A target number adaptive multi-rotor unmanned aerial vehicle sound source positioning method

Abstract

The application discloses a target number adaptive multi-rotor unmanned aerial vehicle sound source positioning method and belongs to the technical field of unmanned aerial vehicle supervision, in particular to a target number adaptive multi-rotor unmanned aerial vehicle sound source positioning method.The application aims to solve the problems that the existing unmanned aerial vehicle acoustic positioning method is easily disturbed by environmental noise and other sound sources in the scene where multiple targets appear at the same time and the target number is unknown, and the problems of false detection and missed detection, difficulty in distinguishing multiple targets at a near angle and insufficient reliability of the direction of arrival estimation result.The process is as follows: a characteristic frequency set is obtained by collecting a microphone array multi-channel signal; a spatial spectrum is constructed based on normalized covariance matrix positive and negative power, a DOA point set is constructed; clustering is performed on the DOA point set to obtain a cluster set; confidence weighted fusion is performed on the spatial spectrum in the cluster to obtain an initial positioning result; peak detection and cluster splitting are performed on the cluster level fusion spatial spectrum to obtain a final positioning result.

Description

Technical Field

[0001] This invention belongs to the field of drone monitoring technology, and particularly relates to a method for sound source localization of multi-rotor drones with adaptive target number. Background Technology

[0002] In recent years, drones have been widely used in many fields such as aerial surveying and mapping, emergency rescue, and logistics delivery. However, with the rapid increase in the number of drones, unauthorized flights are also on the rise, posing significant risks to public safety and low-altitude airspace management. Therefore, researching efficient and reliable drone detection and positioning methods has important engineering application value.

[0003] Existing UAV positioning technologies mainly include radar detection, photoelectric detection, radio frequency detection, and acoustic detection. Among them, radar, photoelectric, and radio frequency technologies are generally affected by environmental conditions, target reflection characteristics, and spectrum control, resulting in high deployment costs, insufficient versatility, and large positioning errors in complex environments. In contrast, acoustic positioning technology utilizes the operating noise generated and propagated by the UAV during flight to achieve target localization. It features flexible deployment, lower cost, and is unaffected by lighting conditions, making it suitable for target detection and localization in most complex scenarios.

[0004] However, in practical applications, existing UAV acoustic positioning methods are susceptible to interference from environmental noise and other sound sources when multiple targets appear simultaneously and the number of targets is unknown. This results in problems such as false detections and missed detections, difficulty in distinguishing multiple targets at close angles, and insufficient reliability of arrival direction estimation results. Summary of the Invention

[0005] The purpose of this invention is to address the problems of existing UAV acoustic localization methods being susceptible to environmental noise and other sound sources in scenarios where multiple targets appear simultaneously and the number of targets is unknown, resulting in false detections and missed detections, difficulty in distinguishing multiple targets at close angles, and insufficient reliability of arrival direction estimation results. Therefore, this invention proposes a target number adaptive sound source localization method for multi-rotor UAVs.

[0006] The specific process of a target number adaptive multi-rotor UAV sound source localization method is as follows:

[0007] Step 1: Acquire multi-channel signals from the microphone array; perform spectral analysis on the acquired multi-channel signals from the microphone array to obtain the detection spectrum; obtain the characteristic frequency set based on the detection spectrum;

[0008] Step 2: Calculate DOA and confidence score based on the feature frequency set; construct DOA point set based on DOA and confidence score; DOA is the estimated direction of arrival.

[0009] Step 3: Perform DBSCAN density clustering on the DOA point set to obtain the cluster set; determine the feature frequency set corresponding to each cluster;

[0010] Step 4: Based on the feature frequency set corresponding to each cluster, obtain the clusters. The initial positioning results;

[0011] Step 5: Cluster The corresponding cluster-level fusion spatial spectrum is subjected to local maximum detection to obtain a candidate peak set, and then the target peak set is determined based on the independent target conditions;

[0012] If the target peak set contains only one target peak, then a cluster is determined. No splitting is needed; the initial localization results obtained in step 4 can be directly used as clusters. The final positioning result;

[0013] If the target peak set contains multiple target peaks, then a cluster is determined. It is necessary to break down the target peaks and refine them by performing three-point parabolic interpolation on each target peak in the target peak set to obtain the final location results of multiple targets; multiple means two or more targets.

[0014] The process continues until all clusters are identified, at which point the final location result is obtained.

[0015] The beneficial effects of this invention are as follows:

[0016] This invention first performs spectral analysis on the multi-channel signals acquired by the microphone array. Within a preset frequency band, characteristic frequencies are selected based on peak height, peak significance, and peak spacing constraints to form a set of characteristic frequencies. Second, the complex spectral coefficients of the reference frequency points matching each frequency are extracted across multiple time frames to construct a multi-channel narrowband snapshot vector sequence. The array spatial covariance matrix is ​​then estimated. Using the positive and negative power operations of the normalized covariance matrix, the signal and noise subspaces are partitioned without pre-setting the number of signal sources, constructing a spatial spectrum and outputting the corresponding azimuth and elevation angle estimates. Simultaneously, the peak significance of the main peak relative to the secondary peak in the spatial spectrum is used as the confidence level of the frequency point estimate, forming a DOA point for angle, frequency, and confidence level. Based on this, the DBSCAN density clustering algorithm is introduced to adaptively cluster the DOA point set. The number of effective clusters obtained by clustering is used as the target number estimate, and outliers are removed as noise points. Then, the single-frequency spatial spectrum corresponding to each cluster is normalized and weighted fusion is performed according to the confidence level to obtain the cluster-level fused spatial spectrum. Peak detection is performed on the fused spatial spectrum to obtain coarse positioning results. The angle estimation accuracy is further improved by interpolation. Finally, multi-peak detection and cluster splitting are performed on the cluster-level fused spatial spectrum. When the preset peak ratio and peak spacing constraints are met, it is determined that there are multiple targets in the cluster, and the cluster is split, and the corresponding DOA estimation results are output to obtain the final positioning result.

[0017] Compared with the prior art, the significant feature of the present invention is that:

[0018] This invention achieves multi-target localization output without relying on a preset number of targets through an overall technical process of "feature frequency screening, blind source spatial spectrum construction by matrix exponentiation, confidence labeling, DBSCAN density clustering adaptive estimation of the number of targets, weighted spectrum fusion of each frequency within the cluster, multi-peak detection and cluster splitting". It can effectively suppress false detections and missed detections caused by environmental noise and other sound source interference, enhance the output stability in complex acoustic environments, and improve the resolution and reliability of the direction of arrival estimation results in near-angle multi-target scenarios.

[0019] This invention obtains multi-frequency point DOA estimation by constructing blind source spatial spectrum based on power operation of normalized covariance matrix, and then performs DBSCAN density clustering on the DOA point set, using the effective cluster number as the target number estimation result. It can realize target number identification and direction estimation output in multi-target scenes without pre-setting the target number.

[0020] This invention employs a blind source space spectrum construction method based on matrix exponentiation. It automatically approximates the signal and noise subspaces by utilizing positive and negative exponentiation operations of the normalized covariance matrix. It does not require prior information on the number of information sources and effectively solves the problem that traditional subspace algorithms are prone to positioning failure due to the estimation error of the number of information sources in scenarios where multiple UAVs have similar models or rotation speeds and cause "same frequency" interference (i.e., the characteristic frequencies of multiple targets fall into the same narrow band analysis interval and aliasing occurs).

[0021] This invention introduces a confidence index based on the significance of spatial spectrum peaks to quantify the reliability of single-frequency direction estimation results, significantly suppressing the impact of outliers generated by environmental noise and interference signals on the final positioning results, and improving the positioning stability and reliability in complex environments.

