A positioning and tracking method and system for low-altitude unmanned aerial vehicles (UAVs)

Abstract

This invention relates to the field of unmanned aerial vehicle (UAV) detection and tracking technology, specifically to a method and system for locating and tracking low-altitude UAVs. The method includes the following steps: acquiring the UAV's acoustic spectrum time-series data and radar spot data; extracting a first acoustic spectrum feature parameter and a second acoustic spectrum feature parameter; extracting a first radar feature parameter and a second radar feature parameter; determining a preliminary position estimate of the UAV based on a first judgment condition; determining an optimized position estimate of the UAV based on a second judgment condition; and tracking the UAV using an improved particle filter algorithm to obtain the UAV's final position and tracking trajectory. This invention, through heterogeneous sensor data fusion, environmentally adaptive feature extraction, multi-level intelligent judgment optimization, and improved particle filter tracking, effectively improves the accuracy, robustness, and real-time performance of UAV detection in complex environments, significantly enhancing regional monitoring and countermeasure capabilities.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) detection and tracking technology, specifically to a positioning and tracking method and system for low-altitude UAVs. Background Technology

[0002] In recent years, with the rapid popularization of drone technology, while bringing convenience, it has also triggered a series of public safety issues such as privacy leaks, airspace security, and even illegal transportation. Therefore, developing efficient and reliable drone detection and tracking technologies has become an urgent need in the field of low-altitude security. Currently, mainstream detection methods mainly rely on single or simple combinations of sensor systems such as radar, photoelectric, and acoustic sensors, aiming to achieve identification and location by analyzing the physical characteristics of targets.

[0003] However, existing UAV detection and tracking solutions still have significant shortcomings. First, radar systems face challenges from multipath effects and ground clutter in complex urban environments, limiting their ability to detect low-altitude, slow-moving, small targets. Acoustic systems are susceptible to environmental noise pollution, and sound wave propagation is easily affected by atmospheric conditions such as temperature, humidity, and air pressure, leading to decreased positioning accuracy. Second, existing multi-sensor fusion methods are mostly simple data-layer or decision-layer fusions, lacking intelligent evaluation of data quality and consistency. When data from different sensors conflict or there are significant differences in signal-to-noise ratio, the reliability of the fusion result is poor, and it may even lead to performance inferior to that of a single sensor. Furthermore, existing tracking algorithms, such as standard Kalman filtering or particle filtering, suffer from significant model mismatch problems when dealing with the highly maneuverable movements of UAVs, making it difficult to simultaneously guarantee tracking accuracy and smoothness.

[0004] To address the aforementioned technical shortcomings, this invention deeply integrates acoustic and radar sensor information and introduces a series of adaptive processing and intelligent judgment mechanisms, aiming to significantly improve the robustness, positioning accuracy, and tracking and prediction capabilities of UAV detection in complex environments. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a positioning and tracking method and system for low-altitude unmanned aerial vehicles (UAVs).

[0006] In a first aspect, the present invention provides a positioning and tracking method for a low-altitude unmanned aerial vehicle (UAV), comprising the following steps:

[0007] Acquire the acoustic spectrum time-series data and radar spot data of the UAV; using the acoustic spectrum time-series data, extract the first acoustic spectrum feature parameter and the second acoustic spectrum feature parameter; using the radar spot data, extract the first radar feature parameter and the second radar feature parameter; based on the first acoustic spectrum feature parameter and the first radar feature parameter, determine the preliminary position estimate of the UAV through a first judgment condition; based on the preliminary position estimate, determine the optimized position estimate of the UAV through a second judgment condition; based on the optimized position estimate, use an improved particle filter algorithm to track the UAV, and obtain the final position and tracking trajectory of the UAV.

[0008] Optionally, acquiring the acoustic spectrum time-series data and radar spot data of the UAV includes: acquiring the acoustic signal of the UAV; performing noise reduction processing on the acoustic signal using an improved joint noise reduction algorithm to obtain a noise reduction result; performing signal enhancement processing based on the noise reduction result to obtain a signal enhancement result; performing windowed short-time Fourier transform processing based on the signal enhancement result to obtain the acoustic spectrum time-series data; acquiring the radar echo signal of the UAV; and preprocessing the radar echo signal to obtain radar spot data, wherein the preprocessing includes pulse compression, Doppler processing, and spot extraction.

[0009] Optionally, extracting the first and second spectral feature parameters using the acoustic spectrum time-series data includes: calculating the cumulative energy distribution of the time spectrum using the acoustic spectrum time-series data; constructing a modified sound velocity model based on the cumulative energy distribution; obtaining the modified sound velocity through the modified sound velocity model; constructing a propagation delay compensation model based on the modified sound velocity; obtaining the first spectral feature parameter through the propagation delay compensation model, wherein the first spectral feature parameter includes the spectral roll-off point; constructing an improved spectral flux calculation model using the acoustic spectrum time-series data; and obtaining the second spectral feature parameter through the improved spectral flux calculation model, wherein the second spectral feature parameter includes the spectral flux.

[0010] Optionally, the modified sound speed model satisfies the following expression:

[0011] ,

[0012] in, To correct for the speed of sound, For ambient temperature, For ambient air pressure, For reference air pressure, This is the nonlinear coefficient of air pressure. For ambient humidity, , These are the first-order and second-order influence coefficients of humidity. This is the sound speed gradient compensation term. As the turbulence disturbance compensation term, the propagation delay compensation model satisfies the following expression:

[0013] ,

[0014] in, The frequency of the spectral roll-off point after compensation. The original spectrum roll-off frequency, This is an estimate of the propagation delay. For the dominant wavelength, This is a correction factor for dispersion effects. To correct for the speed of sound, the improved spectral flux calculation model satisfies the following expression:

[0015] ,

[0016] in, For enhanced spectral flux, This represents the total number of frequency bands. For the first Weighting coefficients for each frequency band for Time of the first Spectral amplitude of each frequency band for Time of the first Spectral amplitude of each frequency band It is a non-linear amplification index. The frame energy normalization factor. The energy normalization index, As a transient enhancement factor, For transient component flux, These are the modulation feature weighting coefficients. This is the modulation spectral flux.

[0017] Optionally, extracting the first radar feature parameter and the second radar feature parameter using the radar spot data includes: extracting micro-Doppler features using the radar spot data; constructing an instantaneous frequency variance calculation model based on the micro-Doppler features; obtaining the first radar feature parameter, which includes the instantaneous frequency variance, through the instantaneous frequency variance calculation model; constructing an improved odd-even mode decomposition coefficient calculation model using the radar spot data; and obtaining the second radar feature parameter, which includes odd-mode coefficients and even-mode coefficients, through the odd-even mode decomposition coefficient calculation model.

