An unmanned aerial vehicle monitoring method and system based on doppler effect of a seismograph

Abstract

This application relates to a method and system for monitoring unmanned aerial vehicles (UAVs) based on the Doppler effect of a seismograph. The method includes: acquiring ground vibration signals of the monitoring area collected by a seismograph, segmenting them, and then performing detrending, denoising, and filtering processing; performing time-frequency analysis on the processed vibration signals to obtain the corresponding time spectrum; extracting UAV signals from the time spectrum using an artificial intelligence algorithm and inputting them into a pre-trained deep learning model to output the time-frequency trajectory region of the UAV signals and their corresponding time range; performing dominant frequency tracking within the time-frequency trajectory region to obtain the instantaneous frequency sequence that changes with time, and extracting frequency offset information representing the Doppler effect of relative motion; constructing Doppler inversion constraints based on the frequency offset information, and obtaining estimation results of the UAV excitation frequency, flight speed, flight altitude, and spatial position through optimization; performing trajectory consistency constraints and anomaly removal on the estimation results, outputting the UAV monitoring results, and generating a monitoring report.

Description

Technical Field

[0001] This application relates to the field of unmanned aerial vehicle (UAV) monitoring technology, and in particular to a UAV monitoring method and system based on the Doppler effect of a seismograph. Background Technology

[0002] Due to their high mobility, low deployment cost, and flexible flight paths, drones have been widely used in surveying and inspection, emergency rescue, logistics transportation, and public safety. At the same time, the demand for real-time monitoring of low-altitude targets is becoming increasingly prominent in scenarios such as airport airspace protection, perimeter protection of important facilities, and security for major events, especially placing higher demands on the detection, identification, and trajectory tracking of "low-altitude, slow-speed, and small" drones.

[0003] Existing drone monitoring methods mainly include radar detection, radio spectrum detection, optoelectronic imaging, and acoustic arrays. Radar is susceptible to clutter in complex terrain, and its effective detection range and identification accuracy for low-altitude small targets are limited by size, power consumption, and cost. Radio detection relies on the electromagnetic radiation characteristics of the drone and its remote control link, and there is a risk of missed detection for targets using frequency hopping, encryption, or autonomous navigation, as well as targets with weak radiation and short-term communication. Optoelectronic imaging is limited by lighting, obstructions, rain, fog, and line-of-sight, and its monitoring capability decreases at night or in poor visibility. Acoustic arrays are susceptible to wind noise, traffic noise, and environmental reflections, and their positioning accuracy and robustness are insufficient at long distances or in complex terrain. Therefore, achieving stable drone monitoring under multi-interference, non-line-of-sight, and all-weather conditions remains a significant challenge. Summary of the Invention

[0004] To overcome, to some extent, the problem of insufficient stability in UAV monitoring methods for detecting, identifying, and tracking UAVs under complex interference, non-line-of-sight, and all-weather conditions, this application provides a UAV monitoring method and system based on the Doppler effect of seismographs.

[0005] The proposed solution is as follows:

[0006] According to a first aspect of the embodiments of this application, a method for monitoring using a drone based on the Doppler effect of a seismograph is provided, comprising: The ground vibration signal of the monitoring area is acquired by the seismograph, and the ground vibration signal is segmented according to the sliding time window to obtain the segmented vibration signal; The segmented vibration signal is subjected to detrending, denoising and filtering to obtain the processed vibration signal; Perform time-frequency analysis on the processed vibration signal to obtain the corresponding time spectrum; The drone signal in the time spectrum is extracted by artificial intelligence algorithm, the drone signal is input into a pre-trained deep learning model, and the time-frequency trajectory region of the drone signal in the time spectrum and its corresponding time range are output. Dominant frequency tracking is performed within the time-frequency trajectory region to obtain an instantaneous frequency sequence that varies with time, and frequency offset information characterizing the Doppler effect of relative motion is extracted based on the instantaneous frequency sequence. Based on the frequency shift information, a Doppler inversion constraint is constructed, and the estimation results of the UAV excitation frequency, flight speed, flight altitude and spatial position are obtained through optimization solution; The estimated results are subjected to trajectory consistency constraints and anomaly removal, and the UAV monitoring results are output and a monitoring report is generated.

[0007] Preferably, performing time-frequency analysis on the processed vibration signal to obtain the corresponding time spectrum includes: The processed vibration signal is windowed and framed according to a sliding time window. Perform a short-time Fourier transform on each frame of vibration signal to obtain the corresponding spectral information for each frame; The spectral information is combined in chronological order to form the time spectrum; The time spectrum is subjected to amplitude compression and normalization to obtain a standardized time spectrum for input into the UAV signal recognition model.

[0008] Preferably, the UAV signal in the time spectrum is extracted using an artificial intelligence algorithm, the UAV signal is input into a pre-trained deep learning model, and the time-frequency trajectory region of the UAV signal in the time spectrum and its corresponding time range are output, including: The time-frequency spectrum is subjected to multi-scale feature extraction using artificial intelligence algorithms to generate a probability map representing the probability of the existence of UAV signals; The probability map is input into a pre-trained deep learning model, which then extracts candidate time-frequency trajectory regions from the probability map based on a preset confidence threshold. Merging and deduplication processes are performed on overlapping or adjacent candidate time-frequency trajectory regions to determine the target time-frequency trajectory region; The corresponding time range is determined based on the start and end positions of the target time-frequency trajectory region on the time axis.

[0009] Preferably, the time-spectrum is subjected to multi-scale feature extraction using artificial intelligence algorithms to generate a probability map representing the probability of the existence of the UAV signal, including: A multi-scale time-frequency representation set is constructed for the time spectrum, and the multi-scale time-frequency representation set includes at least time-frequency representations with different time resolutions and different frequency resolutions; The multi-scale time-frequency representation set is input into a feature extraction network with shared parameters for feature encoding to obtain the corresponding multi-scale feature map. The multi-scale feature map is fused to output a pixel-level probability map of the presence of UAV signals. The probability map is used to characterize the probability of each time-frequency location belonging to the presence of UAV signals.

[0010] Preferably, dominant frequency tracking is performed within the time-frequency trajectory region to obtain an instantaneous frequency sequence that varies with time, and frequency shift information characterizing the Doppler effect of relative motion is extracted based on the instantaneous frequency sequence, including: Within the time-frequency trajectory region, peak search is performed on the frequency amplitude distribution corresponding to each moment along the time direction to determine the dominant frequency position at that moment; The instantaneous frequency sequence is obtained by connecting the dominant frequency positions at each moment in chronological order. The instantaneous frequency sequence is smoothed and jump points are removed to obtain a continuous main frequency change trajectory; Based on the upward and downward trends of the dominant frequency change trajectory, the frequency offset information is extracted. The frequency offset information is used to characterize the approach and departure motions of the UAV relative to the seismometer.

