Low-altitude unmanned aerial vehicle target identification and early warning method based on multi-source perception

Abstract

This invention discloses a method for low-altitude UAV target identification and early warning based on multi-source perception, relating to the field of low-altitude security and control technology. The method includes: acquiring low-altitude perception data from radar, photoelectric, and passive radio frequency sensors, and preprocessing it to form standardized multi-source observation data; establishing a source quality assessment model to calculate source quality factors and performing spatiotemporal alignment of the data; outputting the target fusion trajectory state quantity and confidence level through gated probabilistic correlation and adaptive weighted fusion, and extracting multi-dimensional features such as radar micro-Doppler, photoelectric shape, and radio frequency fingerprint to construct a fusion feature vector input classifier to identify target categories; establishing a threat scoring function by combining trajectory prediction and behavioral pattern analysis, outputting threat level and confidence level, and executing graded early warning based on dynamic thresholds. This invention solves the problems of unstable detection by a single sensor, low identification accuracy, and delayed early warning in low-altitude environments, improving the robustness and predictability of low-altitude UAV target control.

Description

Technical Field

[0001] This invention belongs to the field of low-altitude safety control and intelligent sensing technology, specifically relating to a method for low-altitude UAV target identification and early warning based on multi-source sensing. Background Technology

[0002] As the application of low-altitude airspace continues to deepen in scenarios such as urban operation, logistics distribution, emergency rescue, and public safety, the number and frequency of low-altitude drones have increased significantly. Low-altitude safety and control has become an important part of the three-dimensional spatial governance of smart cities. However, low-altitude scenarios are characterized by low altitude, strong clutter, small targets, high maneuverability, strong obstruction, and multiple electromagnetic interferences, making it difficult to detect, track, identify, and warn of targets in actual operation. On the one hand, drones are mixed with non-target altitudes such as building edge echoes, ground object reflections, and flocks of birds, leading to missed detections, false detections, and discontinuous flight paths. On the other hand, drone flight paths are variable and behavior patterns are complex, making it difficult to reflect their true intentions in a single observation.

[0003] To address the aforementioned issues, existing technologies typically employ single sensing methods such as radar, photoelectric sensors, or passive radio frequency sensors for detection and identification, or enhance coverage by overlaying multi-source information. However, single-source sensing is susceptible to clutter, obstruction, and electromagnetic interference in complex low-altitude environments, resulting in insufficient stability. Furthermore, existing multi-source fusion schemes often rely on fixed weights, empirical rules, or simple voting, lacking adaptive modeling for the asynchronous nature of multiple sources, differences in accuracy, and dynamic changes in reliability. When some sensing sources degrade or become missing, fusion drift and false alarms are likely to occur. Summary of the Invention

[0004] To address the aforementioned problems in the existing technology, this invention provides a method for low-altitude unmanned aerial vehicle (UAV) target identification and early warning based on multi-source sensing. The technical problem to be solved by this invention is achieved through the following technical solution: This invention provides a method for low-altitude unmanned aerial vehicle (UAV) target identification and early warning based on multi-source sensing, comprising: Low-altitude sensing data from different sensing sources are acquired, and standardized multi-source observation data are obtained after preprocessing. A source quality assessment model is established based on the multi-source observation data to calculate the source quality factor, and the multi-source observation data is further spatiotemporally aligned. The aligned observation data from each sensing source is matched with the target trajectory, and the matched multi-source observation data is adaptively weighted and fused to output the target fused trajectory state and fusion confidence. Based on the target fusion trajectory state quantity, feature extraction is performed on the matched multi-source observation data and a fusion feature vector is constructed. The fusion feature vector is then input into a classifier to obtain the identified target category and the corresponding identification confidence level.

