing devices in a monitoring area, and preprocessing of the task instructions and the sensing signals to obtain standardized modal data corresponding to the task instructions and each sensing signal.

[0039] In the present application, the sensing device can be radar, radio transmitter and receiver, optical detection device, camera, position sensor and climate sensor, etc., the sensing signal can be electromagnetic signal (radar signal, TDOA signal, unmanned aerial vehicle communication signal), sensing signal (image, video and sound obtained by camera), environment signal (geographic coordinates collected by position sensor, wind speed and rainfall collected by climate sensor), etc., the task instruction can be user input instruction, which can be voice data or text data, if the input is voice data, it can be converted to text before processing, in the present embodiment, the task instruction can be used as the control parameter of the unmanned aerial vehicle detection, which can include the core target and / or rule of the present detection countermeasure task, for example, the task instruction can be "no-fly zone range: 5 kilometers around the airport runway, height 0-1000 meters. Any unmanned aerial vehicle entering the area is the highest threat (level 5) and countermeasures are taken", the unmanned aerial vehicle detection countermeasure method of the present application can dynamically and reasonably determine the threat level of the invading unmanned aerial vehicle according to the core target and rule of the task instruction.

[0040] Specifically, in this step, the acquired task instruction and each sensing signal are first cleaned, converted or standardized, for example, the radar signal can be subjected to fast Fourier transform (FFT) to filter out noise, extract target frequency and amplitude, and converted into structured data; the video stream is down-sampled (from 30 FPS to 15 FPS) to reduce bandwidth demand, and frame extraction is performed to generate metadata for each frame; the environment signal is formatted, the wind speed is converted from m / s to km / h (5 m / s→18 km / h), the validity of the geographic coordinates is checked, and structured data is generated; and for the task instruction, the task instruction JSON can be first parsed, the key fields are extracted, the instruction integrity is verified, and structured data is generated; then the above processed data is converted into a unified format (JSON or Protobuf) to obtain respective standardized modal data for subsequent use.

[0041] S120, using a distributed stream processing framework, distributing the task instruction and the standardized modal data corresponding to the plurality of sensing devices into different data topics according to a preset topic rule.

[0042] In this step, the standardized modal data corresponding to electromagnetic, image, audio and text can be pushed into different data topics (such as predefined electromagnetic topic, sensing topic, situation topic, etc.) of the distributed stream processing framework to realize asynchronous message transmission, for example, the standardized modal data of radar signal and the standardized modal data of video can be pushed into "electromagnetic topic" and "sensing topic" in the distributed stream processing framework respectively, and the standardized modal data corresponding to environment signal and task instruction is pushed into "situation topic".

[0043] In the present application, a distributed stream processing framework (such as Kafka) is introduced, which can be combined with various databases (such as in-memory databases, time series databases, vector databases, etc.) to support real-time and high-concurrency stream data processing. The modules and models of each consumer object can directly pull data from the topics of Kafka, and can be independently expanded and upgraded. The data is stored in the topic for a period of time (for example, 7 days), and when retraining is needed, the data stream at any time point in the past can be played back, which is more beneficial to the data transmission and processing of dynamic and rapidly changing unmanned scenarios.

[0044] S130, respectively, the task instruction and the standardized modal data corresponding to each sensing signal under the data topic of the distributed stream processing framework are subjected to feature extraction and high-level semantic extraction, to obtain a task context parameter and a high-dimensional feature vector corresponding to each sensing signal.

[0045] In this step, the standardized modal data under each data topic can be processed in parallel. Specifically, the standardized modal data of the task instruction is subjected to feature extraction and high-level semantic extraction to obtain a task context parameter, and the standardized modal data corresponding to each sensing signal is subjected to feature extraction and high-level semantic extraction in parallel to obtain a high-dimensional feature vector corresponding to each sensing signal.

[0046] As shown in Figure 2 S130 specifically includes the following steps S131-S132:

[0047] S131, the standardized modal data corresponding to the task instruction and each sensing signal is subjected to feature extraction in parallel to obtain a plurality of corresponding modal features.

[0048] In this step, each standardized modal data is subjected to feature extraction. Specifically, the electromagnetic signal can be subjected to fast Fourier transform (FFT) and wavelet transform to analyze the frequency structure and signal intensity features, to generate a spectrum vector (i.e., the extracted modal feature) of about 128 dimensions. The image and video data in the perception signal can be subjected to target detection (to extract the bounding box information of the unmanned aerial vehicle) and feature extraction by the YOLOv8 algorithm to obtain the modal features of the perception signal. The audio data can be converted into a 40-dimensional spectral time sequence feature (i.e., the modal feature of the audio data) by using the MFCC (Mel-Frequency Cepstral Coefficients) algorithm. The text data (task instruction) can be processed by the LLaMA large language model to generate a 1024-dimensional semantic embedding vector (i.e., the modal feature of the text data).

[0049] Preferably, the modal features of each data or signal extracted by the above feature extraction can be uniformly formatted and then written into the feature topic of Kafka as subsequent fusion input.

[0050] In S132, high-level semantic information extraction is performed on all the modal features to obtain a high-dimensional feature vector corresponding to each sensing signal and a task context parameter.

[0051] In this step, high-level semantic information is extracted from the modal features of electromagnetic signals, images, audio, and text to generate unified features and improve the behavior understanding ability of the UAV.

[0052] For electromagnetic signals, their modal features (128-dimensional spectral vectors) are input into a one-dimensional convolutional neural network (1D-CNN) to extract 64-dimensional semantic features. An attention mechanism is used in combination with the formula to highlight key frequencies (such as remote control signal frequency bands); where, is the i-th dimension of the spectral vector, is the attention parameter, is the weight.

[0053] For image and video data in the perception signal, a ResNet50 model is used to extract high-dimensional image semantic features from their modal features, with a dimension of about 512, and then compressed to 256 dimensions through a Transformer encoder. The attention is focused on the key areas of the UAV (such as rotors, etc.), reducing background interference and enhancing semantic context.

[0054] For audio data, its modal features (40-dimensional spectral temporal features) can be processed through a temporal convolution network to obtain 32-dimensional voiceprint features of the UAV, improving the discrimination.

[0055] For text data, the LLaMA large model can extract feature words (such as "intercept", "DJI", "dangerous", etc.) according to the formula and compress them to 128 dimensions to obtain the task context parameters after high-level semantic extraction; where, is the original embedding, and are task-related parameters.