[0022] This invention integrates consistent signal features of the same target across multiple characteristic frequencies through confidence-weighted cluster-level spatial spectrum incoherent fusion, enhancing the peak value of the true target and suppressing spurious peaks, effectively improving the localization reliability in near-angle multi-target scenarios. To address the cluster malmerging problem caused by close target angles, a multi-peak detection and cluster splitting mechanism is further employed to effectively reduce missed detections and improve the accuracy of multi-target localization. Attached Figure Description

[0023] Figure 1 This is a flowchart of the present invention;

[0024] Figure 2 The detection spectrum is shown in Example 1;

[0025] Figure 3 This is a DOA point set diagram for Example 1;

[0026] Figure 4 The fusion spectrum thermogram of Target 1 in Example 1;

[0027] Figure 5 The fusion spectrum thermogram of target 2 in Example 1;

[0028] Figure 6 This is a comparison chart of the positioning results in Example 1;

[0029] Figure 7 This is a DOA point set diagram for Example 2;

[0030] Figure 8 The fusion spectrum thermogram of target 1 in Example 2;

[0031] Figure 9 The fusion spectrum thermogram of target 2 in Example 2;

[0032] Figure 10 The thermal spectrum of target 3 in Example 2;

[0033] Figure 11 This is a comparison chart of the positioning results in Example 2. Detailed Implementation

[0034] Specific Implementation Method 1: The specific process of the target number adaptive multi-rotor UAV sound source localization method in this implementation method is as follows:

[0035] Step 1: Acquire multi-channel signals from the microphone array; perform spectral analysis on the acquired multi-channel signals from the microphone array to obtain the detection spectrum; obtain the characteristic frequency set based on the detection spectrum;

[0036] Step 2: Calculate DOA and confidence score based on the feature frequency set; construct DOA point set based on DOA and confidence score; DOA is the estimated direction of arrival.

[0037] Step 3: Perform DBSCAN density clustering on the DOA point set to obtain the cluster set; determine the feature frequency set corresponding to each cluster;

[0038] Step 4: Based on the feature frequency set corresponding to each cluster, obtain the clusters. The initial positioning results;

[0039] Step 5: Cluster The corresponding cluster-level fusion spatial spectrum is subjected to local maximum detection to obtain a candidate peak set, and then the target peak set is determined based on the independent target conditions;

[0040] If the target peak set contains only one target peak, then a cluster is determined. No splitting is needed; the initial localization results obtained in step 4 can be directly used as clusters. The final positioning result;

[0041] If the target peak set contains multiple target peaks, then a cluster is determined. It is necessary to split the target peaks and refine them by three-point parabolic interpolation according to step 44 to obtain the final positioning results of multiple targets; multiple means two or more targets.

[0042] The process continues until all clusters are identified, at which point the final location result is obtained.

[0043] Specific Implementation Method Two: This implementation method differs from Specific Implementation Method One in that: in step 1, multi-channel signals from the microphone array are acquired; spectral analysis is performed on the acquired multi-channel signals from the microphone array to obtain the detection spectrum; a set of characteristic frequencies is obtained based on the detection spectrum; the specific process is as follows:

[0044] Step 11: Acquire multi-channel signals from the microphone array; the specific process is as follows:

[0045] At sampling rate Below, the microphone array (composed of) is used for acquisition. An array consisting of 1 microphone (the first one) The first channel Discrete-time signal at each sampling point ;

[0046] in,

[0047] For channel index, , This represents the total number of microphones.

[0048] For sampling point index, , This represents the total number of samples taken per channel.

[0049] Step 12: Perform spectral analysis on the multi-channel signal of the microphone array acquired in Step 11 to obtain the detection spectrum; the specific process is as follows:

[0050] Select any channel of the array As a reference channel, the discrete-time signals of all sampling points in the reference channel are multiplied by a window function of the same length, and then subjected to a Fast Fourier Transform to obtain the frequency domain signal. ;

[0051] Calculate frequency domain signals amplitude spectrum To prevent meaningless minimum values ​​from occurring during logarithmic operations, a minimum positive number is introduced. , amplitude spectrum Converting to logarithmic form yields the detection spectrum. The expression is:

[0052]

[0053] in, The amplitude spectrum of the frequency domain signal; To prevent the formula from containing extremely small positive numbers with a logarithm of zero, ;

[0054] Calculate the frequency resolution of the Fast Fourier Transform; expressed as:

[0055]

[0056] in, Sampling rate, The number of points for the Fast Fourier Transform;

[0057] Step 13: Based on the detection spectrum The characteristic frequency set is obtained; the specific process is as follows:

[0058] In the preset frequency band Internal detection spectrum Peak detection is performed to obtain all peak values. Each peak value corresponds to one feature frequency, and all feature frequencies constitute a feature frequency set. , For the first One characteristic frequency;

[0059] Preset frequency band satisfy , Sampling rate;

[0060] This is the preset minimum frequency band value. This is the preset maximum value for the frequency band.

[0061] The other steps and parameters are the same as in Specific Implementation Method 1.

[0062] Specific Implementation Method Three: This implementation method differs from Specific Implementation Method One or Two in that: the preset frequency band... Internal detection spectrum Peak detection is performed to obtain all peak values. Each peak value corresponds to one feature frequency, and all feature frequencies constitute a feature frequency set. , For the first One characteristic frequency;

[0063] The specific process is as follows:

[0064] In the preset frequency band The local maximum detection algorithm is used to detect the spectrum. Peak detection is performed to obtain a candidate peak set; each candidate peak in the candidate peak set corresponds to a frequency position. Frequency position The detection spectral amplitude value ;

[0065] The candidate peak set is filtered based on the set constraints to obtain the valid peaks that meet the constraints; each valid peak corresponds to a frequency position, and the frequency positions corresponding to all valid peaks constitute the feature frequency set. , For the first One characteristic frequency;

[0066] The constraints are set as follows:

[0067] 1) Set the minimum peak height threshold;

[0068] The detection spectral amplitude value corresponding to any candidate peak Compare with the minimum peak height threshold, retain candidate peaks whose detected spectral amplitude is greater than the minimum peak height threshold, and delete the rest;

[0069] 2) Set a significance threshold;

[0070] For any candidate peak retained in 1) In the detection spectrum China-Israel candidate summit Centered on the frequency, search in both the decreasing and increasing frequency directions;

[0071] At the candidate peak On the left (decreasing direction), search until the detection spectrum is not lower than the candidate peak for the first time. The position of the amplitude, or search to the lower limit of the preset frequency band (preset frequency band) lower limit The lowest detected spectral amplitude within the left search range is taken as the left reference spectral amplitude, denoted as... ;

[0072] At the candidate peak On the right (increasing direction), search until the detection spectrum is not lower than the candidate peak for the first time. The position of the amplitude, or search to the upper limit of the preset frequency band (preset frequency band) upper limit The lowest detected spectral amplitude within the right-hand search range is taken as the right-hand reference spectral amplitude, denoted as... ;

[0073] Take the larger of the two values ​​as the candidate peak. Local reference spectrum amplitude:

[0074]

[0075] Candidate peak The peak significance is:

[0076]

[0077] when When the value is greater than the significance threshold, retain the candidate peak. Otherwise, delete.

[0078] The significance threshold is 10 dB;

[0079] 3) Set the peak spacing threshold;

[0080] Calculate the absolute value of the frequency difference between any two candidate peaks among all the retained candidate peaks in step 2);

[0081] If the absolute value of the frequency difference is greater than the peak spacing threshold, then two candidate peaks are retained.

[0082] If the absolute value of the frequency difference is less than or equal to the peak spacing threshold, then only the candidate peak with the larger detection spectral amplitude among the two candidate peaks is retained, and the candidate peak with the smaller detection spectral amplitude is deleted.

[0083] The peak spacing threshold is .

[0084] Other steps and parameters are the same as in specific implementation method one or two.