[0005] In one embodiment of the present invention, the step of acquiring low-altitude sensing data from different sensing sources and obtaining standardized multi-source observation data after preprocessing includes: Simultaneously acquire low-altitude sensing data collected by different sensing sources, including radar, photoelectric and passive radio frequency sensors; Differential preprocessing is performed on the low-altitude sensing data from each sensing source to obtain preliminary observation vectors; A timestamp is added to the preliminary observation vector, and the corresponding noise covariance matrix and sensor state vector are calculated to form standardized multi-source observation data.

[0006] In one embodiment of the present invention, the step of establishing a source quality assessment model based on the multi-source observation data to calculate the source quality factor, and further performing spatiotemporal alignment of the multi-source observation data, includes: A source quality assessment model is established, and the source quality factor is calculated based on the signal-to-noise ratio, visibility, interference intensity, and historical consistency of the observation data of each sensing source in the multi-source observation data: ; In the formula, For sensing source The source quality factor, For the Sigmoid function, For weight parameters, , , , These are the normalized signal-to-noise ratio, visibility, interference intensity, and historical consistency index, respectively. For time indexing, , Indicates radar, Indicates photoelectric, Indicates a passive radio frequency sensor; Interpolation compensation is performed on the multi-source observation data based on the timestamp to achieve time alignment, and spatial alignment under different sensing source coordinate systems is achieved through extrinsic parameter calibration and coordinate transformation.

[0007] In one embodiment of the present invention, the steps of matching aligned observation data from various sensing sources with the trajectory of the target, adaptively weighting and fusing the matched multi-source observation data, and outputting the target fused trajectory state and fusion confidence include: Calculate the trajectory of the target Aligned sensory source The observation data at time residual and covariance And based on Mahalanobis distance Determine the aligned sensory source Does the observation data fall within the trajectory? Gated domain; Calculate the aligned sensory source Observational data With trajectory Association probability and combined with source quality factor Generate sensor source fusion weights and trajectory For the aligned sensing source Observational data Fusion weights ; Based on the fusion weight Establish equivalent observation vector and equivalent observation covariance ; Based on the Kalman filter algorithm, according to the equivalent observation vector and equivalent observation covariance Determine the target fusion trajectory state at the current moment. And based on residuals Fusion weights with sensor sources Calculate the output fusion confidence score .

[0008] In one embodiment of the present invention, the step of extracting features from the matched multi-source observation data and constructing a fused feature vector includes: Using the target fusion trajectory state quantity As a constraint, in the matched sensory source The observation segments associated with the target are extracted from the observation data, and feature vectors are obtained by extracting features from the observation segments of each sensing source. Based on sensing source The mean vector of the eigenvectors and standard deviation vector For the sensing source The feature vectors are normalized and mapped to obtain normalized features. ; In the time window The normalized features are aggregated to obtain a fused feature vector. Indicates the length of the time window.

[0009] In one embodiment of the present invention, the target fused trajectory state quantity is used. As a constraint, in the matched sensory source The steps of extracting observation segments associated with the target from the observation data, and extracting features from the observation segments of each sensing source to obtain feature vectors include: Using the target fusion trajectory state quantity To constrain this, observation segments associated with the target are extracted from the aligned radar observation data, aligned electro-optical observation data, and aligned radio frequency observation data, respectively. , and ; From the observation segments respectively , and In the process, micro-Doppler and scattering feature vectors are extracted. Photoelectric shape and infrared thermal feature vector and radio frequency standards and fingerprint feature vectors .

[0010] In one embodiment of the present invention, the fused feature vector is represented as: ; in, Indicates transpose. , , , They are respectively , , The corresponding fusion weights and satisfy , Indicates time window The micro-Doppler and scattering eigenvectors described inside Quantity, Indicates time window The photoelectric shape and infrared thermal feature vector described inside Quantity, Indicates time window Internal radio frequency standard and fingerprint feature vector The quantity.