[0056] Preferably, to measure the contribution of all modalities at the same time, in this embodiment, a consistent loss function is used to guide the training of each model in all modalities. The loss function is:

[0057] ;

[0058] where, is the mapping feature of modality m, is the mapping feature of modality n, for modal distribution divergence, for regularization weight.

[0059] S140, using a spatio-temporal alignment multi-modal fusion algorithm, dynamically adjusting a confidence score strategy based on the task context parameter, fusing the high-dimensional feature vectors corresponding to each sensing signal to obtain multi-modal fusion features containing the state of the UAV.

[0060] Specifically, as shown in Figure 3 S140 includes the following steps S141-S143:

[0061] S141, using a Kalman filter algorithm to perform spatio-temporal alignment processing on all high-dimensional feature vectors.

[0062] In this step, the high-dimensional feature vectors corresponding to each sensing signal after semantic extraction, such as the high-dimensional feature vectors corresponding to electromagnetic signals, perception signals, and environmental signals, can be subjected to time consistency processing, and the high-dimensional feature vectors of the electromagnetic signals, perception signals, and environmental signals after time consistency processing are mapped to the same spatial coordinate system to complete spatio-temporal alignment, thereby ensuring the consistency of information in the time and spatial scales.

[0063] S142, using a Bayesian confidence estimation method to score the confidence of the high-dimensional feature vectors corresponding to each sensing signal after spatio-temporal alignment processing based on the task context parameter.

[0064] In this embodiment, the task context parameter (such as "night", "city") can be used as prior knowledge, and the Bayesian confidence estimation method calculates the posterior probability according to this prior knowledge and the likelihood probability of the sensor data, i.e. the confidence of the sensor providing accurate data in the current environment. Specifically, the confidence of each modal data is quantified using the formula to dynamically adjust the fusion weight to optimize the target state vector (such as position, speed, threat level, etc.); wherein, represents modal data (i.e. high-dimensional feature vectors corresponding to each sensing signal), S represents sensor characteristics such as radar accuracy, and E represents environmental factors such as cities, farmland, etc. The likelihood probability can be modeled using a Gaussian mixture model:

[0065] ;

[0066] It can be understood that the detection accuracy of different task scenarios is different due to the different characteristics of the sensing device, and the weight is dynamically adjusted during fusion. For example, radio detection is good at long-distance signal detection (such as unmanned aerial vehicle remote control signal), and is more suitable for discovering targets in the initial stage of detection through spectrum identification and communication analysis at night or in foggy weather, so the weight will increase in this scene and / or stage. Radar provides high-precision distance and speed information, and is more suitable for positioning and tracking in open areas, so the weight will increase in this scene and / or stage. Audio is usually used to detect unmanned aerial vehicle voiceprints, and video provides target classification, so the weight will usually increase in close-range confirmation and complex environments (such as urban high-rise dense areas).

[0067] S143, filtering out high-dimensional feature vectors with confidence scores lower than a preset threshold from the high-dimensional feature vectors corresponding to each sensing signal, and fusing the remaining high-dimensional feature vectors after filtering out using an extended Kalman filtering algorithm according to the confidence scores, to obtain a multi-modal fusion feature containing the state of the unmanned aerial vehicle.

[0068] In this step, a preset threshold (such as 0.8) is used for filtering to remove noise data. After filtering out high-dimensional feature vectors with confidence scores lower than a preset threshold from the high-dimensional feature vectors corresponding to each sensing signal, the remaining high-dimensional feature vectors corresponding to each sensing signal can be fused using an extended Kalman filtering algorithm. The state of the unmanned aerial vehicle can include key state indicators such as the real-time position, speed, heading angle, and threat level of the unmanned aerial vehicle. The multi-modal fusion feature is a comprehensive description of the current state of the unmanned aerial vehicle, and can also be written into the Kafka trajectory topic for subsequent use.

[0069] S150, predicting and generating the flight trajectory of the unmanned aerial vehicle in a future preset time period using a pre-trained neural network model according to the historical trajectory data of the unmanned aerial vehicle and the multi-modal fusion feature.

[0070] In this embodiment, the neural network model can be an LSTM (Long Short-Term Memory) model. The historical trajectory data stored in the time series database can be called, and the flight trajectory of the unmanned aerial vehicle in a future preset time period can be predicted using a pre-trained LSTM model according to the historical trajectory data and the multi-modal fusion feature.

[0071] S160, threat level determination and behavior anomaly detection based on the task context parameters and the flight trajectory, generating an alarm information, generating a security countermeasure strategy based on the alarm information and executing it.

[0072] In this step, the multi-dimensional analysis (threat level and behavior anomaly detection) and the rapid response mechanism are combined to quickly detect and generate alarm information to prompt the user for early warning, and a security countermeasure strategy can be generated according to the alarm information, and the interference device in the monitoring area is controlled to execute the security countermeasure strategy generated based on the alarm information, thereby ensuring high efficiency in a complex scene.

[0073] As shown in Figure 4 The threat level determination and behavior anomaly detection based on the task context parameters and the flight trajectory in step S160 to generate the alarm information specifically includes the following steps S161-S163:

[0074] S161, the threat degree of the intruding UAV is obtained based on the task context parameters and the flight trajectory by using the Drools rule engine, so as to determine the threat level.

[0075] In the present application, the threat degree can reflect the number, position and flight speed of the intruding UAV.

[0076] In this step, the threat degree of the flight data of the intruding UAV is obtained by using the Drools rule according to the preset rule or the rule contained in the task context parameters, so as to determine the threat level, for example: if the speed of the UAV is greater than 50 meters per second and the flight height is lower than 100 meters, it is determined as a high threat target.

[0077] Specifically, the threat level can be generated based on the Drools rule engine by fuzzy logic, combining real-time flight trajectory parameters (position, speed, heading angle, etc.) and historical behavior patterns according to the formula ; wherein x is a comprehensive threat index (which can be a weighted sum of speed and height), a and b are threat level threshold values, is a membership degree, which is used to quantify the fuzzy degree of "low threat" or "high threat", when the threat level or anomaly score exceeds the threshold value, for example greater than 0.8, the related alarm information is pushed.

[0078] S162, the isolated forest algorithm is used to identify the abnormal flight behavior of the flight trajectory, to obtain abnormal flight behavior data, and a pre-trained classifier is used to identify the pattern intention of the abnormal flight behavior data, to obtain the behavior intention of the UAV.

