Unmanned aerial vehicle sound detection and early warning method and system

Abstract

The invention relates to the technical field of sound detection, in particular to an unmanned aerial vehicle sound detection and early warning method and system. The sensor network layout is pre-optimized based on digital twinning and sound wave simulation, efficient coverage and redundant backup are achieved, deployment blind areas are reduced, the system detection basic capacity is improved, signal preprocessing and event detection are carried out at the node end, data lightweight and intelligent packaging are achieved, transmission and cloud burden is relieved, and the system reliability is improved. The real-time performance and robustness of the system are enhanced, cloud multi-node data fusion, sound source separation and deep learning feature comparison are utilized, unmanned aerial vehicle and non-unmanned aerial vehicle sound sources are effectively distinguished, the recognition precision is improved, false alarms are reduced, digital twin space information and multi-node synchronization signals are combined, unmanned aerial vehicle three-dimensional geodetic coordinate calculation and continuous trajectory generation are achieved, and the method has the advantages of being high in practicability and easy to popularize. Accurate tracking and behavior prediction are supported, dynamic threat assessment is performed based on a flight path and a behavior mode, and graded early warning and early response are realized.

Description

Technical Field

[0001] This invention relates to the field of sound detection technology, and in particular to a method and system for sound detection and early warning of unmanned aerial vehicles (UAVs). Background Technology

[0002] Low-altitude, slow-speed, and small unmanned aerial vehicles (UAVs) are characterized by their small size, low-altitude flight, and strong stealth. Existing radar, electro-optical, and radio UAV detection methods have problems such as low-altitude blind spots, building obstruction, and environmental radio frequency interference in urban environments, making it difficult to detect, identify, and issue early warnings for UAVs. Summary of the Invention

[0003] Therefore, the technical problem to be solved by the present invention is to overcome the difficulty of detection, identification and early warning of drones in the prior art.

[0004] To address the aforementioned technical problems, this invention provides a method for drone sound detection and early warning, comprising: Based on a pre-built digital twin of the protected area and a sound wave propagation simulation algorithm, a sensor node network is constructed, and after the sensor node network is deployed in the field, the sound wave signal of each node is acquired in real time. Edge computing and data encapsulation are performed on the acoustic signal of each node, and then uploaded to the cloud; By performing sound source separation and feature extraction on the synchronous data of multiple nodes in the protected area through the cloud, a voiceprint feature vector is obtained, and a pre-trained recognition model is used to determine whether the voiceprint feature vector belongs to a drone. After determining that the voiceprint feature vector belongs to a drone, the drone's geodetic coordinates are located collaboratively based on the synchronized data of multiple nodes in the protected area and the digital twin of the protected area. The drone flight trajectory is generated based on the continuously output sequence of drone location points, and risk assessment and early warning are performed based on the drone flight trajectory.

[0005] Preferably, the construction of the sensor node network based on the pre-built digital twin of the protected area and the sound wave propagation simulation algorithm includes: Using geographic information systems, building information models, and drone aerial photography data, a digital twin of the protected area is constructed, including a three-dimensional model of the terrain, landforms, and buildings. A sound wave propagation simulation algorithm is run on the digital twin of the protected area to simulate the sound wave propagation when a typical UAV sound source emits sound from different locations. The signal-to-noise ratio and coverage blind spots of each potential sensor node are predicted. A multi-objective optimization algorithm is used to generate the optimal three-dimensional coordinate deployment network of the sensor nodes.

[0006] Preferably, the real-time acquisition of the acoustic signal of each node includes: Continuously collect environmental sound from each node and construct the environmental noise baseline spectrum for each node through statistical modeling; The gain of the preamplifier is adjusted in real time based on the ambient noise baseline spectrum by utilizing the dynamic gain controller built into each node. Using a microphone array and employing a broadband constant beamwidth beamforming algorithm, a directional receiving beam is formed according to a preset key monitoring direction; The clock chip in each node ensures that all microphone channels are sampled synchronously at the microsecond level, and the acquired multi-channel digital audio stream is stamped with timestamps and location stamps to obtain the sound wave signal.

