An unmanned aerial vehicle DOA estimation algorithm based on time-frequency feature extraction and spatial spectrum estimation

Abstract

The application provides a UAV DOA estimation algorithm based on time-frequency feature extraction and spatial spectrum estimation, solves the problems of high UAV detection and positioning cost, and comprises the following steps: S1, signal acquisition and time-frequency diagram generation; S2, using an improved YOLOv7-Tiny model to position the diagram transmission signal in the denoised time-frequency diagram; S3, high signal-to-noise ratio segment extraction; S4, five-element array spatial spectrum estimation; and S5, spatial spectrum filtering and DOA effectiveness verification. The application has the advantages of low cost, convenient deployment and the like.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of unmanned aerial vehicle positioning, and particularly relates to an unmanned aerial vehicle DOA estimation algorithm based on time-frequency feature extraction and spatial spectrum estimation. BACKGROUND

[0002] At present, unmanned aerial vehicles are widely used in civil, industrial and military fields, but the increasing number of unmanned aerial vehicles has caused great threats to individual privacy, social property and public safety. Therefore, it is necessary to develop a system capable of determining the direction of the target, which is usually achieved by thermal direction finding, optical direction finding, radar direction finding and radio frequency direction finding. Thermal direction finding and optical direction finding are easily affected by object overlap and background noise, and have high implementation cost. Radar direction finding has low reliability in urban areas and cannot track small unmanned aerial vehicles. Radio frequency direction finding is based on received signal strength and has low cost and is easy to deploy in lightweight airborne devices.

[0003] In order to solve the problems existing in the prior art, people have carried out long-term exploration and proposed various solutions. For example, Chinese patent document discloses a UAV DOA estimation method and device based on an antenna array [202010220530.2], which includes the following steps: collecting the radio frequency signals communicated between the unmanned aerial vehicle and the remote controller passively through the antenna array of the equilateral polygon, and pre-processing the radio frequency signals; performing frequency domain accumulation on the multiple groups of radio frequency signals received by the antennas of each side to obtain accumulated signals; calculating the starting frequency and the cutoff frequency of the UAV signal by using the cross-correlation algorithm on the accumulated signals with the UAV signal; calculating the UAV signal strength of the UAV signal according to the starting frequency and the cutoff frequency; selecting two antennas with the largest UAV signal strength and adjacent positions, and calculating the actual signal strength ratio of the signal strengths of the two antennas; and estimating the DOA angle of the unmanned aerial vehicle according to the actual signal strength ratio and the antenna directional diagram.

[0004] The above-mentioned scheme solves the problem of unmanned aerial vehicle detection and positioning to some extent, but the scheme still has many problems, such as high cost, inconvenient deployment, etc. SUMMARY

[0005] The purpose of the present application is to solve the above-mentioned problems, and to provide a UAV DOA estimation algorithm based on time-frequency feature extraction and spatial spectrum estimation, which has low cost and is convenient to deploy.

[0006] In order to achieve the above-mentioned purpose, the following technical scheme is adopted: a UAV DOA estimation algorithm based on time-frequency feature extraction and spatial spectrum estimation, comprising the following steps:

[0007] S1: signal acquisition and time-frequency diagram generation;

[0008] S2: Use the improved YOLOv7-Tiny model to locate the image transmission signal in the denoised time-frequency map;

[0009] S3: High signal-to-noise ratio segment extraction;

[0010] S4: Spatial spectrum estimation of a five-element array;

[0011] S5: Spatial Spectrum Filtering and DOA Validity Verification.

[0012] In the above-mentioned UAV DOA estimation algorithm based on time-frequency feature extraction and spatial spectrum estimation, step S1 uses a five-element array antenna to collect signals and covers the 2.4GHz, 5.2GHz and 5.8GHz frequency bands for round-robin sampling.

[0013] In the above-mentioned UAV DOA estimation algorithm based on time-frequency feature extraction and spatial spectrum estimation, step S1 includes the following steps:

[0014] S11: The RF front-end acquires five-channel IQ signals with a sampling rate of 150 MS / s. The DSP chip performs a short-time Fourier transform (STFT) on the IQ signal of any channel and uses a Hamming window filter.