[0079] In this step, the behavioral patterns can include hovering (the drone repeatedly flies in a circular or elliptical trajectory within a specific area, exhibiting a periodic trajectory pattern), hovering (maintaining relative stillness at a specific location), cruising (the drone flies at a stable speed along a fixed direction or path, with a trajectory that is a uniform straight line or a smooth curve), slamming (the drone moves at high speed towards a specific target or direction, with high speed and strong directionality, such as approaching at high speed or moving away quickly), and maneuvering (performing complex trajectory changes, such as sharp turns, evasion, or formation adjustments, with irregular and highly dynamic trajectories). The hovering mode is usually used for continuous surveillance or standby and is related to reconnaissance intentions; the hovering mode is usually used for reconnaissance, surveillance, or waiting for instructions; while cruising is often used for long-distance movement or area coverage; slamming is usually related to intentions such as surprise attack or escape; and maneuvering is usually related to intentions such as evasion, detection, or coordinated action. During anomaly detection, if the drone is a non-cooperative drone and the drone's behavioral pattern intention is abnormal, such as hovering for reconnaissance or rapid escape, it is identified as abnormal behavior information and an alarm can be triggered.

[0080] S163. If the threat level reaches a preset level, and / or the drone's behavior is abnormal, generate an alarm message.

[0081] In this embodiment, the alarm information includes threat level and abnormal behavior information (i.e., the intention of drone behavior identified as abnormal).

[0082] Understandably, in this invention, the order of steps S161 for determining the threat level and step S162 for identifying abnormal flight behavior can be interchanged.

[0083] Furthermore, in some embodiments, after generating alarm information, the threat level and abnormal behavior information in the alarm information can be written into the corresponding topic of the distributed stream processing framework, such as the threat level topic and behavior analysis topic, respectively, as a reference for countermeasure control strategies. The threat level and abnormal behavior information in the distributed stream processing framework can also be pushed externally via the WebSocket protocol for display in the human-machine interface alarm pop-up to alert the user, and control commands containing security countermeasure strategies can be pushed to external jamming devices via MQTT or HTTP interfaces for countermeasures. This embodiment uses the WebSocket / MQTT protocol for data transmission, and the fast jamming mechanism based on the WebSocket / MQTT protocol can shorten the response time and effectively deal with the threat of high-speed moving drones.

[0084] It can be known from the above scheme that the unmanned aerial vehicle detection countermeasure method in this embodiment can obtain standardized modal data by uniformly preprocessing the real-time task instruction and the sensing signals of multiple different sensing devices in the monitoring area, distribute different standardized modal data to different data topics according to a preset topic rule by using a distributed stream processing framework, then issue the standardized modal data in different data topics to downstream consumer objects to perform feature extraction and high-level semantic extraction in parallel, obtain task context parameters and high-dimensional feature vectors corresponding to each sensing signal, and use a spatiotemporal alignment multi-modal fusion algorithm to dynamically adjust a confidence score strategy based on the task context parameters, fuse the high-dimensional feature vectors corresponding to each sensing signal, and obtain multi-modal fusion features containing the state of the unmanned aerial vehicle. Then, the flight trajectory of the unmanned aerial vehicle in a future preset time period is predicted according to the historical trajectory data of the unmanned aerial vehicle and the multi-modal fusion features by using a pre-trained neural network model. Finally, threat level determination and behavior anomaly detection are performed based on the task context parameters and the flight trajectory, an alarm information is generated, a security countermeasure strategy is generated based on the alarm information and is executed. It can be known that the unmanned aerial vehicle detection countermeasure method in this embodiment can detect intruding unmanned aerial vehicles according to multi-source data, can improve the adaptability of detection in complex environments, can avoid false positives or false negatives, and can improve the data transmission efficiency because the multi-source data feature extraction is performed in parallel. Moreover, the distributed stream processing framework is introduced, the data acquisition and transmission efficiency is greatly improved, which is more conducive to data transmission and processing in dynamic and rapidly changing unmanned aerial vehicle environments. In the detection process, the task instruction is also used as a control parameter for unmanned aerial vehicle detection to guide feature fusion and decision-making of different sensing signals, forming a dynamic, intelligent and taskable perception-decision system, which can improve the accuracy and robustness of unmanned aerial vehicle detection in complex environments.

[0085] It should be noted that the size of the serial number of each step in the above embodiment does not mean the order of execution, and the execution order of each process should be determined according to its function and inherent logic, and should not constitute any limitation on the implementation process of the embodiment of the present application.

[0086] Reference Figure 5 , Figure 5 A flowchart of an unmanned aerial vehicle detection countermeasure method provided by the second embodiment of the present application is shown in the figure. As shown in the figure, the method includes the following steps S210-S270:

[0087] S210, a task instruction and sensing signals of multiple different sensing devices in a monitoring area are acquired in real time, and the task instruction and the sensing signals are preprocessed to obtain standardized modal data corresponding to each sensing signal.

[0088] This step is the same as or similar to step S110, and will not be described here.

[0089] S220, distributing the task instruction and the standardized modal data corresponding to the plurality of sensing devices into different data topics according to preset topic rules by using a distributed stream processing framework.

[0090] This step is the same as or similar to step S120, and will not be described here.

[0091] S230, respectively extracting features and high-level semantics from the task instruction and the standardized modal data corresponding to each sensing signal issued under the data topic in the distributed stream processing framework, to obtain a task context parameter and a high-dimensional feature vector corresponding to each sensing signal.

[0092] This step is the same as or similar to step S130, and will not be described here.

[0093] S240, using a spatiotemporal alignment multi-modal fusion algorithm to dynamically adjust a confidence score strategy based on the task context parameter, and fusing the high-dimensional feature vector corresponding to each sensing signal to obtain a multi-modal fusion feature containing the state of the unmanned aerial vehicle.

[0094] This step is the same as or similar to step S140, and will not be described here.

[0095] S250, using a pre-trained neural network model to predict and generate a flight trajectory of the unmanned aerial vehicle in a future preset time period according to historical trajectory data of the unmanned aerial vehicle and the multi-modal fusion feature.

[0096] This step is the same as or similar to step S150, and will not be described here.