[0011] Preferably, the step of performing peak search on the frequency amplitude distribution corresponding to each moment within the time-frequency trajectory region to determine the dominant frequency position at that moment includes: Within the time-frequency trajectory region, the frequency amplitude sequence corresponding to each moment is obtained; Local extrema detection is performed on the frequency amplitude sequence and an amplitude threshold is set to filter out candidate peak values ​​that are lower than the amplitude threshold; The peak with the largest amplitude among the candidate peaks that have been filtered out is selected as the dominant frequency position at that moment; When the main frequency position shifts between adjacent time points exceed a preset jump threshold, the main frequency position is corrected based on the continuity constraint of the main frequency position between adjacent time points in order to suppress the main frequency jump caused by false detection peaks.

[0012] Preferably, the estimation results of the UAV's excitation frequency, flight speed, flight altitude, and spatial position are obtained through optimization, including: Based on the deployment location of the seismograph, a geometric relationship model of the UAV relative to the seismograph is established, and the excitation frequency, flight speed, flight altitude and spatial position of the UAV are set as parameters to be estimated. Based on the geometric relationship model and the parameters to be estimated, predictive frequency offset information is generated; The predicted frequency offset information is matched with the frequency offset information to calculate the frequency offset residual; An optimization problem is constructed with the goal of minimizing the frequency offset residual, and constraints are set on the range of values ​​and the parameter increment of the parameter to be estimated. The optimization problem is solved iteratively until the preset convergence condition is met, and the estimated results of the UAV's excitation frequency, flight speed, flight altitude and spatial position are output.

[0013] Preferably, setting the range constraints and increment constraints for the parameter to be estimated includes: Based on the UAV model library or preset flight safety rules, set corresponding allowable value ranges for the flight speed and the flight altitude, respectively. Based on the spatial boundary of the monitoring area and the effective detection radius of the seismograph, an allowable search range is set for the spatial location; The parameter increment upper limit is set for the flight speed, flight altitude and spatial position corresponding to adjacent sliding time windows to form the parameter increment constraint; When the estimation result corresponding to any sliding time window violates the allowed value range or the upper limit of the parameter increment, the estimation result corresponding to the sliding time window is judged as abnormal and a re-estimation process is triggered.

[0014] Preferably, the estimation results are subjected to continuity constraints and anomaly removal, including: Perform a trajectory consistency check on the flight speed, flight altitude, and spatial position output by adjacent sliding time windows to obtain a consistency evaluation result; When the consistency evaluation result does not meet the preset consistency conditions, the estimation result of the corresponding sliding time window is marked as an outlier. Interpolation repair or re-estimation is performed on the anomaly points to obtain the corrected monitoring trajectory; The corrected monitoring trajectory is subjected to a smoothing filter, and the smoothed monitoring trajectory is output as the monitoring result of the UAV.

[0015] According to a second aspect of the embodiments of this application, a drone monitoring system based on the Doppler effect of a seismograph is provided, comprising: Processor and memory; The processor and memory are connected via a communication bus: The processor is used to call and execute the program stored in the memory; The memory is used to store a program, which is at least used to execute a UAV monitoring method based on the Doppler effect of a seismograph as described in any of the above.

[0016] The technical solution provided in this application may include the following beneficial effects: This technical solution performs sliding window segmentation preprocessing and time-frequency analysis on ground vibration signals acquired by seismometers. It then utilizes artificial intelligence algorithms and deep learning models to automatically locate the time-frequency trajectory region and corresponding time range of UAV signals within the time spectrum, improving the accuracy of UAV signal detection and positioning under complex background vibration interference. Furthermore, it performs dominant frequency tracking within the trajectory region and extracts time-varying frequency offset information, enabling a quantifiable characterization of the Doppler effect. This allows for joint estimation of the UAV's excitation frequency, flight speed, flight altitude, and spatial position based on inversion constraints and optimization solutions. Finally, by combining trajectory consistency constraints and anomaly removal, it suppresses estimation jumps caused by short-term noise, improving the continuity and stability of monitoring results and automatically generating traceable monitoring reports.

[0017] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0018] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0019] Figure 1 This is a flowchart illustrating a UAV monitoring method based on the Doppler effect of a seismograph, provided in one embodiment of this application. Figure 2 This is a schematic diagram of the structure of an unmanned aerial vehicle (UAV) monitoring system based on the Doppler effect of a seismograph, provided in one embodiment of this application.

[0020] Reference numerals: Processor-21; Memory-22. Detailed Implementation

[0021] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0022] Example 1 Traditional drone monitoring methods, such as radar detection, radio spectrum detection, photoelectric imaging, and acoustic arrays, are easily affected by clutter in complex terrain, limiting their ability to detect small low-altitude targets and posing a risk of missed detections. Furthermore, they are susceptible to interference from lighting, obstructions, rain, fog, line-of-sight, wind noise, traffic noise, and environmental reflections, making it a significant challenge to achieve stable drone monitoring under conditions of multiple interferences, non-line-of-sight, and all-weather conditions.

[0023] In response, this application proposes a UAV monitoring method based on the Doppler effect of seismographs, referring to... Figure 1 ,include: S1. Acquire ground vibration signals in the monitoring area collected by the seismograph, and segment the ground vibration signals according to the sliding time window to obtain segmented vibration signals; S2. Perform detrending, denoising, and filtering on the segmented vibration signal to obtain the processed vibration signal; S3. Perform time-frequency analysis on the processed vibration signal to obtain the corresponding time spectrum; S4. Extract the drone signal from the time spectrum using artificial intelligence algorithms, input the drone signal into a pre-trained deep learning model, and output the time-frequency trajectory region of the drone signal in the time spectrum and its corresponding time range. S5. Perform main frequency tracking within the time-frequency trajectory region to obtain the instantaneous frequency sequence that changes with time, and extract frequency offset information representing the Doppler effect of relative motion based on the instantaneous frequency sequence. S6. Construct Doppler inversion constraints based on frequency offset information, and obtain the estimation results of UAV excitation frequency, flight speed, flight altitude and spatial position through optimization solution; S7. Perform trajectory consistency constraints and anomaly removal on the estimation results, output the UAV monitoring results and generate a monitoring report.

[0024] For ease of understanding, the following explains some key terms in this embodiment: A seismograph is a sensor used to sense and record ground vibrations. It collects ground vibration information by converting weak mechanical vibrations of the ground into electrical signals.

[0025] The Doppler effect refers to the phenomenon that the frequency of a wave received by a receiver changes when there is relative motion between the wave source and the receiver. In drone monitoring, the frequency of sound waves or vibration waves generated by the drone propellers shifts during propagation due to the relative motion between the drone and the seismometer.

[0026] Sliding time windows are a signal processing technique that performs localized processing or analysis of a signal by sliding a window of fixed length across a time series. This method helps capture local characteristics of a signal and adapts to its non-stationarity.

[0027] Time-frequency analysis (TF-F) is a signal processing method used to simultaneously analyze the characteristics of a signal in the time and frequency domains. Through TF-F, the time spectrum of a signal can be obtained, thus revealing the variation of the signal's frequency components over time.

[0028] The time spectrum is the result of time-frequency analysis, usually represented as a two-dimensional image. One dimension represents time, and the other dimension represents frequency. The color or brightness of the image represents the signal energy or amplitude at that time and frequency point.

