A signal autonomous detection and classification method based on single vector sensor

Abstract

The present application relates to the field of signal processing, in particular to a kind of signal autonomous detection and classification method based on single vector sensor;The method first utilizes the sound pressure and particle velocity data received by vector sensor to calculate the coherence coefficient of each frequency point and extract phase, obtain phase fluctuation factor by analyzing the statistical characteristics of phase change with time;Then set fluctuation factor threshold, obtain the preliminary detection result of signal using the difference of signal and noise phase fluctuation;Finally, the secondary detection of time length and bandwidth is carried out through the preliminary detection result of signal, and the detection result is classified into narrowband line spectrum signal and wideband signal;Compared with other signal detection techniques based on spectral energy, the present application can provide more robust signal autonomous detection result, and the detection result can be further used for target direction finding and identification.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of signal processing, in particular to a signal autonomous detection and classification method based on a single vector sensor. BACKGROUND

[0002] Deep-sea submersible buoy is an important sonar device for long-term continuous monitoring of marine environment, and is an important part of the ocean observatory network. Because the submersible buoy has good concealment and deployment mobility, it can also be used for collecting the radiated acoustic signals of underwater ship targets, and for establishing a deep-sea underwater target database. Compared with shallow water, the deep-sea water area is much wider, and the number of ship targets within the detection range of the submersible buoy is small. In order to improve the data storage capacity of the submersible buoy and the utilization rate of power, and prolong the working time of the submersible buoy, a matching underwater target autonomous detection system can be designed, which starts the data acquisition module when the system detects the existence of the target. Therefore, the underwater target autonomous detection system is a water acoustic signal processing module that always keeps working in the submersible buoy, which requires that the algorithm of the autonomous detection system is robust, the calculation amount is moderate, and the power consumption is low.

[0003] The submersible buoy system can carry various types of sonar devices or sonar arrays, among which the vector sensor is a very important type of carried device. The vector sensor can measure the sound pressure and particle velocity at a point in space at the same time, and the use of sound pressure and particle velocity data can realize multi-target direction finding, and the velocity spectrum processing and sound intensity averaging based on the vector signal can suppress environmental noise. Therefore, the hardware and signal processing software design of the submersible buoy system carrying the vector sensor has high theoretical research and engineering application value, and the target autonomous detection research based on single vector sensor and vector sensor array has been carried out.

[0004] In terms of signal processing methods, the target detection method based on deep learning and other artificial intelligence fields is not mature enough in the application of underwater acoustic engineering, so it is still necessary to use traditional underwater acoustic signal processing methods to realize underwater acoustic target detection. The classical acoustic vector signal detection method mainly includes direction finding and target line spectrum feature extraction, and finally determines whether the target exists by setting a threshold. The above method based on energy decision is easily affected by environmental noise and sonar device self-noise, which causes sudden changes in signal energy and false alarms. In order to ensure the accuracy and reliability of target detection results, suitable for the above application background and needs, the present application proposes a kind of underwater acoustic signal autonomous detection and classification method based on single acoustic vector sensor. SUMMARY

[0005] To support the design and implementation of the signal processing module for autonomous signal detection in a submersible buoy system, this invention proposes a signal autonomous detection and classification method based on a single vector sensor. This method first calculates the coherence coefficient at each frequency point and extracts the phase using sound pressure and particle velocity data received by the vector sensor. The phase fluctuation factor is obtained by analyzing the statistical characteristics of phase changes over time. Then, a fluctuation factor threshold is set, and the difference between signal and noise phase fluctuations is used to obtain preliminary signal detection results. Finally, secondary detection of duration and bandwidth is performed based on the preliminary detection results, and the detection results are classified as narrowband line spectrum signals and broadband signals. Compared to other signal detection techniques based on spectral energy, this invention provides more robust autonomous signal detection results, which can be further used for target direction finding and identification.

[0006] The technical solution adopted in this invention is a signal autonomous detection and classification method based on a single vector sensor, which consists of the following steps:

[0007] S1 uses sound pressure and particle velocity data received by a single-vector sensor to calculate the coherence coefficient at each frequency point and extract the phase. The phase fluctuation factor is obtained by analyzing the variance of the phase over time; specifically as follows:

[0008] S1.1 segments the sound pressure and particle velocity data acquired by the single-vector sensor: The single-vector sensor acquires particle velocity x-channel sampling data S x Sampling data of particle velocity y channel y Sampling data of particle velocity z-channel z Sound pressure channel sampling data S p Afterwards, with the window length w l The above sampled data is truncated in seconds, and the time corresponding to the i-th data segment is t. i , i is the data segment number, i = 1, 2, 3, ..., S, S is the total number of data segments.