[0007] Preferably, the step of performing edge computing and data encapsulation on the acoustic signal of each node and uploading it to the cloud includes: Using an adaptive filter running on the edge computing chip at each node, the acquired acoustic signal is subjected to real-time spectral subtraction or Wiener filtering based on the noise baseline spectrum. The convolutional neural network is used to analyze the denoised acoustic signal in real time to detect whether the acoustic signal contains transient events with a similarity higher than a preset threshold to the pre-built drone voiceprint feature library. When similar transient events are detected, the nodes are used to package the multi-channel audio data for N seconds before and after the event, as well as the calculated low-dimensional feature vector. For other non-event periods, only the heartbeat signal and statistical summary are packaged. Data packets are prioritized based on the confidence level of the edge AI, and each node's communication module selects different transmission strategies to upload the data packets to the cloud based on network status and data packet priority.

[0008] Preferably, the step of separating the sound source and extracting features from the synchronous data of multiple nodes in the protected area through the cloud to obtain a voiceprint feature vector, and using a pre-trained recognition model to determine whether the voiceprint feature vector belongs to a drone includes: The synchronized data from multiple nodes in the protected area is decomposed into several independent sound source components via the cloud. The separated drone sound source components are input into a pre-trained deep neural network to obtain a deep acoustic signature feature vector. The deep voiceprint feature vector is input into the binary classification model in the pre-trained recognition model to determine whether the deep feature vector belongs to the drone. If the drone is identified as a drone, the deep voiceprint feature vector is input into the metric learning model under the meta-learning framework to identify the drone model.

[0009] Preferably, after determining that the voiceprint feature vector belongs to a UAV, the step of collaboratively locating the UAV's geodetic coordinates based on synchronized data from multiple nodes in the protected area and the digital twin of the protected area includes: Generalized cross-correlation calculations are performed on the sound source components of the UAV across multiple nodes, and the multipath effect is suppressed by the quadratic correlation method to estimate the time difference of sound arrival at different nodes. By combining the three-dimensional coordinates of the nodes in the digital twin of the protected area, the time difference information is substituted into the hyperbola equation system for iterative solution, and the geodetic coordinates of the UAV in three-dimensional space are output.

[0010] Preferably, the step of generating a drone flight trajectory based on the continuously output sequence of drone location points, and performing risk assessment and early warning based on the drone flight trajectory includes: The continuously output sequence of drone position points is input into the filter to smooth and predict the drone's motion state, thereby generating the drone's flight trajectory. Real-time trajectory is matched and calculated against a pre-set database of risk behavior patterns. By integrating risk behavior pattern matching degree, drone model risk level, and intrusion area sensitivity, a weighted scoring model is used to calculate real-time risk scores and dynamically classify risk levels.

[0011] Preferably, the method further includes: The flight trajectory and early warning information of drones are visualized in real time on the situation display and command and control terminal.

[0012] The present invention also provides a drone sound detection and early warning system, comprising: Multiple distributed acoustic detection nodes are deployed in the field based on a sensor node network constructed using a pre-built digital twin of the protected area and a sound wave propagation simulation algorithm. These nodes are used to acquire the sound wave signals of each node in real time. After performing edge computing and data encapsulation on the acoustic signal of each node, the data is uploaded to the cloud. A cloud server is used to perform sound source separation and feature extraction on synchronized data from multiple nodes in the protected area to obtain voiceprint feature vectors. A pre-trained recognition model is then used to determine whether the voiceprint feature vectors belong to a drone. After determining that the voiceprint feature vector belongs to a UAV, the UAV's geodetic coordinates are collaboratively located based on synchronized data from multiple nodes in the protected area and the digital twin of the protected area. Furthermore, the system generates drone flight trajectories based on the continuously output sequence of drone location points, and performs risk assessments and early warnings based on the drone flight trajectories.

[0013] Preferably, the system further includes: The situation display and command and control terminal is used to visualize and display the flight trajectory and early warning information of UAVs in real time.