[0018] Optionally, determining the preliminary position estimate of the UAV based on the first acoustic spectrum feature parameters and the first radar feature parameters, and through a first judgment condition, includes: performing uncertainty quantization on the preliminary acoustic spectrum position and the preliminary radar position based on the first acoustic spectrum feature parameters and the first radar feature parameters to obtain an uncertainty quantization result; calculating the Euclidean distance between the preliminary acoustic spectrum position and the preliminary radar position based on the uncertainty quantization result; determining a first preset threshold using the signal-to-noise ratio of the acoustic spectrum data and the radar data; comparing the magnitude of the Euclidean distance and the first preset threshold; if the Euclidean distance is less than the first preset threshold, then using weighted fusion to output the preliminary position estimate; if the Euclidean distance is not less than the first preset threshold, then selecting the position of the data source with a high signal-to-noise ratio as the preliminary position estimate.

[0019] Optionally, based on the preliminary position estimate, determining the optimized position estimate of the UAV through a second judgment condition includes: obtaining a quantized value of the position uncertainty of the preliminary position estimate; determining a second preset threshold; comparing the quantized value of the position uncertainty with the second preset threshold; if the quantized value of the position uncertainty is not higher than the second preset threshold, then the preliminary position estimate is used as the optimized position estimate; if the quantized value of the position uncertainty is higher than the second preset threshold, then triggering a position optimization process based on the second judgment condition: calculating the consistency score between the acoustic spectrum and radar data; determining a third preset threshold; comparing the consistency score with the third preset threshold; if the consistency score is lower than the third preset threshold, then performing a position re-estimation; if the consistency score is not lower than the third preset threshold, then using the preliminary position estimate as the optimized position estimate.

[0020] Optionally, based on the optimized position estimation, the improved particle filter algorithm is used to track the UAV to obtain the final position and tracking trajectory of the UAV, including: obtaining the initialization result of the particle filter state based on the optimized position estimation; designing an improved particle filter algorithm framework based on the initialization result, the framework including a joint likelihood function, an adaptive weight adjustment mechanism, noise optimization, and a long short-term memory network; designing a fusion strategy model according to the particle filter algorithm framework; and obtaining the final position and tracking trajectory of the UAV through the fusion strategy model.

[0021] Optionally, the joint likelihood function incorporates feature matching degree and temporal consistency constraints, satisfying the following expression:

[0022] ,

[0023] in, For particles The joint likelihood value, As an adaptive weighting factor, , These are the baseline likelihood functions based on acoustic spectrum and radar, respectively. , These are the acoustic spectrum feature vector and radar feature vector measured at the current moment, respectively. , Particles The corresponding acoustic spectrum and radar feature prediction vector, The variance of the acoustic spectrum feature error. The fusion strategy model, representing the vector inner product, satisfies the following expression:

[0024] ,

[0025] in, This is the final tracking trajectory status; State estimation for particle filter output; The state predicted by the LSTM network; For fusion weights; This is the Kalman gain matrix, used to fuse the latest observations. ; This is the observation matrix.

[0026] Secondly, the present invention provides a positioning and tracking system for a low-altitude unmanned aerial vehicle (UAV). The system uses a positioning and tracking method for a low-altitude UAV, comprising: an acquisition module for acquiring acoustic spectrum time-series data and radar spot data of the UAV; an analysis module for extracting a first acoustic spectrum feature parameter and a second acoustic spectrum feature parameter using the acoustic spectrum time-series data; extracting a first radar feature parameter and a second radar feature parameter using the radar spot data; determining a preliminary position estimate of the UAV based on the first acoustic spectrum feature parameter and the first radar feature parameter, according to a first judgment condition; determining an optimized position estimate of the UAV based on the preliminary position estimate, according to a second judgment condition; tracking the UAV using an improved particle filter algorithm based on the optimized position estimate; and an output module for outputting the final position and tracking trajectory of the UAV.

[0027] Compared with the prior art, the beneficial effects of the present invention include:

[0028] 1. Improved the accuracy and robustness of UAV detection and positioning in complex environments. By fusing data from two heterogeneous sensors, acoustic spectrum and radar, a multi-source information complementary perception system was constructed, and a dynamic adaptive threshold and data consistency optimization mechanism was designed to effectively overcome the performance degradation problem of existing sensors in noisy, cluttered, or signal attenuated environments;

[0029] 2. The accuracy and environmental adaptability of feature extraction have been optimized. In acoustic spectrum feature extraction, a modified sound velocity model and propagation delay compensation were introduced, which significantly reduced the impact of atmospheric conditions on acoustic positioning. In radar feature extraction, an improved instantaneous frequency variance and polarization parity decomposition coefficient were adopted to enhance the characterization ability of UAV micro-motion characteristics and scattering mechanisms, making the features more discriminative.

[0030] 3. Intelligent judgment and adaptive optimization of the location estimation process were achieved. By designing multi-level judgment conditions, the fusion strategy or the optimization based on graph neural networks is dynamically selected according to data quality, such as signal-to-noise ratio, uncertainty, and consistency, thus avoiding the error amplification caused by blind fusion when data conflicts or poor quality occur.

[0031] 4. Enhanced real-time tracking and predictive capabilities. The improved particle filter algorithm integrates joint likelihood function and adaptive process noise adjustment, and combines LSTM trajectory prediction for weighted output. This not only improves the tracking accuracy of UAV maneuvers but also enables effective prediction of future trajectories, providing forward-looking information for countermeasure decisions.

[0032] 5. Improved engineering practicality and automation level. From sensor deployment, data preprocessing, feature extraction to fusion tracking, the entire process takes into account the environmental variability and computational efficiency in practical applications. Through adaptive parameter adjustment, model optimization, and automated judgment mechanisms, the reliance on external calibration and manual intervention is reduced, making it more suitable for long-term and wide-ranging regional monitoring applications. Attached Figure Description

[0033] Figure 1 This is a flowchart illustrating a positioning and tracking method for a low-altitude unmanned aerial vehicle (UAV) according to an embodiment of the present invention.

[0034] Figure 2 This is a schematic diagram of the structure of a positioning and tracking system for a low-altitude unmanned aerial vehicle (UAV) according to an embodiment of the present invention. Detailed Implementation

[0035] Specific embodiments of the present invention will now be described in detail. It should be noted that the embodiments described herein are for illustrative purposes only and are not intended to limit the invention. In the following description, numerous specific details are set forth in order to provide a thorough understanding of the invention. However, it will be apparent to those skilled in the art that these specific details are not necessary to practice the invention. In other instances, well-known circuits, software, or methods have not been specifically described to avoid obscuring the invention.

[0036] Throughout this specification, references to "an embodiment," "an embodiment," "an example," or "an example" mean that a particular feature, structure, or characteristic described in connection with that embodiment or example is included in at least one embodiment of the invention. Therefore, the phrases "in an embodiment," "in an embodiment," "an example," or "an example" appearing in various places throughout the specification do not necessarily refer to the same embodiment or example. Furthermore, specific features, structures, or characteristics can be combined in one or more embodiments or examples in any suitable combination and / or sub-combination. Moreover, those skilled in the art will understand that the illustrations provided herein are for illustrative purposes and are not necessarily drawn to scale.

