Line array underwater target detection method based on Eckart filtering and peak screening

Abstract

The application provides a linear array underwater target detection method based on Eckart filtering and peak value screening, which comprises the following steps: firstly, using a wideband diagonal loading MVDR beam forming method to process time domain array signals to obtain a FRAZ matrix; then, screening the FRAZ matrix through a peak value energy-number screening method and a feature weighting method based on Eckart filtering; then, sequentially screening and weighting the processed FRAZ matrix through a peak value cumulative energy screening method and a peak value number screening method; and finally, outputting a bearing history graph; compared with a traditional method, the method can effectively weaken the influence of noise and sidelobes on the underwater weak target detection performance, reduces false alarm, and improves the signal-to-noise ratio of the weak target.

Description

Technical Field

[0001] This invention belongs to the field of sonar array signal processing technology, and relates to a method for detecting underwater targets in linear arrays based on Eckart filtering and peak selection. Background Technology

[0002] A fundamental problem faced by sonar linear array systems is detecting the direction of a sound source target. To achieve stable target detection, sonar linear array systems typically process the broadband acoustic signals received by the sensor array. The basic process of broadband target detection is as follows: First, the time domain is converted into a frequency domain array signal using a short-time Fourier transform; then, beamforming processing is performed on the frequency domain data to obtain the beam scanning azimuth spectrum of each frequency point within a time segment, i.e., the frequency azimuth matrix (FRAZ matrix) of the time segment; then, the FRAZ matrix is ​​superimposed along the frequency dimension to obtain the azimuth spectrum of the corresponding time segment; finally, the azimuth spectrum of the entire time period is normalized and stitched together in chronological order to obtain an azimuth history map showing the change of the sound source target direction over time. The azimuth history map can enhance the target detection effect based on the trace correlation principle. However, due to the limitations of beamforming methods, the beam scanning azimuth spectrum contains a significant amount of sidelobes and noise. When underwater environments are filled with strong noise and acoustic signals that are not targets (interference), noise and interference can significantly hinder the detection of weak targets. Therefore, in broadband target detection, in addition to selecting a suitable beamforming method, it is also necessary to introduce post-processing methods tailored to the characteristics of the FRAZ matrix to further enhance the signal, reduce noise, and decrease sidelobes, thereby improving the target detection performance.

[0003] There are many current methods for target detection using broadband beamforming combined with post-processing, such as:

[0004] Reference 1 (Bono M. et al. Subband Energy Detection in Passive Array Processing[J]. University of Texas at Austin Technical Report,2000:1-6.) proposes to perform subband peak energy detection (SPED) after conventional beamforming (CBF). This method can improve the target resolution and is referred to as the CBF-SPED method. However, the detection results of this method have relatively high noise and sidelobe energy.

[0005] Reference 2 (Lou Wanxiang et al. A weak target detection method based on sub-band [J]. Ship Science and Technology, 2021, 43(19): 149-152.) proposes a method that combines CBF with SPED and Eckart filtering, which can further improve the signal-to-noise ratio. This method is called CBF-SPED-Eckart method. However, while Eckart filtering enhances the target signal-to-noise ratio, it also strengthens the side lobes and increases the false alarm rate.

[0006] Reference 3 (Shefeng Yan et al. Broadband Passive Signal Detection Through Hydrophone Array Optimization and Sped Processing[C]. 2002 China-Japan Acoustics Conference, 2002: 1-4.) proposes a method combining Minimum Variance Distortionless Response (MVDR) beamforming and SPED, denoted as the MVDR-SPED method. The detection results of this method have lower noise and sidelobes compared to the CBF-SPED method, but the MVDR method has the problem of poor robustness.

[0007] Reference 4 (Fan Wentao et al. A beam spectrum feature weighted underwater weak target detection method [J]. Signal Processing, 2022, 38(01):195-201.) proposes a feature weighted post-processing method based on Eckart filtering after diagonally loaded MVDR beamforming. Essentially, it utilizes diagonally loaded MVDR beamforming, SPED and Eckart filtering, which improves robustness and array gain. This method is referred to as the MVDR-SPED-Eckart method. However, this method still has the problem of high false alarm rate. Summary of the Invention

[0008] To address the aforementioned problems, the objective of this invention is to propose a linear array underwater target detection method based on Eckart filtering and peak selection. This method first uses a broadband diagonally loaded MVDR beamforming method to process the time-domain array signal to obtain the FRAZ matrix. Then, the FRAZ matrix is ​​processed through peak energy-number selection and a feature weighting method based on Eckart filtering. Subsequently, the FRAZ matrix is ​​subjected to peak cumulative energy selection, peak number selection, and weighting processing in sequence, and finally, a azimuth history map is output.