[0011] In one embodiment of the present invention, after the steps of extracting features from the matched multi-source observation data and constructing a fused feature vector based on the target fused trajectory state quantity, and inputting the fused feature vector into a classifier to obtain the identified target category and the corresponding identification confidence, the method further includes: Using the state transition matrix, the future trajectory state of the target at the current moment is predicted. Target fusion trajectory state variables for each step: ; In the formula, For the index of the predicted steps, , To predict the number of steps, This is the state transition matrix; Determine the target and protection zone within the prediction time domain Minimum approximation distance: ; In the formula, For the predicted target fusion trajectory state quantity Positional components in It is a distance function; Based on the predicted target fusion trajectory state quantity Extraction of speed acceleration angular velocity of heading ,high and trajectory curvature index vector To calculate threat score ; Based on threat score Threat level output with tiered threshold And based on the threat score Threat confidence is calculated based on the difference between the threshold values ​​for each level. ; According to the threat level and the threat confidence level Issue an early warning.

[0012] In one embodiment of the present invention, ; In the formula, For the Sigmoid function, For distance-risk mapping function, For category risk mapping function, , , , , , , , These are preset constants, , , , Speeds acceleration trajectory curvature and historical behavioral risks The normalized value, Indicates the confidence level of identification. Indicates the target category.

[0013] Compared with the prior art, the beneficial effects of the present invention are as follows: (1) This invention utilizes radar, photoelectric sensors, and passive radio frequency for multi-source sensing. It establishes a source quality assessment model to quantify the reliability of each sensing source in real time and combines spatiotemporal alignment and adaptive weighted fusion mechanisms. Compared with the shortcomings of existing technologies where a single sensor is susceptible to weather, lighting, or electromagnetic interference, leading to missed detections, false alarms, and discontinuous tracks, this invention can fully leverage the complementary advantages of radar's all-weather capability, photoelectric high resolution, and passive radio frequency detection. Even when the performance of a certain sensing source deteriorates, it can still maintain stable tracking capability by relying on other sensing sources, significantly reducing the missed detection rate and false alarm rate, and ensuring the continuity of the target track.

[0014] (2) This invention extracts features from the matched multi-source observation data and constructs a fusion feature vector. It aggregates the micro-Doppler features of radar, the shape and infrared features of photoelectric devices, and the standard and fingerprint features of radio frequency across modes, breaking through the limitations of traditional simple superposition or fixed weight voting. Through normalization processing and dynamic weight allocation, it can more comprehensively characterize UAV targets, which is conducive to improving the accuracy of target classification and recognition.

[0015] (3) When some sensing sources degrade or are missing, existing multi-source fusion schemes are prone to fusion drift and false alarms. This invention introduces a sensor degradation detection mechanism, defining reliability indicators and overall health. When the quality of a certain sensing source is detected to be below a threshold ( When the radar is interfered with or the photoelectric sensor is blocked, the weight redistribution and warning threshold adjustment are automatically triggered. This can smoothly reduce the contribution of the faulty sensor in scenarios where some sensing sources fail, such as radar interference or photoelectric interference, thus ensuring the robustness and reliability of the warning decision in complex environments.

[0016] (4) The present invention is based on the predicted target fusion trajectory state quantity Calculate threat score Furthermore, a tiered early warning strategy was designed, which no longer only alerts based on the current location but can also predict the future. By combining time suppression and spatial merging rules to detect the approach risk of targets within the protected area, high-risk targets can be detected in advance, and duplicate and redundant alarms can be effectively filtered out, which significantly improves the predictability and efficiency of security decisions.

[0017] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0018] Figure 1 This is a flowchart of a low-altitude UAV target identification and early warning method based on multi-source perception provided in an embodiment of the present invention; Figure 2This is a schematic diagram of a low-altitude UAV target recognition and early warning method based on multi-source perception provided in an embodiment of the present invention. Detailed Implementation

[0019] The present invention will be further described in detail below with reference to specific embodiments, but the implementation of the present invention is not limited thereto.