[0037] Please see Figure 1 The present invention provides a positioning and tracking method for a low-altitude unmanned aerial vehicle (UAV), the method comprising the following steps:

[0038] S1. Acquire the acoustic spectrum time series data and radar spot data of the UAV.

[0039] In one embodiment, an acoustic sensor array and a radar system are first deployed in the monitoring area. The acoustic sensor array adopts a uniform linear array layout, and the sensor spacing is set according to the half-wavelength principle of the typical acoustic frequency of the UAV to optimize spatial sampling and avoid aliasing. The radar system is a frequency-modulated continuous wave radar with an operating frequency band of X-band (8 GHz~12 GHz) to balance resolution and atmospheric attenuation. The monitoring area is pre-calibrated environmentally, including measuring background noise and clutter interference data, and establishing a geographic coordinate system for data fusion and location mapping.

[0040] Furthermore, the acoustic spectrum timing data of the UAV is obtained.

[0041] Specifically, the acoustic sensor array synchronously acquires raw acoustic signals at a sampling rate of 44.1 kHz, ensuring coverage of the typical acoustic spectrum range generated by the UAV propellers and motors, such as 50 Hz to 15 kHz. During the acquisition process, environmental parameters, including temperature, humidity, and air pressure, are monitored in real time, and timestamps and sensor location information are embedded. Then, the raw acoustic signals are preprocessed, including noise reduction and signal enhancement. The noise reduction employs an improved joint noise reduction algorithm based on wavelet transform and independent component analysis. This algorithm, based on standard wavelet transform-independent component analysis, introduces an adaptive threshold function and an optimized separation process for independent component analysis, addressing the performance limitations of existing methods in non-stationary noise environments. Specific improvements include:

[0042] 1. An adaptive threshold function based on local signal-to-noise ratio is used to dynamically adjust the wavelet denoising intensity.

[0043] 2. Introduce sound source characteristic constraints in the separation stage of independent component analysis to improve the separation effect of UAV sound signals.

[0044] The proposed joint noise reduction algorithm significantly improves noise reduction performance while maintaining signal integrity by jointly optimizing wavelet thresholding and independent component analysis separation processes. The algorithm's adaptive threshold function... Defined as:

[0045] ,

[0046] in, The noise standard deviation estimated by the median is obtained by calculating the absolute deviation of the median from the first-level detail coefficients after wavelet decomposition. The signal length is directly taken from the number of sampling points in a single data frame acquired by the acoustic sensor. To determine the local signal-to-noise ratio, a sliding window of length 256 points is used. The ratio of signal power to noise power is calculated on the preprocessed acoustic signal. The noise power is obtained through statistics from silent segments. This is the signal-to-noise ratio offset, derived from statistical analysis of a large amount of experimental data. To adjust the parameters, optimization was performed on the validation set using a grid search method. The optimized objective function for independent component analysis was obtained. as follows:

[0047] ,

[0048] in, The separation matrix is ​​obtained by iteratively updating it using the natural gradient descent algorithm. It is a nonlinear function, adjusted according to the non-Gaussian properties of the UAV acoustic signal. For the first Individual components The probability density function prior is obtained by fitting a Gaussian mixture model based on a historical database of UAV acoustic signals. The observed signal is obtained by assembling a hybrid signal matrix from multi-channel acoustic sensor data after wavelet denoising preprocessing. For the Frobenius norm, It is the identity matrix. This is the matrix transpose symbol. , The regularization parameter is obtained through cross-validation. It is through the separation matrix Transformed observation signal The Each independent component The total number of drone sound sources is represented. The signal enhancement process removes low-frequency wind noise and high-frequency interference through bandpass filtering (50 Hz~15 kHz) and applies gain control to compensate for propagation loss. The preprocessed data undergoes windowed Short-Time Fourier Transform (STFT) processing. The specific steps are as follows: the long-duration acoustic signal is divided into multiple short time periods, with a frame length of 2048 sampling points; a Blackman-Harris window function is applied to each frame to reduce spectral leakage; a sliding window with a 75% overlap rate ensures temporal continuity; a Fast Fourier Transform is performed on each windowed frame to obtain the spectrum for that time period. Finally, the STFT processing results are organized into acoustic spectrum time-series data.

[0049] Furthermore, radar spot data of the drone is acquired.

[0050] Specifically, the radar system transmits frequency-modulated continuous wave signals at a pulse repetition frequency of 10 kHz and receives echo signals. During acquisition, radar attitude and position data are recorded simultaneously. Then, the echo signals undergo preprocessing, including pulse compression, Doppler processing, and target extraction. The pulse compression uses matched filtering to improve range resolution; the Doppler processing extracts velocity information using Fast Fourier Transform (FFT) and applies Moving Target Indication (MTI) filtering to suppress static clutter; and the target extraction uses a constant false alarm rate (CFAR) detection algorithm to identify valid target targets, forming target data containing range, azimuth, velocity, and signal-to-noise ratio. After preprocessing, radar target data is obtained.

[0051] In this embodiment, the acoustic sensor array acquired a raw acoustic signal lasting 5 seconds at a sampling rate of 44.1 kHz, with an ambient temperature of 25 degrees Celsius, humidity of 60%, and air pressure of 1013 hPa. After preprocessing, the signal-to-noise ratio (SNR) was improved from the initial 5 dB to 18 dB. A short-time Fourier transform yielded a time-spectrum matrix size of 536 × 1025, corresponding to a time resolution of 46.4 milliseconds and a frequency resolution of 21.5 Hz. The radar system operated in the X-band at a pulse repetition frequency of 10 kHz, acquiring echo signals from the same time period. After pulse compression, the range resolution reached 0.15 meters, and the Doppler processing velocity resolution was 0.2 m / s. 32 valid points were extracted using constant false alarm rate (CFAR) detection. The point data included range, azimuth, velocity, and SNR fields, with the maximum SNR being 22 dB.

[0052] S2. Using the aforementioned acoustic spectrum time-series data, extract the first acoustic spectrum feature parameters and the second acoustic spectrum feature parameters.

[0053] In one embodiment, based on the acoustic spectrum time-series data obtained in step S1, a first acoustic spectrum feature parameter is calculated through a multi-stage feature extraction process. The first acoustic spectrum feature parameter includes the spectral roll-off point, and its extraction process comprehensively considers the physical characteristics of sound wave propagation and the influence of environmental factors.

[0054] Specifically, a refined time-frequency analysis is first performed on the acoustic spectrum time-series data. Based on the windowed short-time Fourier transform in step S1, a multi-resolution spectrum analysis framework is adopted. By introducing an adaptive window length adjustment mechanism, the limitations of the existing fixed window length in the analysis of non-stationary signals are solved.

[0055] Furthermore, the cumulative energy distribution of the time-domain spectrum is calculated. This cumulative energy distribution is obtained by integrating along the time axis based on the calculated time-domain spectrum, and is used to quantify the time-domain cumulative characteristics of the signal energy.