[0014] The technical solution of the present invention has the following advantages over the prior art: The UAV sound detection and early warning method described in this invention firstly optimizes the sensor network layout based on digital twins and acoustic simulation to achieve efficient coverage and redundant backup, reduce deployment blind spots, and enhance the system's basic detection capabilities. Secondly, by performing signal preprocessing and event detection at the node end, data lightweighting and intelligent encapsulation are achieved, reducing the burden on transmission and the cloud, and enhancing the system's real-time performance and robustness. Then, by utilizing cloud-based multi-node data fusion, sound source separation, and deep learning feature comparison, the method effectively distinguishes between UAV and non-UAV sound sources, improving recognition accuracy and reducing false alarms. Next, by combining digital twin spatial information and multi-node synchronization signals, the method realizes the calculation of UAV three-dimensional geodetic coordinates and the generation of continuous trajectories, supporting accurate tracking and behavior prediction. Finally, based on flight trajectory and behavior patterns, the method performs dynamic threat assessment to achieve graded early warning and early response, enhancing the proactive defense and handling capabilities against "black flight" UAVs. Attached Figure Description

[0015] To make the content of this invention easier to understand, the invention will be further described in detail below with reference to specific embodiments and accompanying drawings, wherein: Figure 1 This is a flowchart illustrating the implementation of a drone sound detection and early warning method provided by the present invention; Figure 2 This is a structural block diagram of a drone sound detection and early warning system provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the structure of a distributed acoustic detector provided in one embodiment of the present invention. Detailed Implementation

[0016] The core of this invention is to provide a method and system for drone sound detection and early warning, which effectively improves the accuracy of drone detection and early warning.

[0017] To enable those skilled in the art to better understand the present invention, the invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are merely 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.

[0018] Please refer to Figure 1. Figure 1 The flowchart illustrates the implementation of a drone sound detection and early warning method provided by this invention; the specific operation steps are as follows: S101: Based on the pre-built digital twin of the protected area and the sound wave propagation simulation algorithm, a sensor node network is constructed, and after the sensor node network is deployed in the field, the sound wave signal of each node is acquired in real time. S102: Perform edge computing and data encapsulation on the acoustic signal of each node, and upload it to the cloud; S103: By performing sound source separation and feature extraction on the synchronous data of multiple nodes in the protected area through the cloud, a voiceprint feature vector is obtained, and a pre-trained recognition model is used to determine whether the voiceprint feature vector belongs to a drone. S104: After determining that the voiceprint feature vector belongs to a UAV, the UAV's geodetic coordinates are located collaboratively based on the synchronized data of multiple nodes in the protected area and the digital twin of the protected area. S105: Generate the drone flight trajectory based on the continuously output sequence of drone location points, and perform risk assessment and early warning based on the drone flight trajectory.

[0019] Based on the above embodiments, this embodiment will provide a detailed description of step S101: In some embodiments, the sensor node network is constructed based on a pre-built digital twin of the protected area and a sound wave propagation simulation algorithm, including: Using Geographic Information System (GIS), Building Information Modeling (BIM) and UAV aerial photography data, a digital twin of the protected area is constructed, which includes a three-dimensional model of the terrain, landforms and buildings. A sound wave propagation simulation algorithm is run on the digital twin of the protected area to simulate the propagation of sound waves after reflection, diffraction and attenuation of sound waves from a typical UAV sound source at different locations. The algorithm predicts the signal-to-noise ratio and coverage blind spots of each potential sensor node, and uses a multi-objective optimization algorithm to generate the optimal three-dimensional coordinate deployment network of the sensor nodes.

[0020] It should be noted that the sound wave propagation simulation algorithm can be an acoustic ray tracing algorithm; It should be noted that the multi-objective optimization algorithm can be a genetic algorithm, with the objectives of maximizing global coverage, maximizing redundancy, and minimizing overall cost.

[0021] In some embodiments, the real-time acquisition of the acoustic signal of each node includes: Continuously collect environmental sound from each node and construct the environmental noise baseline spectrum for each node through statistical modeling; The gain of the preamplifier is adjusted in real time based on the ambient noise baseline spectrum using the built-in dynamic gain controller of each node. It should be noted that the sensitivity is increased during quiet periods and overload is prevented during noisy periods to ensure that the ADC (analog-to-digital converter) always operates in the linear region. By using a microphone array and employing a broadband constant beamwidth beamforming algorithm, a directional receiving beam is formed according to the preset key monitoring direction, effectively suppressing interference noise from non-monitoring directions and improving the initial signal-to-noise ratio; it should be noted that the key monitoring direction can be the boundary of the no-fly zone. The clock chip in each node ensures that all microphone channels achieve microsecond-level synchronous sampling, laying the foundation for subsequent accurate time delay positioning. The acquired multi-channel digital audio stream is then stamped with timestamps and location stamps to obtain the sound wave signal.