[0015] ;

[0016] ;

[0017] S12: The window length and sliding step size are set to 1024 and 512 respectively. The power spectrum is then converted by taking the modulus and square of the spectrum, denoted as... And perform a segmented averaging operation:

[0018] ;

[0019] in, This refers to the time sequence number of the time-frequency data. The length of the power spectrum data is set to 32;

[0020] S13: Convert time-frequency data to decibel form:

[0021] ;

[0022] S14: The size of the obtained time-frequency diagram is... The frequency range and time length are 150 MHz and 56 ms, respectively.

[0023] S15: Perform frequency domain centering operation on the time-frequency graph;

[0024] S16: The time-frequency graph size is compressed to [size missing] using bilinear interpolation. The data is stored using unsigned 16-bit integers. The average decibel value is calculated for each time-frequency graph, and this value is set as a threshold. Points with decibel values ​​less than the threshold are taken with the threshold value, while other points are taken with the original decibel value.

[0025] S17: Normalize the denoised time-frequency graph to 0~255:

[0026] ;

[0027] S18: Convert the denoised time-frequency graph into a grayscale image and input it into the improved YOLOv7-Tiny model.

[0028] In the above-mentioned UAV DOA estimation algorithm based on time-frequency feature extraction and spatial spectrum estimation, the improved YOLOv7-Tiny model in step S2 adds a P2 small target detection layer to the neck part of the original model. The model confidence is set to 0.5, the cross-union ratio is set to 0.5, and the image transmission signal localization is completed after inference.

[0029] In the aforementioned UAV DOA estimation algorithm based on time-frequency feature extraction and spatial spectrum estimation, step S2 outputs the initial frequency. Cutoff frequency Minimum time Maximum time The output is an unsigned 8-bit integer data ranging from 0 to 255.

[0030] In the above-mentioned UAV DOA estimation algorithm based on time-frequency feature extraction and spatial spectrum estimation, step S3 includes the following steps:

[0031] S31: The bandwidth of the area where the image transmission signal is located is... Duration is For size Histogram analysis of the decibel values ​​in the matrix was performed, and the minimum decibel value was recorded as [value missing]. The maximum decibel value is ;

[0032] S32: Set the resolution of the image transmission signal relative to the background threshold to 2 dB, and iterate through the interval in 1 dB steps. to Count the number of pixels that satisfy the decibel values ​​in each interval, and record the interval with the largest number of pixels. That is, the decibel range corresponding to the image transmission signal;

[0033] S33: Calculate the time corresponding to the high signal-to-noise segment.

[0034] In the above-mentioned UAV DOA estimation algorithm based on time-frequency feature extraction and spatial spectrum estimation, step S33 includes the following steps:

[0035] S331: Calculate the center frequency of the image transmission signal ;

[0036] S332: Traversal frequency is The time is within the range pixels , ,remember The corresponding score is ;

[0037] S333: If The corresponding decibel value is not within the range ,but If the value is 0, proceed to step S335; otherwise, proceed to step 334.

[0038] S334: Statistics exist The decibel value within the neighborhood is in the range The number of pixels, Take this value;

[0039] S335: Request The time series corresponding to the maximum And mapped to actual time :

[0040] .

[0041] In the above-mentioned UAV DOA estimation algorithm based on time-frequency feature extraction and spatial spectrum estimation, step S3 includes the following steps:

[0042] S34: Regarding time The five channels of the interval IQ signal are subjected to 1024-point Fast Fourier Transform (FFT) respectively, and the FFT result is divided by 1024 to achieve quantization.

[0043] S35: Perform frequency domain centering operation and truncate the frequency to be in the range of 10 ... A segment of an interval;

[0044] S36: Based on the phase correlation in the time and frequency domains, arrange the five extracted frequency domain signal segments in columns. matrix Then, the data is input into the five-element array spatial spectrum estimation algorithm.