[0009] S1.2 The autospectrum and cross-spectrum of each vector signal data segment are calculated using the classical spectral estimation method (Hu Guangshu. Digital Signal Processing: Theory, Algorithms and Implementation [M]. Tsinghua University Press Co., Ltd., 2007: 3.). The spectral analysis frequency is k. j Let j represent the j-th frequency point, j = 1, 2, 3, ..., M, and M represent the total number of frequency points. Autospectral estimation result I pp (i,j), I xx (i,j), I yy (i,j), I zz (i,j) represent time t respectively. i The corresponding data segment at frequency k jAutocorrelation results of the sound pressure channel, the particle velocity x channel, the particle velocity y channel and the particle velocity z channel at time t px (i,j), I py (i,j), I pz (i,j) respectively represent the autocorrelation results of the sound pressure channel, the particle velocity x channel, the particle velocity y channel and the particle velocity z channel at time t i The corresponding data segments at frequency point k j The autocorrelation results of the sound pressure channel and the particle velocity x channel, the sound pressure channel and the particle velocity y channel, the sound pressure channel and the particle velocity z channel at time t i The corresponding data segments at frequency point k j The coherence coefficients at frequency point k

[0010]

[0011] wherein C px (i,j) represents the coherence coefficient of the sound pressure channel and the particle velocity x channel at time t i The corresponding data segments at frequency point k j The coherence coefficient of the sound pressure channel and the particle velocity y channel at time t py The corresponding data segments at frequency point k i The coherence coefficient of the sound pressure channel and the particle velocity z channel at time t j The corresponding data segments at frequency point k pz The coherence coefficient of the sound pressure channel and the particle velocity z channel at time t i The corresponding data segments at frequency point k j The coherence coefficient of the sound pressure channel and the particle velocity z channel at time t

[0012] S1.3 extracts the phase of the coherence coefficient C px (i,j), C py (i,j) and C pz (i,j) to obtain the phase ph px (i,j) of the coherence coefficient C px (i,j), the phase ph py (i,j) of the coherence coefficient C py (i,j), the phase ph pz (i,j) of the coherence coefficient C pz (i,j).

[0013] S1.4 represents the extracted coherence coefficient phase information as follows:

[0014]

[0015] wherein PH x , PH y and PH zLet be the phase matrix storing the coherence coefficients of the sound pressure channel and the x, y, and z channels of particle velocity, respectively. Each row of the matrix is ​​arranged in time order t. i ,t i+1 ...store k1 to k M The phase of the corresponding M frequency points.

[0016] S1.5 sets a duration threshold T1, which should satisfy T1≥10*w l If the window is long w l =1s, the duration threshold T1 should be set to 10s or more. When the accumulated t i To t i+N-1 The corresponding N data segments satisfy Nw l When T1 is greater than or equal to T1, the phase matrix of the coherence coefficients is decomposed into the following form:

[0017]

[0018] in, and These represent the phase matrices of the coherence coefficients of the sound pressure channel and the particle velocity x, y, z channels at N (N < S) ​​data segments and M frequency points after satisfying the duration threshold T1.

[0019] S1.6 Calculate the standard deviation of the phase for each column of each matrix in formula (3) to obtain the standard deviation matrix STD:

[0020]

[0021] Where std represents the standard deviation operation, and the element std in the standard deviation matrix STD in formula (4) is... ph-px (1) is the matrix in formula (3). The first column vector [ph px (1,1),ph px (2,1),...,ph px (i,1),...,ph px (N,1)] T The standard deviation of all elements in the matrix can be used to obtain the other elements in the standard deviation matrix STD.

[0022] Taking the reciprocal of each element of the standard deviation matrix STD in formula (4), and defining the reciprocal of the phase standard deviation as the phase fluctuation factor, we obtain the phase fluctuation factor matrix F:

[0023]

[0024] Based on the difference between the signal and the noise phase fluctuation characteristics, the signal and the noise can be distinguished by the size of the phase fluctuation factor, i.e. the frequency with a smaller phase fluctuation factor is the noise frequency, and the frequency with a larger phase fluctuation factor is the signal frequency. According to the set phase fluctuation factor threshold, the signal characteristics of a certain frequency can be determined.

