A method and device for generating a data set of underwater acoustic passive tracking targets in real time

Abstract

The present invention relates to a method and device for generating a data set of a target for passive underwater acoustic tracking in real time. The method receives an underwater acoustic signal through a passive sonar array, and after confirming the target to be tracked, continuously generates a data set of the target to be tracked in real time. The specific steps include: beam forming; dividing the frequency domain sub-band interval; joint detection to confirm the target; obtaining a single beam sub-band LOFAR; converting and accumulating frequency domain data; generating a target LOFAR spectrogram; storing the feature matrix and label information in a multi-dimensional array; and repeating the above steps. The present invention overcomes the defects of insufficient uniformity of generated target samples in the process of passive underwater acoustic target tracking, which are easily contaminated by other target signals in the frequency domain; and solves the problem of complex sample information storage in underwater acoustic data sets. The present invention can be used in a passive sonar tracking system based on deep learning to generate a data set of a tracked target in real time for subsequent detection and tracking.

Description

Technical Field

[0001] The invention relates to the field of passive tracking of underwater acoustic targets, and in particular to a method for generating a data set of underwater acoustic passive tracking targets in real time. Background Art

[0002] Passive tracking of underwater acoustic targets is an important research direction in underwater acoustic engineering and sonar signal processing. Passive sonar itself does not emit any signal, but only relies on receiving acoustic signals to analyze the marine environment. In recent years, deep learning has been widely used in speech recognition, sound classification and other fields. Using the dataset of underwater acoustic targets to train a neural network model to extract the characteristic information of the target provides a new idea for tracking underwater acoustic targets.

[0003] The quality of the target dataset is related to the accuracy of the neural network model recognition. However, the problem with passive underwater acoustic target tracking is that the types and numbers of targets are random, and there may be multiple targets in the received signal, which is not only a huge challenge for tracking, but also brings great difficulties to the generation of datasets for different targets.

[0004] After analysis, the existing sound dataset generation method is directly applied to the real-time generation of underwater acoustic passive tracking target dataset, but it still has the following defects:

[0005] (1) The number of signal targets from the passive sonar array is uncertain, the start and end times are uncertain, and the target signals are mixed in the time domain, frequency domain, and azimuth. The existing labeling methods cannot accurately label the targets.

[0006] (2) When different targets are close to each other, their frequency domain data are in the same azimuth range. Simple spatial filtering cannot distinguish different targets. The target data samples generated are not uniform enough and may contain a large amount of interference from other targets, as well as a large amount of useless interference from other frequency bands.

[0007] (3) Most samples in existing sound datasets are stored separately from audio and labels, which has a complex structure and increases the difficulty of storage and reading. Summary of the invention

[0008] To this end, the purpose of the present invention is to overcome the defects of the above-mentioned existing sound data set generation methods and provide a method and device for generating underwater acoustic passive tracking target data sets in real time.

[0009] A method for generating a data set of underwater acoustic passive tracking targets in real time is implemented by the following steps:

[0010] S1, the passive sonar array receives the signal, and after filtering, the signal is processed using beamforming to generate a multi-beam LOFAR spectrum and azimuth energy distribution map.

[0011] S2, divide the frequency sub-bands and confirm the target to be tracked. Compare the peak (maximum) in the azimuth energy distribution diagram and the spectrum energy distribution in the multi-beam LOFAR spectrum diagram to confirm the target to be tracked and its azimuth, and confirm the frequency sub-band interval in which the target's frequency band range is located.

[0012] S3, for the target to be tracked confirmed in S2, select the single-beam LOFAR data of the corresponding azimuth according to its azimuth and frequency sub-band interval, and intercept the corresponding single-beam sub-band LOFAR data B′(θ s ).

[0013] S4, the single-beam sub-band LOFAR data of the target obtained in S3 is subjected to inverse Fourier transform to obtain time domain data, the time domain data is accumulated three times, and then short-time Fourier transform is used to generate a two-dimensional time-frequency matrix containing target feature information and orientation information, namely, the target LOFAR spectrogram.

[0014] S5, the time frequency two-dimensional matrix of the target obtained in S4 is added with the target label information to generate a multidimensional array and stored in the data set.

[0015] S6, repeats S1-S5, and generates data sets of all tracked targets during the tracking process.