[0045] In the above-mentioned UAV DOA estimation algorithm based on time-frequency feature extraction and spatial spectrum estimation, step S4 includes the following steps:

[0046] S41: Calculate the frequency domain signal Spatial autocorrelation matrix , for The conjugate transpose of ;

[0047] S42: Obtained using the orthogonal triangular QR decomposition method eigenvalues and its corresponding eigenvector , After that, ask amplitude And arrange them from largest to smallest, with the number of signal sources set to 1, that is, in the extraction When selecting from the noisy subspace, Take the four eigenvectors corresponding to the non-maximum values ​​and arrange them column-wise. matrix , It satisfies an orthogonal relationship with the signal subspace;

[0048] S43: Traverse the pitch angle in steps of 1° within the range of 0 to 89°. , Traverse the azimuth angles within the range of 0 to 359° with a step size of 1°. , Let the azimuth vector of the five-element array antenna be:

[0049] ;

[0050] First, calculate the time delay vector:

[0051] ;

[0052] in, The radius of the five-element array antenna. The speed of electromagnetic waves;

[0053] Next, calculate the direction vector:

[0054] ;

[0055] in, For complex units, This refers to the current operating frequency band of the antenna.

[0056] S44: Calculate the spatial spectral function, and finally find... amplitude :

[0057] .

[0058] In the above-mentioned UAV DOA estimation algorithm based on time-frequency feature extraction and spatial spectrum estimation, step S5 includes the following steps:

[0059] S51: Set threshold ,like Maximum value and The ratio of the average to the mean is greater than If the algorithm outputs a valid DOA, it is considered valid; otherwise, it outputs the DOA obtained from the previous calculation.

[0060] S52: Set threshold ,remember The point where the maximum value is located is ,like exist The absolute values ​​of the gradients in the eight directions within the punctured neighborhood and The ratio of the maximum value is greater than If the algorithm outputs a valid DOA, it is considered valid; otherwise, it outputs the DOA obtained from the previous calculation.

[0061] S53: Assuming the distance between the UAV and the antenna remains almost constant, map the two calculated DOA values ​​to a Cartesian coordinate system:

[0062] ;

[0063] Set threshold ,like If the algorithm outputs a valid DOA, it is considered valid; otherwise, it outputs the DOA obtained from the previous calculation.

[0064] Compared with existing technologies, the advantages of this invention are as follows: It employs a five-element array antenna for rotating sampling in the 2.4GHz, 5.2GHz, and 5.8GHz frequency bands, effectively covering the commonly used communication and image transmission frequency bands of mainstream civilian UAVs, thus improving the system's detection compatibility and acquisition probability against target UAVs; it utilizes an improved YOLOv7-Tiny model to directly detect and locate image transmission signals on the time-frequency map, significantly enhancing the robust perception capability of UAV image transmission signals in complex electromagnetic environments; and it employs an optimized spatial spectrum estimation method to achieve high-resolution DOA estimation, with a verification mechanism designed to ensure the reliability and stability of the final output results. Attached Figure Description

[0065] Figure 1 This is a schematic diagram of the structure of the five-element array antenna of the present invention;

[0066] Figure 2 This is a schematic diagram of the direction finding of the five-element array antenna of the present invention;

[0067] Figure 3 This is a time-frequency diagram in decibel form according to the present invention;

[0068] Figure 4 This is the denoised time-frequency diagram of the present invention;

[0069] Figure 5 This is a YOLO model positioning and signal transmission diagram of the present invention;

[0070] Figure 6 This is the amplitude diagram of the spatial spectral function of the present invention. Detailed Implementation

[0071] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0072] like Figures 1-6 As shown, a UAV DOA estimation algorithm based on time-frequency feature extraction and spatial spectrum estimation includes the following steps:

[0073] S1: Perform signal acquisition and time-frequency diagram generation;

[0074] S2: Use the improved YOLOv7-Tiny model to locate the image transmission signal in the denoised time-frequency map;

[0075] S3: High signal-to-noise ratio segment extraction;

[0076] S4: Spatial spectrum estimation of a five-element array;

[0077] S5: Spatial Spectrum Filtering and DOA Validity Verification.