[0085] Specific Implementation Method Four: This implementation method differs from Specific Implementation Methods One to Three in that: in step 2, the DOA and confidence score are calculated based on the feature frequency set; a DOA point set is constructed based on the DOA and confidence score; the DOA is the direction of arrival estimation result; the specific process is as follows:

[0086] Step 21: Use Short-Time Fourier Transform (STFT) to extract the time-frequency domain of the multi-channel signal to obtain narrowband data; the specific process is as follows:

[0087] Step 211: Calculate the frequency resolution of the short-time Fourier transform. ; indicates as:

[0088]

[0089] in, Sampling rate; The frame length of the short-time Fourier transform;

[0090] Calculate the frequency resolution of the Short-Time Fourier Transform (STFT) to determine the adaptive narrowband analysis interval; since the analysis window of the STFT itself has a certain main lobe width, each discrete frequency point in the time-frequency domain essentially corresponds to a fixed bandwidth. The narrowband analysis range can adaptively cover the transient frequency drift caused by UAV load fluctuations, flight attitude changes or environmental interference, ensuring that the single frequency point data contains the complete target signal energy.

[0091] Step 212: Based on the multi-channel short-time Fourier transform time-frequency data, construct the multi-channel narrowband snapshot vector corresponding to each characteristic frequency; the specific process is as follows:

[0092] Short-time Fourier transform is performed on the discrete-time signals acquired by each channel in the microphone array to obtain the time-frequency data corresponding to each channel;

[0093] For each feature frequency in the feature frequency set Searching within the frequency sequence to find frequencies matching the characteristic frequencies. The frequency point with the smallest absolute value of the difference is used as the reference frequency point for the current characteristic frequency;

[0094] Frequency sequence is ;

[0095] Based on time-frequency data, the complex spectral coefficients (which are derived from the time-frequency data) of all channels at the same time frame and the same reference frequency are extracted to construct a multi-channel narrowband snapshot vector for a single time frame.

[0096]

[0097] in, For the first An M-dimensional column vector in a time frame; For time frame indexing, , For the number of snapshots;

[0098] This indicates that the first channel is in time frame. Characteristic frequencies The complex spectral coefficients at the corresponding reference frequency;

[0099] This indicates that the second channel is in the time frame. Characteristic frequencies The complex spectral coefficients at the corresponding reference frequency;

[0100] Indicates the first Each channel in time frame Characteristic frequencies The complex spectral coefficients at the corresponding reference frequency; This indicates the transpose;

[0101] Step 22: Estimate the spatial covariance matrix based on the multi-channel narrowband snapshot vector sequence; the specific process is as follows:

[0102] Under narrowband conditions, based on characteristic frequency The spatial covariance matrix is ​​constructed from the multi-channel narrowband snapshot vectors of each time frame; it is represented as:

[0103]

[0104] in, Represents characteristic frequency The spatial covariance matrix under; Indicates conjugate transpose;

[0105] Step 23: Estimate the noise power based on the spatial covariance matrix, and construct the normalized covariance matrix based on the noise power;

[0106] Calculate matrix powers based on the normalized covariance matrix; calculate the spatial spectral function based on matrix powers; solve for the DOA based on the spatial spectral function; the specific process is as follows:

[0107] The matrix power approximation method automatically approximates the signal subspace and noise subspace by performing power operations on the covariance matrix, thereby forming sharp spectral peaks when the number of sources is unknown.

[0108] Step 231: Estimate noise power based on the spatial covariance matrix, and construct a normalized covariance matrix based on the noise power; the specific process is as follows:

[0109] Step 2311: Calculate the current feature frequency Lower space covariance matrix The smallest eigenvalue , As an estimate of noise power; the specific process is as follows:

[0110] right 3D covariance matrix Perform eigenvalue decomposition to obtain A number of non-negative real eigenvalues ​​will The non-negative real eigenvalues ​​are arranged in descending order as follows:

[0111]

[0112] in, for The first non-negative real eigenvalue after arranging the non-negative real eigenvalues ​​in descending order; for The second non-negative real eigenvalue after arranging the non-negative real eigenvalues ​​in descending order; for The nth non-negative real eigenvalues ​​arranged in descending order One non-negative real eigenvalue; for The nth non-negative real eigenvalues ​​arranged in descending order One non-negative real eigenvalue;

[0113] Will As characteristic frequency Noise power estimate under ;

[0114] Step 2312: Based on feature frequencies Noise power estimate under Construct the normalized covariance matrix ; indicates as:

[0115]

[0116] in, To prevent extremely small positive numbers with a denominator of zero, ;

[0117] Step 232: Based on the normalized covariance matrix Calculate matrix powers; the specific process is as follows:

[0118] Take the power parameter Calculate the normalized covariance matrix respectively. positive power and negative powers ;

[0119] Step 233: Calculate the spatial spectral function based on matrix exponentiation ; indicates as:

[0120]

[0121] in, It is a spatial guiding vector;

[0122] The spatial guidance vector The solution process is as follows:

[0123] In this invention, a three-dimensional rectangular coordinate system is established using the reference point of the microphone array as the origin. (Satisfies the right-hand rule); where, shaft and The axis is located in the horizontal plane Inside, The axis is perpendicular to the horizontal plane and points upward; The positive axis direction is used as the starting reference direction for the azimuth angle. The positive axis direction is located in the positive azimuth rotation direction; the reference point is selected as the array geometric center or any preset microphone position; the position coordinates of each microphone in the array constitute the array position matrix. The direction of the target sound source is indicated by azimuth. The pitch angle is represented by the azimuth angle. This represents the angle between the projection of the target direction vector onto the horizontal plane XOY and the positive X-axis, with a value range of [value missing]. Pitch angle This represents the angle between the target direction vector and the horizontal plane XOY, with a value range of [value missing]. ;

[0124] In azimuth Pitch angle Within the range , A two-dimensional spatial grid is scanned using a step size, and a spatial guiding vector is constructed at each grid point scanned. :

[0125]

[0126] in, A matrix representing the coordinates of all microphones; Indicates azimuth and elevation angles The corresponding unit vector; Indicates the azimuth angle corresponding to the grid point; This indicates the pitch angle corresponding to the grid point; Indicates the speed of sound; Represents the imaginary unit. Superscript This indicates the transpose;

[0127] Step length for Step length for ;

[0128] Step 234: Based on spatial spectral function Solve for DOA; the specific process is as follows:

[0129] Take the spatial spectral function The maximum value corresponding to As characteristic frequency Arrival direction estimation results The expression is:

[0130]

[0131] in, Represents the spatial spectral function The azimuth and elevation angles corresponding to the maximum values ;

[0132] Represents characteristic frequency The following are the estimated directions of arrival;

[0133] Step 24: Calculate the confidence level based on the spatial spectral function;

[0134] Step 25: Construct a set of DOA points based on DOA and confidence level.

[0135] The other steps and parameters are the same as those in one of the specific implementation methods one to three.

[0136] Specific Implementation Method Five: This implementation method differs from Specific Implementation Methods One to Four in that: in step 24, the confidence level is calculated based on the spatial spectral function; the specific process is as follows:

[0137] To quantify the reliability of the above DOA estimation results, a confidence level is constructed based on the significance of the spatial spectrum peak. Let the spatial spectral function be... The maximum peak value is:

[0138]

[0139] With the maximum peak point Centered on the grid point, remove all grid points in the azimuth direction to the left and right of the center. Each grid point is then removed from the center in the pitch direction. A grid of points forms the neighborhood of the main peak. The maximum value in the remaining spatial spectrum after removing the neighborhood of the main peak is searched and taken as the second highest peak value. ,based on and The confidence level is calculated and expressed as:

[0140]

[0141] in, To prevent extremely small positive numbers with a denominator of zero, ; Characteristic frequency The confidence level is below.

[0142] The other steps and parameters are the same as in any of the specific implementation methods one to four.