[0009] The technical solution of the present invention includes the following steps:

[0010] Step 1) Read the time-domain array signal acquired by the uniform linear array, and then divide the time-domain array signal of each channel into segments with a duration T. slot , Duration T of overlapping adjacent segments over Segmentation is performed to obtain the segmented time-domain array signals for each channel;

[0011] Step 2) Randomly select the time-domain array signal of each channel within a segment, divide it into multiple sub-bands through Discrete Fourier Transform (DFT), select the sub-bands within the target detection frequency range, and obtain the array element domain-frequency domain signal matrix;

[0012] Step 3) Perform diagonal loading MVDR beamforming on the pairwise domain-frequency domain signal matrix (see reference 4 for details of the diagonal loading method) to obtain the FRAZ matrix;

[0013] Step 4) Take the absolute value of the obtained FRAZ matrix and normalize it by dividing it by the smallest element value in the entire matrix to obtain the normalized matrix. Subtract one more to get the intermediate matrix Then the intermediate matrix Subband peak energy screening is performed, and the non-peak energies are set to zero to obtain the subband peak energy screening matrix.

[0014] Step 5) Screening matrix for sub-band peak energy In the beam domain, sub-band peak energy is filtered, selecting the top 20% of peaks in each sub-band, while setting the other peaks to zero, thus obtaining the beam domain peak matrix.

[0015] Step 6) Calculate the peak matrix in the beam domain. By filtering the accumulated energy and peak count in the beam domain, the azimuths with the highest accumulated energy and peak count across all scanning azimuths are selected, while the peak counts for other azimuths are set to zero, thus obtaining the accumulated energy and peak count filtering matrix.

[0016] Step 7) Use the Two-Pass Split Window (TPSW) method (see Reference 5: J. Zhu, et al. An Improved Background Normalization Algorithm for Noise Resilience in Low Frequency[J]. Journal of Marine Science and Engineering, 9, 2021, pp. 803(1-10).) to normalize the matrix. Perform subband noise estimation and subtract one from the noise estimation result to obtain the subband noise matrix.

[0017] Step 8) Filter the matrix using accumulated energy and peak count. and sub-band noise matrix Obtain the weight matrix based on Eckart filtering. (See reference 4 for details), derived from the weight matrix Screening matrix with cumulative energy and peak number The Hadamard product is used to weight the features to obtain the feature weight matrix.

[0018] Step 9) Weight the feature matrix Accumulated energy screening is performed in the beam domain. Among the locations where accumulated energy exists, the locations with the top 20% of accumulated energy are selected to form a feature-weighted energy screening matrix. And the feature weighting matrix The peak values ​​corresponding to the azimuths selected through cumulative energy filtering and several adjacent azimuths are all set to zero, resulting in a feature-weighted zero-set matrix.

[0019] Step 10) Set the feature weighted matrix to zero. Peak count screening is performed in the beam domain. Among the locations where energy peaks exist, the locations with the highest number of peaks in the top 20% are selected to form a feature-weighted peak count screening matrix.

[0020] Step 11) Use the TPSW method to normalize the matrix. Perform azimuth noise estimation and subtract one from the noise estimation result to obtain the azimuth noise matrix.

[0021] Step 12) Convert the azimuth noise matrix Squaring each element and then combining it with the feature-weighted energy screening matrix and feature weighted number screening matrix The sum of the Hadamard product is weighted to obtain the FRAZ matrix. The sum of each column of the FRAZ matrix (superimposed along the frequency dimension) is used to obtain the azimuth spectrum. Then, the spectral intensity of each azimuth in the azimuth spectrum is divided by the maximum spectral intensity in the azimuth spectrum to obtain the normalized azimuth spectrum of the segmented time-domain array signal.