[0029] A deep learning model is a trained model that can identify unique time-frequency patterns or trajectories of drone signals from the time spectrum. This model acquires the ability to distinguish drone signals by learning from a large number of samples of drone signals and background noise.

[0030] Frequency offset information characterizes the specific quantitative data of the Doppler effect, that is, the deviation of the instantaneous frequency of the UAV signal from its excitation frequency. This information directly reflects the relative velocity between the UAV and the seismograph. This embodiment provides a UAV monitoring method based on the seismograph Doppler effect.

[0031] First, ground vibration signals from the monitoring area are acquired using seismographs. Seismographs are deployed within the monitoring area to continuously acquire ground vibration signals. These raw signals can be continuous time-series data. For ease of subsequent processing, this continuous signal can be manually or automatically divided into a series of signal segments with fixed or variable durations. For example, the continuous signal can be simply truncated into segments every N seconds, or segmented according to changes in signal energy, thus obtaining segmented vibration signals.

[0032] Furthermore, the segmented vibration signal undergoes detrending, denoising, and filtering. The received segmented vibration signal typically contains various interferences and noise. Detrending can eliminate DC components or slowly changing trends in the signal by subtracting the average value or fitting a low-order polynomial. Denoising can employ a simple thresholding method, treating signals below a certain amplitude threshold as noise and removing them. Filtering can use simple low-pass, high-pass, or band-pass filters; for example, by setting a fixed frequency range, frequency components outside that range are directly filtered out, thus obtaining the processed vibration signal.

[0033] Subsequently, time-frequency analysis is performed on the processed vibration signal to obtain the corresponding time spectrum. Time-frequency analysis of the processed vibration signal reveals the changes in its frequency components over time. One approach is to use Fourier transform to directly transform the entire signal to the frequency domain, but this method cannot reflect the frequency changes over time. Another approach is to use Short-Time Fourier Transform (STFT), dividing the signal into several overlapping or non-overlapping short frames, performing a Fourier transform on each frame, and then arranging the spectral information of each frame in chronological order to form the time spectrum. This time spectrum can be stored in the form of a two-dimensional matrix, where rows represent time, columns represent frequency, and matrix elements represent the energy or amplitude at the corresponding time and frequency points.

[0034] Next, the drone signal is extracted from the time-frequency spectrum using artificial intelligence algorithms. This drone signal is then input into a pre-trained deep learning model, which outputs the time-frequency trajectory region of the drone signal within the time-frequency spectrum and its corresponding time range. Within the identified time-frequency trajectory region, the dominant frequency of the drone signal needs to be further precisely located. One approach is to perform a simple maximum value search on the frequency amplitude distribution at each time point within this region, identifying the frequency point with the largest amplitude as the dominant frequency at that moment. Connecting these dominant frequency points in chronological order yields the instantaneous frequency sequence. Frequency offset information can be obtained by comparing this instantaneous frequency sequence with a preset reference frequency (e.g., the excitation frequency of the drone in a stationary state). For example, the difference between the instantaneous frequency and the reference frequency can be calculated directly.

[0035] Then, based on this frequency shift information, Doppler inversion constraints are constructed, and the estimated results of the UAV's excitation frequency, flight speed, flight altitude, and spatial position are obtained through optimization. There is a physical relationship between the frequency shift information and the UAV's motion state (speed, altitude, and position) and its excitation frequency. A simplified physical model can be established, setting the UAV's excitation frequency, flight speed, flight altitude, and spatial position as the parameters to be estimated. Based on this model, the theoretical frequency shift information can be predicted according to these unknown parameters.

[0036] Then, by comparing the actual extracted frequency offset information with the predicted frequency offset information, an error function is constructed. This error function can be optimized using a simple gradient descent method or an exhaustive search method to find the parameter combination that minimizes the error, thereby obtaining the estimation results of the UAV's excitation frequency, flight speed, flight altitude, and spatial position.

[0037] Finally, trajectory consistency constraints and anomaly removal are applied to the estimation results, and the UAV monitoring results are output and a monitoring report is generated. The estimation results obtained from the optimization solution may contain instantaneous errors or instabilities. To improve the reliability of the monitoring results, simple post-processing can be performed on these results. Trajectory consistency constraints can be achieved by checking whether the estimation results of adjacent time windows change within a reasonable range; for example, if the velocity or position changes too much in adjacent time windows, inconsistency is considered to exist. Anomaly removal can employ simple statistical methods; for example, estimation results that deviate excessively from the average value can be directly deleted.

[0038] Ultimately, the processed, smoother, and more consistent trajectory data is output as the drone monitoring results, and a monitoring report containing information such as time, location, and speed can be generated.

[0039] This technical solution performs sliding window segmentation preprocessing and time-frequency analysis on ground vibration signals acquired by seismometers. It then utilizes artificial intelligence algorithms and deep learning models to automatically locate the time-frequency trajectory region and corresponding time range of UAV signals within the time spectrum, improving the accuracy of UAV signal detection and positioning under complex background vibration interference. Furthermore, it performs dominant frequency tracking within the trajectory region and extracts time-varying frequency offset information, enabling a quantifiable characterization of the Doppler effect. This allows for joint estimation of the UAV's excitation frequency, flight speed, flight altitude, and spatial position based on inversion constraints and optimization solutions. Finally, by combining trajectory consistency constraints and anomaly removal, it suppresses estimation jumps caused by short-term noise, improving the continuity and stability of monitoring results and automatically generating traceable monitoring reports.

[0040] Example 2 It should be noted that time-frequency analysis is performed on the processed vibration signal to obtain the corresponding time spectrum, including: The processed vibration signal is windowed and framed according to a sliding time window. Perform a short-time Fourier transform on each frame of vibration signal to obtain the corresponding spectral information for each frame; The spectral information is combined in chronological order to form the time spectrum; The temporal spectrum is subjected to amplitude compression and normalization to obtain a standardized temporal spectrum for input into the UAV signal recognition model.

[0041] Specifically, the processed vibration signal is windowed and framed according to a sliding time window, aiming to divide the continuous vibration signal into a series of short-time signal segments with overlapping parts. This process reduces spectral leakage by applying a window function (such as a Hamming window or Hanning window), ensuring a smooth transition of each signal frame in the time domain, thus providing a stable data foundation for subsequent frequency analysis. The settings of the sliding time window, such as the window length and overlap rate, need to be optimized according to the characteristics of the UAV signal and the required time-frequency resolution to balance time resolution and frequency resolution. Subsequently, a short-time Fourier transform is performed on each frame of vibration signal to obtain the corresponding spectral information. The short-time Fourier transform is a common method for converting signals from the time domain to the time-frequency domain. It calculates the frequency components of the signal and their corresponding amplitudes within that time period by performing a Fourier transform on each windowed and framed signal segment. In this way, the frequency distribution of the signal at different time points can be obtained, thereby revealing the dynamic spectral characteristics of the signal. Then, the spectral information is combined in chronological order to form the time spectrum. This step involves arranging the spectral information of all frames in chronological order within the original signal, constructing a two-dimensional matrix. The horizontal axis of this matrix represents time, and the vertical axis represents frequency. The values ​​in the matrix represent the signal energy or amplitude at the corresponding time and frequency points. The time-frequency spectrum visually demonstrates the variation of signal frequency components over time and is crucial for identifying the Doppler effect characteristics of UAVs. Based on this, amplitude compression and normalization are performed on the time-frequency spectrum to obtain a standardized time-frequency spectrum used as input to the UAV signal recognition model. Amplitude compression typically employs logarithmic transformation (e.g., conversion to decibels) to reduce the dynamic range of the signal amplitude, making low-energy UAV signal characteristics more prominent in the time-frequency spectrum while suppressing the influence of high-energy background noise. Normalization maps the compressed time-frequency spectrum amplitude to a preset fixed range (e.g., 0 to 1), eliminating dimensional differences between different time-frequency spectra and ensuring that the data input to the UAV signal recognition model has a consistent scale, thereby improving the model's training efficiency and generalization ability.