[0097] S260, taking one unmanned aerial vehicle as one node, recording the multi-modal fusion feature and the predicted flight trajectory of the unmanned aerial vehicle in a graph network structure to obtain an unmanned aerial vehicle relationship graph, and analyzing the unmanned aerial vehicle relationship graph by using a graph neural network to infer a cooperation relationship of a plurality of unmanned aerial vehicles and generate a global situation map of the monitoring area.

[0098] In this embodiment, one UAV is taken as one node, the node attribute is multi-modal fusion feature and predicted flight trajectory, and a relationship graph of multiple UAVs in a monitoring area is constructed. The connection and interaction between the nodes of the UAV relationship graph can be analyzed by a graph neural network (GNN), and the cooperation score or relationship strength (for example, through an attention mechanism weight) between the UAVs is calculated. A high score value of the relationship edge means that there may be formation, following, communication and other cooperative behaviors between them. A three-dimensional space coordinate is established, the position coordinates and cooperative relationship of each UAV are mapped into the three-dimensional space coordinate, and a global situation map is generated. Whether multiple UAVs exist cooperative reconnaissance (the behavior of multiple UAVs flying in dispersion to cover a large area), formation attack (the behavior of multiple UAVs maintaining a close array and moving at high speed to the same target) or dispersion interference behavior can be more directly known through the global situation map, and the method is suitable for the detection of multiple targets in complex application scenarios (such as cluster cooperation, potential cooperative intrusion and the like).

[0099] In S270, threat level determination and behavior anomaly detection are performed based on the task context parameters and flight trajectory, alarm information is generated according to the threat level, abnormal behavior information and global situation map, and a security countermeasure strategy is generated and executed based on the alarm information.

[0100] In this step, the alarm information is generated based on the threat level, abnormal behavior information and global situation map, which not only can detect and identify the intention of a single UAV, but also can identify the relationship and cluster behavior between multiple UAVs, so as to provide comprehensive and clear situation awareness for decision makers, and facilitate the rapid and accurate formulation of a security countermeasure strategy (such as interference, evidence collection, physical attack and the like).

[0101] Further, in some embodiments, the UAV detection and countermeasure method can further include: adaptively adjusting the weight of each feature in the multi-modal fusion feature based on the historical trajectory data corresponding to the scene of the UAV, that is, when the high-dimensional feature vectors corresponding to each sensing signal are fused, the fusion of the high-dimensional feature vectors corresponding to each sensing signal can be dynamically adjusted according to the task context parameters, and the feature weight can be adaptively adjusted according to the scene characteristics of the historical data to enhance the adaptability to complex environments. Specifically, the weight of each modality (radar, radio, audio and video) can be adaptively adjusted according to the scene characteristics (for example, open area or urban dense area) of the historical data, for example, the weight of audio and video is increased in urban areas, and the specific implementation formula is as follows:

[0102] ;

[0103] wherein, is the weight of modality m, is the modality credibility score, The scene correlation function is obtained based on rules or a machine learning model.

[0104] It can be seen that, in the above scheme, the unmanned aerial vehicle detection countermeasure method of the embodiment can improve the adaptability of detection in a complex environment, and the multi-source data feature extraction is performed in parallel, which can improve the data transmission efficiency, and the distributed stream processing framework is introduced, the data acquisition and transmission efficiency is greatly improved, which is more conducive to the data transmission and processing of dynamic and rapidly changing unmanned aerial vehicle scenes, and the task instruction is used as the control parameter of the unmanned aerial vehicle detection in the detection process, which guides the feature fusion and decision of different sensing signals, forms a dynamic, intelligent and taskable perception-decision system, and can improve the accuracy and robustness of unmanned aerial vehicle detection in a complex environment; and the global situation map is generated by establishing the unmanned aerial vehicle relationship graph for cooperation relationship analysis, the relationship and cluster behavior among multiple unmanned aerial vehicles are identified, a comprehensive and clear situation awareness is provided for decision makers, and a safe countermeasure strategy can be quickly and accurately formulated.

[0105] Figure 6 The flowchart of the unmanned aerial vehicle detection countermeasure method provided by the third embodiment of the present application is shown in Figure 6 The unmanned aerial vehicle detection countermeasure method provided by the embodiment adds the following steps S310-S320 after step S160 of the first embodiment or step S270 of the second embodiment.

[0106] S310, reconstructing a trajectory containing a behavior mode sequence of the unmanned aerial vehicle by using a HMM model based on historical trajectory data of the unmanned aerial vehicle.

[0107] In this step, the historical trajectory data of the unmanned aerial vehicle is analyzed in depth by using a HMM (Hidden Markov Model), and the behavior mode (for example: hovering, cruising, circling, impacting, maneuvering, etc.) of the unmanned aerial vehicle at different time points is inferred and excavated, and then a reconstructed trajectory that can better reflect the intention of the unmanned aerial vehicle is generated by modeling the behavior mode sequence (such as "hovering -> cruising -> circling") composed of the behavior modes at different time points; for example, the "circling" behavior may be related to the reconnaissance intention, and the reconstructed trajectory can highlight its periodic characteristics.

[0108] Specifically, in the embodiment, the state transition probability and the observation probability of the HMM are defined as follows:

[0109] ;

[0110] wherein, is the hidden state at time t, i.e. the behavior mode, is the state transition probability matrix, is the observation probability, To observe data (e.g. position, velocity). Understandably, the HMM model used in this step has been pre-trained using a large amount of historical trajectory data to learn the state transition matrix A and the observation probability B; for example, the "cruise" state may correspond to the observation data of the uniform straight trajectory, and the "maneuver" state corresponds to the high dynamic trajectory change.

[0111] S320, incrementally train the pre-trained neural network model according to the historical data corresponding to the historical trajectory data of the unmanned aerial vehicle and the reconstructed trajectory.

[0112] In this step, the reconstructed trajectory containing the mined behavior pattern sequence is used as an additional input to the LSTM model, and the LSTM model is incrementally trained with historical data, so that the model can dynamically update the historical pattern library according to new data, while retaining old data, reducing the computational overhead of model retraining.

[0113] During incremental training, the model can focus on key behavior patterns by learning, and establish strong correlation between specific behavior patterns and subsequent flight trajectories to enhance the accuracy of future trajectory prediction; for example, when new data indicates that "high-speed evacuation" often occurs after "circling", the LSTM model will adjust the prediction probability to enhance the accuracy of future trajectory prediction; wherein the historical data includes historical behavior pattern labels, multi-modal fusion features, historical predicted flight data, and historical future real flight data.