[0078] Specifically, in step S1, a five-element array antenna is used, uniformly distributed on a circle with a radius of 0.11 m. The five-element array antenna performs signal acquisition and covers the 2.4 GHz, 5.2 GHz, and 5.8 GHz frequency bands in a rotating sampling manner. In the UAV DOA obtained by the direction-finding algorithm, the pitch angle is the angle between the UAV and the normal to the antenna array surface, and the azimuth angle is the angle between the projection of the line connecting the UAV and the center of the antenna array surface onto the array surface and the x-axis of the antenna array surface.

[0079] In detail, step S1 includes the following steps:

[0080] S11: The control antenna's receiving frequency band rotates through the operating frequency bands of the three UAVs mentioned above. When the UAV's flight position is within the antenna's maximum radiation area, the RF front-end acquires five-channel IQ signals with a sampling rate of 150 MS / s. The DSP chip performs a short-time Fourier transform (STFT) on the IQ signal of any channel and uses a Hamming window filter.

[0081] ;

[0082] ;

[0083] S12: The window length and sliding step size are set to 1024 and 512 respectively. The power spectrum is then converted by taking the modulus and square of the spectrum. To reduce the length of the time-frequency data, it is denoted as... And perform a segmented averaging operation:

[0084] ;

[0085] in, This refers to the time sequence number of the time-frequency data. The length of the power spectrum data is set to 32;

[0086] S13: Convert time-frequency data to decibel form:

[0087] ;

[0088] S14: The size of the obtained time-frequency diagram is... The frequency range and time length are 150 MHz and 56 ms, respectively.

[0089] S15: Perform frequency domain centering operation on the time-frequency graph;

[0090] S16: To improve the positioning efficiency and accuracy of the image transmission signal, bilinear interpolation is used to compress the time-frequency map size to [size missing]. The data is stored using unsigned 16-bit integers. The average decibel value of each time-frequency graph is calculated and set as a threshold. Points with a decibel value less than the threshold are taken with the threshold value, while other points are taken with the original decibel value. This filters out the interference of low-energy noise on the subsequent neural network model inference.

[0091] S17: Normalize the denoised time-frequency graph to 0~255:

[0092] ;

[0093] S18: Convert the denoised time-frequency graph into a grayscale image and input it into the improved YOLOv7-Tiny model.

[0094] Furthermore, in step S2, the improved YOLOv7-Tiny model adds a P2 small target detection layer to the neck part of the original model, sets the model confidence to 0.5, sets the crossover ratio to 0.5, and completes the localization of the image transmission signal after inference.

[0095] Furthermore, step S2 outputs the starting frequency. Cutoff frequency Minimum time Maximum time The output is an unsigned 8-bit integer data ranging from 0 to 255.

[0096] In addition, step S3 includes the following steps:

[0097] S31: The bandwidth of the area where the image transmission signal is located is... Duration is For size Histogram analysis of the decibel values ​​in the matrix was performed, and the minimum decibel value was recorded as [value missing]. The maximum decibel value is ;

[0098] S32: To avoid the impact of background noise on the extraction of high signal-to-noise signal segments, the resolution of the image transmission signal relative to the background threshold is set to 2 dB, and the interval is traversed in 1 dB increments. to Count the number of pixels that satisfy the decibel values ​​in each interval, and record the interval with the largest number of pixels. That is, the decibel range corresponding to the image transmission signal;

[0099] S33: Calculate the time corresponding to the high signal-to-noise segment.

[0100] Meanwhile, step S33 includes the following steps:

[0101] S331: Calculate the center frequency of the image transmission signal ;

[0102] S332: Traversal frequency is The time is within the range pixels , ,remember The corresponding score is ;

[0103] S333: If The corresponding decibel value is not within the range ,but If the value is 0, proceed to step S335; otherwise, proceed to step 334.