[0022] Based on the above embodiments, this embodiment will provide a detailed description of step S102: In some embodiments, the step of performing edge computing and data encapsulation on the acoustic signal of each node and uploading it to the cloud includes: By using an adaptive filter running on the edge computing chip of each node, the collected acoustic signal is subjected to real-time spectral subtraction or Wiener filtering based on the noise baseline spectrum, thus initially removing the steady-state noise of the environment. The convolutional neural network is used to analyze the denoised acoustic signal in real time to detect whether the acoustic signal contains transient events with a similarity higher than a preset threshold to the pre-built drone voiceprint feature library. It should be noted that the convolutional neural network can be a lightweight one-dimensional convolutional neural network, and transient events with a similarity higher than a preset threshold in the pre-built drone voiceprint feature library can be, for example, rotor harmonics or motor humming. When similar transient events are detected, the nodes are used to package the multi-channel audio data for N seconds before and after the event, as well as the calculated low-dimensional feature vector. For other non-event periods, only the heartbeat signal and statistical summary are packaged. It should be noted that low-dimensional feature vectors can be zero-crossing rates or spectral centroids; Data packets are prioritized based on the confidence level of the edge AI, and each node's communication module selects different transmission strategies to upload the data packets to the cloud based on network status and data packet priority.

[0023] It should be noted that data can be categorized into three priorities: high suspicion, medium suspicion, or environmental data. It should be noted that the node's communication module (4G / 5G / CAT-1) selects different transmission strategies based on network status and data packet priority. For example, high-probability data packets use the TCP protocol and multiple network slices to ensure reliable transmission, while environmental data uses the UDP protocol to tolerate a small amount of packet loss. It should be noted that when the network is interrupted, the node starts a local cache, prioritizing the caching of high-priority data. Once the network is restored, the cached data is re-uploaded to the cloud in priority order.

[0024] Based on the above embodiments, this embodiment will provide a detailed description of step S103: In some embodiments, the step of performing sound source separation and feature extraction on synchronized data from multiple nodes in the protected area via the cloud to obtain a voiceprint feature vector, and then using a pre-trained recognition model to determine whether the voiceprint feature vector belongs to a drone, includes: The synchronized data from multiple nodes in the protected area is decomposed into several independent sound source components via the cloud. It should be noted that after receiving synchronized data from multiple nodes in the same area, the cloud first uses independent component analysis or spatiotemporal clustering algorithms to decompose the mixed signal into several statistically independent sound source components, effectively separating out pure drone sounds, car sounds, wind sounds, etc. The separated drone sound source components are input into a pre-trained deep neural network to obtain a deep acoustic signature feature vector. It should be noted that the isolated suspected drone sound source components are input into a pre-trained deep neural network. This network, pre-trained on a large-scale audio dataset, possesses powerful time-frequency domain feature extraction capabilities and outputs a high-dimensional, abstract, and robust deep voiceprint feature vector. It should be noted that the drone voiceprint feature database is the data foundation for deep learning training. Deployed in a cloud server, it processes and analyzes the sound signals of drones collected by signal acquisition devices, extracts voiceprint feature information, uses neural network model algorithms to identify drones, and establishes a drone voiceprint feature database. When a detection is finally confirmed (such as through photoelectric confirmation), the voiceprint features and tags of that event will be automatically added to the training set. The recognition model is incrementally learned periodically, so that the voiceprint database can dynamically evolve and adapt to new drone models. The deep voiceprint feature vector is input into the binary classification model in the pre-trained recognition model to determine whether the deep feature vector belongs to the drone, thereby minimizing false alarms. If the drone is identified as a drone, the deep voiceprint feature vector is input into the metric learning model under the meta-learning framework to identify the drone model.

