Low altitude unmanned aerial vehicle detection and intelligent situation awareness method and system

Abstract

The application discloses a thunder and vision integrated low-altitude unmanned aerial vehicle detection and intelligent situation awareness method and system, belongs to the low-altitude security and multi-sensor fusion technical field, and the method comprises the steps of acquiring radar track information of multiple targets; performing coarse classification identification and threat level evaluation on the targets based on the radar track information, marking key targets, and generating a guidance instruction; turning the camera based on the guidance instruction, acquiring target azimuth / elevation angle information, performing asymmetric weighted fusion on the radar track information and the target azimuth / elevation angle information, and generating a high-precision fusion track; collecting multispectral images for fine classification identification to obtain dual-spectrum images with type identification information; displaying the fused target situation information; and the target situation information at least comprises the high-precision fusion track, the dual-spectrum images with type identification information, identification confidence and threat level. The target classification precision is improved, and accurate classification under an extremely low false alarm rate is realized.

Description

Technical Field

[0001] This application belongs to the field of low-altitude security and multi-sensor fusion technology, and specifically relates to a radar-visual integrated method and system for low-altitude UAV detection and intelligent situational awareness. Background Technology

[0002] With the rapid development of the low-altitude economy, effectively preventing and controlling unauthorized drone flights in core areas has become a key focus of security. Low-altitude target detection mainly relies on well-known technologies such as radar detection, photoelectric detection, coordinate transformation, and deep learning recognition. The main steps include: the radar performs a panoramic scan of the designated airspace, and outputs the distance, azimuth, and speed information of the target after detection; the radar sends the target coordinates to the photoelectric turntable, which guides the turntable to adjust its lens to the corresponding angle; the photoelectric system uses a visual recognition algorithm to determine whether the target is a drone; the display and control terminal displays the radar track and photoelectric image in separate windows, and the final confirmation is performed manually.

[0003] However, the existing technical solutions have the following drawbacks in practical applications: First, the false alarm rate of classification and recognition is relatively high. It mainly relies on the static feature recognition of photoelectric images. In situations such as long distances, poor lighting, or blurred target features, it cannot effectively distinguish between drones and birds, resulting in poor recognition reliability of the system in complex low-altitude environments. Second, the trajectory tracking accuracy is low. Existing solutions mostly rely on radar to guide photoelectric signals in one direction, failing to utilize the high-precision angle measurement advantage of photoelectric signals to correct the angle measurement error of the radar. Third, the situation assessment dimension is one-sided. Its threat assessment algorithm is based solely on distance and does not consider dynamic factors such as recognition confidence, target maneuvering intentions, and estimated arrival time, resulting in insufficient decision support capabilities. Fourth, the model's generalization ability is poor. Due to the high cost and limited quantity of real-world drone test samples in low-altitude environments, existing deep learning models are prone to underfitting or overfitting when facing new or rare drone models. Summary of the Invention

[0004] To address the aforementioned issues, this application provides a low-altitude integrated situational awareness method and system for a radar-visual integrated UAV reconnaissance and surveillance system. Through a comprehensive situational awareness process involving radar reconnaissance and tracking, photoelectric confirmation imaging, classification and identification via an information fusion system, and secondary fusion, the system achieves improved detection, tracking, and identification accuracy of key targets (UAVs / birds) in complex low-altitude environments, ultimately enhancing the system's low-altitude integrated situational awareness capabilities. The method includes: Acquire radar track information for multiple targets; Based on the radar track information, targets are coarsely classified and identified, and threat levels are assessed. Targets whose identification confidence meets a preset threshold and whose threat level reaches a preset level are marked as key targets, and guidance instructions are generated. The camera automatically adjusts based on guidance commands to acquire target azimuth / elevation information, and performs asymmetric weighted fusion of radar track information with target azimuth / elevation information to generate a high-precision fused track. Multispectral images are acquired and subjected to precise classification and identification to obtain bispectral images with type identification information; The fused target situation information is displayed; the target situation information includes at least a high-precision fused track, a bispectral image with type identification information, identification confidence level, and threat level.

[0005] Optionally, acquiring radar track information of multiple targets includes: By using unmanned aerial vehicle (UAV) reconnaissance radar to scan low-altitude airspace to obtain distance, angle, and speed information of moving targets, and then processing the data through coordinate transformation and TWS tracking, radar track information of multiple targets is generated.