[0022] Step 13) Repeat steps 2 to 12. After processing all the segmented time-domain array signals, stitch together the normalized azimuth spectra of all the obtained segmented time-domain array signals in chronological order to obtain a azimuth history diagram showing the change of the azimuth of the sound source target over time.

[0023] The beneficial results of this invention are:

[0024] 1. Compared with traditional methods, the method proposed in this invention can effectively reduce the impact of noise and sidelobes on the underwater weak target detection performance, reduce false alarms, and improve the signal-to-noise ratio of weak targets.

[0025] 2. The method proposed in this invention is simple in principle and easy to implement in engineering. Attached Figure Description

[0026] Figure 1 This is a flowchart illustrating the implementation of the present invention;

[0027] Figure 2 The results of field experiments using specific embodiments of the present invention are as follows: (a) A position history map obtained by processing field data using the existing CBF-SPED method; (b) A position history map obtained by processing field data using the existing CBF-SPED-Eckart method; (c) A position history map obtained by processing field data using the existing MVDR-SPED method; (d) A position history map obtained by processing field data using the existing MVDR-SPED-Eckart method; (e) A position history map obtained by processing field data using the method of the present invention; (f) A comparison of the normalized position spectrum obtained by processing the first segment of field data using five methods (CBF-SPED method, CBF-SPED-Eckart method, MVDR-SPED method, MVDR-SPED-Eckart method, and the method of the present invention); (g) ... Figure 2 (f) Enlarged view at azimuth 80° to 110°. Detailed Implementation

[0028] Figure 1 This is a flowchart illustrating the implementation of the present invention. The specific implementation includes the following steps:

[0029] Step 1: Read the time-domain array signal acquired by the uniform linear array, and then divide the time-domain array signal of each channel into segments with a duration T. slot , Duration T of overlapping adjacent segments over The signal is segmented to obtain the time-domain array signal of each channel.

[0030] Step 1-1) If the sampling rate of the time-domain array signal is greater than 10 times the signal bandwidth, then the signal needs to be downsampled. The time-domain array signal matrix after downsampling is Y:

[0031]

[0032] y m(n) represents the element value in the m-th row and n-th column of the time-domain array signal matrix Y, where M is the number of elements in the uniform linear array, and N is the number of time-domain sampling points after downsampling.

[0033] N = T total ×f s

[0034] Among them, T total f is the duration of the signal to be processed. s The sampling rate of the downsampled signal is typically no greater than 4 times the signal bandwidth and no less than 2 times the signal bandwidth.

[0035] If the sampling rate of the time-domain array signal is no more than 10 times the signal bandwidth, then the time-domain array signal matrix Y after downsampling is the same as the matrix formed by the time-domain array signal acquired by the directly read uniform linear array.

[0036] Step 1-2) Set the segmented signal duration to T slot The overlap time between adjacent segments is T. over ;

[0037] Typically, T slot The value range of T is 1 to 3 seconds. over The value range is 0 to 0.9T. slot s, in this embodiment, T slot For 3s, T over It takes 2.7 seconds;

[0038] Number of signal sampling points within the segment N slot Represented as

[0039] N slot =T slot ×f s

[0040] The number of overlapping points N between segments over Represented as

[0041] N over =T over ×f s

[0042] If the number of sampling points is not an integer, it will be rounded to an integer.

[0043] Note that the number of signal sampling points N within the segment slot The number of overlapping points N between segments over None of them can be greater than the number of time-domain sampling points N after downsampling;

[0044] If the duration of the last segment is less than T slot To ensure frequency resolution, the last segment is discarded, resulting in a total of N segments.time-slot Represented as

[0045]

[0046] When the total number of segments is N time-slot If the value is not an integer, round it down to the nearest integer.

[0047] nth slot (n slot =1,2,…,N time-slot The time-domain array signal matrix within each segment is

[0048] in, The position of the starting point of the segmented signal in the time-domain sampling point N:

[0049]

[0050] Step 2: Take any segment and divide the time-domain array signal of each channel into multiple sub-bands using DFT. Select the sub-bands within the target detection frequency range to obtain the array element domain-frequency domain signal matrix.