[0025] It should be noted that the metric learning model learns a feature space that makes the features of drones of the same model close to each other and those of different models far apart. By calculating the distance to known samples in the voiceprint database, it can achieve accurate model identification with small samples or even zero samples.

[0026] Based on the above embodiments, this embodiment will provide a detailed description of step S104: In some embodiments, after determining that the voiceprint feature vector belongs to a UAV, the step of collaboratively locating the UAV's geodetic coordinates based on synchronized data from multiple nodes in the protected area and the digital twin of the protected area includes: Generalized cross-correlation calculations are performed on the sound source components of the UAV across multiple nodes, and the multipath effect is suppressed by the quadratic correlation method to estimate the time difference of sound arrival at different nodes. By combining the three-dimensional coordinates of the nodes in the digital twin of the protected area, the time difference information is substituted into the hyperbola equation system for iterative solution, and the geodetic coordinates of the UAV in three-dimensional space are output.

[0027] Based on the above embodiments, this embodiment will provide a detailed description of step S105: In some embodiments, generating a drone flight trajectory based on a continuously output sequence of drone location points, and performing risk assessment and early warning based on the drone flight trajectory includes: The continuously output sequence of drone position points is input into the filter to smooth and predict the drone's motion state, thereby generating the drone's flight trajectory. It should be noted that the filter can be an extended Kalman filter or a particle filter, and the motion state includes position, velocity, and acceleration; Real-time trajectory is matched and calculated against a pre-set database of risk behavior patterns. It should be noted that the preset risk behavior pattern library can include straight-line penetration, circumnavigation reconnaissance, hovering gaze, and logistics routes; By integrating risk behavior pattern matching degree, drone model risk level, and intrusion area sensitivity, a weighted scoring model is used to calculate real-time risk scores and dynamically classify risk levels, such as low, medium, high, and emergency.

[0028] Based on the above embodiments, the drone sound detection and early warning method further includes: The flight trajectory and early warning information of drones are visualized in real time on the situation display and command and control terminal.

[0029] It should be noted that: On the digital twin base of the display screen, the drone can be rendered as a 3D model and its historical and predicted trajectories can be plotted in real time. After clicking on the drone target, an information panel pops up, dynamically displaying its real-time voiceprint spectrum, the identified model, threat level, power signal prediction curve, and the system's recommended response strategy. The system automatically calls upon public safety cameras or pan-tilt units located near the trajectory, uses the PTZ automatic tracking algorithm to lock onto the target, and displays the video stream in a picture-in-picture format on the situation map, achieving integrated verification of sound, light, electricity, and electricity.

[0030] It should be noted that: Based on the threat level and pre-set contingency plans, alarm information is distributed in a differentiated manner to managers at different levels (such as patrol posts, command centers, and superior units) through API interfaces, App push notifications, SMS messages, etc. For high-level threats, the system automatically plans the nearest interception route for security personnel on the digital twin map and highlights nearby available countermeasures (such as jamming gun deployment points).

[0031] After obtaining the highest level of authorization, it automatically or semi-automatically sends instructions to the networked countermeasures equipment control system to carry out directional radio frequency interference or navigation signal deception against the target drone, completing a closed loop from detection to disposal.

[0032] Please refer to Figure 2 , Figure 2 A structural block diagram of a drone sound detection and early warning system provided in an embodiment of the present invention; specifically, it may include: Multiple distributed acoustic detection nodes (distributed acoustic detection) are deployed in the field based on a sensor node network constructed using a pre-built digital twin of the protected area and a sound wave propagation simulation algorithm. This network is used to acquire the sound wave signal from each node in real time. After performing edge computing and data encapsulation on the acoustic signal of each node, the data is uploaded to the cloud. A cloud server (central cloud) is used to perform sound source separation and feature extraction on synchronized data from multiple nodes in the protected area to obtain voiceprint feature vectors. A pre-trained recognition model is then used to determine whether the voiceprint feature vectors belong to a drone. After determining that the voiceprint feature vector belongs to a UAV, the UAV's geodetic coordinates are collaboratively located based on synchronized data from multiple nodes in the protected area and the digital twin of the protected area. Furthermore, the system generates drone flight trajectories based on the continuously output sequence of drone location points, and performs risk assessments and early warnings based on the drone flight trajectories.