[0025] S2 sets the phase fluctuation factor threshold, and obtains the preliminary detection result of the signal by using the difference between the signal and the noise phase fluctuation;

[0026] S2.1 calculates the phase fluctuation factor of the noise data, and sets the maximum value C th The phase fluctuation factor threshold is set for all frequencies, and the detection result obtained based on the threshold value is represented as

[0027]

[0028] wherein D i (j) represents the detection result, 0 represents that the detection result is noise, 1 represents that the detection result is the target signal, F(1,j) represents the element in the first row of the phase fluctuation factor matrix F, F(2,j) represents the element in the second row of the phase fluctuation factor matrix F, and F(3,j) represents the element in the third row of the phase fluctuation factor matrix F.

[0029] It should be noted that, since the acoustic channel conditions of different water areas are different, the phase fluctuation factor threshold needs to be supported by certain noise data analysis results or prior information to ensure the robustness of the detection result. If long-time signal monitoring is required, the phase fluctuation factor threshold based on noise data analysis needs to be updated at least once a day.

[0030] S2.2 obtains t i to t i+N-1 time period, and stores the detection result D i (j) and updates the data. The data updating method is that the data segment corresponding to t i+N is stacked in, and the data segment corresponding to t i is stacked out. At this time, the data time period to be processed is updated to t i+1 to t i+N , and the S1 process is repeated on the updated data segment to obtain the preliminary detection result D i+1 (j). The preliminary detection result matrix R can be obtained by repeating S2.1.

[0031]

[0032] The processing data is updated in the data segment stacking and unstacking manner, and the preliminary detection result of the updated data is stored in each row of the detection result matrix in turn.

[0033] The primary detection result of whether a certain frequency is a signal in different time periods is obtained by the preliminary detection, and whether the signal of a certain frequency is a wideband or narrowband signal needs to be determined by secondary detection. The secondary detection needs to use the continuity of the line spectrum signal in time and the continuity of the wideband signal in frequency domain to realize signal classification, so it is necessary to accumulate multiple preliminary detection results to form a preliminary detection result matrix R. Each row of the preliminary detection result matrix R corresponds to different processing time, and each column corresponds to different signal frequency. When the processing time length accumulated by the preliminary detection result matrix R meets the secondary detection condition, the secondary detection can be started based on the matrix R.

[0034] S3. The preliminary detection result of the signal obtained by S2 is subjected to secondary detection, and the secondary detection result is classified into narrowband line spectrum signal and wideband signal;

[0035] S3.1 With the accumulation of the preliminary detection result matrix R, when the dimension of the preliminary detection result matrix R reaches GxM, G is the preliminary detection times, and the preliminary detection result matrix R stores the detection results of the time from t i to t i+G+N-1 , the length of the processed data time is (G+N)w l . A time length threshold T2 for judging whether to start secondary detection is set, which should meet T2≥2*T1. If T1=20s, T2 should be set to not less than 40s. When (G+N)w l ≥T2, the preliminary detection result matrix R G that meets the secondary detection condition is:

[0036]

[0037] The sum of each column element in the matrix R G and the ratio of the accumulated preliminary detection times G in the matrix are calculated respectively:

[0038]

[0039] The matrix R G is composed of 0 and 1, and the position corresponding to the element 1 indicates that the frequency is the target signal in the time. The ratio calculated by formula (9) is the proportion of each frequency point being determined as the target signal in G times of preliminary detection. The ratio CT(j) is defined as the time length proportion parameter, which is between 0-1. The parameter describes the time continuity of the target signal at the frequency k j .

[0040] For radiation signals from underwater targets, line spectrum signals typically have a longer time continuity and a larger CT(j) value compared to broadband signals. We set time proportion parameter thresholds for narrowband line spectrum signals and broadband signals respectively. That is, the time proportion parameter threshold for narrowband line spectrum signals needs to be set larger than that for broadband signals. In this way, the frequency of the line spectrum signal can be extracted by filtering the time proportion parameter CT(j).

[0041] S3.2 Narrowband line spectrum detection.

[0042] Set the threshold T for the duration percentage parameter of narrowband line spectrum signals. L When the length proportion parameter CT(j) ≥ T L When, it indicates the frequency k j For the line spectrum signal frequency, the line spectrum detection result at this frequency is recorded as 1; if the duration percentage parameter CT(j) < T L This indicates the frequency k j If the frequency is not a line spectrum signal frequency, record the line spectrum detection result for that frequency as 0. Duration percentage parameter threshold T L The setting range is between 0 and 1. The specific value can be confirmed after debugging and testing data. For example, setting T... L Setting it to 0.8 means that only frequency points that account for more than 80% of the duration are considered line spectrum signals.