[0051] Step 2-1) with the nth slot Time-domain array signal matrix within each segment For example, perform N operations on the time-domain data of each array element. slot Point DFT transformation to the frequency domain, i.e., for Each row of data is processed by N slot Point DFT yields the frequency domain array signal matrix.

[0052]

[0053] in

[0054]

[0055]

[0056] xm(fk) represents the frequency domain array signal matrix. The value of the element in the m-th row and k-th column, f k Here, k represents the kth sub-band, and K represents the total number of sub-bands.

[0057] Step 2-2) The target detection frequency is set to a maximum of f. high The lowest detection frequency is f low :

[0058]

[0059] Among them, DULA denoted as the element spacing, and c as the underwater sound velocity;

[0060] Select f k The frequency point within this frequency range is f d :

[0061] f low ≤f d ≤f high ,d=d L ,…,d H

[0062] Where d represents the selected frequency point f d The frequency point f obtained after DFT k The position index (sub-band sequence number) in the middle, d L d represents the position index (subband number) corresponding to the minimum frequency point within the target detection frequency range. H This indicates the position index (sub-band number) corresponding to the maximum frequency point within the target detection frequency range;

[0063] Obtain the element-frequency domain signal matrix

[0064]

[0065] Step 3: Perform diagonal loading MVDR beamforming on the pairwise domain-frequency domain signal matrix to obtain the FRAZ matrix.

[0066] Step 3-1) First, design an azimuth scan vector Θ as the scanning azimuth range for the subsequent diagonally loaded MVDR beamforming method:

[0067]

[0068] Where n scan N is the azimuth number. scan This represents the total number of directions scanned. Indicates the nth scan The azimuth number corresponds to the azimuth; in this embodiment, the scanning azimuth range is set to 0–180 degrees, with an interval of 0.5 degrees, then N scan =361, θ1=0,

[0069] Step 3-2) Using the diagonally loaded MVDR beamforming method, respectively, the pairwise domain-frequency domain signal matrix is ​​transformed. Process each column of data:

[0070] With array element domain-frequency domain signal matrix Data in column d (Similar examples below)

[0071]

[0072] calculate covariance matrix

[0073]

[0074] Frequency point f d azimuth θ nscan The corresponding guide vector is

[0075]

[0076] Frequency point f d ,position The corresponding beamforming output is

[0077]

[0078] Among them, I M×M denoted as an M-order identity matrix; λ represents the diagonal loading factor, which is set to 1000 in this embodiment.

[0079] Then the element-domain frequency domain signal matrix After diagonal loading MVDR beamforming processing, the nth beam can be obtained. slot FRAZ matrix within each segment

[0080]

[0081] Step 4: Take the absolute value of the obtained FRAZ matrix, and normalize it by dividing it by the smallest element value in the entire matrix to obtain the normalized matrix. Subtract one more to get the intermediate matrix Then the intermediate matrix Subband peak energy screening is performed, and the non-peak energies are set to zero to obtain the subband peak energy screening matrix.

[0082] Normalized matrix Represented as

[0083]

[0084] Indicates frequency point f d ,position The corresponding normalized beamforming output;

[0085] intermediate matrix Represented as

[0086]

[0087] For the intermediate matrix matrix elements,

[0088] With frequency point f d For example, the specific steps for subband peak energy screening are as follows:

[0089]

[0090]

[0091] Step 5: Screening matrix for sub-band peak energy In the beam domain, sub-band peak energy is filtered, selecting the top 20% of peaks in each sub-band, while setting the remaining peaks to zero, thus obtaining the beam domain peak matrix.

[0092] With frequency point f d For example

[0093] Given a row vector, calculate the number of non-zero elements, which is the number of energy peaks in the sub-band, and we get n. peak Then to Sort the elements in the array in ascending order and record the index of each element before sorting. Then, take the last element at the position of 0.2n after sorting. peak (when 0.2n) peak If the value is not an integer, round it to the nearest whole number. The element at the corresponding index position is retained, and all other elements are set to zero;

[0094] Step 6: Calculate the peak matrix in the beam domain. The cumulative energy and peak count in the beam domain are filtered to select the azimuths with the highest cumulative energy and the highest peak count across all scanning azimuths. Peak counts for other azimuths are set to zero, resulting in a cumulative energy and peak count filtering matrix.