[0020] Figure 1 This is a flowchart of a low-altitude UAV target recognition and early warning method based on multi-source perception provided in an embodiment of the present invention. Figure 2 This is a schematic diagram illustrating a process of a low-altitude unmanned aerial vehicle (UAV) target identification and early warning method based on multi-source perception provided in an embodiment of the present invention. Figures 1-2 As shown, this embodiment of the invention provides a method for low-altitude UAV target identification and early warning based on multi-source perception, including: S1. Acquire low-altitude sensing data from different sensing sources, and obtain standardized multi-source observation data after preprocessing.

[0021] In this embodiment, low-altitude sensing data from different sensing sources refers to the data set obtained by three different types of sensors—radar, photoelectric, and passive radio frequency—detecting the same low-altitude area within the same time period.

[0022] Specifically, step S1 includes: S11. Simultaneously acquire low-altitude sensing data collected by different sensing sources, including radar, photoelectric and passive radio frequency sensors.

[0023] S12. Differentiated preprocessing of low-altitude sensing data from each sensing source to obtain preliminary observation vectors.

[0024] S13. Add timestamps to the preliminary observation vectors, calculate the corresponding noise covariance matrix and sensor state vectors to form standardized multi-source observation data.

[0025] In this embodiment, radar, photoelectric, and passive radio frequency sensors are simultaneously acquired at discrete moments. The collected low-altitude sensing data are denoted as follows: , , Differentiated preprocessing is performed on these heterogeneous data: for radar echo data... By estimating the background clutter power And conduct testing and judgment. To extract candidate targets; for photoelectric data, through a function Generate preprocessed images and output the candidate box set. For radio frequency data, energy statistics are calculated. and with threshold Comparison. Finally, timestamps are appended to all preprocessed low-altitude sensing data. And calculate the noise covariance matrix. and sensor state vector To form standardized multi-source observation data .

[0026] S2. Establish a source quality assessment model based on multi-source observation data to calculate the source quality factor, and further perform spatiotemporal alignment of the multi-source observation data.

[0027] In step S2, a source quality assessment model is established. Based on the signal-to-noise ratio, visibility, interference intensity, and historical consistency of the observation data of each sensing source in the multi-source observation data, the source quality factor is calculated. ; In the formula, For sensing source The source quality factor, For the Sigmoid function, For weight parameters, , , , These are the normalized signal-to-noise ratio, visibility, interference intensity, and historical consistency index, respectively. For time indexing, , Indicates radar, Indicates photoelectric, This indicates a passive radio frequency sensor.

[0028] To address the issue of asynchronous sampling by multiple sensors, interpolation compensation is performed on multi-source observation data based on timestamps to achieve time alignment. ; In the formula, The observation data is time-aligned. and They are time points and Observational data, These are the interpolation coefficients.

[0029] To address the issue of coordinate system inconsistency, spatial alignment between different sensing source coordinate systems is achieved through extrinsic parameter calibration and coordinate transformation. ; in, To unify the coordinate system The coordinates of the point below, For sensing source Point coordinates in a coordinate system To start from the source of perception To a unified coordinate system The rotation matrix, To start from the source of perception arrive Translation matrix; The covariance in a unified coordinate system is: ; in, For sensing source Covariance in coordinate system Let be the Jacobian matrix for coordinate transformation.

[0030] S3. Match the aligned observation data from each sensing source with the target trajectory, and perform adaptive weighted fusion on the matched multi-source observation data to output the target fused trajectory state and fusion confidence.

[0031] Specifically, step S3 includes: S31. Calculate the target's trajectory. Aligned sensory source The observation data at time residual and covariance And based on Mahalanobis distance Determine the aligned sensory source Does the observation data fall within the trajectory? Gated domain.

[0032] For example, ; in, To fuse trajectory state variables, This is the measurement function.

[0033] ; in, To predict covariance, For measuring the Jacobian.

[0034] .

[0035] when It was determined that the signal came from the sensing source. The observation data falls into the trajectory gating region. The threshold value is used for gate control. As an observation dimension, The significance level is indicated by .

[0036] S32. Calculate the aligned sensory source Observational data With trajectory Association probability and combined with source quality factor Generate sensor source fusion weights and trajectory Aligned sensory sources Observational data Fusion weights .