[0114] As can be seen, the unmanned aerial vehicle detection countermeasure method of the embodiment can also perform trajectory modeling according to historical trajectory data of unmanned aerial vehicle detection, obtain a reconstructed trajectory, and incrementally train an LSTM model with the reconstructed trajectory corresponding to the historical trajectory data and historical data to optimize the model and enhance the accuracy of future trajectory prediction. Understandably, the historical data includes data composed of behavior pattern labels, multi-modal fusion features, predicted flight data, and real flight data obtained by each unmanned aerial vehicle detection, and the model can be optimized by retrieving the data when the above data stored after multiple detections reaches a preset data amount, or the model can be optimized according to the above obtained data after each detection.

[0115] Reference Figure 7 , Figure 7 is a structural schematic diagram of the unmanned aerial vehicle detection countermeasure system in the first embodiment of the application. In the embodiment shown in the drawing, the unmanned aerial vehicle detection countermeasure system includes a data preprocessing unit 101, a data communication unit 102, a feature extraction unit 103, a feature fusion unit 104, a trajectory prediction unit 105, a detection countermeasure unit 106, and an incremental training unit 107. The functions of each unit are described in detail as follows:

[0116] The data preprocessing unit 101 is configured to acquire task instructions and sensing signals of a plurality of different sensing devices in a monitoring area in real time, and preprocess the task instructions and the sensing signals to obtain standardized modal data corresponding to the task instructions and the sensing signals respectively.

[0117] The data communication unit 102 is configured to distribute the task instructions and the standardized modal data corresponding to the plurality of sensing devices into different data topics according to preset topic rules by using a distributed stream processing framework.

[0118] The feature extraction unit 103 is configured to perform feature extraction and high-level semantic extraction on the task instructions and the standardized modal data corresponding to the sensing signals respectively, which are issued by the data topics in the distributed stream processing framework, to obtain a task context parameter and a high-dimensional feature vector corresponding to each sensing signal.

[0119] The feature fusion unit 104 is configured to perform fusion on the high-dimensional feature vectors corresponding to the sensing signals by using a spatiotemporal alignment multi-modal fusion algorithm, to obtain multi-modal fusion features containing a state of the unmanned aerial vehicle, based on dynamic adjustment of a confidence score strategy according to the task context parameter.

[0120] The trajectory prediction unit 105 is configured to predict a flight trajectory of the unmanned aerial vehicle in a future preset time period according to historical trajectory data of the unmanned aerial vehicle and the multi-modal fusion features by using a pre-trained neural network model.

[0121] The detection countermeasure unit 106 is configured to perform threat level determination and behavior anomaly detection based on the task context parameter and the flight trajectory, to generate alarm information, and to generate a security countermeasure strategy and perform the security countermeasure strategy based on the alarm information.

[0122] The incremental training unit 107 is configured to perform trajectory modeling on the historical trajectory data of the unmanned aerial vehicle by using a HMM model to obtain a reconstructed trajectory containing a behavior mode sequence of the unmanned aerial vehicle, and to perform incremental training on the pre-trained neural network model by using an incremental learning mechanism according to historical data corresponding to the historical trajectory data of the unmanned aerial vehicle and the reconstructed trajectory.

[0123] In some embodiments, the data communication unit 102 is further configured to:

[0124] write the threat level and the abnormal behavior information in the alarm information into a topic corresponding to the distributed stream processing framework;

[0125] push the threat level and the abnormal behavior information in the distributed stream processing framework to the outside by using a WebSocket protocol for display, to be used for an alarm pop-up window of a human-computer interaction interface.

[0126] In some embodiments, the feature fusion unit 104 is specifically configured to:

[0127] aligning all high-dimensional feature vectors in space-time by using Kalman filtering algorithm;

[0128] scoring the high-dimensional feature vectors corresponding to each of the time-space aligned sensor signals based on the task context parameters by using Bayesian confidence estimation method;

[0129] filtering out high-dimensional feature vectors with confidence scores lower than a preset threshold from the high-dimensional feature vectors corresponding to each of the sensor signals, and fusing the remaining high-dimensional feature vectors after filtering by using extended Kalman filtering algorithm according to the confidence scores, to obtain multi-modal fusion features containing the state of the UAV; wherein the sensor signals include electromagnetic signals, perception signals and environmental signals; and the state of the UAV includes real-time position, speed and heading angle of the UAV.

[0130] In some embodiments, the detection countermeasure unit 106 is specifically configured to:

[0131] obtaining the threat level of the invading UAV based on the task context parameters and the flight trajectory by using a Drools rule engine, so as to determine the threat level; wherein the threat level reflects the number, position and flight speed of the invading UAV;

[0132] performing abnormal flight behavior identification on the flight trajectory by using an Isolation Forest algorithm, to obtain abnormal flight behavior data, and performing pattern intention identification on the abnormal flight behavior data by using a pre-trained classifier, to obtain the behavior intention of the UAV;

[0133] generating an alarm information if the threat level reaches a preset level and / or the behavior intention of the UAV is abnormal.

[0134] In some embodiments, the detection countermeasure unit 106 is further specifically configured to:

[0135] taking one UAV as one node, recording the multi-modal fusion features and the predicted flight trajectory of the UAV in a graph network structure, to obtain a UAV relationship graph, and analyzing the UAV relationship graph by using a graph neural network, to infer the cooperative relationship of multiple UAVs, and generate a global situation map of the monitoring area;

[0136] generating an alarm information according to the threat level, abnormal behavior information and global situation map, generating a security countermeasure strategy based on the alarm information and executing the security countermeasure strategy.

[0137] In some embodiments, the feature fusion unit 104 is further specifically configured to:

[0138] adapting the weights of each feature in the multi-modal fusion features based on the historical trajectory data corresponding to the scene of the UAV.

[0139] The unmanned aerial vehicle detection countermeasure system provided by the application can improve data processing efficiency, enhance adaptability to complex environments, and optimize an LSTM model through behavior reconstruction and verification, thereby improving unmanned aerial vehicle behavior prediction accuracy and supporting dynamic analysis and real-time decision optimization of unmanned aerial vehicle behavior in complex scenarios.