[0104] S334: Statistics exist The decibel value within the neighborhood is in the range The number of pixels, Take this value;

[0105] S335: Request The time series corresponding to the maximum And mapped to actual time :

[0106] .

[0107] As can be seen, step S3 includes the following steps:

[0108] S34: Regarding time The five channels of the interval IQ signal are subjected to 1024-point Fast Fourier Transform (FFT) respectively. At the same time, the FFT result is divided by 1024 to achieve quantization, so as to prevent overflow when calculating the spatial autocorrelation matrix later.

[0109] S35: Perform frequency domain centering operation and truncate the frequency to be in the range of 10 ... A segment of an interval;

[0110] S36: Based on the phase correlation in the time and frequency domains, arrange the five extracted frequency domain signal segments in columns. matrix Then, the data is input into the five-element array spatial spectrum estimation algorithm.

[0111] Clearly, in step S4, the five-element array spatial spectrum estimation algorithm is designed for circular array antennas and is an improvement on the linear array antenna multiple signal classification (MUSIC) algorithm. It is achieved by decomposing the signal subspace and the noise subspace, and specifically includes the following steps:

[0112] S41: Calculate the frequency domain signal Spatial autocorrelation matrix , for The conjugate transpose of ;

[0113] S42: Obtained using the orthogonal triangular QR decomposition method eigenvalues and its corresponding eigenvector , After that, ask amplitude And arrange them from largest to smallest; since the extracted frequency domain signal segments usually only contain the image transmission signal of the target UAV, the number of signal sources is set to 1, that is, in the extraction When selecting from the noisy subspace, Take the four eigenvectors corresponding to the non-maximum values ​​and arrange them column-wise. matrix , It satisfies an orthogonal relationship with the signal subspace;

[0114] S43: Traverse the pitch angle in steps of 1° within the range of 0 to 89°. , Traverse the azimuth angles within the range of 0 to 359° with a step size of 1°. , Let the azimuth vector of the five-element array antenna be:

[0115] ;

[0116] First, calculate the time delay vector:

[0117] ;

[0118] in, The radius of the five-element array antenna. The speed of electromagnetic waves;

[0119] Next, calculate the direction vector:

[0120] ;

[0121] in, For complex units, This refers to the current operating frequency band of the antenna.

[0122] S44: Calculate the spatial spectral function, and finally find... amplitude :

[0123] .

[0124] Preferably, the spatial spectral function typically exhibits a peak at the DOA of the signal, therefore, The pitch and azimuth angles corresponding to the maximum values ​​can be considered as the DOA of the UAV. When the pitch angle is 0, the azimuth angle does not exist. However, since the maximum radiation area of ​​the five-element array antenna is 30°, and the input frequency domain signal of the five-element array spatial spectrum estimation algorithm may have strong noise signals or weak image transmission signals, step S5 uses the following steps to filter out erroneous DOAs:

[0125] S51: Set threshold ,like Maximum value and The ratio of the average to the mean is greater than If the algorithm outputs a valid DOA, it is considered valid; otherwise, it outputs the DOA obtained from the previous calculation.

[0126] S52: Set threshold ,remember The point where the maximum value is located is ,like exist The absolute values ​​of the gradients in the eight directions within the punctured neighborhood and The ratio of the maximum value is greater than If the algorithm outputs a valid DOA, it is considered valid; otherwise, it outputs the DOA obtained from the previous calculation.

[0127] S53: Assuming the distance between the UAV and the antenna remains almost constant, map the two calculated DOA values ​​to a Cartesian coordinate system:

[0128] ;

[0129] Set threshold ,like If the algorithm outputs a valid DOA, it is considered valid; otherwise, it outputs the DOA obtained from the previous calculation.

[0130] Steps S51 and S52 are mainly to prevent the extracted frequency domain signal from being too weak, resulting in a small maximum value or a relatively flat peak in the spatial spectrum function. Step S53 is mainly to prevent the presence of strong noise signals in the extracted frequency domain signal from causing the DOA of the algorithm's output noise signal. When the elevation angle output by the algorithm exceeds the maximum radiation area, the antenna orientation is adjusted to ensure the accuracy of the DOA estimation.