[0037] For example, define the aligned perceptual source Observational data With trajectory The probability of association is: ; in, For the candidate observation data set that passes the gate, For from the source of perception Observational data Association weights, This represents the weight of the missed detection hypothesis.

[0038] The perceptual source fusion weights generated from the source quality factors are expressed as follows: ; Combining association probability Forming a trajectory For the above aligned sensory sources Observational data Fusion weights: .

[0039] S33, Based on Fusion Weights Establish equivalent observation vector and equivalent observation covariance .

[0040] In step S33, the equivalent observation vector Represented as: ; in, For from the source of perception And through the gated observation data set, For from the source of perception Observational data After time alignment, at time Observations; The equivalent observation covariance can be approximated as: ; in, For from the source of perception Observational data After completing time alignment and mapping to moments The corresponding observation noise covariance.

[0041] S34. Based on the Kalman filter algorithm, according to the equivalent observation vector and equivalent observation covariance Determine the target fusion trajectory state at the current moment. And based on residuals Fusion weights with sensor sources Calculate the output fusion confidence score .

[0042] Specifically, the updated target fusion trajectory state is represented as follows: ; in, For Kalman gain, To measure the Jacobian; trajectory At any moment The fusion confidence is expressed as: ; In the formula, This indicates a cropping operation.

[0043] S4. Based on the target fusion trajectory state quantity, feature extraction is performed on the matched multi-source observation data and a fusion feature vector is constructed. The fusion feature vector is then input into the classifier to obtain the identified target category and the corresponding recognition confidence.

[0044] In this embodiment, step S4 includes: S41, Using target fusion trajectory state variables As a constraint, in the matched sensory source The observation segments associated with the target are extracted from the observation data, and feature vectors are obtained by extracting features from the observation segments of each sensing source.

[0045] Specifically, the target fusion trajectory state quantity To constrain this, observation segments associated with the target are extracted from the aligned radar observation data, aligned electro-optical observation data, and aligned radio frequency observation data, respectively. , and Subsequently, differential feature extraction was performed on the observation segments from each sensing source: from radar observation segments Extracting micro-Doppler and scattering eigenvectors From photoelectric observation fragments Extracting photoelectric shape and infrared thermal feature vectors From radio frequency observation segments Extracting radio frequency standard and fingerprint feature vector .

[0046] S42, Based on sensing source The mean vector of the eigenvectors and standard deviation vector For the source of perception The feature vectors are normalized to eliminate dimensional differences, resulting in normalized features. : .

[0047] S43, in the time window The normalized features are aggregated to obtain a fused feature vector. Indicates the length of the time window.

[0048] To enhance the stability of features in the time domain, further adjustments are made to the time window. Internal normalization characteristics Perform sliding aggregation. Specifically, calculate the time window for each sensing source. The characteristic mean within: , , ,in, Indicates time window Internal micro Doppler and scattering eigenvectors Quantity, Indicates time window Internal photoelectric shape and infrared thermal feature vector Quantity, Indicates time window Internal radio frequency standard and fingerprint feature vector The quantity.

[0049] The constructed fusion feature vector is represented as ,in, Indicates transpose. , , , They are respectively , , The corresponding fusion weights and satisfy Finally, the fused feature vector is input into the classifier. The resulting class probability vector is: ; in, For the number of categories, For the target to belong to the first The probability of a class; The output target category is: ; The output recognition confidence level is: ; in, To identify categories, To identify confidence levels.