[0140] Reference Figure 8 , Figure 8 The schematic block diagram of the unmanned aerial vehicle detection countermeasure system provided by the second embodiment of the application is shown in FIG. 2. As shown in FIG. 2, the unmanned aerial vehicle detection countermeasure system of the embodiment comprises a server 301, a plurality of different types of sensor devices 302 arranged in a monitoring area, a plurality of different types of interference devices 303 arranged in the monitoring area, and a terminal 304. Figure 8

[0141] The plurality of sensor devices 302 are configured to collect information of an unmanned aerial vehicle invading the monitoring area in real time from different dimensions to generate sensor signals. The sensor devices 302 can include radars, radio transmitters and receivers, cameras, position sensors, climate sensors, etc. The sensor signals can include electromagnetic signals (radar signals, TDOA signals, unmanned aerial vehicle communication signals), perception signals (images, videos, and sounds obtained by cameras), environmental signals (geographic coordinates collected by position sensors, wind speed and rainfall collected by climate sensors), etc.

[0142] The server 301 is configured to acquire a task instruction and sensor signals of a plurality of sensor devices 302 in a monitoring area in real time, and pre-process the task instruction and the sensor signals to obtain standardized modal data corresponding to the task instruction and each sensor signal. The standardized modal data corresponding to the task instruction and each sensor signal are subjected to feature extraction and high-level semantic extraction in parallel to obtain a task context parameter and a high-dimensional feature vector corresponding to each sensor signal. A time-space aligned multi-modal fusion algorithm is used to dynamically adjust a confidence score strategy based on the task context parameter, and the high-dimensional feature vectors corresponding to each sensor signal are fused to obtain multi-modal fusion features containing unmanned aerial vehicle states. A pre-trained neural network model is used to predict and generate a flight trajectory of the unmanned aerial vehicle in a preset time period in the future according to historical trajectory data of the unmanned aerial vehicle and the multi-modal fusion features. Threat level determination and behavior anomaly detection are performed based on the task context parameter and the flight trajectory to generate an alarm information, and a security countermeasure strategy is generated based on the alarm information.

[0143] The terminal 304 is configured to send a task instruction and display alarm information for a decision maker to view.

[0144] ​A plurality of interference devices 303 are connected with the server 301 through an MQTT or HTTP interface, and are used to execute the security countermeasures pushed by the server 301; wherein the interference device 303 can be a radio frequency jammer, a decoy induction module, etc., through the fast interference mechanism of the MQTT protocol, the system response time can be shortened, and the response speed is improved by about 40% compared with the prior art, effectively coping with the threat of high-speed moving unmanned aerial vehicles.

[0145] Specifically, referring to Figure 9 , Figure 9 is a schematic diagram of the internal architecture of the server 301 in the unmanned aerial vehicle detection countermeasure system of the embodiment. In the embodiment, the server 301 mainly includes an input layer, a processing layer and a storage layer. The function flow executed by the above-mentioned server 301 is realized in the processing layer. At the beginning, the input layer can obtain sensing signals (electromagnetic signals, sensing signals, environmental signals) and task instructions through a special hardware interface, TCP or API, and push them into different topics (such as electromagnetic topics, sensing topics, and situation topics) of the distributed stream processing framework of the processing layer after preprocessing. After feature extraction in parallel in the processing layer, feature fusion is performed, multi-modal fusion features are obtained, flight trajectories are predicted through a model, and threat level determination, behavior anomaly detection and global situation map generation are performed. The model can also be incrementally trained through historical backtracking, that is, incrementally trained using historical data.

[0146] In the embodiment, a plurality of databases (in-memory databases, time series databases, vector databases, SQL structured databases, document databases, spatial databases, etc.) are arranged in the server 301, so as to be used in cooperation with the distributed stream processing framework, and hot data (real-time trajectories), warm data (time series data) and cold data (historical trajectories, semantic embeddings) are stored in layers, so as to optimize query efficiency and data management.

[0147] In the embodiment, a unified database semantic module is also designed to facilitate data flow between different databases. The read, write, batch read and batch write interfaces of different databases are unified through injection inversion technology, and the heterogeneity between different databases is shielded.

[0148] When data flows between databases, such as Figure 10As shown, the application sends a data access request (such as reading real-time sensor values, writing drone trajectories), which is passed to the database semantic module through "data request", and the request type can include single read / write and batch read / write operations; the semantic module receives the request and parses it into standardized operation instructions ("semantic analysis" arrow), and through IoC (inversion of control) injection, the semantic module dynamically selects the corresponding database operation interface, shielding the differences between the underlying SQL / non-SQL databases. For example, reading hot data (Redis) or cold data (MySQL) uses the same read interface and does not need to consider the database type; the abstract interface layer converts the instructions of the semantic module into a general data operation protocol for driving call to ensure compatibility with the driver layer, and the general data operation protocol abstracts the details of database operations (such as SQL statements, database commands) to provide a unified entry for the driver layer; the driver layer then calls the corresponding database driver according to the operation type, such as JDBC, database client.

[0149] Because the data input by each sensor device 302 is the latest data, it is stored in the memory data as hot data, supporting high-frequency read / write, including returning real-time sensor values, real-time trajectories of drones, and data such as the state of the drone. In order to reduce the pressure of hot layer data on the memory, part of the data will be stored in the time series database at regular intervals, and data migration can be performed according to the time stamp timeout. Different types of drone data can be retained for different times, and the time series database can be stored on a solid state hard disk type hardware device to accelerate the efficiency of IO. It can also use the time series database itself caching technology to provide 1-2G cache space in the memory, provide query hit rate through column query and data compression, and accelerate query efficiency. For traditional structured data, historical data, and document data, traditional cold layer technology is used for persistent storage. The data in the cold layer includes historical trajectories, logs, storage semantic embedding, HNSW index, geofencing, map information, and task data.

[0150] In the above technical solution, the unmanned aerial vehicle detection countermeasure system provided by the embodiment can improve data processing efficiency, enhance adaptability to complex environments, and optimize the LSTM model through behavior reconstruction and verification to improve the accuracy of unmanned aerial vehicle behavior prediction, support dynamic analysis and real-time decision optimization of unmanned aerial vehicle behavior in complex scenarios, and through the injection inversion technology, unify the read, write, batch read, and batch write interfaces of different databases, support long-time continuous high-frequency read / write, and improve the data access speed by about 50%. The disaster recovery capability of cross-database migration ensures the continuity of system operation, and the fault recovery time is greatly shortened.