[0006] Optionally, the step of performing coarse classification and threat level assessment of targets based on the radar track information includes: After preprocessing radar track data using data augmentation techniques, coarse target classification and identification are performed based on a hybrid deep learning network. The identification confidence is calculated using a confidence smoothing algorithm based on a time-domain sliding window. At the same time, threat level assessment is performed based on a comprehensive threat level quantification evaluation function driven by multiple factors.

[0007] Optionally, the hybrid deep learning network includes a TCN channel and a Transformers channel: The TCN channel uses a multi-scale parallel structure to extract the temporal local features of the radar track; The Transformers channel uses a multi-head self-attention mechanism to extract long-term global correlation features of radar tracks; The features extracted from the TCN channel and the Transformers channel are fused together, and the target classification probability is output through multi-strategy pooling, feature concatenation and enhancement, and fully connected layer mapping.

[0008] Optionally, the step of asymmetrically weighted fusion of radar track information and target azimuth / elevation information to generate a high-precision fused track includes: The target azimuth / elevation information includes photoelectric azimuth, photoelectric elevation, and photoelectric image information; The radar track information includes radar azimuth, radar elevation, and range; Receive photoelectric azimuth, photoelectric elevation, and photoelectric image information transmitted by the dual-spectrum photoelectric instrument; perform weighted fusion of the photoelectric azimuth with the radar azimuth in the radar track information to obtain a fused azimuth; perform weighted fusion of the photoelectric elevation with the radar elevation in the radar track information to obtain a fused elevation. Based on the fused azimuth angle, the fused elevation angle, and the range information in the radar track information, a multi-level coordinate transformation is performed to generate a high-precision fused track.

[0009] Optionally, the step of performing multi-level coordinate transformation based on the fused azimuth angle, the fused elevation angle, and the range information in the radar track information to generate a high-precision fused track includes: The azimuth angle, elevation angle, and radar distance information are extracted and fused to form polar coordinate data. The polar coordinate data is then transformed from the sensor coordinate system to the carrier coordinate system, the geodetic rectangular coordinate system, and the geographic coordinate system in a multi-level coordinate transformation to generate a high-precision fused trajectory, which is then sent to the terminal control system.

[0010] Optionally, the acquisition of multispectral images for precise classification and identification to obtain bispectral images with type identification information includes: Multispectral images are obtained by imaging the target in visible light and infrared light using a dual-spectral photoelectric instrument. The target detection model is invoked to perform secondary visual feature recognition and verification on the multispectral image, and the category label of the target is output. The category label, along with the photoelectric azimuth and photoelectric elevation angles bearing the same timestamp, are used as type identification information. The bispectral image with the type identification information is then sent to the terminal control system via a switch.

[0011] This application also provides a radar-visual integrated low-altitude UAV detection and intelligent situational awareness system, which includes: Unmanned aerial vehicle (UAV) reconnaissance radar is used to acquire radar track information of multiple targets; The information fusion system is used to perform coarse classification and threat level assessment of targets based on the radar track information. Targets with a recognition confidence level meeting a preset threshold and a threat level reaching a preset level are marked as key targets. The target track is sent to the integrated control terminal module, which generates guidance commands and sends them to the dual-spectral photoelectric instrument. The system also receives target azimuth / elevation information and performs asymmetric weighted fusion of the radar track information and target azimuth / elevation information to generate a high-precision fused track. Finally, it receives multispectral images for fine classification and identification, obtaining a dual-spectral image with type identification information. The dual-spectral optoelectronic instrument includes a visible light sensor and an infrared sensor, which are used to receive guidance commands, automatically adjust the lens based on the guidance commands, acquire multispectral images and target azimuth / elevation angle information and send them to the information fusion system; The integrated control terminal is used to display the fused target situation information; the target situation information includes at least a high-precision fused track, a bispectral image with type identification information, identification confidence level, and threat level; A network switch is used for information exchange and transmission between various components.

[0012] This application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the methods provided in the above embodiments.

[0013] This application also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the methods provided in the above embodiments.