[0042] Example 3 It should be noted that the UAV signal in the time spectrum is extracted using artificial intelligence algorithms, and the UAV signal is input into a pre-trained deep learning model. The model outputs the time-frequency trajectory region of the UAV signal in the time spectrum and its corresponding time range, including: By using artificial intelligence algorithms to extract multi-scale features from the time spectrum, a probability map representing the probability of the existence of drone signals is generated. The probability map is input into a pre-trained deep learning model, which then extracts candidate time-frequency trajectory regions from the probability map based on a pre-set confidence threshold. Merging and deduplication processes are performed on overlapping or adjacent candidate time-frequency trajectory regions to determine the target time-frequency trajectory region; The corresponding time range is determined based on the start and end positions of the target time-frequency trajectory region on the time axis.

[0043] Specifically, multi-scale feature extraction is performed on the time-spectrum to analyze the data from different granularities or perspectives, capturing the characteristics of UAV signals at different time scales and frequency resolutions. This can be achieved by applying a series of filters or convolution kernels of different sizes, allowing effective identification of UAV signals at other scales even if they are not prominent at a single scale. After feature extraction, the system generates a probability map, a two-dimensional matrix where each element corresponds to a time-frequency point in the time-spectrum, and the value of the element represents the probability that the time-frequency point belongs to a UAV signal. This probability value is typically between 0 and 1, with a higher value indicating a higher likelihood that the point is a UAV signal. After generating the probability map, candidate time-frequency trajectory regions are extracted from the probability map based on a preset confidence threshold. This confidence threshold is a preset value, such as 0.5, 0.7, or higher, used to filter out time-frequency points in the probability map that have a high probability of containing UAV signals. All time-frequency points with probability values ​​higher than this threshold are marked as potential UAV signal points, forming several discrete or continuous regions in the time-spectrum, which are the candidate time-frequency trajectory regions. This step effectively filters out low-confidence noise or interference, thus narrowing down the scope of subsequent processing. To further optimize the identification results, overlapping or adjacent candidate time-frequency trajectory regions are merged and deduplicated to determine the target time-frequency trajectory region. Due to noise or signal intermittency, the initially extracted candidate regions may be fragmented or redundant. The merging process aims to treat candidate regions that are very close in the time-frequency plane (e.g., less than a preset time-frequency distance threshold) or overlap on the time axis as part of the same UAV signal and integrate them into a larger, more complete region. The deduplication process ensures that each UAV signal corresponds to only one unique and definite "target time-frequency trajectory region". This can be achieved through connected component analysis, morphological operations, or density-based clustering algorithms to obtain more accurate and continuous UAV signal time-frequency trajectories. Finally, the corresponding time range is determined based on the start and end positions of the target time-frequency trajectory region on the time axis. For each target time-frequency trajectory region determined after merging and deduplication, the system identifies the minimum and maximum time points of that region on the time axis. These two time points constitute the corresponding time range of the drone signal, precisely indicating the start and end times of the drone signal within the monitoring period.

[0044] Furthermore, multi-scale feature extraction of the time-frequency spectrum is performed using artificial intelligence algorithms to generate a probability map representing the probability of the presence of UAV signals, including: Construct a multi-scale time-frequency representation set for the time spectrum, which includes time-frequency representations with different time resolutions and different frequency resolutions. The multi-scale time-frequency representation sets are respectively input into a feature extraction network with shared parameters for feature encoding to obtain the corresponding multi-scale feature maps; Scale fusion is performed on the multi-scale feature maps to output a pixel-level probability map of the existence of UAV signals. The probability map is used to characterize the probability of the existence of UAV signals at each time and frequency location.

[0045] Specifically, constructing a multi-scale time-frequency representation set aims to capture the time-frequency features of UAV signals from multiple observation granularities. For example, signals with short durations and rapid frequency changes can be captured using a time-frequency representation with high time resolution and low frequency resolution; while signals with long durations, slow frequency changes, but wide bandwidth can be captured using a time-frequency representation with low time resolution and high frequency resolution. This approach ensures that regardless of the time-frequency characteristics exhibited by the UAV signal, it can be effectively represented and analyzed. In practice, this can be achieved by resampling and smoothing the original time spectrum with different parameters, or by transforming it using wavelet basis functions at different scales, to generate time-frequency maps with different time-frequency resolution characteristics. Further, the multi-scale time-frequency representation set is input into a feature extraction network with shared parameters for feature encoding, resulting in corresponding multi-scale feature maps. Using a feature extraction network with shared parameters allows the network to learn generalized feature representations that generalize to inputs at different scales, thereby effectively reducing the number of model parameters, improving training efficiency, and enhancing the model's robustness to features at different scales. The feature encoding process transforms the original time-frequency representation into a more abstract and discriminative feature vector or feature map. These features better characterize the essential attributes of the UAV signal and suppress noise interference. For example, the feature extraction network can be a convolutional neural network (CNN), where the kernels and biases of each layer remain consistent when processing time-frequency representations at different scales. Based on this, scale fusion is performed on the multi-scale feature maps to output a pixel-level UAV signal existence probability map. This probability map represents the probability that each time-frequency location belongs to the UAV signal. The purpose of scale fusion is to effectively integrate complementary information extracted from different scales to form a more comprehensive and robust feature representation. Through fusion, the limitations of single-scale features can be overcome, improving the ability to recognize complex UAV signals. The final output pixel-level probability map, where each pixel value represents the probability that the corresponding time-frequency point belongs to the UAV signal, provides a refined basis for subsequent accurate identification of the time-frequency trajectory region of the UAV signal. Specific methods for scale fusion may include, but are not limited to: concatenating feature maps of different scales along the channel dimension and then processing them through an additional convolutional layer; or summing or weighting the feature maps of different scales element-wise; or introducing an attention mechanism to dynamically allocate fusion weights based on the importance of the feature maps. The fused feature map is then passed through a classification layer (e.g., a 1x1 convolutional layer followed by a sigmoid activation function) to output the probability value of each pixel.