[0033] Based on the above embodiments, and using a pre-built digital twin of the protected area and a sound wave propagation simulation algorithm, the sensor node network is constructed as follows: Using geographic information systems, building information models, and drone aerial photography data, a digital twin of the protected area is constructed, including a three-dimensional model of the terrain, landforms, and buildings. A sound wave propagation simulation algorithm is run on the digital twin of the protected area to simulate the sound wave propagation when a typical UAV sound source emits sound from different locations. The signal-to-noise ratio and coverage blind spots of each potential sensor node are predicted. A multi-objective optimization algorithm is used to generate the optimal three-dimensional coordinate deployment network of the sensor nodes.

[0034] like Figure 3 Based on the above embodiments, the distributed acoustic detection node includes: High-sensitivity microphone array: used to continuously collect ambient sound from each node and construct the ambient noise baseline spectrum for each node through statistical modeling. The preamplifier gain is adjusted in real time based on the ambient noise baseline spectrum using a built-in dynamic gain controller. A broadband constant beamwidth beamforming algorithm is used to form a directional receiving beam based on the preset key monitoring direction.

[0035] Multi-channel synchronous data acquisition card (multi-channel data acquisition): used for analog-to-digital conversion, with a built-in clock chip to ensure microsecond-level synchronous sampling of all microphone channels.

[0036] Edge computing unit (data processing): Contains a built-in lightweight AI model used to add timestamps and location stamps to the acquired multi-channel digital audio streams to obtain sound wave signals. Edge computing and data encapsulation are performed on the acoustic signal of each node, and then uploaded to the cloud.

[0037] Power system and communication module: Responsible for power supply and communication with the cloud via 4G / 5G network.

[0038] Based on the above embodiments, the edge computing unit is used for: Using a built-in adaptive filter, the acquired acoustic signal is subjected to real-time spectral subtraction or Wiener filtering based on the noise baseline spectrum. The convolutional neural network is used to analyze the denoised acoustic signal in real time to detect whether the acoustic signal contains transient events with a similarity higher than a preset threshold to the pre-built drone voiceprint feature library. When similar transient events are detected, the nodes are used to package the multi-channel audio data for N seconds before and after the event, as well as the calculated low-dimensional feature vector. For other non-event periods, only the heartbeat signal and statistical summary are packaged. Data packets are prioritized based on the confidence level of the edge AI, and different transmission strategies are selected based on network status and data packet priority to upload the data packets to the cloud through the communication module.

[0039] Based on the above embodiments, the cloud includes: Data processing engine: It is used to decompose the synchronous data of multiple nodes in the protected area into several independent sound source components through the cloud, and input the separated drone sound source components into a pre-trained deep neural network to obtain a deep acoustic feature vector.

[0040] Voiceprint feature database: Stores and manages deep voiceprint feature vectors for various types of UAVs, and supports dynamic evolution.

[0041] AI recognition and localization engine: It is used to input the deep voiceprint feature vector into the binary classification model in the pre-trained recognition model to determine whether the deep feature vector belongs to a drone. If it is determined to be a drone, the deep voiceprint feature vector is input into the metric learning model under the meta-learning framework to identify the drone model.

[0042] Based on the above embodiments, the AI ​​recognition and positioning engine is also used to perform generalized cross-correlation calculations on the drone's sound source components among multiple nodes, and to suppress multipath effects through a quadratic correlation method, estimate the time difference of sound reaching different nodes, and combine the three-dimensional coordinates of the nodes in the digital twin of the protected area to substitute the time difference information into the hyperbolic equation system for iterative solution, and output the ground coordinates of the drone in three-dimensional space.

[0043] Based on the above embodiments, the AI ​​recognition and positioning engine is also used for: The continuously output sequence of drone position points is input into the filter to smooth and predict the drone's motion state, thereby generating the drone's flight trajectory. Real-time trajectory is matched and calculated against a pre-set database of risk behavior patterns. By integrating risk behavior pattern matching degree, drone model risk level, and intrusion area sensitivity, a weighted scoring model is used to calculate real-time risk scores and dynamically classify risk levels.