[0143] Specific Implementation Method Six: This implementation method differs from Specific Implementation Methods One to Five in that: in step 25, a DOA point set is constructed based on DOA and confidence level; the specific process is as follows:

[0144] Each characteristic frequency Arrival direction estimation results With characteristic frequency Confidence level By combining these points, a single-frequency DOA point can be obtained.

[0145] A set of DOA points with confidence attributes is constructed based on single-frequency DOA points;

[0146] Represented as:

[0147]

[0148] in, This represents a set of DOA points that includes confidence attributes;

[0149] Represents the set of characteristic frequencies The set of DOA points with confidence attributes, which consists of single-frequency DOA points corresponding to all feature frequencies.

[0150] Indicates confidence level The DOA point of the attribute.

[0151] The other steps and parameters are the same as those in any of the specific implementation methods one to five.

[0152] Specific Implementation Method Seven: This implementation method differs from Specific Implementation Methods One through Six in that: in step 3, DBSCAN density clustering is performed on the DOA point set to obtain a cluster set; the characteristic frequency set corresponding to each cluster is determined; the specific process is as follows:

[0153] Step 31: For the DOA point set containing confidence attributes Perform DBSCAN density clustering to obtain a set of clusters; the specific process is as follows:

[0154] Set the neighborhood radius and minimum number of samples;

[0155] To solve the azimuth angle in The problem of wraparound truncation at the boundary is defined with confidence level. DOA points of attributes With confidence level DOA points of attributes The angular Euclidean distance between them in two-dimensional directional space is expressed as:

[0156]

[0157]

[0158] in,

[0159] Indicates confidence level The DOA point of the attribute; Indicates confidence level The DOA point of the attribute;

[0160] Represents characteristic frequency The following are the estimated directions of arrival; Represents characteristic frequency The following are the estimated directions of arrival;

[0161] Represents characteristic frequency The confidence level is below; Represents characteristic frequency The confidence level is below;

[0162] Indicates the difference in azimuth angle;

[0163] DOA points with confidence attributes and The angular Euclidean distance between them;

[0164] Since the range of azimuth angle values ​​is It exhibits a looping characteristic where the beginning and end are connected. and difference Directly taking the difference will be mistakenly interpreted as a difference between the two. The Euclidean distance between DOA points can avoid this problem in density clustering;

[0165] Point set based on the angular Euclidean distance between DOA points Perform DBSCAN density clustering and use the number of clusters generated as the target number for adaptive estimation. And obtain the cluster set. ;

[0166] in, Indicates the first cluster; Indicates the second cluster; Indicates the first A cluster; Indicates the first A cluster;

[0167] Points that do not belong to any cluster are marked as noise outliers and removed;

[0168] Step 32: Calculate the feature frequency set corresponding to each cluster; the specific process is as follows:

[0169] Record each cluster The feature frequencies corresponding to all DOA points are included to form a cluster. The corresponding set of characteristic frequencies This is used to call the corresponding frequency of spatial spectrum data during subsequent cluster-level spatial spectrum fusion.

[0170] in, Cluster The corresponding set of characteristic frequencies; Indicates belonging to a cluster Single-frequency DOA points with confidence attributes; .

[0171] The other steps and parameters are the same as those in one of the specific implementation methods one to six.

[0172] Specific Implementation Method Eight: This implementation method differs from Specific Implementation Methods One through Seven in that: in step 4, the clusters are obtained based on the feature frequency set corresponding to each cluster. The initial positioning results; the specific process is as follows:

[0173] Step 41: For clusters The spatial spectrum corresponding to each characteristic frequency Amplitude normalization is performed to obtain the normalized spatial spectrum. ; indicates as:

[0174]

[0175] in, Represents the spatial spectral function; Represents the spatial spectral function Global maximum peak value across all scan grids (for clusters) Each of them Each has a corresponding spatial spectrum , It is a spatial spectral function In all (Maximum value on the grid); To prevent extremely small positive numbers with a denominator of zero, ;

[0176] Step 42: For clusters Normalized spatial spectrum corresponding to each characteristic frequency Weighted incoherent fusion is performed to obtain the cluster-level fusion spatial spectrum. ; indicates as:

[0177]

[0178] in, Represents characteristic frequency The weights below;

[0179] Characteristic frequency The weights below From characteristic frequency Confidence level Confirmed; expressed as:

[0180]

[0181] Step 43: In the cluster-level fusion spatial spectrum The maximum value is detected to obtain the angle corresponding to the maximum peak point; the specific process is as follows:

[0182] Preset step size , ;

[0183] In azimuth Pitch angle Perform a spatial grid scan within the range to obtain discrete azimuth angles. and pitch angle ;

[0184] Determine the cluster-level fusion spatial spectrum Values ​​on the spatial grid And solve for the coarse positioning angle corresponding to the maximum peak point. ; indicates as:

[0185]

[0186]

[0187] in,

[0188] Indicates azimuth and elevation angles The corresponding cluster-level fusion spatial spectrum;

[0189] express The grid index corresponding to the maximum value;

[0190] Indicates when grid index Pick The corresponding azimuth and elevation angles;

[0191] This indicates the coarse positioning angle corresponding to the maximum peak point;

[0192] Step 44: To overcome the limitation of the angle grid step size on positioning accuracy, coarse positioning angles are calculated for the points corresponding to the maximum peak points. Perform three-point parabolic interpolation refinement, cluster Output direction of arrival estimation Arrival direction estimation As a cluster The initial positioning results.

[0193] The other steps and parameters are the same as those in any of the specific implementation methods one to seven.

[0194] Specific Implementation Method Nine: This implementation method differs from Specific Implementation Methods One through Eight in that: in step 44, to overcome the limitation of the angle grid step size on positioning accuracy, a coarse positioning angle corresponding to the maximum peak point is used. Perform three-point parabolic interpolation refinement, cluster Output direction of arrival estimation Arrival direction estimation As a cluster The initial positioning results; the specific process is as follows:

[0195] Step 441: Fix ,Pick , , ,but ;

[0196] in, This represents the discrete grid index of the pitch angle corresponding to coarse positioning; This represents the spectral amplitude value of the grid cell preceding the maximum peak point; This represents the spectral amplitude at the point of maximum peak. This represents the spectral amplitude value of the grid one azimuth angle after the point of maximum peak. Indicates the cluster-level fusion spatial spectrum in the grid index The function value at that location; Indicates the cluster-level fusion spatial spectrum in the grid index The function value at that location; Indicates the cluster-level fusion spatial spectrum in the grid index The function value at that location; This represents the grid offset for azimuth parabolic interpolation; This represents the high-precision azimuth estimate output after interpolation refinement; This represents the coarse positioning angle of the azimuth obtained from discrete grid scanning; Indicates the angle scan step size;

[0197] Step 442: Fix ,Pick , , ,but ;

[0198] in, This indicates the azimuth discrete grid index corresponding to coarse positioning; This represents the spectral amplitude value of the grid at the pitch angle preceding the maximum peak point; This represents the spectral amplitude at the point of maximum peak. This represents the spectral amplitude value of the grid at the pitch angle following the point of maximum peak. Indicates the cluster-level fusion spatial spectrum in the grid index The function value at that location; Indicates the cluster-level fusion spatial spectrum in the grid index The function value at that location; Indicates the cluster-level fusion spatial spectrum in the grid index The function value at that location; This represents the grid offset for parabolic pitch interpolation. This represents the high-precision pitch angle estimate output after interpolation refinement. This represents the pitch angle and coarse positioning angle obtained from discrete grid scanning; Indicates the angle scan step size;

[0199] Step 443: Coarsely locate the angle corresponding to the maximum peak point. Perform three-point parabolic interpolation refinement, cluster Output direction of arrival estimation Arrival direction estimation This serves as the initial localization result.

[0200] The other steps and parameters are the same as those in one of the specific implementation methods one to eight.