[0095] Step 6-1) Calculate the peak matrix in the beam domain. Summing each column yields a row vector that records the accumulated energy in each direction;

[0096] Sort the elements in this row vector in ascending order and record the index of each element before sorting. Then, take the last 0.2N elements after sorting. scan (when 0.2N) scan If the value is not an integer, round it off, and retain the logical index of the cumulative energy of these elements in the row vector.

[0097] Step 6-2) Calculate the peak matrix in the beam domain. Change all elements greater than zero to 1 to obtain a new matrix. Sum each column of the matrix to obtain a row vector that records the number of peaks in each direction.

[0098] Sort the elements in this row vector in ascending order and record the index of each element before sorting. Then, take the last 0.2N elements after sorting. scan (when 0.2N) scan (If the value is not an integer, round it to the nearest integer) and retain the number of peak values ​​of these elements in the row vector.

[0099] Step 6-3) Retrieve the cumulative energy logical index Logical index of peak count The intersection of these points yields the logical indexes for accumulated energy and peak count. Preserving the peak matrix in the beam domain Logical index of cumulative energy and peak count The peak value in the corresponding direction is set, while the peak values ​​in other directions are set to zero, thus obtaining a selection matrix for cumulative energy and peak number.

[0100] Step 7: Use the TPSW method to normalize the matrix. Perform subband noise estimation and subtract one from the noise estimation result to obtain the subband noise matrix.

[0101] With frequency point f d For example

[0102]

[0103] position The subband noise estimate is obtained by processing the data contained in two nearby rectangular windows as follows:

[0104] First, define a window interval W containing the data location index. t-index for

[0105] W t-index =[n scan -G,…,n scan -E-1,n scan -E,n scan +E,n scan +E+1,…,n scan +G]

[0106] Where G represents the interval between n and n within the window. scan The farthest directional index and n scanThe absolute value of the difference between n and n, E represents the difference between n and n within the window interval. scan The nearest directional index and n scan The absolute value of the difference;

[0107] Then, calculate the azimuth θ. nscan Average spectral intensity within the nearby window interval

[0108] in, Corresponding normalized matrix Mid-frequency point f d Down, direction The beamforming output is located at n. scan +G>N scan Or n scan -G < 0, i.e., θ nscan When the scan area is close to the minimum or maximum value of the scanning azimuth range, only values ​​within [0, N] are taken in the window. scan Data within the specified range;

[0109] Define a new sequence:

[0110]

[0111] Where r is a threshold adjustment parameter:

[0112]

[0113] In this embodiment, the relevant parameters E of TPSW are 3, G is 14, and C represents the adjustment factor, which is usually 1.14.

[0114] obtain frequency point f d ,position Within the nearby window range average u t (f d ,n scan )for

[0115]

[0116] Subtract one from the above result, and then iterate through all frequency points f. d and direction Obtain the subband noise matrix

[0117] Step 8: Filter the matrix using accumulated energy and peak count. Sub-band noise matrix Obtain the weight matrix based on Eckart filtering. From the weight matrix Screening matrix with cumulative energy and peak number The Hadamard product is used to weight the features to obtain the feature weight matrix.

[0118] Step 8-1) Using frequency point f d For example, a matrix can be selected using accumulated energy and the number of peaks. Sub-band noise matrix Obtain the weight matrix based on Eckart filtering. At frequency point f d ,position The element value at position is denoted as

[0119]

[0120] Step 8-2) From the weight matrix And cumulative energy and peak number screening matrix The feature weighting matrix is ​​obtained by performing Hadamard product weighting.

[0121] Step 9: Weight the feature matrix Accumulated energy screening is performed in the beam domain. Among the locations where accumulated energy exists, the locations with the top 20% of accumulated energy are selected to form a feature-weighted energy screening matrix. The feature-weighted energy screening matrix The peak values ​​corresponding to the azimuths selected through cumulative energy filtering and several adjacent azimuths are all set to zero, resulting in a feature-weighted zero-set matrix.