[0056] Furthermore, based on the cumulative energy distribution, a modified sound velocity model is used to determine the spectral roll-off point, defined as the frequency point where the cumulative energy reaches a specific proportion of 85% of the total energy. This proportion is obtained through statistical analysis of a large amount of experimental data. The modified sound velocity model introduces a humidity gradient compensation term and a pressure nonlinearity correction factor into the standard sound velocity formula, significantly improving the accuracy of sound velocity calculations under complex atmospheric conditions. The expression of the modified sound velocity model is as follows:

[0057] ,

[0058] in, To correct for the speed of sound, The ambient temperature is measured by a digital temperature sensor deployed in an array of acoustic sensors at a rate of 0.1. Accuracy is obtained through direct measurement; The ambient air pressure is obtained by acquiring data at a sampling rate of 1Hz using a high-precision air pressure sensor and then smoothing it using a Kalman filter. Used as a reference pressure, serving as a standardized benchmark value; The nonlinear coefficient for air pressure was obtained through sound speed measurement experiments under different altitude conditions. The ambient humidity is obtained through continuous monitoring using a capacitive humidity sensor. , The first and second order influence coefficients of humidity are obtained by polynomial fitting based on the humidity gradient experimental data; The sound velocity gradient compensation term is calculated by measuring the three-dimensional temperature field gradient through a distributed array of temperature sensors deployed within the monitoring area. This is the turbulence disturbance compensation term, calculated using wind speed fluctuation data along the sound wave propagation path.

[0059] Furthermore, based on the corrected sound velocity obtained from the corrected sound velocity model, propagation delay compensation is performed on the spectral roll-off frequency value. The compensation process employs a combination of time delay estimation and frequency domain correction, specifically implemented through the following expression of the propagation delay compensation model:

[0060] ,

[0061] in, The compensated spectral roll-off frequency is output as the first spectral characteristic parameter. The original spectrum roll-off point frequency was accurately extracted from the cumulative energy distribution curve using cubic spline interpolation. The propagation delay estimate is calculated by combining the generalized cross-correlation function with the geometric configuration of the sensor array; The dominant wavelength is calculated by combining the wavelength corresponding to the peak frequency of the acoustic spectrum with the dispersion relation. This is the dispersion effect correction factor, calculated using a dispersion characteristic model of sound waves propagating in the atmosphere; To correct the speed of sound, the above corrected speed of sound model is used.

[0062] Furthermore, the first acoustic spectrum feature parameters are output through the propagation delay compensation model.

[0063] Furthermore, a second spectral characteristic parameter is extracted, which includes spectral flux. Its calculation process incorporates a multi-band weighting mechanism and a transient component enhancement strategy based on traditional spectral flux calculation to better characterize the dynamic characteristics of the UAV's acoustic signal. The expression for the improved spectral flux calculation model is as follows:

[0064] ,

[0065] in, The enhanced spectral flux is output as the second spectral characteristic parameter. The total number of frequency bands is set to 256 based on the spectral resolution. For the first The weighting coefficients for each frequency band are allocated using the information entropy weighting method based on the energy distribution characteristics of UAV acoustic signals in different frequency bands. for Time of the first Spectral amplitude of each frequency band for Time of the first The spectral amplitude of each frequency band is obtained from the STFT results after amplitude calibration. The nonlinear amplification index was determined through signal dynamic range optimization experiments. The frame energy normalization factor is obtained by normalizing the total energy of the current frame by combining the energy statistics of historical frames. The energy normalization index is obtained through statistical analysis of a large amount of acoustic signal data. The transient enhancement factor is dynamically adjusted through the output of a transient detection algorithm based on wavelet transform. The transient component flux is obtained by extracting high-frequency transient components through wavelet packet decomposition and combining them with Hilbert transform. These are the modulation feature weighting coefficients, obtained through modulation spectrum analysis and optimization. The modulation spectral flux is calculated by extracting the periodic modulation components from the acoustic spectrum.

[0066] Furthermore, the improved spectral flux calculation model outputs the second acoustic spectrum characteristic parameters.

[0067] In this embodiment, the cumulative energy distribution is calculated based on the acoustic spectrum time-series data, and the sound velocity value is obtained as 346.2 m / s using a modified sound velocity model. The original spectral roll-off frequency is 8.2 kHz, which is corrected to 8.35 kHz after propagation delay compensation and used as the first acoustic spectrum feature parameter. Simultaneously, the enhanced spectral flux is calculated, with 256 frequency band weighting coefficients weighted by information entropy. The frame energy normalization factor is 0.85, the transient enhancement factor is 0.32, and the final second acoustic spectrum feature parameter value is 0.67.

[0068] S3. Using the radar spot data, extract the first radar feature parameters and the second radar feature parameters.

[0069] In one embodiment, the radar spot data obtained in step S2 is first used to calculate the first radar feature parameters through a multi-level feature extraction process. The first radar feature parameters include the instantaneous frequency variance of the micro-Doppler features. The extraction process takes into account the micro-motion characteristics of the UAV rotor and the time-frequency analysis of the radar signal.

[0070] Specifically, the radar spot data is first subjected to refined time-frequency analysis. Based on the pulse compression and Doppler processing in step S1, an adaptive kernel time-frequency distribution method is adopted. By introducing a kernel function parameter optimization mechanism, the limitations of existing time-frequency analysis methods in terms of cross-term suppression and resolution balance are solved.

[0071] Furthermore, based on the results of the refined time-frequency analysis, micro-Doppler features are extracted. These micro-Doppler features are characterized by the frequency modulation components in the radar echo signal caused by the UAV rotor rotation and body vibration. A micro-Doppler separation algorithm based on variational mode decomposition is used to decompose the composite radar signal into multiple intrinsic mode functions and screen out the mode components related to rotor micro-motion.

[0072] Furthermore, based on the micro-Doppler features, the instantaneous frequency variance of the micro-Doppler features is calculated and used as the first radar feature parameter. The variance calculation incorporates time-varying weights and frequency domain confidence assessments on top of the standard variance formula to improve the feature's ability to distinguish UAV types and motion states. The expression for the improved instantaneous frequency variance calculation model is as follows:

[0073] ,

[0074] in, The enhanced instantaneous frequency variance is output as the first radar characteristic parameter. The number of sampling points for the instantaneous frequency sequence is determined by the number of time-frequency analysis points in the radar signal processing chain; For the first The time-varying weights of each sampling point are dynamically calculated based on the local signal-to-noise ratio of the time-frequency spectrum and the gradient rate of change of the instantaneous frequency at that moment. For the first The instantaneous frequency of each sampling point is accurately extracted from the micro-Doppler mode components by combining Hilbert transform with the Tig-Kaiser energy operator; For weighted average frequency, through and The weighted average was calculated. The time-frequency distribution variance weighting coefficients are obtained through optimization of the singular value decomposition results of the time-frequency distribution matrix. The variance of the time-frequency distribution is calculated using the variance of the Wigner-Weil distribution matrix; The confidence level adjustment factor is determined based on the envelope smoothness of the micro-Doppler mode components. The frequency domain confidence evaluation term is calculated by combining the spectral flatness of the instantaneous frequency sequence and the peak-to-sidelobe ratio of the autocorrelation function.