[0043] S3.3 Broadband Signal Detection.

[0044] Set the threshold T for the duration percentage parameter of broadband signal detection. B The threshold value of the duration percentage parameter T B With T L The settings are the same, but it should be noted that because the time continuity of broadband signals is not as strong as that of line spectrum signals, T... B It needs to be set to be higher than T L Small. Use D. B (j), j = 1, 2, ..., M represents the broadband signal detection result, if T B ≤CT(j)<T L , frequency k j The test result at point D was recorded as D. B (j) = 1; if CT(j) < T B , frequency k j The test result at point D was recorded as D. B (j) = 0. Let vector D... B By summing the elements and calculating the ratio to M, the bandwidth percentage parameter can be obtained.

[0045]

[0046] The bandwidth ratio parameter CF describes the ratio of signal bandwidth to the analysis frequency bandwidth. A larger CF parameter indicates a larger signal bandwidth, thus it can be used to filter out broadband signals. The bandwidth ratio parameter threshold F is set. B The setting range is 0 to 1, and F can be calculated based on the ratio of the expected minimum detection bandwidth to the analysis frequency bandwidth. B If CF≥F B This indicates that a broadband signal has been detected, and the broadband signal detection result is 1; if CF < F B If the result is 0, it indicates that no broadband signal was detected, and the detection results for all frequencies are replaced with 0.

[0047] S3.4 Classification of narrowband line spectrum and broadband signals.

[0048] By recording the frequencies corresponding to narrowband line spectrum detection in S3.2 and S3.3, the frequency information of narrowband line spectrum signals and broadband signals can be obtained respectively, thereby realizing the classification of narrowband and broadband signals.

[0049] S4 data stream update.

[0050] After processing steps S1 to S3, the processing of t was achieved. i To t i+G+N-1 The detection and classification of narrowband and wideband signals in the G+N segment of data at time t corresponds to the next signal detection and classification. i+1 The process begins with S1, which calculates the phase fluctuation factor of the data; S2, which uses the phase fluctuation factor to perform preliminary signal detection; and S3, which classifies narrowband and broadband signals based on the preliminary detection results.

[0051] This invention performs two data updates, S2.2 and S4. The purpose of the S2.2 data update is to acquire detection results at different times and store them in a matrix to form a preliminary detection result matrix. The purpose of the S4 data update is to perform the next signal detection and classification on the subsequently acquired vector data after completing the entire process of signal detection and classification.

[0052] The present invention has the following beneficial effects:

[0053] 1.The application proposes a signal detection and classification method suitable for single vector sensor, which utilizes the phase fluctuation difference of signal and noise coherence coefficient in sound pressure and vibration velocity vector channel data, and adopts variance quantitative analysis to analyze the fluctuation degree of phase, so as to realize signal detection based on phase fluctuation characteristics. On the basis of signal detection results, the differences of narrowband and broadband signals in time domain and frequency domain are utilized to realize the classification of the two. Compared with energy detection method, the detection and classification algorithm proposed in the application is based on long-time phase feature analysis of data, which can effectively avoid false alarm and misjudgment caused by transient interference and other occasional conditions, realize stable signal detection, and also can perform signal classification, which is not possessed by other vector signal detection algorithms.

[0054] 2.The application comprehensively compares the phase fluctuation factor and threshold of sound pressure channel and all particle vibration velocity channel coherence coefficients (see formula (5) and (6)) in the signal detection process, and fully utilizes all data information collected by the vector sensor.

[0055] 3.The signal processing method adopted in the application is mainly the classical spectrum estimation algorithm based on Fourier theory, and the whole detection and classification process is mainly composed of preliminary detection and secondary detection, which is clear in process, small in calculation amount and supports real-time data processing, and has high engineering application value.