[0151] It should be noted that the specific implementation process of the unmanned aerial vehicle detection countermeasure system and each unit or component can be clearly understood by those skilled in the art, and the corresponding description in the foregoing method embodiments can be referred to. For the convenience and brevity of description, it will not be repeated here.

[0152] In one embodiment, a computer device is provided, and an internal structure diagram of the computer device can be as shown in Figure 11 The computer device includes a processor, a memory, a network interface and a database connected through a system bus. The processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile and / or volatile storage medium, an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operating system and the computer program in the non-volatile storage medium to run. The network interface of the computer device is configured to communicate with external devices through network connection. Further, the computer device can also include a display screen and an input device (such as a mouse, a keyboard, etc.) for interaction.

[0153] Specifically, the processor in the computer device implements the steps of the unmanned aerial vehicle detection countermeasure method provided in the first embodiment, the second embodiment and the third embodiment when executing the computer program.

[0154] In one embodiment, the present application can also provide a computer readable storage medium having a computer program stored thereon, and the computer program is executed by the processor to implement the steps of the unmanned aerial vehicle detection countermeasure method provided in the first embodiment, the second embodiment and the third embodiment.

[0155] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer readable storage medium, and when executed, can include the processes of the above-mentioned embodiment methods. Any reference to memory, storage, database or other medium used in the embodiments provided by the present application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration but not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0156] Those skilled in the art can clearly understand that, for the convenience and brevity of description, only the division of the above-mentioned functional units and modules is exemplified, and in actual application, the above-mentioned functions can be completed by different functional units and modules according to needs, that is, the internal structure of the system is divided into different functional units or modules to complete all or part of the above-described functions.

[0157] The above embodiments are only used to illustrate the technical solutions of the present application, but not limit it. Although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that the technical solutions recorded in the foregoing embodiments can be modified, or some technical features can be replaced by equivalents. These modifications or replacements do not change the essence of the corresponding technical solutions, and should be included in the protection scope of the present application.

Claims

1. A method for detecting and countering unmanned aerial vehicles (UAVs), characterized in that, The drone detection and countermeasures method includes: The task instructions and sensing signals from multiple different sensing devices in the monitoring area are acquired in real time, and the task instructions and sensing signals are preprocessed to obtain standardized modal data corresponding to the task instructions and each sensing signal. Feature extraction and high-level semantic extraction are performed in parallel on the standardized modal data corresponding to the task instructions and each sensor signal, respectively, to obtain the task context parameters and the high-dimensional feature vectors corresponding to each sensor signal. Using a spatiotemporally aligned multimodal fusion algorithm, the confidence scoring strategy is dynamically adjusted based on the task context parameters, and the high-dimensional feature vectors corresponding to each sensor signal are fused to obtain multimodal fusion features that include the UAV state. The flight trajectory of the UAV within a preset time period is predicted and generated using a pre-trained neural network model based on the UAV's historical trajectory data and the multimodal fusion features. Threat level determination and abnormal behavior detection are performed based on the mission context parameters and flight trajectory, alarm information is generated, and security countermeasures are generated and executed based on the alarm information; The spatiotemporally aligned multimodal fusion algorithm dynamically adjusts the confidence scoring strategy based on the task context parameters, fuses the high-dimensional feature vectors corresponding to each sensor signal, and obtains multimodal fusion features that include the UAV state, including: The Kalman filter algorithm is used to perform spatiotemporal alignment on all high-dimensional feature vectors. A Bayesian confidence estimation method is used to score the confidence of the high-dimensional feature vectors corresponding to each spatiotemporally aligned sensor signal based on the task context parameters. Specifically, this includes: using the task context parameters as prior knowledge, employing the Bayesian confidence estimation method based on the prior knowledge and the likelihood probability of the sensor signal using the formula... The posterior probability is calculated to obtain the confidence score of the high-dimensional feature vector corresponding to each sensor signal; where, This represents the high-dimensional feature vector corresponding to each sensing signal, where S represents sensor characteristics and E represents environmental factors. Let the likelihood probability be denoted as '('). High-dimensional feature vectors with confidence scores below a preset threshold are filtered out from the high-dimensional feature vectors corresponding to each sensor signal. Then, the extended Kalman filter algorithm is used to fuse the remaining high-dimensional feature vectors after filtering based on the confidence scores to obtain multimodal fusion features that include the UAV state. The sensor signals include electromagnetic signals, perception signals, and environmental signals. The UAV state includes the UAV's real-time position, speed, and heading angle.

2. The UAV detection and countermeasure method as described in claim 1, characterized in that, After obtaining the standardized modal data corresponding to the task instructions and each sensor signal, the process further includes: Using a distributed stream processing framework, the task instructions and standardized modal data corresponding to multiple sensing devices are distributed to different data topics according to preset topic rules.

3. The UAV detection and countermeasure method as described in claim 2, characterized in that, After generating the alarm information, the following is also included: The threat level and abnormal behavior information in the alarm information are written into the topic corresponding to the distributed stream processing framework; The threat level and abnormal behavior information in the distributed stream processing framework are pushed to the outside for display using the WebSocket protocol, so as to be used for alarm pop-ups in the human-computer interaction interface.

4. The UAV detection and countermeasure method as described in claim 1, characterized in that, The threat level determination and abnormal behavior detection based on the task context parameters and flight trajectory, and the generation of alarm information, include: Based on the task context parameters and flight trajectory, the Drools rule engine is used to obtain the threat level of the intruding drones, thereby determining the threat level; wherein, the threat level reflects the number, location, and flight speed of the intruding drones; The isolated forest algorithm is used to identify abnormal flight behavior on the flight trajectory to obtain abnormal flight behavior data. A pre-trained classifier is then used to identify the pattern intent of the abnormal flight behavior data to obtain the UAV's behavioral intent. If the threat level reaches a preset level, and / or the drone's behavior is abnormal, an alarm message will be generated.

5. The UAV detection and countermeasure method as described in claim 1, characterized in that, The process of predicting and generating the drone's flight trajectory within a preset future time period using a pre-trained neural network model based on the drone's historical trajectory data and the multimodal fusion features further includes: Based on each UAV as a node, a graph network structure is used to record the multimodal fusion features and predicted flight trajectories of the UAVs, thus obtaining a UAV relationship graph. By analyzing the UAV relationship graph using a graph neural network, the cooperative relationships of multiple UAVs are inferred, and a global situation map of the monitoring area is generated.