[0122] Step 9-1) Weight the feature matrix Summing each column yields a row vector that records the accumulated energy in each direction;

[0123] Step 9-2) Calculate the number of non-zero elements n in this row vector. e20 Then, sort the elements in this row vector in a forward order and record the index of each element before sorting. Take the last 0.2n element after sorting. e20 (when 0.2n) e20 (If the value is not an integer, round it off). Record the feature-weighted energy filtering logical index of these elements in the row vector. Preserving feature weighting matrix Feature-weighted energy filtering logical index The peak value of the corresponding direction is set, while the peak values ​​of other directions are set to zero, thus obtaining the feature-weighted energy screening matrix.

[0124] Step 9-3) Weight the feature matrix The peak values ​​corresponding to the azimuth selected through cumulative energy filtering and several adjacent azimuths are all set to zero. The specific implementation is as follows:

[0125]

[0126] Among them, l one Indicates in The azimuth index corresponding to the azimuth with a logic value of 1; denoted as the feature weighting matrix after the zeroing operation is... This represents a feature-weighted zeroing matrix.

[0127] Step 10: Set the feature weighted zero matrix Peak count screening is performed in the beam domain. Among the locations where energy peaks exist, the locations with the highest number of peaks in the top 20% are selected to form a feature-weighted peak count screening matrix.

[0128] Step 10-1) Set the feature weighted matrix to zero. The elements greater than zero are reassigned to 1 to obtain a new matrix. Then, each column is summed to obtain a row vector that records the number of peaks in each direction.

[0129] Step 10-2) Calculate the number of non-zero elements n in this row vector. n20 Then, sort the elements in this row vector in a forward order and record the index of each element before sorting. Take the last 0.2n element after sorting. n20 (when 0.2n) n20 (If the element is not an integer, round it to the nearest integer). Then, place these elements in the feature-weighted zeroing matrix. The element at the corresponding index position is retained, and the others are set to zero, resulting in a feature-weighted number filtering matrix.

[0130] Step 11: Use the TPSW method to normalize the matrix. Perform azimuth noise estimation and subtract one from the noise estimation result to obtain the azimuth noise matrix.

[0131] The relevant parameters for azimuth noise estimation are the same as those for sub-band noise estimation in step 7, therefore the same symbols E, G, C, A, and r are used for azimuth. For example

[0132]

[0133] The superscript T indicates transpose;

[0134] Frequency point f d The azimuth noise estimate is obtained from the data contained in two nearby rectangular windows after the following processing:

[0135] First, define a window interval W containing the data sub-band index. f-index for

[0136] W f-index =[dG,…,dE-1,dE,d+E,d+E+1,…,d+G]

[0137] Where G represents the absolute value of the difference between the sub-band index that is farthest from d within the window interval and d, and E represents the absolute value of the difference between the sub-band index that is closest to d within the window interval and d.

[0138] Then, calculate the frequency point f. d The average spectral intensity within the nearby window interval:

[0139]

[0140] in, Corresponding normalized matrix Mid-frequency point f d Down, direction Beamforming output at that point. For d+G>d H Or dG < d L That is, f d When the value is close to the minimum or maximum value of the target detection frequency range, only values ​​within [d] are taken in the window. L ,d H Data within the specified range;

[0141] Define a new sequence:

[0142]

[0143] A = G - E + 1

[0144] Get location At frequency f d Within the nearby window range The average value is

[0145]

[0146] Subtract one from the above result, and then iterate through all frequency points f. d and direction Obtain the azimuth noise matrix

[0147] Step 12: Convert the azimuth noise matrix Squaring each element and then combining it with the feature-weighted energy screening matrix and feature weighted number screening matrix The sum of the Hadamard product is used to obtain the FRAZ matrix. The sum of each column of the FRAZ matrix (superimposed along the frequency dimension) is used to obtain the azimuth spectrum. Then, the spectral intensity of each azimuth in the azimuth spectrum is divided by the maximum spectral intensity in the azimuth spectrum to obtain the normalized azimuth spectrum of the segmented time-domain array signal.

[0148] Step 13: Repeat steps 2 to 12. After processing all the segmented time-domain array signals, stitch together the normalized azimuth spectra of all the obtained segmented time-domain array signals in chronological order to obtain a azimuth history diagram showing the change of the sound source target's azimuth over time throughout the entire period.