[0201] Specific Implementation Method Ten: This implementation method differs from Specific Implementation Methods One to Nine in that: in step 5, the cluster... Corresponding cluster-level fusion spatial spectrum Local maxima detection is performed to obtain a candidate peak set, and then the target peak set is determined based on the independent target condition;

[0202] If the target peak set contains only one target peak, then a cluster is determined. No splitting is needed; the initial localization results obtained in step 4 can be directly used as clusters. The final positioning result;

[0203] If the target peak set contains multiple target peaks, then a cluster is determined. If there are multiple near-angle targets, cluster splitting is required. For each target peak in the target peak set, three-point parabolic interpolation is performed to refine the target peaks according to step 44 to obtain the final positioning results of multiple targets (1 target peak corresponds to 1 target). Multiple targets are greater than or equal to 2. The final positioning result is obtained after all clusters are judged.

[0204] The specific process is as follows:

[0205] Step 51: For clusters Corresponding cluster-level fusion spatial spectrum Local maxima detection is performed to obtain a set of candidate peaks, represented as:

[0206]

[0207] For candidate peak set Sort the candidate peaks by spectral amplitude from largest to smallest, and select the candidate peaks with the largest spectral amplitudes as the candidate peak set. The first target peak, and initialize the candidate peak set. target peak set ; indicates as:

[0208]

[0209] in, To set candidate peaks Candidate Peak After sorting the spectral amplitudes from largest to smallest, the candidate peak with the largest spectral amplitude is selected.

[0210] To select from candidate peak sets The target peak set obtained through screening;

[0211] Then, iterate through the candidate peaks in sequence. Other candidate peaks Determine candidate peaks Does it meet the independence condition? If the candidate peak If the independent objective condition is met, then the candidate peak will be... It was determined to be an independent target peak and a candidate peak was selected. Add to target peak set Otherwise, candidate peak Not included in the target peak set ;

[0212] After the traversal is complete, a set of candidate peaks is obtained. The target peak set;

[0213] If the target peak set If a cluster contains only one target peak, it is considered a cluster. No splitting is needed; the initial localization results obtained in step 4 can be directly used as clusters. The final positioning result;

[0214] If the target peak set If it contains multiple target peaks, then it is determined to be a cluster. There are multiple near-angle targets, requiring cluster splitting; target peak set Each target peak in the azimuth and elevation directions is refined using three-point parabolic interpolation in step 44 to obtain the final positioning results for multiple targets (one target peak corresponds to one target); multiple targets are two or more.

[0215] Step 52: Repeat step 51 until all clusters are identified and the final location result is obtained. ;

[0216] in, For the final target quantity;

[0217] The independent condition is that both the peak ratio condition and the peak spacing condition are satisfied simultaneously.

[0218] Peak ratio condition: candidate peak The ratio of the amplitude value to the amplitude value of the first target peak is not less than the peak-to-peak ratio threshold. (The preferred range is 0.5-0.8);

[0219] Peak spacing condition: candidate peak Two-dimensional spatial angular Euclidean distances between all confirmed target peaks ;

[0220] The angular space Euclidean distance threshold can be set according to the angular scan step size, array aperture, target frequency, and near-angle resolution requirements.

[0221] The other steps and parameters are the same as those in any of the specific implementation methods one to nine.

[0222] To make the technical solution and advantages of the present invention clearer, a simulation environment containing the sound sources of two multi-rotor UAVs and three multi-rotor UAVs was constructed for experiments. It should be understood that the two embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the invention. The specific parameter settings and processing procedures for the experiment are as follows:

[0223] Example 1:

[0224] I. Simulation conditions and parameter settings; the process is as follows:

[0225] 1: Sampling and propagation parameters: Sampling rate set to The simulation duration is 1.0s, and the speed of sound is... ;

[0226] 2. Microphone Array Deployment: A four-element three-dimensional array is used; element coordinates (unit: m) are: ;

[0227] 3. UAV Sound Source Model: Based on the tip passing frequency (BPF) characteristics of multi-rotor UAVs, two UAV sound sources with different fundamental frequencies and spatial positions are constructed to simulate a near-angle multi-target scenario; the fundamental frequency calculation formula is:

[0228]

[0229] Drone 1 parameters: Number of leaves Harmonic number 8, harmonic attenuation index Position coordinates ;

[0230] Drone 2 parameters: Number of leaves Harmonic number 8, harmonic attenuation index Position coordinates ;

[0231] 4: Target Propagation Model: The received signal of each array element is synthesized from distance attenuation, propagation delay, and superimposed noise: For each array element... Calculate distance Delay The attenuation coefficient is ;in, Indicates the first The location coordinates of the drone's sound source; Indicates the first The position coordinates of each microphone element.

[0232] The array receives signals consisting of superimposed signals from two UAVs, random disturbance noise at the sound source (simulating the UAV's own noise), and Gaussian white noise with a standard deviation of 0.02 (simulating ambient background noise).

[0233] II. Algorithm execution steps; the process is as follows:

[0234] Step 1: Feature frequency set extraction:

[0235] Step 11: Apply a Hanning window to the signal of channel 0, perform a Fast Fourier Transform (FFT) with 44100 points, and calculate the logarithmic amplitude spectrum to obtain the detection spectrum. ,like Figure 2 ;

[0236] Step 12: Set minimum peak height constraint: minimum peak height threshold is 20dB; Set peak significance: significance threshold is 10dB; Set minimum peak spacing: the absolute value of the frequency difference between two candidate peaks must be greater than 13. ;

[0237] Step 13: Extract the feature frequency set in the range of 200Hz-1400Hz. ;

[0238] Step 2: Calculate DOA and confidence score based on the feature frequency set; the specific process is as follows:

[0239] Step 21: Perform STFT (Frame Length Transmission) on the multi-channel signal. Points, overlapping 2048 points, number of time frames ), calculate the frequency resolution of STFT Selecting on the discrete frequency axis The frequency point with the smallest absolute difference is used as the reference frequency point. The complex spectral coefficients of all channels at the reference frequency point are extracted to form the multi-channel narrowband snapshot vector of a single time frame.

[0240]

[0241] Step 22: Construct the estimated covariance matrix:

[0242]

[0243] in, For the first An M-dimensional column vector in a time frame; For time frame indexing, , For the number of snapshots;

[0244] Step 23: Calculate the current feature frequency The smallest eigenvalue of the covariance matrix under the given conditions As a noise power estimate; construct a normalized matrix. :

[0245]

[0246] in, To prevent extremely small positive numbers with a denominator of zero;

[0247] Step 24: Calculate the matrix power; the process is as follows:

[0248] Take the power parameter Calculate the positive powers of the normalized matrices respectively. and negative powers ;

[0249] Step 25: At the azimuth angle Pitch angle Within the range The grid is scanned using a step size; at each grid point scanned, a corresponding spatial guide vector is constructed. :

[0250]

[0251] in, For microphone coordinate array; For azimuth and elevation angles The corresponding unit vector; The azimuth angle of the target drone; The pitch angle of the target drone; Indicates the speed of sound; Represents the imaginary unit. Superscript This indicates the transpose;

[0252] Based on spatial guide vectors and matrix powers, calculate the matrix power approximation spatial spectral function at grid points:

[0253]

[0254] Among them, superscript Indicates conjugate transpose;

[0255] After traversing the entire grid, extract the spatial spectral function. The maximum peak value, and the angle corresponding to the maximum peak value. That is, it serves as the estimated DOA value at that characteristic frequency point;

[0256] Step 26: Construct the confidence score and DOA point set; the process is as follows:

[0257] Let the spatial spectral function be... The maximum peak value is:

[0258]

[0259] Centered on the discrete angle grid where the maximum peak point is located, the neighborhood of the main peak formed by the two grid points to the left and right in the azimuth direction and the two grid points above and below in the elevation direction is removed. Then, the maximum value is searched in the remaining spatial spectrum after removing the neighborhood of the main peak, which is taken as the second highest peak. The confidence level is defined as:

[0260]

[0261] in, To prevent extremely small positive numbers with a denominator of zero, ; Characteristic frequency The confidence level is below;

[0262] The estimated DOA results are combined with their corresponding confidence levels to form DOA points with confidence level labels. Finally, the results for each frequency point are summarized to form a DOA point set, such as... Figure 3 ;

[0263] Step 3: Estimate the density clustering and target number of DOA point sets; the process is as follows:

[0264] Step 31: Two DOA points , The angular spatial Euclidean distance is:

[0265]

[0266]

[0267] in, Indicates confidence level The DOA point of the attribute; Indicates confidence level The DOA point of the attribute;

[0268] Represents characteristic frequency The following are the estimated directions of arrival; Represents characteristic frequency The following are the estimated directions of arrival;

[0269] Represents characteristic frequency The confidence level is below; Represents characteristic frequency The confidence level is below;

[0270] Indicates the difference in azimuth angle;

[0271] Step 32: DBSCAN density clustering: Parameter setting neighborhood radius is... The minimum sample size is 2; DBSCAN density clustering obtains cluster sets and noisy outliers, and the number of non-empty clusters is used as an adaptive estimate of the target number. Under the data conditions of this embodiment, two non-empty valid clusters are obtained. , Therefore, the target number is estimated. .

[0272] Step 4: Cluster-level spatial spectrum fusion and DOA output; the process is as follows:

[0273] Step 41: For each characteristic frequency within each cluster Corresponding spatial spectrum Perform maximum value normalization:

[0274]

[0275] Step 42: Perform incoherent superposition to obtain the weighted fusion spectrum; represented as:

[0276]

[0277]

[0278] Step 43: In the cluster-level fusion spatial spectrum Local maxima detection is performed to determine the maximum peak point of the discrete grid in the fused spatial spectrum, and the corresponding angle is used as the coarse localization result. Then, the azimuth and elevation angles corresponding to the peak point are refined using three-point parabolic interpolation. Output initial positioning results , .

[0279] Step 5: Use 0.7 times the peak amplitude of the maximum peak spectrum of the cluster-level fused spatial spectrum on the discrete angle grid as the peak ratio threshold, and set the minimum peak spacing to 3 angle grid points. Since the azimuth and elevation scanning steps in this embodiment are both... Therefore, the minimum peak spacing Upon verification, only one target peak exists in each of the two cluster-level fused spectra, indicating no cluster erroneous merging. Therefore, cluster splitting is unnecessary, and the initial localization results are retained. Output: , ,like Figure 6 .

[0280] III. Comparison of positioning results; specifically:

[0281]

[0282] Example 2:

[0283] I. Simulation conditions and parameter settings; the specific process is as follows:

[0284] 1: Set the sampling rate to The simulation duration is 1.0s, and the speed of sound is... ;

[0285] 2: A four-element three-dimensional array is used. The element coordinates (unit: m) are: ;

[0286] 3: Based on the tip passing frequency (BPF) characteristics of multi-rotor UAVs, three UAV sound sources with different fundamental frequencies and spatial positions are constructed to simulate a near-angle multi-target scenario; the fundamental frequency calculation formula is:

[0287]

[0288] Drone 1 parameters: Number of leaves Harmonic number 8, harmonic attenuation index Position coordinates ;

[0289] Drone 2 parameters: Number of leaves Harmonic number 8, harmonic attenuation index Position coordinates ;

[0290] Drone 3 parameters: Number of leaves Harmonic number 8, harmonic attenuation index Position coordinates ;

[0291] 4: The received signal of each array element is synthesized from distance attenuation, propagation delay, and superimposed noise: for each array element Calculate distance Delay The attenuation coefficient is ;in, Indicates the first The location coordinates of the drone's sound source; Indicates the first The position coordinates of each microphone array element;

[0292] The array receives signals consisting of three superimposed UAV signals, random disturbance noise from the sound source (simulating UAV noise itself), and Gaussian white noise with a standard deviation of 0.02 (simulating ambient background noise).

[0293] II. Algorithm execution steps; the specific process is as follows:

[0294] Step 1: Feature frequency set extraction; the process is as follows:

[0295] Step 11: Apply a Hanning window to the signal of channel 0, perform a Fast Fourier Transform (FFT) with 44100 points, and calculate the logarithmic amplitude spectrum to obtain the detection spectrum. ;

[0296] Step 12: Set minimum peak height constraint: minimum peak height threshold is 20dB; Set peak significance: significance threshold is 10dB; Set minimum peak spacing: the absolute value of the frequency difference between two candidate peaks must be greater than 13. ;

[0297] Step 13: Extract the feature frequency set in the range of 200 Hz-1400 Hz. ;

[0298] Step 2: Calculate DOA and confidence score based on the feature frequency set; the specific process is as follows:

[0299] Step 21: Perform STFT (frame length over time) on the multi-channel signal. Points, overlapping 2048 points, number of time frames ), calculate the frequency resolution of the STFT. Selecting on the discrete frequency axis... The frequency point with the smallest absolute difference is used as the reference frequency point. The complex spectral coefficients of all channels at the reference frequency point are extracted to form the multi-channel narrowband snapshot vector of a single time frame.

[0300]

[0301] Step 22: Estimate the covariance matrix:

[0302]

[0303] in, For the first An M-dimensional column vector in a time frame; For time frame indexing, , For the number of snapshots;

[0304] Step 23: Calculate the current feature frequency The smallest eigenvalue of the covariance matrix under the given conditions As a noise power estimate; construct a normalized matrix. ; indicates as:

[0305]

[0306] in, To prevent extremely small positive numbers with a denominator of zero;

[0307] Step 24: Obtain the power parameter Calculate the positive powers of the normalized matrices respectively. and negative powers ;

[0308] Step 25: At the azimuth angle Pitch angle Within the range The grid is scanned using a step size. At each grid point scanned, a corresponding spatial guiding vector is constructed. :

[0309]

[0310] in, For microphone coordinate array; For azimuth and elevation angles The corresponding unit vector; The azimuth angle of the target drone; The pitch angle of the target drone; Indicates the speed of sound; Represents the imaginary unit. Superscript This indicates the transpose;

[0311] Based on the spatial guiding vector and matrix exponentiation, calculate the matrix exponentiation approximate spatial spectral function at this grid point:

[0312]

[0313] Among them, superscript Indicates conjugate transpose;

[0314] After traversing the entire grid, extract the spatial spectral function. The maximum peak value, and the angle corresponding to that maximum peak value. That is, it serves as the estimated DOA value at that characteristic frequency point;

[0315] Step 26: Construct the confidence score calculation and DOA point set; the process is as follows:

[0316] Let the spatial spectral function be... The maximum peak value is:

[0317]

[0318] Centered on the discrete angle grid where the maximum peak point is located, the neighborhood of the main peak formed by the two grid points to the left and right in the azimuth direction and the two grid points above and below in the elevation direction is removed. Then, the maximum value is searched in the remaining spatial spectrum after removing the neighborhood of the main peak, which is taken as the second highest peak. The confidence level is defined as:

[0319]

[0320] in, To prevent extremely small positive numbers with a denominator of zero, ; Characteristic frequency The confidence level is below.