[0075] Furthermore, the first radar characteristic parameter is output through the improved instantaneous frequency variance calculation model.

[0076] Furthermore, a second radar characteristic parameter is extracted, which includes the odd-even mode decomposition coefficients of the polarization scattering matrix. The calculation process introduces adaptive scattering mechanism discrimination and azimuth compensation based on the existing polarization decomposition to more accurately characterize the scattering characteristics of the UAV target. The expression for the improved odd-even mode decomposition coefficient calculation model is as follows:

[0077] ，

[0078] ,

[0079] in, These are the odd modulus coefficients. These are even-mode coefficients, which together serve as the output of the second radar characteristic parameter. , These are the polarization scattering matrix elements for horizontal transmission and horizontal reception, and vertical transmission and vertical reception, respectively, extracted from the fully polarization scattering matrix measured by the radar system. is the Frobenius norm, used to calculate the overall difference between matrix elements; It is an asymmetric scattering enhancement factor based on cross-polarization components. The ratio of the amplitude to the common polarization component is calculated. The incident angle compensation coefficient is calculated using the relative geometric relationship between the radar and the target, combined with an electromagnetic scattering model. Angle of incidence The compensation function is obtained by fitting the angle response function based on the physical optics model; The symmetric scattering enhancement factor is calculated based on the copolarization correlation coefficient and scattering entropy. This is the azimuth compensation coefficient, calculated using the azimuth of the target relative to the radar line of sight. Obtained by combining array antenna pattern compensation; Azimuth The compensation function is obtained by fitting the measured antenna pattern data using a polynomial.

[0080] It should be noted that the data for the above parameters can all be extracted and calculated from the full polarization scattering matrix trace data acquired by the radar.

[0081] Furthermore, the second radar characteristic parameters are output through the improved odd-even mode decomposition coefficient calculation model.

[0082] In this embodiment, adaptive kernel time-frequency analysis is performed on the radar spot data, and variational mode decomposition is used to obtain four eigenmode functions. Micro-Doppler features are extracted from these, with an instantaneous frequency sequence length of 128 points. The time-varying weights are calculated based on the local signal-to-noise ratio. The enhanced instantaneous frequency variance is calculated to be 45.6 Hz², which is used as the first radar characteristic parameter. Fully polarimetric scattering matrix analysis shows... The component is 0.35. The component is 0.28, and the odd and even mode decomposition coefficients are 0.72 and 0.65, respectively, which together serve as the second radar characteristic parameter.

[0083] S4. Based on the first acoustic spectrum feature parameters and the first radar feature parameters, and through the first judgment condition, determine the preliminary position estimate of the UAV.

[0084] In one embodiment, based on the first acoustic spectrum feature parameters extracted in step S2 and the first radar feature parameters extracted in step S3, the preliminary acoustic spectrum position and the preliminary radar position are calculated respectively through a multi-source data registration and spatiotemporal alignment process. The calculation of the preliminary acoustic spectrum position adopts a direction-of-arrival positioning algorithm based on generalized cross-correlation time delay estimation, and the UAV spatial coordinates are retrieved by inverting the arrival time difference of the acoustic signal received by the acoustic sensor array. The calculation of the preliminary radar position adopts a point-track association positioning algorithm based on polar coordinate-Cartesian coordinate transformation, and the UAV spatial coordinates are directly calculated by the distance, azimuth and elevation angle measured by the radar.

[0085] Furthermore, uncertainty quantification is performed on the preliminary acoustic spectrum position and the preliminary radar position. The uncertainty quantification is achieved by constructing an acoustic spectrum positioning error ellipsoid and a radar positioning error ellipsoid, respectively. The principal axis direction of the acoustic spectrum positioning error ellipsoid is determined by the geometric configuration of the acoustic sensor array, and the principal axis length is calculated by the Cramer-Rao lower bound of the time delay estimation. The principal axis direction of the radar positioning error ellipsoid is determined by the radar beam pointing, and the principal axis length is calculated by the eigenvalue decomposition results of the range resolution, angular resolution, and their covariance matrix.

[0086] Furthermore, based on the quantized preliminary acoustic spectrum position and the preliminary radar position, the preliminary position estimate of the UAV is determined through a first judgment condition.

[0087] Specifically, the Euclidean distance between the initial position of the acoustic spectrum and the initial position of the radar is first calculated, that is, the L2 norm of the difference between the coordinates of the two positions is taken.

[0088] Furthermore, a first preset threshold is determined. This first preset threshold is no longer a fixed value, but is dynamically adjusted according to environmental conditions and data quality. The expression for its determination model is as follows:

[0089] ,

[0090] in, The first preset threshold, The base distance threshold is determined by the statistical quantiles of the acoustic spectrum and radar positioning error in historical data; The signal-to-noise ratio adjustment coefficient is obtained through experimental calibration based on positioning accuracy at different signal-to-noise ratio levels. and The signal-to-noise ratios of the acoustic spectrum data and radar data are respectively obtained through the signal-to-noise ratio calculation models in steps S1 and S3. This is the signal-to-noise ratio normalization factor, used to balance the impact of the signal-to-noise ratio; The uncertainty adjustment coefficient is obtained by using the uncertainty of the position estimate from the Monte Carlo simulation. and The variances of the acoustic spectrum and radar position estimation are respectively obtained by propagating the covariance matrix of the positioning algorithm; This is the maximum permissible variance, set according to system performance requirements.

[0091] Furthermore, comparing the Euclidean distance with the first preset threshold yields the following results:

[0092] If the Euclidean distance is less than the first preset threshold, a weighted fusion output is used to estimate the initial position.

[0093] ,

[0094] in, For the initial position estimate of the output, Preliminary position of the sound spectrum. Preliminary radar location. , The weights for acoustic spectrum and radar are respectively, and are dynamically calculated based on their respective position estimation variance and signal-to-noise ratio;

[0095] If the Euclidean distance is greater than or equal to the first preset threshold, then the location of the data source with a high signal-to-noise ratio is selected as the preliminary location estimate, i.e.:

[0096] like ,but = ;

[0097] like ,but = .

[0098] Furthermore, the initial position estimate of the UAV is output. The preliminary position estimate is the initial position value of the UAV obtained by fusing acoustic spectrum and radar initial positioning information, which is to be further optimized.

[0099] It should be noted that the first judgment condition is based on whether the Euclidean distance between the acoustic spectrum and the radar preliminary position is less than a dynamic first preset threshold, which determines whether to use weighted fusion or to select the position of a single data source as the preliminary position estimate.