[0014] This application provides a low-altitude integrated situational awareness method and system for a radar-visual integrated UAV reconnaissance and surveillance system, comprising: acquiring radar track information of multiple targets; performing coarse classification and threat level assessment of the targets based on the radar track information, marking targets with identification confidence levels meeting preset thresholds and threat levels reaching preset levels as key targets, and generating guidance commands; automatically adjusting the camera based on the guidance commands to acquire target azimuth / elevation information, performing asymmetric weighted fusion of the radar track information and target azimuth / elevation information to generate a high-precision fused track; acquiring multispectral images for fine classification and identification to obtain a bispectral image with type identification information; and displaying the fused target situational information; wherein the target situational information includes at least a high-precision fused track, a bispectral image with type identification information, identification confidence level, and threat level.

[0015] A dual-channel feature extraction architecture coupled with temporal convolutional networks (TCNs) and transformers enables accurate classification of targets such as drones and birds in complex environments; The confidence smoothing algorithm based on time-domain sliding window effectively filters out instantaneous classification jitter caused by sensor noise or environmental flicker by assigning higher weights to near-end recognition data, while ensuring recognition sensitivity. The multi-factor collaborative driving comprehensive threat level quantitative evaluation function introduces the identification confidence level as a dynamic weight into the evaluation matrix, and achieves "conservative compensation" when the perceived data is ambiguous, so as to ensure that the dynamic threat judgment is accurate. The asymmetric weighted fusion mechanism and multi-level coordinate transformation eliminate radar angle measurement bias, achieving high-precision and high-smoothness geographic coordinate output of the target; The parallel communication distribution architecture based on multicast mechanism enables "one-to-many" transmission and reception of radar raw data stream, photoelectric angle stream, image feature stream and fused situational command within the local area network. This parallel listening mechanism ensures low-latency synchronization of high-frequency sensing data and control commands between the information fusion system and multiple control terminals, significantly improving the system's fault tolerance performance under multi-sensor and multi-task concurrency.

[0016] Other features and advantages of this application will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the application. The objectives and other advantages of this application may be realized and obtained by means of the structures pointed out in the description, claims and drawings. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating an integrated radar-visual low-altitude UAV detection and intelligent situational awareness method provided in an embodiment of this application.

[0019] Figure 2 This is a block diagram of an integrated radar-visual low-altitude UAV detection and intelligent situational awareness system provided in an embodiment of this application.

[0020] Figure 3 This is a diagram illustrating the overall architecture of an integrated radar-visual low-altitude UAV detection and intelligent situational awareness system provided in this application embodiment.

[0021] Figure 4 A flowchart for target identification and threat level assessment provided in the embodiments of this application.

[0022] Figure 5 The weighted fusion and coordinate transformation process provided in this application embodiment. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0024] In recent years, with the rapid development of the low-altitude economy, preventing unauthorized drone flights targeting core areas, no-fly zones, and sensitive facilities has become a key focus and challenge for low-altitude security. Currently, existing low-altitude target detection and identification systems mainly rely on the following technologies: (1) Radar detection technology Radar detection systems utilize the principle of electromagnetic wave reflection to achieve all-weather, large-area active search of a given airspace. A typical tracking method is Track-While-Scan (TWS) technology. This system can maintain continuous airspace search while predicting the trajectory and updating the status of multiple detected targets, outputting motion parameters such as target range, azimuth, and velocity.

[0025] (2) Photoelectric detection technology Optoelectronic detection systems acquire high-resolution images of targets using visible light or infrared sensors, offering advantages such as high angular accuracy and intuitive target identification. However, their field of view is limited, resulting in low efficiency for independent large-scale airspace searches. Therefore, in practical engineering applications, radar systems are typically required to provide target guidance information to achieve accurate identification and tracking of specific targets.

[0026] (3) Coordinate transformation technology To achieve effective fusion of multi-sensor data, existing systems typically transform the target coordinates output by each sensor from its local coordinate system to a unified spatial reference system, including the geodetic rectangular coordinate system and the geographic coordinate system (such as the WGS-84 coordinate system). This coordinate transformation process is a fundamental step in multi-sensor information fusion.