[0044] Based on the above embodiments, the drone sound detection and early warning system also includes: The situation display and command control terminal (situation display terminal) is used to visualize and display the flight trajectory and early warning information of UAVs in real time.

[0045] Based on the above embodiments, the situation display and command control terminal includes: Digital Twin Visualization Engine: Used to render drones as 3D models on a digital twin base of a display screen, and to plot their historical and predicted trajectories in real time; After clicking on the drone target, an information panel will pop up, dynamically displaying its real-time voiceprint spectrum, the identified model, threat level, power signal prediction curve, and the system's recommended response strategy; The system automatically calls upon public safety cameras or pan-tilt units at high points near the trajectory, uses the PTZ automatic tracking algorithm to lock onto the target, and displays the video stream in a picture-in-picture format on the situation map, achieving "sound, light, electricity and gas" linkage verification.

[0046] Alarm and Response Management Module: Based on threat level and preset plans, the module distributes alarm information in a differentiated manner to managers at different levels (such as patrol posts, command centers, and superior units) through API interfaces, App push, SMS, etc. For high-level threats, the system automatically plans the shortest interception path for security personnel on the digital twin map and highlights nearby available countermeasures devices (such as jamming gun deployment points).

[0047] The countermeasures control interface is used to automatically or semi-automatically send instructions to the networked countermeasures control system after obtaining the highest authorization, to carry out directional radio frequency interference or navigation signal deception against the target drone, completing a closed loop from detection to disposal.

[0048] Obviously, the above embodiments are merely illustrative examples for clear explanation and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations here. However, obvious variations or modifications derived therefrom are still within the scope of protection of this invention.

Claims

1. A method for sound detection and early warning of unmanned aerial vehicles (UAVs), characterized in that, include: Based on a pre-built digital twin of the protected area and a sound wave propagation simulation algorithm, a sensor node network is constructed, and after the sensor node network is deployed in the field, the sound wave signal of each node is acquired in real time. Edge computing and data encapsulation are performed on the acoustic signal of each node, and then uploaded to the cloud; By performing sound source separation and feature extraction on the synchronous data of multiple nodes in the protected area through the cloud, a voiceprint feature vector is obtained, and a pre-trained recognition model is used to determine whether the voiceprint feature vector belongs to a drone. After determining that the voiceprint feature vector belongs to a UAV, the UAV's geodetic coordinates are located collaboratively based on the synchronized data of multiple nodes in the protected area and the digital twin of the protected area. The drone flight trajectory is generated based on the continuously output sequence of drone location points, and risk assessment and early warning are performed based on the drone flight trajectory.

2. The method for sound detection and early warning of unmanned aerial vehicles according to claim 1, characterized in that, The sensor node network constructed based on the pre-built digital twin of the protected area and the sound wave propagation simulation algorithm includes: Using geographic information systems, building information models, and drone aerial photography data, a digital twin of the protected area is constructed, which includes three-dimensional models of terrain, landforms, and buildings. A sound wave propagation simulation algorithm is run on the digital twin of the protected area to simulate the sound wave propagation when a typical UAV sound source emits sound from different locations. The signal-to-noise ratio and coverage blind spots of each potential sensor node are predicted. A multi-objective optimization algorithm is used to generate the optimal three-dimensional coordinate deployment network of the sensor nodes.

3. The method for drone sound detection and early warning according to claim 1 or 2, characterized in that, The real-time acquisition of acoustic signals from each node includes: Continuously collect environmental sound from each node and construct the environmental noise baseline spectrum for each node through statistical modeling; The gain of the preamplifier is adjusted in real time based on the ambient noise baseline spectrum by utilizing the dynamic gain controller built into each node. Using a microphone array and employing a broadband constant beamwidth beamforming algorithm, a directional receiving beam is formed according to a preset key monitoring direction; The clock chip in each node ensures that all microphone channels are sampled synchronously at the microsecond level, and the acquired multi-channel digital audio stream is stamped with timestamps and location stamps to obtain the sound wave signal.