[0056] 4.The signal processing results such as self-spectrum, cross-spectrum, narrowband line spectrum frequency and signal bandwidth generated in the detection and classification link of the application can also be used in subsequent target direction finding and positioning, and as an important vector signal preprocessing module, it has a relatively wide application prospect. BRIEF DESCRIPTION OF DRAWINGS

[0057] Figure 1 SUMMARY

[0058] Figure 2 LOFAR analysis results of each channel data of the vector sensor

[0059] Figure 3 Phase fluctuation factor of sound pressure and vector channel data coherence coefficient

[0060] Figure 4 Signal detection results

[0061] Figure 5 Signal classification results DETAILED DESCRIPTION

[0062] The practical application of the application will be described below in combination with the drawings:

[0063] The signal detection and classification method proposed in this invention is used to process measured data from a single vector sensor. The LOFAR (Low Frequency Analysis Recording) analysis results of the measured sound pressure and vector channel data are as follows: Figure 2 As shown in the diagram. The following operational procedures will describe the detection and classification of the narrowband line spectrum and broadband signals from the measured data.

[0064] S1 uses the sound pressure and vibration velocity data measured by the vector sensor to calculate the coherence coefficient of each frequency point and extract the phase. The phase fluctuation factor is obtained by analyzing the statistical characteristics of the phase change over time.

[0065] S1.1 segments the vector signal acquired by the single-vector sensor. Input x-channel data S x y channel data S y Z-channel data S z Sound pressure p-channel data S p With window length w l =1s to perform non-overlapping truncation of the input data. For example... Figure 2 As shown, the measured data duration is 450 seconds, so it can be divided into 450 data segments, with each data segment corresponding to a time of 1 second, 2 seconds, 3 seconds, ..., 450 seconds.

[0066] S1.2 The autospectrum and crossspectrum of each measured data segment are calculated using the classical spectral estimation method. The selected frequency range is 400 to 700 Hz. The coherence coefficient of each frequency point is calculated according to formula (1).

[0067] S1.3 Extract the phase of the coherence coefficient.

[0068] S1.4 Store the extracted coherence coefficient phase information according to formula (2).

[0069] S1.5 sets the duration threshold to T1 = 20s. When the phase of the coherence coefficient of the 20 data segments corresponding to the 1st to the 20th seconds is accumulated, the threshold condition of the duration threshold T1 is met, and the coherence coefficient phase matrix as shown in formula (3) can be obtained.

[0070] S1.6 Calculate the standard deviation of each column element in the three phase matrices obtained in step S1.5, and then take the reciprocal of each standard deviation to obtain the phase fluctuation factor, resulting in the phase fluctuation factor matrix as shown in formula (5). Figure 3 The phase fluctuation factor matrix displays the coherence coefficients and phase fluctuation factors of the x, y, and z vector channels and the p sound pressure channel data, respectively, for each row. The magnitude of the phase fluctuation factor can be used to determine whether the frequency is the target signal.

[0071] S2 sets the phase fluctuation factor threshold, and obtains the preliminary detection result of the signal by using the difference between the signal and the noise phase fluctuation;

[0072] S2.1 sets the phase fluctuation factor threshold of all frequencies to C by analyzing the measured noise data th = 2, and performs signal detection according to formula (6).

[0073] S2.2 obtains the detection result in the first 20s time period after the threshold detection of S2.1, stores the detection result and updates the data, the data segment corresponding to the 21st is stacked, the data segment corresponding to the first is stacked, at this time, the processing data is updated to the data segment corresponding to the second to the 21st time period, the S1 process is repeated to obtain the detection result, and the detection result matrix is obtained by repeating the detection process.

[0074] S3 performs secondary detection by using the preliminary detection result of the signal, and classifies the detection result into narrowband line spectrum signals and wideband signals;

[0075] S3.1 sets the time threshold for determining whether the secondary detection is started to T2 = 40s, which indicates that the detection of the data segment corresponding to the first 40s time has been completed, and 21 detection results are stored in the detection result matrix, and the detection matrix dimension is 21x301. The time length ratio parameter is calculated, that is, the mean value of each column element of the detection matrix is calculated, the dimension of the time length ratio parameter vector is 1x301, and the secondary detection is started to classify the target.

[0076] S3.2 narrowband line spectrum detection.

[0077] The time length ratio parameter threshold of the narrowband line spectrum signal is set to T L = 0.8, the frequency point corresponding to the time length ratio parameter greater than or equal to the time length ratio parameter threshold is the line spectrum signal frequency, and the line spectrum detection result of the frequency is recorded as 1; the frequency point corresponding to the time length ratio parameter less than the time length ratio parameter threshold is not the line spectrum signal frequency, and the line spectrum detection result of the frequency is recorded as 0.

[0078] S3.3 wideband signal detection.