[0046] Example 4 It should be noted that dominant frequency tracking is performed within the time-frequency trajectory region to obtain an instantaneous frequency sequence that varies with time. Based on this instantaneous frequency sequence, frequency shift information characterizing the Doppler effect of relative motion is extracted, including: Within the time-frequency trajectory region, peak search is performed on the frequency amplitude distribution corresponding to each moment along the time direction to determine the dominant frequency position at that moment. By connecting the dominant frequency positions at each moment in chronological order, an instantaneous frequency sequence is obtained. The instantaneous frequency sequence is smoothed and jump points are removed to obtain a continuous trajectory of the main frequency change; Based on the upward and downward trends of the dominant frequency change trajectory, frequency offset information is extracted. This frequency offset information is used to characterize the approach and departure motions of the UAV relative to the seismometer.

[0047] Specifically, within the time-frequency trajectory region, a peak search is performed on the frequency amplitude distribution corresponding to each moment along the time direction to determine the dominant frequency position at that moment. This step aims to identify the most significant energy concentration point of the UAV signal in the frequency dimension at a specific time point, i.e., its dominant frequency component. In implementation, for each time slice (i.e., the frequency amplitude distribution at a certain moment) in the time-frequency trajectory region, a peak detection algorithm can be applied to identify the frequency point with the largest amplitude or exceeding a preset threshold as the dominant frequency position at that moment. For example, all frequency points within the time slice can be traversed, and the frequency point with the largest amplitude can be selected as the dominant frequency. The dominant frequency positions at each moment are connected in chronological order to obtain the instantaneous frequency sequence. Once the dominant frequency position at each moment is determined, these discrete dominant frequency points are concatenated in their corresponding chronological order to form a continuous sequence. This sequence intuitively reflects the dynamic process of the dominant frequency of the UAV signal changing over time, which is the basis for subsequent analysis of the Doppler effect. The instantaneous frequency sequence is smoothed and abrupt transition points are removed to obtain a continuous dominant frequency change trajectory. Due to environmental noise, signal interference, or random errors during peak search, the original instantaneous frequency sequence may contain irregular fluctuations or sudden jumps. To improve the accuracy of subsequent analysis, the sequence needs to be processed. Smoothing can be achieved using methods such as moving average filtering, Savitzky-Golay filtering, or Kalman filtering to eliminate high-frequency noise and make the frequency change trend clearer. Jump point removal can be achieved by setting a frequency change rate threshold, detecting statistical outliers (e.g., based on standard deviation), or combining physical constraints (e.g., frequency change limits corresponding to the maximum acceleration of the UAV) to identify and remove or correct abrupt frequency points that do not conform to actual physical laws. Removed jump points can be repaired by interpolation (e.g., linear interpolation, spline interpolation) to obtain a more continuous, stable, and realistic dominant frequency change trajectory. Based on the upward and downward trends of the dominant frequency change trajectory, the frequency offset information is extracted. This frequency offset information is used to characterize the approach and departure motions of the UAV relative to the seismometer. The core manifestation of the Doppler effect is that when there is relative motion between the sound source (drone) and the receiver (seismograph), the received frequency shifts. When the drone approaches the seismograph, the received frequency increases relative to its inherent excitation frequency, exhibiting an upward trend; when the drone moves away from the seismograph, the received frequency decreases, exhibiting a downward trend. By analyzing the smoothed dominant frequency trajectory, a reference frequency can be determined (e.g., the excitation frequency of the drone at rest or in a specific state, or the average frequency along the trajectory). The difference between the instantaneous frequency and this reference frequency is then calculated; this is the frequency shift information. The time series of this frequency shift information directly quantifies the degree and direction of the drone's approach or departure from the seismograph.

[0048] Specifically, peak search is performed on the frequency amplitude distribution corresponding to each moment within the time-frequency trajectory region to determine the dominant frequency position at that moment, including: Within the time-frequency trajectory region, the frequency amplitude sequence corresponding to each time moment is obtained; Local extrema detection is performed on the frequency amplitude sequence and an amplitude threshold is set to filter out candidate peaks below the amplitude threshold; The peak with the largest amplitude among the candidate peaks that have been filtered out is selected as the dominant frequency position at that moment; When the main frequency position shifts between adjacent time points exceed the preset jump threshold, the main frequency position is corrected based on the continuity constraint of the main frequency position between adjacent time points in order to suppress the main frequency jump caused by false detection peaks.

[0049] Specifically, within the time-frequency trajectory region, acquiring the frequency amplitude sequence corresponding to each moment means that within the identified UAV signal time-frequency trajectory region, for each discrete time point, the system extracts all frequency components and their corresponding amplitude information to form a one-dimensional frequency amplitude sequence. This sequence reflects the distribution of signal energy at different frequencies at a specific moment and is the basic data for dominant frequency identification. Subsequently, local extremum detection is performed on the frequency amplitude sequence, and an amplitude threshold is set to filter out candidate peaks below the amplitude threshold. Local extremum detection aims to identify all potential peak points in the frequency amplitude sequence, which represent frequencies where signal energy is relatively concentrated at that moment. Common local extremum detection methods include finding frequency points with amplitudes greater than their left and right neighboring points. To distinguish between real signal peaks and spurious peaks caused by noise or weak interference, an amplitude threshold is preset. Any detected local extremum whose amplitude is below the preset threshold is considered an invalid peak and filtered out. This amplitude threshold can be dynamically adjusted based on experience, signal characteristics, or noise levels to ensure that only sufficiently strong signal components are considered subsequently. Next, the peak with the largest amplitude is selected from the candidate peaks that have passed the screening as the dominant frequency position at that moment. After local extremum detection and amplitude threshold screening, there may still be multiple candidate peaks that meet the conditions. In this step, the system selects the frequency point with the largest amplitude (i.e., the strongest energy) from these remaining candidate peaks as the dominant frequency position at the current moment. This strategy is based on the characteristic that UAV signals usually exhibit the most concentrated energy on the time-frequency graph, thereby ensuring that the identified dominant frequency is the most significant signal component at the current moment. Finally, when the dominant frequency position shift between adjacent moments exceeds a preset jump threshold, the dominant frequency position is corrected based on the continuity constraint of the dominant frequency position between adjacent moments to suppress the dominant frequency jump caused by falsely detected peaks. When a UAV is flying in the air, the Doppler frequency shift it generates is usually continuous and smooth. Therefore, there should be no drastic jump in the dominant frequency position between adjacent moments. This application introduces a preset jump threshold to determine whether the difference between the dominant frequency position identified at the current moment and the dominant frequency position at the previous moment (or the next moment) is too large. If the offset exceeds the threshold, the current master frequency is considered to be potentially a false detection. In this case, the system will correct the error based on the continuity constraint of the master frequency positions at adjacent time points. For example, it can re-evaluate the candidate peaks at the current time point, prioritizing peaks closer to the master frequency positions at adjacent time points; or, if a reasonable peak cannot be found, interpolation or extrapolation methods can be used to estimate the current master frequency position based on the master frequency trends at previous stable time points. This correction mechanism effectively utilizes the physical continuity of UAV signal frequency changes, significantly enhancing the robustness of master frequency tracking and avoiding false identifications caused by noise or transient interference.