[0131] In summary, the principle of this embodiment is as follows: the acquired IQ signal is converted into a time-frequency map through a short-time Fourier transform (STFT), and then the improved YOLOv7-Tiny model is used to locate the image transmission signal in the denoised time-frequency map. Afterwards, high signal-to-noise ratio segments are extracted from the image transmission signal, and their spatial spectrum is estimated to finally obtain the pitch and azimuth angles of the UAV.

[0132] The specific embodiments described herein are merely illustrative of the spirit of the invention. Those skilled in the art to which this invention pertains may make various modifications or additions to the described specific embodiments or use similar methods to substitute them, without departing from the spirit of the invention or exceeding the scope defined by the appended claims.

[0133] Although this paper makes frequent use of terms such as Short-Time Fourier Transform, the possibility of using other terms is not excluded. These terms are used merely for the convenience of describing and explaining the essence of this invention; interpreting them as any additional limitation would contradict the spirit of this invention.

Claims

1. A UAV DOA estimation algorithm based on time-frequency feature extraction and spatial spectrum estimation, characterized in that, Includes the following steps: S1: Perform signal acquisition and time-frequency diagram generation; S2: Use the improved YOLOv7-Tiny model to locate the image transmission signal in the denoised time-frequency map; S3: High signal-to-noise ratio segment extraction; S4: Spatial spectrum estimation of a five-element array; S5: Spatial Spectrum Filtering and DOA Validity Verification.

2. The UAV DOA estimation algorithm based on time-frequency feature extraction and spatial spectrum estimation according to claim 1, characterized in that, In step S1, a five-element array antenna is used for signal acquisition, covering the 2.4GHz, 5.2GHz, and 5.8GHz frequency bands for rotating sampling.

3. The UAV DOA estimation algorithm based on time-frequency feature extraction and spatial spectrum estimation according to claim 2, characterized in that, Step S1 includes the following steps: S11: The RF front-end acquires five-channel IQ signals with a sampling rate of 150 MS / s. The DSP chip performs a short-time Fourier transform (STFT) on the IQ signal of any channel and uses a Hamming window filter. ； ； S12: The window length and sliding step size are set to 1024 and 512 respectively. The power spectrum is then converted by taking the modulus and square of the spectrum, denoted as... And perform a segmented averaging operation: ； in, This refers to the time sequence number of the time-frequency data. The length of the power spectrum data is set to 32; S13: Convert time-frequency data to decibel form: ； S14: The size of the obtained time-frequency diagram is... The frequency range and time duration are 150 MHz and 56 ms, respectively. S15: Perform frequency domain centering on the time-frequency graph; S16: The time-frequency graph size is compressed to [size missing] using bilinear interpolation. The data is stored using unsigned 16-bit integers. The average decibel value is calculated for each time-frequency graph, and this value is set as a threshold. Points with decibel values ​​less than the threshold are taken with the threshold value, while other points are taken with the original decibel value. S17: Normalize the denoised time-frequency graph to 0~255: ； S18: Convert the denoised time-frequency graph into a grayscale image and input it into the improved YOLOv7-Tiny model.

4. The UAV DOA estimation algorithm based on time-frequency feature extraction and spatial spectrum estimation according to claim 1, characterized in that, In step S2, the improved YOLOv7-Tiny model adds a P2 small target detection layer to the neck part of the original model. The model confidence is set to 0.5, the crossover ratio is set to 0.5, and the image transmission signal localization is completed after inference.

5. The UAV DOA estimation algorithm based on time-frequency feature extraction and spatial spectrum estimation according to claim 4, characterized in that, The step S2 outputs the starting frequency. Cutoff frequency Minimum time Maximum time The output is an unsigned 8-bit integer data ranging from 0 to 255.