[0050] Optionally, after extracting features from the matched multi-source observation data and constructing a fused feature vector, and inputting the fused feature vector into a classifier to obtain the identified target category and the corresponding recognition confidence, the method further includes: Using the state transition matrix, the trajectory state variables are fused based on the target at the current time. Predicting the future Target fusion trajectory state variables for each step: ; In the formula, For the index of the predicted steps, , To predict the number of steps, This is the state transition matrix; Determine the target and protection zone within the prediction time domain The minimum approximation distance is used to quantify potential intrusion risk. This distance is determined by minimizing the predicted location from the protected area. The distance function between boundaries is obtained, and the calculation formula is: ; In the formula, For the predicted target fusion trajectory state quantity Positional components in The distance function from a point to a set, where the prediction time domain refers to the future. The time range constituted by the steps Based on the predicted target fusion trajectory state quantity Extraction of speed acceleration angular velocity of heading ,high and trajectory curvature index vector To calculate threat score : ; In the formula, For the Sigmoid function, For distance risk ± mapping function, Distance scale parameters , Indicates the spatial distance between the target and the protected area. For category risk mapping function, , , , , , , , These are preset constants, , , , Speeds acceleration trajectory curvature and historical behavioral risks The normalized value, Indicates the confidence level of identification. Indicates the target category and historical behavioral risk. This can be obtained from statistics such as the number of times the target violated regulations, the number of times it entered sensitive areas, and the duration of its loitering within the past window.

[0051] Finally, based on the threat score Threat level output with tiered threshold : ; In the formula, , , , These represent threat levels from high to low, with the threshold values ​​indicating the level. .

[0052] Based on threat score Threat confidence is calculated based on the difference between the threshold values ​​for each level. According to threat level and threat confidence A warning was issued. This included the threat level. Corresponding threat confidence level Represented as: ; in, For each level threshold, the scaling factor .

[0053] To enable the early warning threshold to adapt to environmental changes, the early warning classification threshold is defined as follows: ( ),in, Based on the threshold, This is a dynamically adjustable item; In order to be with the first The adjustment coefficient corresponding to each threshold. As an environmental risk factor, For overall perceived quality indicators. , is the source importance weight and satisfies The warning level can be mapped based on threshold segmentation: ; in, It is at the warning level. This is the lowest confidence threshold.

[0054] Furthermore, temporal inhibition and spatial merging rules are introduced: the trajectory of the target The last time it was issued was at least at the level of The warning time is Cooldown time is If satisfied If two trajectories exist within the same time window, then the same or lower level warning will not be sent repeatedly. and Its spatial distance is , It is a 2-norm. For trajectory The position vector components, if they satisfy , To merge the distance threshold, two warnings will be merged into a single event output.

[0055] For each sensory source at time Define reliability metrics: ; in, These are the weighting coefficients. For data availability tags, The consistency error between the sensing source and the fused trajectory; when the following conditions are met. ( When the degradation threshold is reached, the sensor is determined. Degradation has occurred. The reallocated fusion weights are: ; in, To adjust the index.

[0056] When the sensing accuracy decreases, a degradation compensation term is added to the above threshold adjustment term: ; in, This is the degradation compensation coefficient. For overall health.

[0057] To avoid frequent and significant changes in weights, a smooth update is introduced. The smoothed weights are: The smoothed threshold is ,in, This is the smoothing coefficient.

[0058] As can be seen from the above embodiments, the beneficial effects of the present invention are as follows: (1) This invention utilizes radar, photoelectric sensors, and passive radio frequency for multi-source sensing. It establishes a source quality assessment model to quantify the reliability of each sensing source in real time and combines spatiotemporal alignment and adaptive weighted fusion mechanisms. Compared with the shortcomings of existing technologies where a single sensor is susceptible to weather, lighting, or electromagnetic interference, leading to missed detections, false alarms, and discontinuous tracks, this invention can fully leverage the complementary advantages of radar's all-weather capability, photoelectric high resolution, and passive radio frequency detection. Even when the performance of a certain sensing source deteriorates, it can still maintain stable tracking capability by relying on other sensing sources, significantly reducing the missed detection rate and false alarm rate, and ensuring the continuity of the target track.

[0059] (2) This invention extracts features from the matched multi-source observation data and constructs a fusion feature vector. It aggregates the micro-Doppler features of radar, the shape and infrared features of photoelectric devices, and the standard and fingerprint features of radio frequency across modes, breaking through the limitations of traditional simple superposition or fixed weight voting. Through normalization processing and dynamic weight allocation, it can more comprehensively characterize UAV targets, which is conducive to improving the accuracy of target classification and recognition.