[0100] In this embodiment, the preliminary acoustic position coordinates are calculated to be 125.3 meters, 87.6 meters, and 15.2 meters, while the preliminary radar position coordinates are 123.1 meters, 85.9 meters, and 14.8 meters. The Euclidean distance between the two is 3.2 meters. In the dynamic first preset threshold calculation, the base distance threshold is 5 meters, the acoustic signal-to-noise ratio is 18 dB, the radar signal-to-noise ratio is 22 dB, the position estimation variances are 0.8 and 0.5 respectively, and the final threshold is 4.1 meters. Since 3.2 meters is less than 4.1 meters, a weighted fusion is used to obtain the preliminary position estimate, with an acoustic weight of 0.45 and a radar weight of 0.55. The fused coordinates are 124.1 meters, 86.7 meters, and 15.0 meters.

[0101] S5. Based on the preliminary position estimate, determine the optimal position estimate of the UAV through the second judgment condition.

[0102] In one embodiment, based on the preliminary location estimate obtained in step S4, a multi-dimensional uncertainty assessment and data consistency verification process is used to determine whether location optimization is necessary. The uncertainty assessment is achieved by calculating the eigenvalues ​​of the covariance matrix of the preliminary location estimate, specifically using a confidence interval estimation method based on the Fisher information matrix to quantify the dispersion of the location estimate in three-dimensional space. Through this uncertainty assessment, the location uncertainty of the preliminary location estimate, i.e., the dispersion degree, is obtained.

[0103] It should be noted that the positional uncertainty obtained above is a quantified value.

[0104] Furthermore, a second preset threshold is determined, which is adaptively adjusted based on the system positioning accuracy requirements and dynamic environmental characteristics. Its calculation model is as follows:

[0105] ,

[0106] in, The second preset threshold; The basic uncertainty threshold was obtained through offline calibration based on the size of the monitoring area and the typical movement speed of the UAV. , These are adaptive adjustment coefficients, reflecting the influence of location variance and time update interval on the threshold, respectively, obtained through regression analysis of historical data; The variance of the preliminary position estimate is calculated from the trace of the covariance matrix of the preliminary position estimate; This is the variance normalization factor, used to balance the magnitude of variance; The data update interval is determined by the timing characteristics of the acoustic spectrum and radar data acquisition and processing chain; This is the time decay constant, set according to the system response speed requirements.

[0107] Further, the position uncertainty quantization value of the preliminary position estimate is compared with the second preset threshold; if the position uncertainty quantization value is higher than the second preset threshold, a position optimization process based on the second judgment condition is triggered; if the position uncertainty quantization value is not higher than the second preset threshold, the entire position optimization process is skipped, and the preliminary position estimate obtained in step S4 is directly used as the final optimized position estimate.

[0108] Regarding the location optimization process for the second judgment condition, specifically, the consistency score between the acoustic spectrum and radar data is first calculated. This consistency score is constructed by fusing multi-dimensional information such as location differences, feature similarity, and temporal continuity. The calculation model is as follows:

[0109] ,

[0110] in, For consistency score; , These are the preliminary acoustic spectrum position vector and the preliminary radar position vector, respectively, obtained from the output of step S4; It is the Euclidean norm; The variance of the position error is calculated using the joint covariance matrix of the acoustic spectrum and radar positioning error. , , These are weighting coefficients, representing the importance of location, feature, and temporal components, respectively, and are obtained through the analytic hierarchy process combined with expert knowledge. , These are the acoustic spectrum feature vector and the radar feature vector, respectively, which are formed by concatenating the first and second feature parameters extracted in steps S2 and S3. The time-series correlation between acoustic spectrum and radar data is obtained by calculating the Pearson correlation coefficient between the acoustic spectrum time-series data and radar spot data within a sliding window.

[0111] Furthermore, a third preset threshold is determined. This third preset threshold is set based on the system's minimum requirements for data consistency, and dynamically fine-tuned considering sensor performance and environmental interference. Its expression is:

[0112] ,

[0113] in, The third preset threshold; The basic consistency score threshold was obtained through statistical analysis of the score distribution of correct and incorrect cases in a large amount of experimental data. The signal-to-noise ratio influencing factor characterizes the degree of impact of sensor data quality on consistency requirements and is obtained through experimental calibration. , These are the signal-to-noise ratios of acoustic spectrum and radar data, respectively. This is the reference value for the system's maximum signal-to-noise ratio.

[0114] Furthermore, the consistency score is compared with the third preset threshold, and the results are as follows:

[0115] If the consistency score is lower than the third preset threshold, a multi-source data fusion model based on a graph neural network is initiated for location re-estimation. The graph neural network model uses acoustic spectrum feature nodes and radar feature nodes as inputs to construct a spatiotemporal heterogeneous graph. Node attributes include feature parameters, initial location values, and timestamps, while edge attributes include sensor type association weights and spatiotemporal proximity. The node feature update rule of the graph neural network adopts an existing multi-head graph attention mechanism.

[0116] After passing through multiple layers, the graph neural network aggregates global information through a readout layer and outputs a re-estimated position, which serves as an optimized position estimate.

[0117] If the consistency score is not lower than the third preset threshold, the preliminary position estimate is directly used as the optimized position estimate.

[0118] It should be noted that the second judgment condition is based on whether the consistency score between the acoustic spectrum and radar data is lower than the third preset threshold, which determines whether to activate the graph neural network for position optimization.

[0119] In this embodiment, the eigenvalues ​​of the covariance matrix of the initial location estimation are 1.8, 1.2, and 0.9, respectively, and the calculated location uncertainty is 1.2. The second preset threshold has a base value of 1.0, which is adjusted to 1.15 after considering the variance factor and time update interval. Since 1.2 is higher than 1.15, the location optimization process is triggered. In the consistency score calculation, the location difference term is 0.75, the feature similarity is 0.82, and the temporal correlation is 0.68, with a weighted total score of 0.76. The third preset threshold is 0.78. Since 0.76 is lower than 0.78, graph neural network optimization is initiated. After three layers of graph attention propagation, the re-estimated coordinates of 127.2 meters, 89.1 meters, and 15.8 meters are output as the optimized location estimate.

[0120] S6. Based on the optimized position estimation, the improved particle filter algorithm is used to track the UAV to obtain the final position and tracking trajectory of the UAV.

[0121] In one embodiment, the particle filter state is first initialized based on the optimized position estimate obtained in step S5. This initialization process employs an adaptive particle distribution strategy, generating an initial particle set by constructing a Gaussian mixture model centered on the optimized position estimate and with position uncertainty as the covariance. The number of particles is dynamically adjusted according to the state space dimension and computational resources. It is important to note that these particles are used to approximate random samples representing the posterior probability distribution of the UAV's motion state.

[0122] Furthermore, an improved particle filter algorithm framework is designed. This framework introduces a joint likelihood function, an adaptive weight adjustment mechanism, and residual-driven process noise optimization on the basis of standard particle filtering. It also integrates Long Short-Term Memory (LSTM) trajectory prediction to achieve high-precision estimation and prediction of UAV motion state.