[0027] (4) Deep learning recognition technology In target recognition, existing technologies often employ visual recognition methods based on deep learning neural networks to perform real-time target detection and classification on images acquired by optoelectronic systems. These methods can achieve automatic recognition of drones to a certain extent, but their recognition performance is highly dependent on image quality and the completeness of the training samples.

[0028] In summary, existing low-altitude detection and identification systems have made some progress in radar search, photoelectric imaging, coordinate transformation, and deep learning recognition. However, they still have significant shortcomings in target classification accuracy, trajectory tracking smoothness, multi-source information fusion efficiency, and robustness in complex environments, necessitating the development of more efficient and accurate integrated situational awareness methods.

[0029] To address the aforementioned technical issues, this application proposes a low-altitude integrated situational awareness method and system for a radar-visual integrated UAV reconnaissance and surveillance system.

[0030] Figure 1A flowchart illustrating a low-altitude integrated situational awareness method for a radar-visual integrated UAV reconnaissance and surveillance system provided in this application embodiment is shown below. Figure 1 As shown, the method includes: Acquire radar track information for multiple targets; Based on the radar track information, targets are coarsely classified and identified, and threat levels are assessed. Targets whose identification confidence meets a preset threshold and whose threat level reaches a preset level are marked as key targets, and guidance instructions are generated. The camera automatically adjusts based on guidance commands to acquire target azimuth / elevation information, and performs asymmetric weighted fusion of radar track information with target azimuth / elevation information to generate a high-precision fused track. Multispectral images are acquired and subjected to precise classification and identification to obtain bispectral images with type identification information; The fused target situation information is displayed; the target situation information includes at least a high-precision fused track, a bispectral image with type identification information, identification confidence level, and threat level.

[0031] Optionally, acquiring radar track information of multiple targets includes: By using unmanned aerial vehicle (UAV) reconnaissance radar to scan low-altitude airspace to obtain distance, angle, and speed information of moving targets, and then processing the data through coordinate transformation and TWS tracking, radar track information of multiple targets is generated.

[0032] Optionally, the step of performing coarse classification and threat level assessment of targets based on the radar track information includes: After preprocessing radar track data using data augmentation techniques, coarse target classification and identification are performed based on a hybrid deep learning network. The identification confidence is calculated using a confidence smoothing algorithm based on a time-domain sliding window. At the same time, threat level assessment is performed based on a comprehensive threat level quantification evaluation function driven by multiple factors.

[0033] Optionally, the hybrid deep learning network includes a TCN channel and a Transformers channel: The TCN channel uses a multi-scale parallel structure to extract the temporal local features of the radar track; The Transformers channel uses a multi-head self-attention mechanism to extract long-term global correlation features of radar tracks; The features extracted from the TCN channel and the Transformers channel are fused together, and the target classification probability is output through multi-strategy pooling, feature concatenation and enhancement, and fully connected layer mapping.

[0034] Optionally, the step of asymmetrically weighted fusion of radar track information and target azimuth / elevation information to generate a high-precision fused track includes: The target azimuth / elevation information includes photoelectric azimuth, photoelectric elevation, and photoelectric image information; The radar track information includes radar azimuth, radar elevation, and range; Receive photoelectric azimuth, photoelectric elevation, and photoelectric image information transmitted by the dual-spectrum photoelectric instrument; perform weighted fusion of the photoelectric azimuth with the radar azimuth in the radar track information to obtain a fused azimuth; perform weighted fusion of the photoelectric elevation with the radar elevation in the radar track information to obtain a fused elevation. Based on the fused azimuth angle, the fused elevation angle, and the range information in the radar track information, a multi-level coordinate transformation is performed to generate a high-precision fused track.

[0035] Optionally, the step of performing multi-level coordinate transformation based on the fused azimuth angle, the fused elevation angle, and the range information in the radar track information to generate a high-precision fused track includes: The azimuth angle, elevation angle, and radar distance information are extracted and fused to form polar coordinate data. The polar coordinate data is then transformed from the sensor coordinate system to the carrier coordinate system, the geodetic rectangular coordinate system, and the geographic coordinate system in a multi-level coordinate transformation to generate a high-precision fused trajectory, which is then sent to the terminal control system.