[0060] (3) When some sensing sources degrade or are missing, existing multi-source fusion schemes are prone to fusion drift and false alarms. This invention introduces a sensor degradation detection mechanism, defining reliability indicators and overall health. When the quality of a certain sensing source is detected to be below a threshold ( When the radar is interfered with or the photoelectric sensor is blocked, the weight redistribution and warning threshold adjustment are automatically triggered. This can smoothly reduce the contribution of the faulty sensor in scenarios where some sensing sources fail, such as radar interference or photoelectric interference, thus ensuring the robustness and reliability of the warning decision in complex environments.

[0061] (4) The present invention is based on the predicted target fusion trajectory state quantity Calculate threat score Furthermore, a tiered early warning strategy was designed, which no longer only alerts based on the current location but can also predict the future. By combining time suppression and spatial merging rules to detect the approach risk of targets within the protected area, high-risk targets can be detected in advance, and duplicate and redundant alarms can be effectively filtered out, which significantly improves the predictability and efficiency of security decisions.

[0062] In the description of this invention, the terms "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to a specific feature, structure, material, or characteristic described in connection with that embodiment or example, which is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. In addition, those skilled in the art can combine and integrate the different embodiments or examples described in this specification.

[0063] Although this application has been described herein in conjunction with various embodiments, other variations of the disclosed embodiments can be understood and implemented by those skilled in the art in carrying out the claimed application by reviewing the accompanying drawings, the disclosure, and the appended claims.

[0064] The above description, in conjunction with specific preferred embodiments, provides a further detailed explanation of the present invention. However, it should not be construed that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art, various simple deductions or substitutions can be made without departing from the inventive concept, and all such modifications and substitutions should be considered within the scope of protection of the present invention.

Claims

1. A method for low-altitude unmanned aerial vehicle (UAV) target identification and early warning based on multi-source sensing, characterized in that, include: Low-altitude sensing data from different sensing sources are acquired, and standardized multi-source observation data are obtained after preprocessing. A source quality assessment model is established based on the multi-source observation data to calculate the source quality factor, and the multi-source observation data is further spatiotemporally aligned. The aligned observation data from each sensing source is matched with the target trajectory, and the matched multi-source observation data is adaptively weighted and fused to output the target fused trajectory state and fusion confidence. Based on the target fusion trajectory state quantity, feature extraction is performed on the matched multi-source observation data and a fusion feature vector is constructed. The fusion feature vector is then input into a classifier to obtain the identified target category and the corresponding identification confidence level.

2. The method according to claim 1, characterized in that, The steps for acquiring low-altitude sensing data from different sensing sources and obtaining standardized multi-source observation data after preprocessing include: Simultaneously acquire low-altitude sensing data collected by different sensing sources, including radar, photoelectric and passive radio frequency sensors; Differential preprocessing is performed on the low-altitude sensing data from each sensing source to obtain preliminary observation vectors; A timestamp is added to the preliminary observation vector, and the corresponding noise covariance matrix and sensor state vector are calculated to form standardized multi-source observation data.

3. The method according to claim 1, characterized in that, The steps of establishing a source quality assessment model based on the multi-source observation data to calculate the source quality factor, and further performing spatiotemporal alignment of the multi-source observation data, include: A source quality assessment model is established, and the source quality factor is calculated based on the signal-to-noise ratio, visibility, interference intensity, and historical consistency of the observation data of each sensing source in the multi-source observation data. ； In the formula, For sensing source The source quality factor, For the Sigmoid function, For weight parameters, , , , These are the normalized signal-to-noise ratio, visibility, interference intensity, and historical consistency index, respectively. For time indexing, , Indicates radar, Indicates photoelectric, Indicates a passive radio frequency sensor; Interpolation compensation is performed on the multi-source observation data based on the timestamp to achieve time alignment, and spatial alignment under different sensing source coordinate systems is achieved through extrinsic parameter calibration and coordinate transformation.