[0050] Example 5 It should be noted that the estimated results of the UAV's excitation frequency, flight speed, flight altitude, and spatial position obtained through optimization include: A geometric relationship model of the UAV relative to the seismograph is established based on the deployment location of the seismograph, and the excitation frequency, flight speed, flight altitude and spatial position of the UAV are set as parameters to be estimated. Predicted frequency offset information is generated based on the geometric relationship model and the parameters to be estimated; The predicted frequency offset information is matched with the frequency offset information to calculate the frequency offset residual; An optimization problem is constructed with the goal of minimizing the frequency offset residual, and constraints are set on the range of values ​​of the parameters to be estimated and the parameter increment. The optimization problem is solved by iterative optimization until the preset convergence condition is met, and the estimated results of the UAV excitation frequency, flight speed, flight altitude and spatial position are output.

[0051] In this embodiment, firstly, a geometric relationship model of the UAV relative to the seismograph is established based on the seismograph's deployment location, and the UAV's excitation frequency, flight speed, flight altitude, and spatial position are set as parameters to be estimated. The geometric relationship model is a mathematical expression describing the relative motion state between the UAV and the seismograph, and its function is to establish a physical connection between the UAV's spatial position, flight speed, and other parameters and the Doppler frequency shift received by the seismograph. This model can be constructed based on the UAV's coordinates in three-dimensional space, its flight speed vector, and the seismograph's known deployment location. For example, the Doppler frequency shift can be derived by calculating the rate of change of distance between the UAV and the seismograph. The UAV's excitation frequency, flight speed, flight altitude, and spatial position are set as parameters to be estimated; these parameters are core variables describing the UAV's motion state and acoustic characteristics, and their accurate determination is crucial for achieving UAV monitoring. Secondly, predicted frequency shift information is generated based on the geometric relationship model and the parameters to be estimated. This step is a forward modeling process. Based on the currently assumed parameters such as the UAV's excitation frequency, flight speed, flight altitude, and spatial position, and combined with the established geometric model, the theoretically expected Doppler frequency shift of the seismograph is calculated. This is equivalent to simulating and predicting the Doppler effect generated by the UAV's motion, providing a benchmark for subsequent comparison with actual observation data. Next, the predicted frequency shift information is matched with the actual frequency shift information to calculate the frequency shift residual. This step aims to quantify the difference between theoretical predictions and actual observations. By matching the generated predicted frequency shift information point-by-point or segment-by-segment with the actual frequency shift information extracted from the seismic signal, the difference between the two can be calculated, i.e., the frequency shift residual. This residual value directly reflects the degree of deviation between the currently estimated parameters and the actual parameters, serving as the basis for the optimization algorithm to adjust parameters to reduce errors. Then, an optimization problem is constructed with the goal of minimizing the frequency shift residual, and constraints are set on the range of values ​​for the estimated parameters and the parameter increment. The optimization problem is to transform UAV parameter estimation into a mathematical solution process. Its core objective is to find a set of parameters to be estimated that minimizes the residual between the predicted and actual frequency offset information. To ensure the physical rationality and stability of the optimization results, constraints are introduced. Range constraints limit the estimated parameters to a practically feasible physical range; for example, flight speed and altitude should be within the allowable range for the UAV model, and spatial position should be within the monitoring area. Parameter increment constraints limit the magnitude of parameter changes within adjacent time windows, preventing unreasonable and drastic jumps in the estimation results, thereby improving the smoothness and continuity of the trajectory. Finally, iterative optimization is used to solve the optimization problem until the preset convergence condition is met, outputting the estimated results of the UAV's excitation frequency, flight speed, flight altitude, and spatial position. Iterative optimization is a numerical method that gradually approximates the optimal solution.In each iteration, the optimization algorithm adjusts the parameters to be estimated based on the current frequency offset residuals and constraints, and recalculates the predicted frequency offset information until the residuals are sufficiently small or the parameter changes are below a preset threshold, thus reaching the preset convergence condition. Commonly used iterative optimization algorithms include, but are not limited to, gradient descent and the Levenberg-Marquardt algorithm. Through this iterative process, the final output is the converged estimation results of the UAV's excitation frequency, flight speed, flight altitude, and spatial position. These results represent the best estimates of the UAV's state under the current model and data. The constraints on the range of values ​​for the parameters to be estimated and the parameter increment constraints include: Based on the drone model library or preset flight safety rules, set corresponding allowable value ranges for flight speed and flight altitude respectively; Based on the spatial boundaries of the monitoring area and the effective detection radius of the seismograph, an allowable search range is set for the spatial location; Set upper limits for parameter increments for the flight speed, flight altitude, and spatial position corresponding to adjacent sliding time windows to form parameter increment constraints; When the estimation result corresponding to any sliding time window violates the allowed value range or the upper limit of parameter increment, the estimation result corresponding to that sliding time window is judged as abnormal and a re-estimation process is triggered.

[0052] Specifically, based on a UAV model database or preset flight safety rules, corresponding allowable value ranges are set for the flight speed and flight altitude, respectively. This measure aims to set reasonable physical boundaries for the UAV's flight speed and altitude. In actual flight, the speed and altitude of a UAV are limited by various factors such as its own design, dynamic characteristics, airspace management, and safety regulations. During implementation, a database containing parameters such as the maximum / minimum flight speed and maximum flight altitude of known UAV models can be established to automatically apply the corresponding parameter range when a specific UAV model is detected. In addition, for general UAVs or specific monitoring scenarios, general safe flight parameters can be preset, such as the maximum flight altitude limit for civilian UAVs. These rules can serve as hard constraints to ensure the physical rationality of the estimation results. In the optimization solver, these allowable value ranges can serve as boundary conditions to limit the search space of the parameters to be estimated. At the same time, based on the spatial boundary of the monitoring area and the effective detection radius of the seismograph, an allowable search range is set for the spatial location. This feature is used to limit the estimation range of the UAV's spatial location, making it conform to the geographical boundaries of the actual monitoring area and the detection capability of the seismograph. The spatial boundaries of the monitoring area can be defined using geographic information system data or manually, defining the latitude, longitude, altitude range, or three-dimensional coordinate boundaries to ensure that the estimated position of the UAV does not exceed these boundaries. The seismograph's detection capability for UAV signals is limited; beyond a certain distance, the signal attenuates to the point of being unrecognizable. Therefore, a maximum effective detection distance can be set based on the seismograph's sensitivity, environmental noise level, and UAV signal strength. The estimated position of the UAV should be within an area centered on the seismograph and with an effective detection radius as its radius. These ranges also serve as constraints in the optimization problem. Furthermore, upper limits for parameter increments are set for the flight speed, flight altitude, and spatial position corresponding to adjacent sliding time windows, constituting the parameter increment constraints. This feature aims to introduce temporal continuity constraints, preventing unreasonable drastic parameter changes between adjacent time windows, thereby improving the smoothness and stability of trajectory estimation. The UAV's motion is continuous; its speed, altitude, and position do not change abruptly in a short period. For each parameter to be estimated, a maximum allowable change between adjacent sliding time windows is set. These incremental upper limits can serve as inequality constraints in optimization problems, or, during iterative optimization, as soft constraints or regularization terms used for the current time window's optimization, incorporating the estimation results of the previous time window. The setting of the incremental upper limits should consider dynamic performance parameters such as the UAV's maximum acceleration, maximum climb / descent rate, and maximum turn rate. When the estimation result corresponding to any sliding time window violates the allowed value range or the parameter incremental upper limit, the estimation result corresponding to that sliding time window is judged as abnormal and a re-estimation process is triggered. This feature provides an error detection and recovery mechanism, ensuring that the final output estimation result always conforms to the preset physical and kinematic constraints.After the optimization solution is completed in each sliding time window, the system immediately checks the estimated results of the UAV excitation frequency, flight speed, flight altitude, and spatial position. If any parameter exceeds its preset allowable range, or if its change exceeds the upper limit of the parameter increment compared to the corresponding parameter in the previous time window, the estimation result is marked as abnormal. When an anomaly is detected, strategies such as re-optimization, backtracking, interpolation, or issuing warnings can be adopted for handling. By setting constraints on the range of values ​​and parameter increments of the parameters to be estimated, this application can effectively solve the problems of convergence difficulties, getting trapped in local optima, and producing estimation results that do not conform to the actual physical meaning that may occur during the optimization solution process. Specifically, by setting allowable ranges for flight speed and flight altitude based on the UAV model library or preset flight safety rules, and setting allowable search ranges for spatial position based on the spatial boundary of the monitoring area and the effective detection radius of the seismograph, the optimization search space can be limited to a physically feasible area, avoiding the generation of estimation values ​​that exceed the actual physical limitations. Furthermore, setting upper limits for parameter increments for flight speed, flight altitude, and spatial position corresponding to adjacent sliding time windows introduces the continuity characteristic of UAV motion, effectively suppressing drastic jumps in the estimation results over time, making the UAV trajectory smoother and more consistent with actual motion patterns. When the estimation result of any sliding time window violates these constraints, the system can promptly identify it as an anomaly and trigger reassessment, thereby improving the robustness and accuracy of the monitoring results and ensuring the reliability and physical rationality of the UAV monitoring results.