[0036] Optionally, the acquisition of multispectral images for precise classification and identification to obtain bispectral images with type identification information includes: Multispectral images are obtained by imaging the target in visible light and infrared light using a dual-spectral photoelectric instrument. The target detection model is invoked to perform secondary visual feature recognition and verification on the multispectral image, and the category label of the target is output. The category label, along with the photoelectric azimuth and photoelectric elevation angles bearing the same timestamp, are used as type identification information. The bispectral image with the type identification information is then sent to the terminal control system via a switch.

[0037] This application effectively overcomes the problem of poor recognition reliability caused by relying solely on static features of photoelectric images in the prior art by constructing a multimodal feature fusion classification mechanism. It significantly reduces the false alarm rate between UAVs and birds, and achieves an extremely low false alarm level while ensuring a high recall rate, thereby greatly improving the credibility of target classification in complex low-altitude environments.

[0038] This application not only improves target positioning accuracy but also outputs a highly smooth and accurate geographic coordinate track. It overcomes the technical limitations of unidirectional radar-to-electro-optical guidance in existing radar-visual linkage modes by establishing a two-way closed-loop collaborative tracking architecture between radar and electro-optical systems. This fully utilizes the inherent high-precision angle measurement advantage of electro-optical detection technology to correct and compensate for radar angle measurement errors in real time, thereby obtaining a highly smooth and accurate target geographic coordinate track and significantly improving the continuity and stability of track tracking.

[0039] This application achieves intelligent, multi-dimensional threat quantification assessment, providing a scientific quantitative basis for command and decision-making. It changes the one-sided assessment method in existing technologies that relies solely on distance for threat determination. By constructing a threat quantification assessment model that integrates multiple dynamic factors such as identification confidence, target maneuvering intent, and estimated arrival time, it effectively solves the problem of inaccurate threat judgment under ambiguous targets. This provides a scientific, comprehensive, and quantifiable basis for command and decision-making, significantly enhancing the system's responsiveness to complex situations.

[0040] This application addresses the challenge of model training in limited-sample environments, improving perceptual robustness in extreme and unknown scenarios. Specifically, it addresses the prominent issues of high cost, limited quantity, and imbalanced class distribution in acquiring real-world UAV test samples in low-altitude environments. By employing targeted limited-sample learning strategies and data augmentation mechanisms, it effectively alleviates the underfitting or overfitting phenomena caused by insufficient samples in existing deep learning models. This significantly improves the model's generalization ability and perceptual robustness in new models, rare aircraft types, and extreme scenarios, ensuring stable and reliable system operation under limited-sample conditions.

[0041] Figure 2 A block diagram of an integrated radar-visual low-altitude UAV detection and intelligent situational awareness system provided in this application embodiment. The system includes: Unmanned aerial vehicle (UAV) reconnaissance radar is used to acquire radar track information of multiple targets; The information fusion system is used to perform coarse classification and threat level assessment of targets based on the radar track information. Targets with a recognition confidence level meeting a preset threshold and a threat level reaching a preset level are marked as key targets. The target track is sent to the integrated control terminal module, which generates guidance commands and sends them to the dual-spectral photoelectric instrument. The system also receives target azimuth / elevation information and performs asymmetric weighted fusion of the radar track information and target azimuth / elevation information to generate a high-precision fused track. Finally, it receives multispectral images for fine classification and identification, obtaining a dual-spectral image with type identification information. The dual-spectral optoelectronic instrument includes a visible light sensor and an infrared sensor, which are used to receive guidance commands, automatically adjust the lens based on the guidance commands, acquire multispectral images and target azimuth / elevation angle information and send them to the information fusion system; The integrated control terminal is used to display the fused target situation information; the target situation information includes at least a high-precision fused track, a bispectral image with type identification information, identification confidence level, and threat level; A network switch is used for information exchange and transmission between various components.

[0042] Optionally, the network switch supports multicast communication protocols. The UAV reconnaissance radar, the dual-spectrum optoelectronic instrument, the information fusion system, and the integrated control terminal interact with each other through a preset multicast address group, enabling the parallel distribution of radar raw data stream, optoelectronic angle stream, image feature stream, and fused situational commands within the local area network.