[0321] The estimated DOA results are combined with their corresponding confidence levels to form DOA points with confidence level labels. Finally, the results for each frequency point are summarized to form a DOA point set, such as... Figure 7 ;

[0322] Step 3: DOA point set density clustering and target number estimation; the process is as follows:

[0323] Step 31: Two DOA points , The angular space Euclidean distance is:

[0324]

[0325]

[0326] in, Indicates confidence level The DOA point of the attribute; Indicates confidence level The DOA point of the attribute;

[0327] Represents characteristic frequency The following are the estimated directions of arrival; Represents characteristic frequency The following are the estimated directions of arrival;

[0328] Represents characteristic frequency The confidence level is below; Represents characteristic frequency The confidence level is below; Indicates the difference in azimuth angle;

[0329] Step 32: DBSCAN clustering; the process is as follows:

[0330] Perform DBSCAN clustering, setting the neighborhood radius as a parameter. The minimum sample size is 2; clustering yields cluster sets and noisy outliers, and the number of non-empty clusters is used as an adaptive estimate of the target number. Under the data conditions of this embodiment, three non-empty valid clusters were obtained. , , Therefore, the target number is estimated. ;

[0331] Step 4: Cluster-level spatial spectrum fusion and DOA output; the process is as follows:

[0332] Step 41: For each cluster Corresponding spatial spectrum Perform maximum value normalization; the process is as follows:

[0333]

[0334] Step 42: Perform incoherent superposition to obtain the weighted fusion spectrum; the process is as follows:

[0335]

[0336]

[0337] Step 43: In the cluster-level fusion spatial spectrum Local maxima detection is performed to determine the maximum peak point of the discrete grid in the fused spatial spectrum, and the corresponding angle is used as the coarse localization result. Then, the azimuth and elevation angles corresponding to the peak point are refined using three-point parabolic interpolation. Output initial positioning results , , ;

[0338] Step 5: Use 0.7 times the peak amplitude of the maximum peak spectrum of the cluster-level fused spatial spectrum on the discrete angle grid as the peak ratio threshold, and set the minimum peak spacing to 3 angle grid points. Since the azimuth and elevation scanning steps in this embodiment are both... Therefore, the minimum peak spacing Upon verification, each cluster-level fusion spectrum contained only one target peak, with no cluster mis-merging, eliminating the need for cluster splitting. The initial localization results were retained, and the output was... , , ,like Figure 11 .

[0339] III. Comparison of positioning results; specifically:

[0340]

[0341] This invention may have other embodiments. Without departing from the spirit and essence of this invention, those skilled in the art can make various corresponding changes and modifications according to this invention, but these corresponding changes and modifications should all fall within the protection scope of the appended claims.

Claims

1. A target-number adaptive sound source localization method for multi-rotor UAVs, characterized in that: The specific process of the method is as follows: Step 1: Acquire multi-channel signals from the microphone array; perform spectral analysis on the acquired multi-channel signals from the microphone array to obtain the detection spectrum; obtain the characteristic frequency set based on the detection spectrum; Step 2: Calculate DOA and confidence score based on the feature frequency set; construct DOA point set based on DOA and confidence score; DOA represents the estimated direction of arrival. Step 3: Perform DBSCAN density clustering on the DOA point set to obtain the cluster set; determine the feature frequency set corresponding to each cluster; Step 4: Based on the feature frequency set corresponding to each cluster, obtain the clusters. The initial positioning results; Step 5: Cluster The corresponding cluster-level fusion spatial spectrum is subjected to local maximum detection to obtain a candidate peak set, and then the target peak set is determined based on the independent target conditions; If the target peak set contains only one target peak, then a cluster is determined. No splitting is needed; the initial localization results obtained in step 4 can be directly used as clusters. The final positioning result; If the target peak set contains multiple target peaks, then a cluster is determined. It is necessary to break down the target peaks and refine them by performing three-point parabolic interpolation on each target peak in the target peak set to obtain the final location results of multiple targets; multiple means two or more targets. The process continues until all clusters are identified, at which point the final location result is obtained.

2. The target number adaptive multi-rotor UAV sound source localization method according to claim 1, characterized in that: The specific process of step 1 is as follows: Step 11: Acquire multi-channel signals from the microphone array; Step 12: Perform spectral analysis on the multi-channel signal of the microphone array acquired in Step 11 to obtain the detection spectrum; Step 13: Based on the detection spectrum Obtain the set of characteristic frequencies; The specific process is as follows: In the preset frequency band Internal detection spectrum Peak detection is performed to obtain all peak values. Each peak value corresponds to one feature frequency, and all feature frequencies constitute a feature frequency set. , For the first One characteristic frequency; Preset frequency band satisfy , Sampling rate; This is the preset minimum frequency band value. This is the preset maximum value for the frequency band.

3. The target number adaptive multi-rotor UAV sound source localization method according to claim 2, characterized in that: The preset frequency band Internal detection spectrum Peak detection is performed to obtain all peak values. Each peak value corresponds to one feature frequency, and all feature frequencies constitute a feature frequency set. , For the first One characteristic frequency; The specific process is as follows: In the preset frequency band The local maximum detection algorithm is used to detect the spectrum. Peak detection is performed to obtain a candidate peak set; each candidate peak in the candidate peak set corresponds to a frequency position. Frequency position The detection spectral amplitude value ; The candidate peak set is filtered based on the set constraints of peak height, peak significance, and peak spacing to obtain valid peaks that meet the constraints; each valid peak corresponds to a frequency position, and the frequency positions corresponding to all valid peaks constitute the characteristic frequency set. , For the first Characteristic frequencies.

4. The target number adaptive sound source localization method for a multi-rotor UAV according to claim 3, characterized in that: The specific process of step 2 is as follows: Step 21: Use short-time Fourier transform to extract the time-frequency domain of the multi-channel signal to obtain narrowband data; the specific process is as follows: Step 211: Calculate the frequency resolution of the short-time Fourier transform. ; Step 212: Based on the multi-channel short-time Fourier transform time-frequency data, construct the multi-channel narrowband snapshot vector corresponding to each characteristic frequency; Step 22: Estimate the spatial covariance matrix based on the multi-channel narrowband snapshot vector sequence; the specific process is as follows: Based on characteristic frequency Construct the spatial covariance matrix from the multi-channel narrowband snapshot vectors of each time frame; Step 23: Estimate the noise power based on the spatial covariance matrix, and construct the normalized covariance matrix based on the noise power; Calculate matrix powers based on the normalized covariance matrix; Calculation of spatial spectral functions based on matrix exponentiation; The DOA is solved based on the spatial spectral function; the specific process is as follows: Step 231: Estimate noise power based on the spatial covariance matrix, and construct a normalized covariance matrix based on the noise power; the specific process is as follows: Step 2311: Calculate the current feature frequency Lower space covariance matrix The smallest eigenvalue , As an estimate of noise power; the specific process is as follows: right 3D covariance matrix Perform eigenvalue decomposition to obtain A number of non-negative real eigenvalues ​​will The non-negative real eigenvalues ​​are arranged in descending order as follows: in, for The first non-negative real eigenvalue after arranging the non-negative real eigenvalues ​​in descending order; for The second non-negative real eigenvalue after arranging the non-negative real eigenvalues ​​in descending order; for The nth non-negative real eigenvalues ​​arranged in descending order One non-negative real eigenvalue; for The nth non-negative real eigenvalues ​​arranged in descending order One non-negative real eigenvalue; Will As characteristic frequency Noise power estimate under ; Step 2312: Based on feature frequencies Noise power estimate under Construct the normalized covariance matrix ; indicates as: in, It is a positive number. ; Step 232: Based on the normalized covariance matrix Calculate matrix powers; the specific process is as follows: Take the power parameter Calculate the normalized covariance matrix respectively. positive power and negative powers ; Step 233: Calculate the spatial spectral function based on matrix exponentiation ; indicates as: in, It is a spatial guiding vector; Step 234: Based on spatial spectral function Solve for DOA; the specific process is as follows: Take the spatial spectral function The maximum value corresponding to As characteristic frequency Arrival direction estimation results The expression is: in, Represents the spatial spectral function The azimuth and elevation angles corresponding to the maximum values ; Represents characteristic frequency The arrival direction estimation results are as follows; Step 24: Calculate the confidence level based on the spatial spectral function; Step 25: Construct a set of DOA points based on DOA and confidence level.