4. The method for sound detection and early warning of unmanned aerial vehicles according to claim 3, characterized in that, The step of performing edge computing and data encapsulation on the acoustic signal of each node and uploading it to the cloud includes: Using an adaptive filter running on the edge computing chip at each node, the acquired acoustic signal is subjected to real-time spectral subtraction or Wiener filtering based on the noise baseline spectrum. The convolutional neural network is used to analyze the denoised acoustic signal in real time to detect whether the acoustic signal contains transient events with a similarity higher than a preset threshold to the pre-built drone voiceprint feature library. When similar transient events are detected, the nodes are used to package the multi-channel audio data for N seconds before and after the event, as well as the calculated low-dimensional feature vector. For other non-event periods, only the heartbeat signal and statistical summary are packaged. Data packets are prioritized based on the confidence level of the edge AI, and each node's communication module selects different transmission strategies to upload the data packets to the cloud based on network status and data packet priority.

5. The method for sound detection and early warning of unmanned aerial vehicles according to claim 4, characterized in that, The process of separating sound sources and extracting features from synchronized data from multiple nodes in the protected area via the cloud to obtain a voiceprint feature vector, and then using a pre-trained recognition model to determine whether the voiceprint feature vector belongs to a drone, includes: The synchronized data from multiple nodes in the protected area is decomposed into several independent sound source components via the cloud. The separated drone sound source components are input into a pre-trained deep neural network to obtain a deep acoustic signature feature vector. The deep voiceprint feature vector is input into the binary classification model in the pre-trained recognition model to determine whether the deep feature vector belongs to the drone. If the drone is identified as a drone, the deep voiceprint feature vector is input into the metric learning model under the meta-learning framework to identify the drone model.

6. The method for sound detection and early warning of unmanned aerial vehicles according to claim 5, characterized in that, After determining that the voiceprint feature vector belongs to a UAV, the method of collaboratively locating the UAV's geodetic coordinates based on synchronized data from multiple nodes in the protected area and the digital twin of the protected area includes: The generalized cross-correlation of the drone's sound source components is calculated among multiple nodes, and the multipath effect is suppressed by the quadratic correlation method to estimate the time difference of sound arrival at different nodes. By combining the three-dimensional coordinates of the nodes in the digital twin of the protected area, the time difference information is substituted into the hyperbola equation system for iterative solution, and the geodetic coordinates of the UAV in three-dimensional space are output.

7. The method for sound detection and early warning of unmanned aerial vehicles according to claim 5, characterized in that, The process of generating a drone flight trajectory based on a continuously output sequence of drone location points, and performing risk assessment and early warning based on the drone flight trajectory, includes: The continuously output sequence of drone position points is input into the filter to smooth and predict the drone's motion state and generate the drone's flight trajectory. Real-time trajectory is matched and calculated against a pre-set database of risk behavior patterns. By integrating risk behavior pattern matching degree, drone model risk level, and intrusion area sensitivity, a weighted scoring model is used to calculate real-time risk scores and dynamically classify risk levels.

8. The method for sound detection and early warning of unmanned aerial vehicles according to claim 1, characterized in that, Also includes: The flight trajectory and early warning information of drones are visualized in real time on the situation display and command and control terminal.

9. A drone sound detection and early warning system, characterized in that, include: Multiple distributed acoustic detection nodes are deployed in the field based on a sensor node network constructed using a pre-built digital twin of the protected area and a sound wave propagation simulation algorithm. These nodes are used to acquire the sound wave signals of each node in real time. After performing edge computing and data encapsulation on the acoustic signal of each node, the data is uploaded to the cloud. A cloud server is used to perform sound source separation and feature extraction on synchronized data from multiple nodes in the protected area to obtain voiceprint feature vectors. A pre-trained recognition model is then used to determine whether the voiceprint feature vectors belong to a drone. After determining that the voiceprint feature vector belongs to a UAV, the UAV's geodetic coordinates are collaboratively located based on synchronized data from multiple nodes in the protected area and the digital twin of the protected area. Furthermore, the system generates drone flight trajectories based on the continuously output sequence of drone location points, and performs risk assessments and early warnings based on the drone flight trajectories.

10. The UAV sound detection and early warning system according to claim 9, characterized in that, Also includes: The situation display and command and control terminal is used to visualize and display the flight trajectory and early warning information of UAVs in real time.

Images