[0079] The time length ratio parameter threshold of the wideband signal detection is set to T B = 0.6, the bandwidth ratio parameter threshold is set to F B = 0.25. The detection result of the frequency corresponding to the time length ratio parameter greater than or equal to T B = 0.6 and less than T L = 0.8 is recorded as 1; the detection result of the frequency corresponding to the time length ratio parameter less than T BA duration percentage parameter of 0.6 corresponds to recording the detection result at the frequency as 0. Store the above detection results and calculate the bandwidth percentage parameter according to formula (10). If the bandwidth percentage parameter is greater than or equal to the bandwidth percentage parameter threshold F B =0.25 indicates that a broadband signal has been detected; if it is less than the bandwidth percentage parameter threshold F B If the value is 0.25, it indicates that no broadband signal was detected, and the detection results for all frequencies are replaced with 0.

[0080] S3.4 Classification of narrowband line spectrum and broadband signals.

[0081] By recording the frequencies corresponding to a result of 1 for both the narrowband line spectrum detection (S3.2) and the broadband signal detection (S3.3), the frequency information of the narrowband line spectrum signal and the broadband signal can be obtained respectively, thereby achieving the classification of narrowband and broadband signals.

[0082] S4 data stream update.

[0083] After processing steps S1 to S3, narrowband and wideband signals are detected and classified once for each of the 40 data segments of each channel corresponding to the time interval from 1 second to 40 seconds. The next signal detection and classification starts from the data segment of each channel in the second second, and then processing steps S1 to S3 are performed again until all data is processed.

[0084] Figure 4 The results show the signal detection of all measured data. The narrowband line spectrum signal is mainly present in the first 200 seconds, while a broadband signal with a frequency band of 550–700 Hz also exists within the same time period. In the 200–400 second time interval, a broadband signal with clear interference fringes can be observed in the analysis frequency band (400–700 Hz), which originates from the broadband radiated noise of the test ship. Based on the above observations and analysis, it can be concluded that preliminary detection results for both narrowband and broadband signals can be obtained.

[0085] Results 4 show that the preliminary detection stage can determine the presence or absence of a signal, but it cannot classify the signal. Therefore, in Figure 4 A second detection is performed based on the signal detection results to achieve signal classification. The classification results are as follows: Figure 5 As shown. Figure 5 The two sub-figures show the accumulation of frequency information over time for the narrowband line spectrum signal and the wideband signal after signal classification, respectively. The narrowband signal detection results will... Figure 4 Line spectrum signals meeting the narrowband signal classification criteria were extracted, specifically appearing as discrete, long-term continuous straight lines in the frequency domain. Wideband signal detection results showed... Figure 4 The classification and extraction results of medium-bandwidth signals are specifically presented as strip-shaped results distributed along the frequency axis.

Claims

1. A single vector sensor based signal autonomous detection and classification method, characterized in that, The method comprises the following steps: S1. Calculate the coherence coefficient of each frequency point and extract the phase using the sound pressure and particle velocity data received by the single vector sensor, and obtain the phase fluctuation factor by analyzing the variance of the phase change over time. Specifically as follows: S1.1 Segment the sound pressure and particle velocity data collected by single vector sensor: the single vector sensor collects particle velocity x channel sampling data S x , particle velocity y channel sampling data S y , particle velocity z channel sampling data S z and sound pressure channel sampling data S p After that, the above sampling data is intercepted with a window length w l second, the time corresponding to the i-th data segment is t i , i is the data segment number, i = 1, 2, 3, ···, S, S is the total number of data segments; S1.2 The autospectrum and cross-spectrum of each vector signal data segment are calculated using classical spectral estimation methods, with the spectral analysis frequency being k. j j represents the j-th frequency point, j = 1, 2, 3, ..., M, where M represents the total number of frequency points; autospectral estimation result I pp (i,j), I xx (i,j), I yy (i,j), I zz (i,j) represent time t respectively. i The corresponding data segment at frequency k j The autospectral and cross-spectral estimation results of the sound pressure channel, particle velocity x channel, particle velocity y channel, and particle velocity z channel at point I. px (i,j), I py (i,j), I pz (i,j) represent time t respectively. i The corresponding data segment at frequency k j Cross-spectrums of sound pressure channel and particle velocity x-channel data segment, cross-spectrums of sound pressure channel and particle velocity y-channel, cross-spectrums of sound pressure channel and particle velocity z-channel; time t calculated using self-spectrums and cross-spectrums. i The corresponding data segment at frequency k j The coherence coefficient at that point can be obtained as follows: where C px (i,j) represents the time t i corresponding data segment at frequency point k j The coherence coefficient of the sound pressure channel and the particle vibration x channel at frequency point k py (i,j) represents the time t i corresponding data segment at frequency point k j The coherence coefficient of the sound pressure channel and the particle vibration y channel at frequency point k pz (i,j) represents the time t i corresponding data segment at frequency point k j The coherence coefficient of the sound pressure channel and the particle vibration z channel at frequency point k S1.3 Extracting the coherence coefficients C px (i,j), C py (i,j), and C pz (i,j) and C px (i,j) and C px (i,j), C py (i,j) and C py (i,j), C pz (i,j) and C pz (i,j) S1.