[0043] In some embodiments, the integrated radar-visual low-altitude UAV detection and intelligent situational awareness method and system provided in this application is implemented through the following six steps. Figure 3 This application provides an embodiment of an integrated radar-visual low-altitude UAV detection and intelligent situational awareness system, as shown in the following diagram. Figure 3 : Step 1: Radar Target Detection and Tracking The UAV reconnaissance radar searches a designated airspace to acquire motion information such as target distance, azimuth, and speed. The system employs TWS (Track While Scan) technology to maintain the search operation while predicting the trajectories and updating the status of multiple detected targets, thus forming radar target tracks.

[0044] Step Two: Target Coarse Identification and Threat Level Assessment The UAV reconnaissance radar sends the target tracking track to the information fusion system through a network switch. The information fusion system performs coarse classification and threat level assessment based on the target radar track data. Figure 4 The flowchart for target identification and threat level assessment provided in the embodiments of this application is as follows: Figure 4 . Specifically, it includes: 1. Data preprocessing based on small sample augmentation: To address the issues of scarce training samples for UAVs in low-altitude airspace and the extreme imbalance between the number of UAV / bird samples, GANs (Generative Adversarial Networks) and data augmentation techniques are used to expand the small sample flight path data of UAVs, thereby improving the model's generalization ability under extreme conditions.

[0045] 2. Hybrid deep learning network recognition: The TCN (Temporal Convolutional Network) channel extracts long-term temporal features of the target trajectory using a multi-scale parallel TCN structure. This channel consists of three parallel branches: short, medium, and long. Each branch consists of two layers of causal convolution and residual connections, and weight normalization is introduced to normalize the weights; The three branches use progressively larger kernel sizes (2, 3, 5) and the same dilation rate. This allows for differentiated configuration of the receptive field, with a total dimension of 128 after splicing the features of each branch.

[0046] [1,2] are the dilation coefficients specified for the first and second convolutional layers, with a one-to-one correspondence between the two layers. All TCN branches share this configuration. First convolutional layer: dilation rate = 1, second convolutional layer: dilation rate = 2. Dilation rate = 1 indicates normal convolution without skipping points, resulting in a smaller viewing range. Dilation rate = 2 indicates sampling at a 1-point interval, allowing for a longer viewing range and adapting to long-term time-series tracks.

[0047] Transformers channel: Utilizes a multi-head self-attention mechanism to capture the global correlation between long-term motion features of the target.

[0048] It consists of two stacked encoder layers, each containing eight self-attention modules (each attention head has a dimension of 16) and a feedforward neural network based on the GELU activation function; Layer Normalization is used between layers to normalize the data, resulting in a final output dimension of 128.

[0049] Fusion recognition module: Multi-strategy pooling: Extracts global statistical features through three parallel mechanisms: attention pooling, mean pooling, and standard deviation pooling; Feature concatenation and enhancement: The three pooling features are concatenated into a 384-dimensional vector, and the dimension is reduced from 384 to 128 through a feature enhancement network with residual connections; Classification output: Feature mapping is performed through 3 fully connected layers (dimensions of 128, 64, and 32 respectively), and finally the target classification probability is output through the Softmax function.

[0050] 3. Calculation of confidence level: Output the class probability of the target using a hybrid deep learning network To achieve stable recognition, the system employs a confidence smoothing algorithm based on a time-domain sliding window. Let... For the first The overall confidence level of the frame, the first Overall confidence level of frames The calculation formula is:

[0051] in, This indicates that the 5 most recent sampling points (i.e., "5-point identification") are taken. Let be the time decay weighting coefficient, and satisfy . By setting The system can quickly identify the target type and filter out false alarms caused by momentary jitter.

[0052] 4. Target threat level assessment: Evaluation index vector : (Timeliness Component): Based on (Time to Closest Point of Approach, estimated time to reach the closest approach point). The calculation formula is as follows: This reflects the urgency of the target harassment; the shorter the time, the closer the component value is to 1.

[0053] (Spatial components): based on (Closest Point of Approach). This reflects the directness of the target trajectory towards the core area of ​​the pre-defined protected target. The smaller the distance, the higher the component value. The larger.

[0054] (Attribute Components): Target type weights given by the recognition algorithm. Drones are considered to pose a higher threat, while birds pose a lower threat.

[0055] (Maneuvering component): Measured using the curvature and rate of change of acceleration of the target's trajectory. The more dramatic the maneuver (such as an abrupt dive or rapid change of direction), the higher its potential hostility weight.