[0123] Specifically, for each particle, a joint likelihood function based on the acoustic spectrum and radar waves is calculated. This joint likelihood function introduces feature matching degree and temporal consistency constraints on the basis of the traditional Gaussian likelihood model, and its expression is as follows:

[0124] ,

[0125] in, For particles The joint likelihood value; The adaptive weighting factor is dynamically calculated based on the ratio of the current acoustic spectrum data signal-to-noise ratio to the radar data signal-to-noise ratio. , These are the baseline likelihood functions based on acoustic spectrum and radar, respectively, calculated using the Gaussian kernel function of the measured values ​​and particle prediction values; , These are the acoustic spectrum feature vector and radar feature vector measured at the current moment, respectively, which are composed of the feature parameters extracted in steps S2 and S3. , Particles The corresponding acoustic spectrum and radar feature prediction vectors are obtained through the feature-state mapping function; The variance of the acoustic spectrum feature error is obtained by statistical analysis of feature matching errors in historical data. This represents the inner product of vectors, used to measure feature similarity.

[0126] Furthermore, the particle weights are normalized, and a system resampling strategy is adopted to avoid particle degradation. The system resampling introduces weight smoothing and particle diffusion operations on the basis of standard polynomial resampling to maintain particle diversity.

[0127] Furthermore, statistical analysis is performed on the residuals of the particle filter, and the process noise covariance matrix is ​​dynamically adjusted. The adjustment model is as follows:

[0128] ,

[0129] in, for The process noise covariance matrix at time step; This is a smoothing factor used to control the decay rate of historical noise information, and is set according to the dynamic characteristics of the system. This is the residual vector, which is the difference between the observed value and the predicted mean of the particle set; The residual sample covariance matrix; The residual weighting coefficient is dynamically adjusted based on the whitening test results of the residual sequence. The state-observation Jacobian matrix is ​​approximated by particle set statistical linearization; The prior state covariance matrix is ​​calculated from the particle set.

[0130] Furthermore, a time series prediction model is introduced, using a Long Short-Term Memory (LSTM) network to predict the UAV trajectory. The LSTM network takes a historical state sequence as input and outputs state predictions for multiple future time steps. These predictions are then weighted and fused with the particle filter output to obtain the final tracking trajectory. The fusion strategy is as follows:

[0131] ,

[0132] in, This is the final tracking trajectory status; State estimation for particle filter output; The state predicted by the LSTM network; For fusion weights, the estimated covariance based on particle filtering and the prediction error of LSTM are adaptively calculated. This is the Kalman gain matrix, used to fuse the latest observations. ; This is the observation matrix.

[0133] Furthermore, the final position and tracking trajectory of the drone are output, and countermeasures are triggered based on the trajectory prediction results. These countermeasures include early warning prompts, interference signal transmission, or automatic tracking and locking.

[0134] It should be noted that the final position is obtained by weighted fusion of the particle filter output state and the LSTM predicted trajectory; the tracking trajectory consists of a sequence of final positions over consecutive time steps.

[0135] In this embodiment, 500 particles are generated using an initial particle filter, and a Gaussian mixture model is constructed based on optimized position estimates of 127.2 m, 89.1 m, and 15.8 m. In the joint likelihood function calculation, the adaptive weighting factor is 0.6, the acoustic spectral feature matching degree is 0.88, and the radar feature matching degree is 0.91. The process noise covariance matrix is ​​adjusted using residual-driven adjustment, with a smoothing factor of 0.8. The LSTM network predicts the trajectory for the next 5 steps based on the past 20 state sequences, with a prediction error of 0.25 m. The final fused output shows the UAV's position coordinates as 128.5 m, 90.3 m, and 16.1 m. The tracking trajectory shows the UAV moving northeast at a speed of 8 m / s, triggering a level-three warning and initiating automatic tracking and locking.

[0136] Please see Figure 2 , Figure 2 This is a schematic diagram of the structure of a positioning and tracking system for a low-altitude unmanned aerial vehicle (UAV) according to an embodiment of the present invention. The system includes an input device, a processor, an output device, and a memory. The input device, processor, output device, and memory are interconnected. The memory stores a computer program, which includes program instructions. The processor is configured to call the program instructions. The system uses the positioning and tracking method for a low-altitude UAV. The output device includes an acquisition module, the processor includes an analysis module, and the output device includes an output module.

[0137] The acquisition module is used to acquire the acoustic spectrum time-series data and radar spot data of the UAV;

[0138] The analysis module is used to extract a first acoustic spectrum feature parameter and a second acoustic spectrum feature parameter using the acoustic spectrum time-series data; extract a first radar feature parameter and a second radar feature parameter using the radar spot data; determine a preliminary position estimate of the UAV based on the first acoustic spectrum feature parameter and the first radar feature parameter through a first judgment condition; determine an optimized position estimate of the UAV based on the preliminary position estimate through a second judgment condition; and perform UAV tracking using an improved particle filter algorithm based on the optimized position estimate.

[0139] The output module is used to output the final position and tracking trajectory of the drone.

[0140] In summary, this invention constructs a high-precision, robust, and intelligently adaptive UAV detection and tracking system by fusing heterogeneous data from acoustic spectrum and radar, combining an environment-adaptive feature extraction model, a multi-level intelligent optimization judgment mechanism based on dynamic thresholds and graph neural networks, and an improved particle filter algorithm that integrates LSTM prediction. This effectively solves the technical challenges of limited sensor performance, low data fusion reliability, and insufficient tracking and prediction accuracy in complex environments.

[0141] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. Although the present invention 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 or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be covered within the scope of the claims and specification of the present invention.

Claims

1. A positioning and tracking method for a low-altitude unmanned aerial vehicle (UAV), characterized in that, The method includes the following steps: Acquire acoustic spectrum time-series data and radar spot data of the UAV; Using the aforementioned acoustic spectrum time-series data, extract the first acoustic spectrum feature parameters and the second acoustic spectrum feature parameters; Using the radar spot data, extract the first radar feature parameters and the second radar feature parameters; Based on the first acoustic spectrum feature parameters and the first radar feature parameters, a preliminary position estimate of the UAV is determined through the first judgment condition. Based on the preliminary position estimate, the optimized position estimate of the UAV is determined by the second judgment condition; Based on the optimized position estimation, the improved particle filter algorithm is used to track the UAV, and the final position and tracking trajectory of the UAV are obtained. Based on the first acoustic spectrum feature parameters and the first radar feature parameters, and through the first judgment condition, the preliminary position estimate of the UAV is determined as follows: Based on the first acoustic spectrum feature parameters and the first radar feature parameters, uncertainty quantization is performed on the preliminary acoustic spectrum position and the preliminary radar position to obtain the uncertainty quantization result; Based on the uncertainty quantification results, the Euclidean distance between the preliminary acoustic spectrum position and the preliminary radar position is calculated; The first preset threshold is determined by using the signal-to-noise ratio of acoustic spectrum data and radar data; The Euclidean distance is compared with the first preset threshold. If the Euclidean distance is less than the first preset threshold, a weighted fusion is used to output a preliminary location estimate. If the Euclidean distance is not less than the first preset threshold, the location of the data source with a high signal-to-noise ratio is selected as the preliminary location estimate. Based on the preliminary position estimate, the optimized position estimate of the UAV is determined by the second judgment condition, including: Obtain the quantized value of the position uncertainty of the preliminary position estimate; Determine the second preset threshold; The position uncertainty quantization value is compared with the second preset threshold; if the position uncertainty quantization value is not higher than the second preset threshold, the preliminary position estimate is used as the optimized position estimate; if the position uncertainty quantization value is higher than the second preset threshold, a position optimization process based on a second judgment condition is triggered. Calculate the consistency score between the acoustic spectrum and radar data; Determine the third preset threshold; The consistency score is compared with the third preset threshold; if the consistency score is lower than the third preset threshold, a position re-estimation is performed; if the consistency score is not lower than the third preset threshold, the preliminary position estimate is used as the optimized position estimate.