4. Express the extracted phase information of the coherence coefficient as a matrix as follows: where PH x , PH y , and PH z are phase matrices storing the coherence coefficients of the sound pressure channels and the particle velocity x, y, z channels, respectively, with each row of the matrix storing the phases for k1 to k M corresponding to the M frequency points in time order t i , t i+1 ,... S1.5 sets a time threshold T1, when the accumulation t i to t i+N-1 The corresponding N data segments satisfy Nw l When Nw≥T1, the phase matrix of the coherence coefficient is split into the following form: wherein, and respectively represent the phase matrix of the coherence coefficients of the sound pressure channel and the particle velocity x, y, z channels at N data segments, M frequency points after satisfying the time length threshold T1, N < S. S1.

6. Calculate the standard deviation of each column of each matrix in formula (3) to obtain the standard deviation matrix STD: Where std represents the standard deviation operation, and the element std in the standard deviation matrix STD in formula (4) is... ph-px (1) is the matrix in formula (3). The first column vector [ph px (1,1),ph px (2,1),...,ph px (i,1),...,ph px (N,1)] T The standard deviation of all elements in the matrix can be used to obtain the other elements in the standard deviation matrix STD; Take the reciprocal of each element of the standard deviation matrix STD in formula (4), and define the reciprocal of the phase standard deviation as the phase fluctuation factor to obtain the phase fluctuation factor matrix F: Based on the difference in phase fluctuation characteristics of signals and noise, the size of the phase fluctuation factor can be used to distinguish signals and noise, that is, the frequency with a smaller phase fluctuation factor is a noise frequency, and the frequency with a larger phase fluctuation factor is a signal frequency. According to the set phase fluctuation factor threshold, the signal characteristics of a certain frequency can be determined; S2. Set the phase fluctuation factor threshold to obtain the preliminary detection result of the signal based on the difference in phase fluctuation between signals and noise; S2.1 Calculate the phase fluctuation factor of the noise data, and find the maximum value C of the phase fluctuation factor th The phase fluctuation factor threshold is set for all frequencies, and the detection result obtained based on the threshold value is represented as where D i (j) denotes the detection result, 0 indicates that the detection result is noise, 1 indicates that the detection result is a target signal, F(1,j) denotes an element in the first row of the phase fluctuation factor matrix F, F(2,j) denotes an element in the second row of the phase fluctuation factor matrix F, and F(3,j) denotes an element in the third row of the phase fluctuation factor matrix F. S2.2 After the preliminary threshold detection of S2.1, t i to t i+N-1 The detection result D of the time period i (j), store the detection result and update the data, the data update method is t i+N The data segment corresponding to the time is stacked, t i The data segment corresponding to the time is stacked, t i+1 to t i+N Repeat the S1 process on the updated data segment and obtain the preliminary detection result D i+1 (j), repeat S2.1 to obtain the preliminary detection result matrix R: Update the processing data in the form of data segment in and out of the stack, and each row of the detection result matrix stores the preliminary detection result of the updated data in turn; S3. Perform secondary detection on the preliminary detection result of the signal obtained in S2, and classify the secondary detection result into narrowband line spectrum signals and wideband signals; S3.1 With the accumulation of the preliminary detection result matrix R, when the dimension of the preliminary detection result matrix R reaches G x M, G is the preliminary detection times, and the preliminary detection result matrix R stores the detection results from t i to t i+G+N-1 The length of the processed data is (G+N)w l ; a time length threshold T2 for determining whether to start secondary detection is set, when (G+N)w l ≥ T2, the preliminary detection result matrix R that meets the secondary detection condition is: G ​ The sum of each column element of the matrix R is calculated G The ratio of the sum of each column element of the matrix R to the accumulated preliminary detection number G in the matrix is calculated Matrix R G is a matrix composed of 0 and 1, where the position of the element with 1 corresponds to the time and frequency that the frequency is the target signal in the time. The ratio calculated by formula (9) is the proportion of each frequency point determined as the target signal in G preliminary detections. Define the ratio CT(j) as the time length proportion parameter, which is between 0-1. This parameter describes the time continuity of the target signal at frequency k j . S3.