[0056] Inaccurate identification and a conservative approach to threat assessment are necessary.

[0057] Construct a comprehensive threat assessment function :

[0058] in: The preset indicator weighting coefficients are dynamically adjusted according to the application scenario. : Identify the confidence level correction factor. When identifying the confidence level... hour, A positive value is used for "conservative compensation" to ensure that the threat assessment of ambiguous targets is accurate. Function: Smoothly maps the threat level calculation results to... Interval.

[0059] Level determination: System preset hard threshold and The calculated Value comparison with threshold: like The threat level is determined to be "high"; like It was determined to be "medium"; like It was judged as "low".

[0060] Step 3: Marking and transmitting key targets: The information fusion system processes the coarse identification and threat level assessment results. When the target's identification confidence is greater than 85% and the threat level is "high", it is marked as a key target. The target's track with identification type, identification confidence and threat level is sent to the control terminal for highlighting. At the same time, the guidance command for the target is sent to the dual-spectrum photoelectric instrument. Step 4: Weighted Fusion and Coordinate Transformation Figure 5 The weighted fusion and coordinate transformation process provided in the embodiments of this application, such as Figure 5 The information fusion system will transmit the target optical azimuth angle sent by the photoelectric instrument. Optical pitch angle Target radar azimuth angle contained in radar track Radar elevation angle Perform weighted fusion processing: ;

[0061] in, Since the angle measurement accuracy of the photoelectric instrument is significantly higher than that of the radar, its angle measurement fusion weight should also be greater than that of the radar angle measurement fusion weight. Here, we set the photoelectric angle measurement fusion weight as follows: Radar angle fusion weights ; The fused polar coordinate data is then processed. : Sensor coordinate system Carrier coordinate system (corrected pitch, roll, and yaw angles); Carrier coordinate system Geodetic rectangular coordinate system; Geodetic rectangular coordinate system WGS-84 (Geographic Coordinate System); Finally, the high-precision trajectory of the key targets is obtained and sent to the terminal control system.

[0062] Step 5: Precise recognition of visible / infrared images based on YOLOv8: The dual-spectral photoelectric instrument performs visible light and infrared imaging on the target to acquire multispectral images, and sends the image information to the information fusion system through a switch. The information fusion system performs secondary visual feature recognition and verification on the image using the YOLOv8 (You Only Look Once version 8, the eighth-generation real-time target detection algorithm) model, and sends the target image with type identification information to the terminal control system through the switch.

[0063] Step Six: Overall Situation Display The terminal control system receives high-precision fused target tracks, dual-spectral images with target type markings, target threat levels, identification confidence levels, and other information, and renders them in real time on the terminal software to display comprehensive low-altitude situational information within the monitored area for command and control personnel.

[0064] This application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the methods provided in the above embodiments.

[0065] This application also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the methods provided in the above embodiments.

[0066] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

[0067] Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for integrated radar-visual low-altitude UAV detection and intelligent situational awareness, characterized in that, The method includes: Acquire radar track information for multiple targets; Based on the radar track information, targets are coarsely classified and identified, and threat levels are assessed. Targets whose identification confidence meets a preset threshold and whose threat level reaches a preset level are marked as key targets, and guidance instructions are generated. The camera automatically adjusts based on guidance commands to acquire target azimuth / elevation information, and performs asymmetric weighted fusion of radar track information with target azimuth / elevation information to generate a high-precision fused track. Multispectral images are acquired and subjected to precise classification and identification to obtain bispectral images with type identification information; The fused target situation information is displayed; the target situation information includes at least a high-precision fused track, a bispectral image with type identification information, identification confidence level, and threat level.

2. The method according to claim 1, characterized in that, The acquisition of radar track information for multiple targets includes: By using unmanned aerial vehicle (UAV) reconnaissance radar to scan low-altitude airspace to obtain distance, angle, and speed information of moving targets, and then processing the data through coordinate transformation and TWS tracking, radar track information of multiple targets is generated.

3. The method according to claim 1, characterized in that, The coarse classification and threat level assessment of targets based on the radar track information includes: After preprocessing radar track data using data augmentation techniques, coarse target classification and identification are performed based on a hybrid deep learning network. The identification confidence is calculated using a confidence smoothing algorithm based on a time-domain sliding window. At the same time, threat level assessment is performed based on a comprehensive threat level quantification evaluation function driven by multiple factors.