4. The method according to claim 3, characterized in that, The steps of matching aligned observation data from various sensing sources with the target's trajectory, adaptively weighting and fusing the matched multi-source observation data, and outputting the target fused trajectory state and fusion confidence include: Calculate the trajectory of the target Aligned sensory source The observation data at time residual and covariance And based on Mahalanobis distance Determine the aligned sensory source Does the observation data fall within the trajectory? Gated domain; Calculate the aligned sensory source Observational data With trajectory Association probability and combined with source quality factor Generate sensor source fusion weights and trajectory For the aligned sensing source Observational data Fusion weights ; Based on the fusion weight Establish equivalent observation vector and equivalent observation covariance ; Based on the Kalman filter algorithm, according to the equivalent observation vector and equivalent observation covariance Determine the target fusion trajectory state at the current moment. And based on residuals Fusion weights with sensor sources Calculate the output fusion confidence score .

5. The method according to claim 4, characterized in that, Based on the target fused trajectory state quantity, the steps of extracting features from the matched multi-source observation data and constructing a fused feature vector include: Using the target fused trajectory state quantity As a constraint, in the matched sensory source The observation segments associated with the target are extracted from the observation data, and feature vectors are obtained by extracting features from the observation segments of each sensing source. Based on sensing source The mean vector of the eigenvectors and standard deviation vector For the sensing source The feature vectors are normalized and mapped to obtain normalized features. ; In the time window The normalized features are aggregated to obtain a fused feature vector. Indicates the length of the time window.

6. The method according to claim 5, characterized in that, Using the target fused trajectory state quantity As a constraint, in the matched sensory source The steps of extracting observation segments associated with the target from the observation data, and extracting features from the observation segments of each sensing source to obtain feature vectors include: Using the target fused trajectory state quantity To constrain this, observation segments associated with the target are extracted from the aligned radar observation data, aligned electro-optical observation data, and aligned radio frequency observation data, respectively. , and ; From the observation segments respectively , and In the process, micro-Doppler and scattering feature vectors are extracted. Photoelectric shape and infrared thermal feature vector and radio frequency standards and fingerprint feature vectors .

7. The method according to claim 6, characterized in that, The fused feature vector is represented as follows: ， in, Indicates transpose. , , , They are respectively , , The corresponding fusion weights and satisfy , Indicates time window The micro-Doppler and scattering eigenvectors described inside Quantity, Indicates time window The photoelectric shape and infrared thermal feature vector described inside Quantity, Indicates time window Internal radio frequency standard and fingerprint feature vector The quantity.

8. The method according to claim 1, characterized in that, Based on the target fusion trajectory state quantity, after extracting features from the matched multi-source observation data and constructing a fusion feature vector, and inputting the fusion feature vector into a classifier to obtain the identified target category and the corresponding identification confidence, the method further includes: Using the state transition matrix, the future trajectory state of the target at the current moment is predicted. Target fusion trajectory state variables for each step: ； In the formula, For the index of the predicted steps, , To predict the number of steps, This is the state transition matrix; Determine the target and protection zone within the prediction time domain Minimum approximation distance: ； In the formula, For the predicted target fusion trajectory state quantity Positional components in It is a distance function; Based on the predicted target fusion trajectory state quantity Extraction of speed acceleration angular velocity of heading ,high and trajectory curvature index vector To calculate threat score ; Based on threat score Threat level output with tiered threshold And based on the threat score Threat confidence is calculated based on the difference between the threshold values ​​for each level. ; According to the threat level and the threat confidence level Issue an early warning.

9. The method according to claim 8, characterized in that, ； In the formula, For the Sigmoid function, For distance-risk mapping function, For category risk mapping function, , , , , , , , These are preset constants, , , , Speeds acceleration trajectory curvature and historical behavioral risks The normalized value, Indicates the confidence level of identification. Indicates the target category.

Images