6. The UAV detection and countermeasure method as described in claim 1, characterized in that, The drone detection and countermeasures method also includes: The weights of each feature in the multimodal fusion feature are adaptively adjusted based on the scene corresponding to the historical trajectory data of the drone.

7. The method for detecting and countering unmanned aerial vehicles (UAVs) as described in any one of claims 1-6, characterized in that, The drone detection and countermeasures method also includes: Based on historical trajectory data of UAVs, trajectory modeling is performed using an HMM model to obtain a reconstructed trajectory containing a sequence of UAV behavior patterns. Based on the historical trajectory data of the UAV and the reconstructed trajectory, the pre-trained neural network model is incrementally trained using an incremental learning mechanism; wherein, the historical data includes historical behavior pattern labels, multimodal fusion features, historical predicted flight data, and historical future real flight data.

8. A drone detection and countermeasure system, characterized in that, The drone detection and countermeasure system includes: The data preprocessing unit is used to acquire task instructions and sensing signals from multiple different sensing devices in the monitoring area in real time, and to preprocess the task instructions and the sensing signals to obtain standardized modal data corresponding to the task instructions and each sensing signal. The feature extraction unit is used to perform feature extraction and high-level semantic extraction in parallel on the standardized modal data corresponding to the task instructions and each sensor signal, respectively, to obtain the task context parameters and the high-dimensional feature vectors corresponding to each sensor signal. The feature fusion unit utilizes a spatiotemporally aligned multimodal fusion algorithm to dynamically adjust the confidence scoring strategy based on the task context parameters, fusing the high-dimensional feature vectors corresponding to each sensor signal to obtain multimodal fused features including the UAV state. Specifically, it uses a Kalman filter algorithm to perform spatiotemporal alignment processing on all high-dimensional feature vectors; using the task context parameters as prior knowledge, it employs a Bayesian confidence estimation method based on the prior knowledge and the likelihood probability of the sensor signals using the formula... The confidence score of the high-dimensional feature vector corresponding to each sensing signal is calculated; where, This represents the high-dimensional feature vector corresponding to each sensing signal, where S represents sensor characteristics and E represents environmental factors. The likelihood probability is used to filter out high-dimensional feature vectors with confidence scores below a preset threshold from the high-dimensional feature vectors corresponding to each sensor signal. Then, the extended Kalman filter algorithm is used to fuse the remaining high-dimensional feature vectors based on the confidence scores to obtain a multimodal fusion feature containing the UAV's state. The sensor signals include electromagnetic signals, perception signals, and environmental signals; the UAV state includes the UAV's real-time position, velocity, and heading angle. The trajectory prediction unit is used to predict and generate the flight trajectory of the UAV within a preset time period in the future based on the historical trajectory data of the UAV and the multimodal fusion features using a pre-trained neural network model. The detection and countermeasure unit is used to determine the threat level and detect abnormal behavior based on the mission context parameters and flight trajectory, generate alarm information, generate security countermeasure strategies based on the alarm information, and execute them.

9. The UAV detection and countermeasure system as described in claim 8, characterized in that, The drone detection and countermeasure system also includes: The data communication unit is used to distribute the task instructions and standardized modal data corresponding to multiple sensing devices to different data topics according to preset topic rules using a distributed stream processing framework; and / or The incremental training unit is used to perform trajectory modeling using an HMM model based on the historical trajectory data of the UAV, to obtain a reconstructed trajectory containing a sequence of UAV behavior patterns, and to incrementally train the pre-trained neural network model using an incremental learning mechanism based on the historical data corresponding to the historical trajectory data of the UAV and the reconstructed trajectory.

10. A drone detection and countermeasure system, characterized in that, It includes a server, several different types of sensors installed in the monitoring area, several different types of jamming devices and terminals installed in the monitoring area, among which, Several of the aforementioned sensing devices are used to collect information about drones in the intrusion monitoring area in real time from different dimensions and generate sensing signals; The server is used to acquire task instructions and sensor signals from multiple sensors in the monitoring area in real time, and preprocess the task instructions and sensor signals to obtain standardized modal data corresponding to the task instructions and each sensor signal. It then performs feature extraction and high-level semantic extraction in parallel on the standardized modal data corresponding to the task instructions and each sensor signal to obtain task context parameters and high-dimensional feature vectors corresponding to each sensor signal. Using a spatiotemporally aligned multimodal fusion algorithm, the server dynamically adjusts the confidence scoring strategy based on the task context parameters and fuses the high-dimensional feature vectors corresponding to each sensor signal to obtain multimodal fusion features that include the UAV's state. Finally, it utilizes a pre-trained neural network model based on the UAV's historical trajectory. The data and the multimodal fusion features predict and generate the flight trajectory of the UAV within a preset time period in the future; based on the task context parameters and the flight trajectory, threat level determination and abnormal behavior detection are performed, alarm information is generated, and a security countermeasure strategy is generated based on the alarm information; wherein, the multimodal fusion algorithm using spatiotemporal alignment dynamically adjusts the confidence scoring strategy based on the task context parameters, fuses the high-dimensional feature vectors corresponding to each sensor signal, and obtains multimodal fusion features containing the UAV state, including: using the Kalman filter algorithm to perform spatiotemporal alignment processing on all high-dimensional feature vectors; using the task context parameters as prior knowledge, the Bayesian confidence estimation method is used to estimate the likelihood probability of the sensor signals based on the prior knowledge and the formula The confidence score of the high-dimensional feature vector corresponding to each sensing signal is calculated; where, This represents the high-dimensional feature vector corresponding to each sensing signal, where S represents sensor characteristics and E represents environmental factors. The likelihood probability is used to filter out high-dimensional feature vectors with confidence scores below a preset threshold from the high-dimensional feature vectors corresponding to each sensor signal. Then, the extended Kalman filter algorithm is used to fuse the remaining high-dimensional feature vectors based on the confidence scores to obtain a multimodal fusion feature containing the UAV's state. The sensor signals include electromagnetic signals, perception signals, and environmental signals; the UAV state includes the UAV's real-time position, velocity, and heading angle. The terminal is used to send task instructions and display alarm information; Several jamming devices are connected to the server via MQTT or HTTP interfaces to execute the security countermeasures pushed by the server.

11. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the unmanned aerial vehicle (UAV) detection and countermeasure method as described in any one of claims 1 to 7.

12. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the unmanned aerial vehicle (UAV) detection and countermeasure method as described in any one of claims 1 to 7.

Images