2. The positioning and tracking method for a low-altitude unmanned aerial vehicle according to claim 1, characterized in that, Acquiring the acoustic spectrum time-series data and radar spot data of the UAV includes: Acquire the acoustic signal of the UAV; use an improved joint noise reduction algorithm to perform noise reduction processing on the acoustic signal to obtain the noise reduction processing result; perform signal enhancement processing based on the noise reduction processing result to obtain the signal enhancement processing result; and perform windowed short-time Fourier transform processing based on the signal enhancement processing result to obtain the acoustic spectrum time series data. The radar echo signal of the UAV is acquired; the radar echo signal is preprocessed to obtain radar spot data, the preprocessing including pulse compression, Doppler processing and spot extraction.

3. The positioning and tracking method for a low-altitude unmanned aerial vehicle according to claim 1, characterized in that, Using the aforementioned acoustic spectrum time-series data, the extraction of the first acoustic spectrum feature parameters and the second acoustic spectrum feature parameters includes: Using the time-series acoustic spectrum data, the cumulative energy distribution of the time spectrum is calculated; based on the cumulative energy distribution, a modified sound velocity model is constructed; the modified sound velocity is obtained through the modified sound velocity model; a propagation delay compensation model is constructed based on the modified sound velocity; and a first acoustic spectrum feature parameter is obtained through the propagation delay compensation model, wherein the first acoustic spectrum feature parameter includes the spectral roll-off point. Using the aforementioned acoustic spectrum time-series data, an improved spectral flux calculation model is constructed; through the improved spectral flux calculation model, a second acoustic spectrum feature parameter is obtained, the second acoustic spectrum feature parameter including spectral flux.

4. The positioning and tracking method for a low-altitude unmanned aerial vehicle according to claim 3, characterized in that, The modified sound speed model satisfies the following expression: ， in, To correct for the speed of sound, For ambient temperature, For ambient air pressure, For reference air pressure, This is the nonlinear coefficient of air pressure. For ambient humidity, , These are the first-order and second-order influence coefficients of humidity. This is the sound speed gradient compensation term. As the turbulence disturbance compensation term, the propagation delay compensation model satisfies the following expression: ， in, The frequency of the spectral roll-off point after compensation. The original spectrum roll-off frequency, This is an estimate of the propagation delay. For the dominant wavelength, This is a correction factor for dispersion effects. To correct for the speed of sound, the improved spectral flux calculation model satisfies the following expression: ， in, For enhanced spectral flux, This represents the total number of frequency bands. For the first Weighting coefficients for each frequency band for Time of the first Spectral amplitude of each frequency band for Time of the first Spectral amplitude of each frequency band It is a non-linear amplification index. The frame energy normalization factor. The energy normalization index, As a transient enhancement factor, For transient component flux, These are the modulation feature weighting coefficients. This is the modulation spectral flux.

5. The positioning and tracking method for a low-altitude unmanned aerial vehicle according to claim 1, characterized in that, Using the radar spot data, extracting the first radar feature parameters and the second radar feature parameters includes: Using the radar spot data, micro-Doppler features are extracted; based on the micro-Doppler features, an instantaneous frequency variance calculation model is constructed; through the instantaneous frequency variance calculation model, a first radar feature parameter is obtained, the first radar feature parameter including the instantaneous frequency variance; Using the radar spot data, an improved odd-even mode decomposition coefficient calculation model is constructed; through the odd-even mode decomposition coefficient calculation model, a second radar characteristic parameter is obtained, which includes odd mode coefficient and even mode coefficient.

6. The positioning and tracking method for a low-altitude unmanned aerial vehicle according to claim 1, characterized in that, Based on the optimized position estimation, the improved particle filter algorithm is used for UAV tracking to obtain the UAV's final position and tracking trajectory, including: Based on the optimized position estimation, the initialization result of the particle filter state is obtained; Based on the initialization results, an improved particle filter algorithm framework is designed, which includes a joint likelihood function, an adaptive weight adjustment mechanism, noise optimization, and a long short-term memory network. Based on the particle filter algorithm framework, a fusion strategy model is designed; The final position and tracking trajectory of the UAV are obtained through the fusion strategy model.

7. A positioning and tracking method for a low-altitude unmanned aerial vehicle according to claim 6, characterized in that, The joint likelihood function incorporates feature matching degree and temporal consistency constraints, satisfying the following expression: ， in, For particles The joint likelihood value, As an adaptive weighting factor, , These are the baseline likelihood functions based on acoustic spectrum and radar, respectively. , These are the acoustic spectrum feature vector and radar feature vector measured at the current moment, respectively. , Particles The corresponding acoustic spectrum and radar feature prediction vector, The variance of the acoustic spectrum feature error. The fusion strategy model, representing the vector inner product, satisfies the following expression: ， in, This is the final tracking trajectory status; State estimation for particle filter output; The state predicted by the LSTM network; For weighting; This is the Kalman gain matrix, used to fuse the latest observations. ; This is the observation matrix.

8. A positioning and tracking system for a low-altitude unmanned aerial vehicle (UAV), the system using a positioning and tracking method for a low-altitude UAV as described in any one of claims 1 to 7, characterized in that, The system includes: The acquisition module is used to acquire the acoustic spectrum time-series data and radar spot data of the UAV; The analysis module is used to extract a first acoustic spectrum feature parameter and a second acoustic spectrum feature parameter using the acoustic spectrum time-series data; extract a first radar feature parameter and a second radar feature parameter using the radar spot data; determine a preliminary position estimate of the UAV based on the first acoustic spectrum feature parameter and the first radar feature parameter through a first judgment condition; determine an optimized position estimate of the UAV based on the preliminary position estimate through a second judgment condition; and perform UAV tracking using an improved particle filter algorithm based on the optimized position estimate. The output module is used to output the final position and tracking trajectory of the drone.

Images