[0149] Figure 2 The results are from experimental data obtained using specific embodiments of the present invention. Figure 2 (a) to Figure 2 (e) The horizontal axis represents the azimuth, in degrees (°), the vertical axis represents the time, in seconds (s), and the color of the color bar represents the energy level of the normalized azimuth spectrum corresponding to the pigment point in the figure, in dB. Figure 2 (f) and Figure 2 (g) The horizontal axis represents the azimuth, with the unit being "degrees (°)" and the vertical axis represents the energy level of the normalized azimuth spectrum, with the unit being "dB".

[0150] During the actual measurement, there were four fixed sound sources near the horizontal direction of the uniform linear array (one weak sound source was located near the 90° azimuth, and the other three sound sources were located near the 70°, 80° and 95° azimuth respectively, which were stronger). There were three strong sound sources with slowly changing azimuth near the 130° azimuth. There were two strong sound sources that appeared briefly in directions far away from the horizontal direction (one appeared at about 13-15s, roughly at the 150° azimuth, and the other appeared at about 24-26s, roughly at the 160° azimuth).

[0151] Figure 2 (a) is the azimuth history map obtained by processing measured data using the CBF-SPED method. The azimuths of the six strong sound sources closer to the array's transverse direction are relatively clear, but the weak sound sources near the 90° azimuth are difficult to observe. Due to their greater distance from the array's transverse direction, the azimuths of the two briefly appearing strong sound sources are also relatively blurry. Furthermore, there is a lot of noise and sidelobes, which is not conducive to target detection. Our goal is to reduce noise and sidelobes through a broadband target detection method to make the target's azimuth clearer and improve the target resolution, so that the azimuth history map can more clearly show the azimuth of the sound source target. Figure 2 (b) is the azimuth history map obtained by processing measured data using the CBF-SPED-Eckart method. Figure 2 (c) A position history map obtained by processing measured data using the MVDR-SPED method. Figure 2(d) is the azimuth history diagram obtained by processing measured data using the MVDR-SPED-Eckart method. It can be seen that the broadband diagonally loaded MVDR method has a higher array gain and weaker noise and sidelobes compared to the broadband CBF method. However, after SPED processing, the target resolution is not significantly improved compared to the CBF-SPED method. Eckart filtering can significantly reduce noise and improve target resolution, but when the target is strong and the target azimuth is close, the accuracy of signal and noise estimation may decrease, which may easily lead to a decrease in the spectral intensity at the target's azimuth. While enhancing the target signal-to-noise ratio, Eckart filtering also strengthens the sidelobes, increasing the false alarm rate. Weak targets are strengthened to varying degrees, but the sidelobes and noise near their azimuth are relatively high. Figure 2 (e) is the azimuth history map obtained by processing the measured data by the method proposed in this invention. After peak energy-number filtering, most of the sidelobes and noise are reduced. After Eckart filtering and subsequent cumulative energy and peak number filtering, the peaks in the azimuth that meet the conditions are selected after increasing the signal-to-noise ratio of the target. Compared with the azimuth history maps obtained by the above four methods, the background is clearer, the target resolution is higher, and weak targets can be detected better. Figure 2 (f) A comparison was made of the normalized azimuth spectra obtained by processing the first segment of the measured data using five different methods. Figure 2 (g) is a magnified view of its spectral energy from 80° to 110°. Because the method proposed in this invention performs SPED processing and subsequently removes a large number of peaks, the spectral energy in most azimuths is 0. In order to... Figure 2 (f) and Figure 2 (g) To better display the normalized azimuth spectrum obtained by the method proposed in this invention, the spectral intensity of the azimuth with a spectral intensity of 0 is set to the minimum value of the color bar used in the azimuth history diagram shown above, i.e., -80dB. The comparison results of the normalized azimuth spectrum show that the method proposed in this invention can display the target's location more clearly, with higher target resolution, and can improve the detection effect of weak targets.