4. The method according to claim 3, characterized in that, The hybrid deep learning network includes a TCN channel and a Transformers channel: The TCN channel uses a multi-scale parallel structure to extract the temporal local features of the radar track; The Transformers channel uses a multi-head self-attention mechanism to extract long-term global correlation features of radar tracks; The features extracted from the TCN channel and the Transformers channel are fused together, and the target classification probability is output through multi-strategy pooling, feature concatenation and enhancement, and fully connected layer mapping.

5. The method according to claim 1, characterized in that, The step of asymmetrically weighted fusion of radar track information and target azimuth / elevation information to generate a high-precision fused track includes: The target azimuth / elevation information includes photoelectric azimuth, photoelectric elevation, and photoelectric image information; The radar track information includes radar azimuth, radar elevation, and range; Receive photoelectric azimuth, photoelectric elevation, and photoelectric image information transmitted by the dual-spectrum photoelectric instrument; perform weighted fusion of the photoelectric azimuth with the radar azimuth in the radar track information to obtain a fused azimuth; perform weighted fusion of the photoelectric elevation with the radar elevation in the radar track information to obtain a fused elevation. Based on the fused azimuth angle, the fused elevation angle, and the range information in the radar track information, a multi-level coordinate transformation is performed to generate a high-precision fused track.

6. The method according to claim 5, characterized in that, The process of performing multi-level coordinate transformation based on the fused azimuth angle, the fused elevation angle, and the range information in the radar track information to generate a high-precision fused track includes: The azimuth angle, elevation angle, and radar distance information are extracted and fused to form polar coordinate data. The polar coordinate data is then transformed from the sensor coordinate system to the carrier coordinate system, the geodetic rectangular coordinate system, and the geographic coordinate system in a multi-level coordinate transformation to generate a high-precision fused trajectory, which is then sent to the terminal control system.

7. The method according to claim 1, characterized in that, The acquired multispectral images are subjected to precise classification and identification to obtain a bispectral image with type identification information, including: Multispectral images are obtained by imaging the target in visible light and infrared light using a dual-spectral photoelectric instrument. The target detection model is invoked to perform secondary visual feature recognition and verification on the multispectral image, and the category label of the target is output. The category label, along with the photoelectric azimuth and photoelectric elevation angles with the same timestamp, are used as type identification information. The bispectral image with the type identification information is then sent to the terminal control system via a switch.

8. A radar-visual integrated low-altitude UAV detection and intelligent situational awareness system, characterized in that, include: Unmanned aerial vehicle (UAV) reconnaissance radar is used to acquire radar track information of multiple targets; The information fusion system is used to perform coarse classification and identification of targets and threat level assessment based on the radar track information. Targets whose identification confidence meets the preset threshold and whose threat level reaches the preset level are marked as key targets. The target track is sent to the integrated control terminal module, and guidance commands are generated and sent to the dual-spectrum photoelectric instrument. The system receives target azimuth / elevation information and performs asymmetric weighted fusion of radar track information with target azimuth / elevation information to generate a high-precision fused track; it also receives multispectral images and performs fine classification and identification to obtain a bispectral image with type identification information. The dual-spectral optoelectronic instrument includes a visible light sensor and an infrared sensor, which are used to receive guidance commands, automatically adjust the lens based on the guidance commands, acquire multispectral images and target azimuth / elevation angle information and send them to the information fusion system; The integrated control terminal is used to display the fused target situation information; The target situation information includes at least a high-precision fused flight track, a bispectral image with type identification information, identification confidence level, and threat level. A network switch is used for information exchange and transmission between various components.

9. The system according to claim 8, characterized in that, The network switch supports multicast communication protocols. The UAV reconnaissance radar, the dual-spectrum photoelectric instrument, the information fusion system, and the integrated control terminal interact with each other through a preset multicast address group, enabling the parallel distribution of radar raw data stream, photoelectric angle stream, image feature stream, and fused situational commands within the local area network.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 7.

11. An electronic 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 program, it implements the method as described in any one of claims 1 to 7.

Images