[0053] Example 6 It should be noted that continuity constraints and outlier removal are applied to the estimation results, including: Perform a trajectory consistency check on the flight speed, flight altitude and spatial position output by adjacent sliding time windows to obtain consistency evaluation results; When the consistency evaluation results do not meet the preset consistency conditions, the estimation results of the corresponding sliding time window will be marked as outliers. Interpolation repair or re-estimation is performed on outliers to obtain the corrected monitoring trajectory; The corrected monitoring trajectory is smoothed and filtered, and the smoothed monitoring trajectory is output as the UAV monitoring result.

[0054] The process involves performing a trajectory consistency check on the flight speed, flight altitude, and spatial position output from adjacent sliding time windows to obtain a consistency evaluation result. This aims to assess the rationality and smoothness of changes in UAV flight parameters (such as flight speed, flight altitude, and spatial position) over a continuous time period. Due to the physical motion characteristics of UAVs, their flight parameters should not undergo drastic or physically incompatible changes within a short period. This check can be performed by calculating the rate of change, acceleration, or positional deviation of flight parameters between adjacent time windows. For example, the difference in flight speed between adjacent time windows can be compared to see if it is within a reasonable range, or the distance and time interval between adjacent spatial positions can be calculated to see if they meet the maximum speed limit of the UAV. The consistency evaluation result can be a Boolean value (consistent / inconsistent) or a quantified inconsistency index. When the consistency evaluation result does not meet the preset consistency conditions, the estimation result of the corresponding sliding time window is marked as an outlier. The preset consistency conditions are the criteria for judging whether the trajectory is reasonable. When the trajectory consistency check result indicates abnormal changes in flight parameters, the estimation result of that time window is considered unreliable and needs to be marked for subsequent processing. The preset consistency conditions may include: a flight speed change rate threshold, a flight altitude change rate threshold, and a spatial position change distance threshold, etc. For example, if the velocity change in adjacent time windows exceeds a certain preset threshold, it is considered that the consistency condition is not met. Marking outliers can be achieved by setting a flag in the data structure or storing them in a separate outlier list. Interpolation repair or reestimation processing is performed on the outliers to obtain the corrected monitoring trajectory. Outlier repair is a key step in improving the continuity and accuracy of the monitoring trajectory. Interpolation repair uses the trend of normal points before and after the outlier to estimate the value of the outlier, and can use methods such as linear interpolation, spline interpolation, and polynomial interpolation. For example, for an outlier velocity point, it can be replaced by the average or weighted average of its preceding and following normal velocity points. Reestimation processing attempts to recalculate the value of the outlier. The optimization solution process for that time window can be rerun, but it may require adjusting optimization parameters or introducing stronger constraints, or using the estimation results of adjacent normal time windows as prior information for assistance. Smoothing filtering is performed on the corrected monitoring trajectory, and the smoothed monitoring trajectory is output as the UAV monitoring result. Even after outlier repair, the trajectory may still contain a certain degree of noise or minor fluctuations. Smoothing filters aim to eliminate these residual noises, making the final monitoring trajectory smoother, more consistent with the laws of physical motion, and improving visual readability. Various filtering algorithms can be employed, such as moving average filtering, Kalman filtering, Gaussian filtering, and Savitzky-Golay filtering. For example, moving average filtering smooths the data by calculating the average of each point and its neighbors. Kalman filtering, on the other hand, can be combined with the UAV's motion model to provide a more accurate estimate and smoothing of the trajectory.

[0055] Example 7 A drone monitoring system based on the Doppler effect of seismographs, referring to Figure 2 ,include: Processor and memory; The processor and memory are connected via a communication bus: The processor is used to call and execute programs stored in memory. A memory for storing a program, which is at least used to execute a UAV monitoring method based on the seismograph Doppler effect as described in any of the above embodiments.

[0056] It is understood that the same or similar parts in the above embodiments can be referred to each other, and the contents not described in detail in some embodiments can be referred to the same or similar contents in other embodiments.

[0057] It should be noted that in the description of this application, the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance. Furthermore, in the description of this application, unless otherwise stated, "a plurality of" means at least two.

[0058] Any process or method described in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing a particular logical function or process, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the function involved, as will be understood by those skilled in the art to which embodiments of this application pertain.

[0059] It should be understood that various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0060] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

[0061] Furthermore, the functional units in the various embodiments of this application can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.

[0062] The storage media mentioned above can be read-only memory, disk, or optical disk, etc.

[0063] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. 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.