Claims

1. A method for detecting underwater targets using a linear array based on Eckart filtering and peak selection, characterized in that, This method consists of the following steps: Step 1) Read the time-domain array signal acquired by the uniform linear array, and then divide the time-domain array signal of each channel into segments with a duration T. slot , Duration T of overlapping adjacent segments over Segmentation is performed to obtain the segmented time-domain array signals for each channel; Step 2) Randomly select the time-domain array signals of each channel within a segment, divide them into multiple sub-bands through discrete Fourier transform, select the sub-bands within the target detection frequency range, and obtain the array element domain-frequency domain signal matrix; Step 3) Perform beamforming on the pairwise domain-frequency domain signal matrix to obtain the FRAZ matrix; Step 4) Take the absolute value of the obtained FRAZ matrix and normalize it by dividing it by the smallest element value in the entire matrix to obtain the normalized matrix. Subtract one more to get the intermediate matrix Then the intermediate matrix Subband peak energy screening is performed, and the non-peak energies are set to zero to obtain the subband peak energy screening matrix. Step 5) Screening matrix for sub-band peak energy In the beam domain, sub-band peak energy is filtered, selecting the top 20% of peaks in each sub-band, while setting the remaining peaks to zero, thus obtaining the beam domain peak matrix. Step 6) Calculate the peak matrix in the beam domain. By filtering the accumulated energy and peak count in the beam domain, the azimuths with the highest accumulated energy and peak count across all scanning azimuths are selected, while the peak counts for other azimuths are set to zero, thus obtaining the accumulated energy and peak count filtering matrix. Step 7) Normalize the matrix Perform subband noise estimation and subtract one from the noise estimation result to obtain the subband noise matrix. Step 8) Filter the matrix using accumulated energy and peak count. Sub-band noise matrix Obtain the weight matrix based on Eckart filtering. From the weight matrix Z nslot Screening matrix with cumulative energy and peak number The Hadamard product is used to weight the features to obtain the feature weight matrix. Step 9) Weight the feature matrix Accumulated energy screening is performed in the beam domain. Among the locations where accumulated energy exists, the locations with the top 20% of accumulated energy are selected to form a feature-weighted energy screening matrix. And the feature weighting matrix The peak values ​​corresponding to the azimuths selected through cumulative energy filtering and several adjacent azimuths are all set to zero, resulting in a feature-weighted zero-set matrix. Step 10) Set the feature weighted matrix to zero. Peak count screening is performed in the beam domain. Among the locations where energy peaks exist, the locations with the highest number of peaks in the top 20% are selected to form a feature-weighted peak count screening matrix. Step 11) Normalize the matrix Perform azimuth noise estimation and subtract one from the noise estimation result to obtain the azimuth noise matrix. Step 12) Convert the azimuth noise matrix Squaring each element and then combining it with the feature-weighted energy screening matrix and feature weighted number screening matrix The sum of the Hadamard product is weighted to obtain the FRAZ matrix. The sum of each column of the FRAZ matrix is ​​used to obtain the azimuth spectrum. Then, the spectral intensity of each azimuth in the azimuth spectrum is divided by the maximum spectral intensity in the azimuth spectrum to obtain the normalized azimuth spectrum of the segmented time-domain array signal. Step 13) Repeat steps 2 to 12. After processing all the segmented time-domain array signals, stitch together the normalized azimuth spectra of all the obtained segmented time-domain array signals in chronological order to obtain a azimuth history diagram showing the change of the azimuth of the sound source target over time.

2. A method for detecting underwater targets in a linear array based on Eckart filtering and peak selection according to claim 1, characterized in that: In step 1), the segmented signal duration T slot The value range is 1 to 3 seconds, and the overlap time T between adjacent segments is... over The value range is 0 to 0.9T. slot s.

3. A method for detecting underwater targets in a linear array based on Eckart filtering and peak selection according to claim 2, characterized in that: In step 1), the segmented signal duration T slot The duration of overlap between adjacent segments is 3 seconds (T). over It takes 2.7 seconds.

4. A method for detecting underwater targets in a linear array based on Eckart filtering and peak selection as described in claim 1, characterized in that: In step 3), the beamforming method is the diagonally loaded MVDR beamforming method.

5. A method for detecting underwater targets in a linear array based on Eckart filtering and peak selection according to claim 1, characterized in that: In step 7), the normalized matrix is... The method for subband noise estimation is the split double-window method.

6. A method for detecting underwater targets in a linear array based on Eckart filtering and peak selection according to claim 1, characterized in that: In step 11), the normalized matrix is... The method for estimating subband noise during azimuth noise estimation is the split double-window method.

Images