6. The UAV DOA estimation algorithm based on time-frequency feature extraction and spatial spectrum estimation according to claim 1, characterized in that, Step S3 includes the following steps: S31: The bandwidth of the area where the image transmission signal is located is... Duration is For size Histogram analysis of the decibel values ​​in the matrix was performed, and the minimum decibel value was recorded as [value missing]. The maximum decibel value is ; S32: Set the resolution of the image transmission signal relative to the background threshold to 2 dB, and iterate through the interval in 1 dB steps. to Count the number of pixels that satisfy the decibel values ​​in each interval, and record the interval with the largest number of pixels. That is, the decibel range corresponding to the image transmission signal; S33: Calculate the time corresponding to the high signal-to-noise segment.

7. The UAV DOA estimation algorithm based on time-frequency feature extraction and spatial spectrum estimation according to claim 6, characterized in that, Step S33 includes the following steps: S331: Calculate the center frequency of the image transmission signal ; S332: Traversal frequency is The time is within the range pixels , ,remember The corresponding score is ; S333: If The corresponding decibel value is not within the range ,but If the value is 0, proceed to step S335; otherwise, proceed to step 334. S334: Statistics exist The decibel value within the neighborhood is in the range The number of pixels, Take this value; S335: Request The time series corresponding to the maximum And mapped to actual time : 。 8. The UAV DOA estimation algorithm based on time-frequency feature extraction and spatial spectrum estimation according to claim 7, characterized in that, Step S3 includes the following steps: S34: Regarding time The five channels of the interval IQ signal are subjected to 1024-point Fast Fourier Transform (FFT) respectively, and the FFT result is divided by 1024 to achieve quantization. S35: Perform frequency domain centering operation and truncate the frequency to be in the range of 10 ... A segment of an interval; S36: Based on the phase correlation in the time and frequency domains, arrange the five extracted frequency domain signal segments in columns. matrix Then, the data is input into the five-element array spatial spectrum estimation algorithm.

9. The UAV DOA estimation algorithm based on time-frequency feature extraction and spatial spectrum estimation according to claim 1, characterized in that, Step S4 includes the following steps: S41: Calculate the frequency domain signal Spatial autocorrelation matrix , for The conjugate transpose of ; S42: Obtained using the orthogonal triangular QR decomposition method eigenvalues and its corresponding eigenvector , After that, ask amplitude And arrange them from largest to smallest, with the number of signal sources set to 1, that is, in the extraction When selecting from the noisy subspace, Take the four eigenvectors corresponding to the non-maximum values ​​and arrange them column-wise. matrix , It satisfies an orthogonal relationship with the signal subspace; S43: Traverse the pitch angle in steps of 1° within the range of 0 to 89°. , Traverse the azimuth angles within the range of 0 to 359° with a step size of 1°. , Let the azimuth vector of the five-element array antenna be: ； First, calculate the time delay vector: ； in, The radius of the five-element array antenna. The speed of electromagnetic waves; Next, calculate the direction vector: ； in, For complex units, This refers to the current operating frequency band of the antenna. S44: Calculate the spatial spectral function, and finally find... amplitude : 。 10. The UAV DOA estimation algorithm based on time-frequency feature extraction and spatial spectrum estimation according to claim 1, characterized in that, Step S5 includes the following steps: S51: Set threshold ,like Maximum value and The ratio of the average to the mean is greater than If the algorithm outputs a valid DOA, it is considered valid; otherwise, it outputs the DOA obtained from the previous calculation. S52: Set threshold ,remember The point where the maximum value is located is ,like exist The absolute values ​​of the gradients in the eight directions within the punctured neighborhood and The ratio of the maximum value is greater than If the algorithm outputs a valid DOA, it is considered valid; otherwise, it outputs the DOA obtained from the previous calculation. S53: Assuming the distance between the UAV and the antenna remains almost constant, map the two calculated DOA values ​​to a Cartesian coordinate system: ； Set threshold ,like If the algorithm outputs a valid DOA, it is considered valid; otherwise, it outputs the DOA obtained from the previous calculation.

Images