[0064] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. A UAV monitoring method based on the Doppler effect of a seismograph, characterized in that, include: The ground vibration signal of the monitoring area is acquired by the seismograph, and the ground vibration signal is segmented according to the sliding time window to obtain the segmented vibration signal; The segmented vibration signal is subjected to detrending, denoising and filtering to obtain the processed vibration signal; Perform time-frequency analysis on the processed vibration signal to obtain the corresponding time spectrum; The drone signal in the time spectrum is extracted by artificial intelligence algorithm, the drone signal is input into a pre-trained deep learning model, and the time-frequency trajectory region of the drone signal in the time spectrum and its corresponding time range are output. Dominant frequency tracking is performed within the time-frequency trajectory region to obtain an instantaneous frequency sequence that varies with time, and frequency offset information characterizing the Doppler effect of relative motion is extracted based on the instantaneous frequency sequence. Based on the frequency shift information, a Doppler inversion constraint is constructed, and the estimation results of the UAV excitation frequency, flight speed, flight altitude and spatial position are obtained through optimization solution; The estimated results are subjected to trajectory consistency constraints and anomaly removal, and the UAV monitoring results are output and a monitoring report is generated.

2. The UAV monitoring method based on the Doppler effect of a seismograph according to claim 1, characterized in that, The step of performing time-frequency analysis on the processed vibration signal to obtain the corresponding time spectrum includes: The processed vibration signal is windowed and framed according to a sliding time window. Perform a short-time Fourier transform on each frame of vibration signal to obtain the corresponding spectral information for each frame; The spectral information is combined in chronological order to form the time spectrum; The time spectrum is subjected to amplitude compression and normalization to obtain a standardized time spectrum for input into the UAV signal recognition model.

3. The UAV monitoring method based on the Doppler effect of a seismograph according to claim 1, characterized in that, The drone signal in the time spectrum is extracted using an artificial intelligence algorithm. The drone signal is then input into a pre-trained deep learning model, which outputs the time-frequency trajectory region of the drone signal in the time spectrum and its corresponding time range, including: The time-frequency spectrum is subjected to multi-scale feature extraction using artificial intelligence algorithms to generate a probability map representing the probability of the existence of UAV signals; The probability map is input into a pre-trained deep learning model, which then extracts candidate time-frequency trajectory regions from the probability map based on a preset confidence threshold. Merging and deduplication processes are performed on overlapping or adjacent candidate time-frequency trajectory regions to determine the target time-frequency trajectory region; The corresponding time range is determined based on the start and end positions of the target time-frequency trajectory region on the time axis.

4. The UAV monitoring method based on the Doppler effect of a seismograph according to claim 3, characterized in that, The time-spectrum is subjected to multi-scale feature extraction using artificial intelligence algorithms to generate a probability map representing the probability of the presence of UAV signals, including: A multi-scale time-frequency representation set is constructed for the time spectrum, and the multi-scale time-frequency representation set includes at least time-frequency representations with different time resolutions and different frequency resolutions; The multi-scale time-frequency representation set is input into a feature extraction network with shared parameters for feature encoding to obtain the corresponding multi-scale feature map. The multi-scale feature map is fused to output a pixel-level probability map of the presence of UAV signals. The probability map is used to characterize the probability of each time-frequency location belonging to the presence of UAV signals.

5. The UAV monitoring method based on the Doppler effect of a seismograph according to claim 1, characterized in that, Dominant frequency tracking is performed within the time-frequency trajectory region to obtain a time-varying instantaneous frequency sequence. Based on the instantaneous frequency sequence, frequency shift information characterizing the Doppler effect of relative motion is extracted, including: Within the time-frequency trajectory region, peak search is performed on the frequency amplitude distribution corresponding to each moment along the time direction to determine the dominant frequency position at that moment; The instantaneous frequency sequence is obtained by connecting the dominant frequency positions at each moment in chronological order. The instantaneous frequency sequence is smoothed and jump points are removed to obtain a continuous main frequency change trajectory; Based on the upward and downward trends of the dominant frequency change trajectory, the frequency offset information is extracted. The frequency offset information is used to characterize the approach and departure motions of the UAV relative to the seismometer.

6. The UAV monitoring method based on the Doppler effect of a seismograph according to claim 5, characterized in that, The step of performing peak search on the frequency amplitude distribution corresponding to each moment within the time-frequency trajectory region to determine the dominant frequency position at that moment includes: Within the time-frequency trajectory region, the frequency amplitude sequence corresponding to each moment is obtained; Local extrema detection is performed on the frequency amplitude sequence and an amplitude threshold is set to filter out candidate peak values ​​that are lower than the amplitude threshold; The peak with the largest amplitude among the candidate peaks that have been filtered out is selected as the dominant frequency position at that moment; When the main frequency position shifts between adjacent time points exceed a preset jump threshold, the main frequency position is corrected based on the continuity constraint of the main frequency position between adjacent time points in order to suppress the main frequency jump caused by false detection peaks.

7. The UAV monitoring method based on the Doppler effect of a seismograph according to claim 1, characterized in that, The estimation results of the UAV's excitation frequency, flight speed, flight altitude, and spatial position are obtained through optimization, including: Based on the deployment location of the seismograph, a geometric relationship model of the UAV relative to the seismograph is established, and the excitation frequency, flight speed, flight altitude and spatial position of the UAV are set as parameters to be estimated. Based on the geometric relationship model and the parameters to be estimated, predictive frequency offset information is generated; The predicted frequency offset information is matched with the frequency offset information to calculate the frequency offset residual; An optimization problem is constructed with the goal of minimizing the frequency offset residual, and constraints are set on the range of values ​​and the parameter increment of the parameter to be estimated. The optimization problem is solved iteratively until the preset convergence condition is met, and the estimated results of the UAV's excitation frequency, flight speed, flight altitude and spatial position are output.

8. The UAV monitoring method based on the Doppler effect of a seismograph according to claim 7, characterized in that, Setting the range constraints and increment constraints for the parameter to be estimated includes: Based on the UAV model library or preset flight safety rules, set corresponding allowable value ranges for the flight speed and the flight altitude, respectively. Based on the spatial boundary of the monitoring area and the effective detection radius of the seismograph, an allowable search range is set for the spatial location; The parameter increment upper limit is set for the flight speed, flight altitude and spatial position corresponding to adjacent sliding time windows to form the parameter increment constraint; When the estimation result corresponding to any sliding time window violates the allowed value range or the upper limit of the parameter increment, the estimation result corresponding to the sliding time window is judged as abnormal and a re-estimation process is triggered.

9. The UAV monitoring method based on the Doppler effect of a seismograph according to claim 1, characterized in that, The estimation results are subjected to continuity constraints and outlier removal, including: Perform a trajectory consistency check on the flight speed, flight altitude, and spatial position output by adjacent sliding time windows to obtain a consistency evaluation result; When the consistency evaluation result does not meet the preset consistency conditions, the estimation result of the corresponding sliding time window is marked as an outlier. Interpolation repair or re-estimation is performed on the anomaly points to obtain the corrected monitoring trajectory; The corrected monitoring trajectory is subjected to a smoothing filter, and the smoothed monitoring trajectory is output as the monitoring result of the UAV.

10. A UAV monitoring system based on the Doppler effect of a seismograph, characterized in that, include: Processor and memory; The processor and memory are connected via a communication bus: The processor is used to call and execute the program stored in the memory; The memory is used to store a program, which is at least used to execute the UAV monitoring method based on the Doppler effect of a seismograph as described in any one of claims 1-9.

Images