2. Narrowband line spectrum detection A time length proportion parameter threshold T for narrowband line spectrum signals is set L When the time length proportion parameter CT(j) ≥ T L , it indicates that the frequency k j is a line spectrum signal frequency, and the line spectrum detection result of the frequency is recorded as 1; if the time length proportion parameter CT(j) < T L , it indicates that the frequency k j is not a line spectrum signal frequency, and the line spectrum detection result of the frequency is recorded as 0; S3.

3. Wideband signal detection Set the time length proportion parameter threshold T for wideband signal detection B , the time length proportion parameter threshold T B is set in the same way as T L , because the time continuity of the wideband signal is not as strong as the line spectrum signal, so T B needs to be set smaller than T L ; D B (j) represents the wideband signal detection result, if T B ≤ CT(j) < T L , the detection result at frequency k j is recorded as D B (j) = 1; if CT(j) < T B , the detection result at frequency k j is recorded as D B (j) = 0; sum all elements of vector D B and calculate the ratio with M, and the bandwidth proportion parameter can be obtained: The bandwidth proportion parameter CF describes the proportion of the signal bandwidth to the analysis frequency bandwidth, and the greater the bandwidth proportion parameter CF indicates the greater the signal bandwidth, so that the parameter can be used to screen out wideband signals; a bandwidth proportion parameter threshold F is set B If CF≥F B , it indicates that a wideband signal is detected, and the wideband signal detection result is 1; if CF<F B , it indicates that no wideband signal is detected, and all the detection results of all frequencies are replaced by 0; S3.

4. Narrowband line spectrum and wideband signal classification Record the corresponding frequencies of S3.2 narrowband line spectrum detection and S3.3, and obtain the frequency information of narrowband line spectrum signals and wideband signals respectively, so as to realize the classification of narrowband and wideband signals; S4. Data stream update After processing steps S1 to S3, the processing of t was achieved. i To t i+G+N-1 The detection and classification of narrowband and wideband signals in the G+N segment of data at time t corresponds to the next signal detection and classification. i+1 The process begins with S1, which calculates the phase fluctuation factor of the data; S2, which uses the phase fluctuation factor to perform preliminary signal detection; and S3, which classifies narrowband and broadband signals based on the preliminary detection results.

2. A signal self-detection and classification method based on the single vector sensor of claim 1, characterized in that: In S1.5, the time length threshold T1 should satisfy T1≥10*w l .

3. A method for signal self-detection and classification based on the single vector sensor of claim 1, characterized in that: In S2.1, due to the different sound channel conditions of different water areas, certain noise data analysis results or prior information support is needed when setting the phase fluctuation factor threshold to ensure the robustness of the detection result.

4. A method for signal self-detection and classification based on the single vector sensor of claim 3, characterized in that: In S2.1, if long-term signal monitoring is required, the phase fluctuation factor threshold based on noise data analysis should be updated at least once a day.

5. A method for signal self-detection and classification based on the single vector sensor of claim 1, characterized in that: In S3.1, the time threshold T2 for determining whether to start secondary detection should satisfy T2≥2*T1.

6. A signal self-detection and classification method based on the single vector sensor of claim 1, characterized in that: In S3.2, the time length proportion parameter threshold T L The setting range of the time length proportion parameter threshold T is between 0 and 1, and the specific value can be confirmed by debugging and measuring data.

7. A method for signal self-detection and classification based on the single vector sensor of claim 6, characterized in that: In S3.2, the time length proportion parameter threshold T L is set to 0.8, that is, only the frequency point with a time length proportion exceeding 80% is considered as a line spectrum signal.

8. A signal self-detection and classification method based on the single vector sensor of claim 1, characterized in that: In S3.3, the bandwidth proportion parameter threshold F B The setting range is 0-1.

9. A signal self-detection and classification method based on the single vector sensor of claim 8, characterized in that: In S3.3, the bandwidth proportion parameter threshold F can be calculated according to the ratio of the minimum bandwidth of the expected detection and the analysis frequency bandwidth B .

Images