5. The target number adaptive sound source localization method for a multi-rotor UAV according to claim 4, characterized in that: The specific process of step 24 is as follows: Let the spatial spectrum function The maximum peak value is: With the maximum peak point Centered on the grid point, remove all grid points in the azimuth direction to the left and right of the center. Each grid point is then removed from the center in the pitch direction. A grid of points forms the neighborhood of the main peak. The maximum value in the remaining spatial spectrum after removing the neighborhood of the main peak is searched and taken as the second highest peak value. ,based on and The confidence level is calculated and expressed as: in, It is a positive number. ; Characteristic frequency The confidence level is below.

6. The target number adaptive multi-rotor UAV sound source localization method according to claim 5, characterized in that: The specific process of step 25 is as follows: Each characteristic frequency Arrival direction estimation results With characteristic frequency Confidence level By combining these elements, a single-frequency DOA point can be obtained. A set of DOA points with confidence attributes is constructed based on single-frequency DOA points.

7. The target number adaptive sound source localization method for a multi-rotor UAV according to claim 6, characterized in that: The specific process of step 3 is as follows: Step 31: For the DOA point set containing confidence attributes Perform DBSCAN density clustering to obtain a set of clusters; the specific process is as follows: Set the neighborhood radius and minimum number of samples; Define with confidence level DOA points of attributes With confidence level DOA points of attributes The angular Euclidean distance between them is expressed as: in, Indicates confidence level The DOA point of the attribute; Indicates confidence level The DOA point of the attribute; Represents characteristic frequency The arrival direction estimation results are as follows; Represents characteristic frequency The arrival direction estimation results are as follows; Represents characteristic frequency The confidence level is below; Represents characteristic frequency The confidence level is below; Indicates the difference in azimuth angle; DOA points with confidence attributes and The angular Euclidean distance between them; Point set based on the angular Euclidean distance between DOA points Perform DBSCAN density clustering and use the number of clusters generated by the clustering as the target number for adaptive estimation. And obtain the cluster set. ; in, Indicates the first cluster; Indicates the second cluster; Indicates the first A cluster; Indicates the first A cluster; Points that do not belong to any cluster are marked as noise outliers and removed; Step 32: Calculate the feature frequency set corresponding to each cluster; the specific process is as follows: Record each cluster The feature frequencies corresponding to all DOA points are included to form a cluster. The corresponding set of characteristic frequencies ; in, Cluster The corresponding set of characteristic frequencies; Indicates belonging to a cluster Single-frequency DOA points with confidence attributes; .

8. The target number adaptive multi-rotor UAV sound source localization method according to claim 7, characterized in that: The specific process of step 4 is as follows: Step 41: For clusters The spatial spectrum corresponding to each characteristic frequency Amplitude normalization is performed to obtain the normalized spatial spectrum. ; indicates as: in, Represents the spatial spectral function; Represents the spatial spectral function The global maximum peak value across all scan grids; It is a positive number. ; Step 42: For clusters Normalized spatial spectrum corresponding to each characteristic frequency Weighted incoherent fusion is performed to obtain the cluster-level fusion spatial spectrum. ; indicates as: in, Represents characteristic frequency The weights below; Characteristic frequency The weights below From characteristic frequency Confidence level Confirmed; expressed as: Step 43: In the cluster-level fusion spatial spectrum The maximum value is detected to obtain the angle corresponding to the maximum peak point; the specific process is as follows: Preset step size , ; In azimuth Pitch angle Perform a spatial grid scan within the range to obtain discrete azimuth angles. and pitch angle ; Determine the cluster-level fusion spatial spectrum Values ​​on the spatial grid And solve for the coarse positioning angle corresponding to the maximum peak point. ; Step 44: Coarsely locate the angle corresponding to the maximum peak point. Perform three-point parabolic interpolation refinement, cluster Output direction of arrival estimation Arrival direction estimation As a cluster The initial positioning results.

9. A target-number adaptive sound source localization method for a multi-rotor UAV according to claim 8, characterized in that: The specific process of step 44 is as follows: Step 441: Fix ,Pick , , ,but ; in, This represents the discrete grid index of the pitch angle corresponding to coarse positioning; This represents the spectral amplitude value of the grid cell preceding the maximum peak point; This represents the spectral amplitude at the point of maximum peak. This represents the spectral amplitude value of the grid one azimuth angle after the point of maximum peak. Indicates the cluster-level fusion spatial spectrum in the grid index The function value at that location; Indicates the cluster-level fusion spatial spectrum in the grid index The function value at that location; Indicates the cluster-level fusion spatial spectrum in the grid index The function value at that location; This represents the grid offset for azimuth parabolic interpolation; This represents the high-precision azimuth estimate output after interpolation refinement; This represents the coarse positioning angle of the azimuth obtained from discrete grid scanning; Indicates the angle scan step size; Step 442: Fix ,Pick , , ,but ; in, This indicates the azimuth discrete grid index corresponding to coarse positioning; This represents the spectral amplitude value of the grid at the pitch angle preceding the maximum peak point; This represents the spectral amplitude at the point of maximum peak. This represents the spectral amplitude value of the grid at the pitch angle following the point of maximum peak. Indicates the cluster-level fusion spatial spectrum in the grid index The function value at that location; Indicates the cluster-level fusion spatial spectrum in the grid index The function value at that location; Indicates the cluster-level fusion spatial spectrum in the grid index The function value at that location; This represents the grid offset for parabolic pitch interpolation. This represents the high-precision pitch angle estimate output after interpolation refinement. This represents the pitch angle and coarse positioning angle obtained from discrete grid scanning; Indicates the angle scan step size; Step 443: Coarsely locate the angle corresponding to the maximum peak point. Perform three-point parabolic interpolation refinement, cluster Output direction of arrival estimation Arrival direction estimation This serves as the initial localization result.

10. A target-number adaptive sound source localization method for a multi-rotor UAV according to claim 9, characterized in that: The specific process of step 5 is as follows: Step 51: For clusters Corresponding cluster-level fusion spatial spectrum Local maxima detection is performed to obtain a set of candidate peaks, represented as: For candidate peak set Sort the candidate peaks by spectral amplitude from largest to smallest, and select the candidate peaks with the largest spectral amplitudes as the candidate peak set. The first target peak, and initialize the candidate peak set. target peak set ; indicates as: in, To set candidate peaks Candidate Peak After sorting the spectral amplitudes from largest to smallest, the candidate peak with the largest spectral amplitude is selected. To select from candidate peak sets The target peak set obtained through screening; Then, iterate through all candidate peaks in sequence. Other candidate peaks Determine candidate peaks Does it meet the independence condition? If the candidate peak If the independent objective condition is met, then the candidate peak will be... It was determined to be an independent target peak and a candidate peak was selected. Add to target peak set Otherwise, candidate peak Not included in the target peak set ; After the traversal is complete, a set of candidate peaks is obtained. The target peak set; If the target peak set If a cluster contains only one target peak, it is considered a cluster. No splitting is needed; the initial localization results obtained in step 4 can be directly used as clusters. The final positioning result; If the target peak set If it contains multiple target peaks, then it is determined to be a cluster. There are multiple near-angle targets, requiring cluster splitting; target peak set Each target peak in the azimuth and elevation directions is refined by three-point parabolic interpolation in step 44 to obtain the final positioning results of multiple targets; multiple means two or more targets. Step 52: Repeat step 51 until all clusters are identified and the final location result is obtained. ; in, For the final target quantity; The independent condition is that both the peak ratio condition and the peak spacing condition are satisfied simultaneously. Peak ratio condition: candidate peak The ratio of the amplitude value to the amplitude value of the first target peak is not less than the peak-to-peak ratio threshold. ; Peak spacing condition: candidate peak Two-dimensional spatial angular Euclidean distances between all confirmed target peaks ; The threshold value is the Euclidean distance in the angular space.

Images