[0016] Preferably, the beam forming method is:

[0017] A passive sonar array consisting of M hydrophones is used to receive water acoustic signals;

[0018] S11, divide the azimuth range of 0°-180° into beams using the sinusoidal equal value division method;

[0019] S12 uses frequency domain broadband beamforming to transform the accumulated 1s time domain signal x from multiple hydrophone channels into i (t) Convert each channel into frequency domain signal x i (f);

[0020] S13, taking the first array element as the reference array element, according to △T(i,θ s )=D i sinθ s / υ c , traverse all array elements and beam pointing angles, and calculate the time difference △T(i,θ s ), where D i is the distance from the rest of the array elements to the reference array element, υ c is the speed of sound, θ s is the beam pointing angle;

[0021] S14, based on the delay calculation of S13, expands to the frequency domain and calculates the frequency domain broadband beamforming weight vector matrix w(i,θ s ,f)=exp(j2πf△T(i,θ s ), and the frequency domain signal x i (f) are multiplied to generate the LOFAR data B(θ s ), as shown in formula (1):

[0022] B(θ s )=w(i,θ s ,f)x i (f) / M (1)

[0023] Then the multi-beam LOFAR diagram is Y = [y(θ1), y(θ2), …, y(θ s )], and obtain the frequency distribution change corresponding to the direction;

[0024] S15, based on the beam LOFAR data B(θ obtained in S14 s ), calculate the beam pointing θ s The azimuthal energy y(θ s ), as shown in formula (2):

[0025] y(θ s )=20lg∑B(θ s ) (2)

[0026] Then the azimuth energy distribution diagram during this period of time is Y = [y(θ1), y(θ2), …, y(θ s )], and obtain the energy distribution changes in different directions within the entire frequency band.

[0027] As a preferred embodiment, S21, screening all the peak (maximum) positions θ in the azimuth energy graph that exceed the threshold s , find the corresponding single-beam LOFAR spectrum in the multi-beam LOFAR spectrum, use the multi-hypothesis association principle to confirm its frequency distribution interval, and divide it into the corresponding frequency sub-band interval. Determine a potential target based on the three characteristics of energy threshold, azimuth and frequency distribution;

[0028] S22, after the potential target is determined, if a target with similar characteristics appears more than three times in a row, the potential target is confirmed as a tracked target.

[0029] Preferably, the time-frequency two-dimensional matrix is ​​generated as follows:

[0030] S41, the single-beam sub-band LOFAR data of the tracked target is transformed by inverse Fourier transform to obtain the time domain data of the target.

[0031] S42, the time domain data of the target is accumulated three times to obtain the time domain data of the target within a longer time range.

[0032] S43, the accumulated target time domain data is transformed by short-time Fourier transform to generate a two-dimensional time-frequency matrix containing target feature information and orientation information.

[0033] Preferably, the frequency sub-band interval ranges from 0-100 Hz, 100-200 Hz, 200-400 Hz, 400-800 Hz, 800-1600 Hz, and 1600-3200 Hz.

[0034] A device for generating a water acoustic passive tracking target data set in real time, comprising:

[0035] Processor and memory;

[0036] The memory is used to store programs;

[0037] The processor is used to execute the program in the memory, so that the device for generating a target data set for passive underwater acoustic tracking in real time executes the above-mentioned method for generating a target data set for passive underwater acoustic tracking in real time.

[0038] Beneficial effects:

[0039] The method for generating a data set for underwater acoustic passive tracking targets in real time provided by the present invention is a data set generation method designed for underwater acoustic passive tracking based on deep learning. When the method is used, the method can generate a data set of tracked targets in real time when the number and type of targets are unknown, and can be used for training various neural network models in deep learning.

[0040] In addition, compared with the existing methods for generating underwater acoustic target data sets, the data set generation method provided by the present invention ensures the uniqueness of the target samples as much as possible through beamforming and frequency sub-band division; the target samples are present in the form of a two-dimensional matrix of time and frequency, and finally form a high-dimensional array with label information and stored in the data set.

[0041] The present invention makes up for the problem that the existing underwater acoustic target data set generation method is not applicable in real-time tracking, increases the amount of target information contained in the sample, and simplifies the storage method of the data set.

[0042] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present disclosure and, together with the description, serve to explain the principles of the present disclosure.

[0044] Figure 1 This is the data set generation flow chart

[0045] Figure 2 This is a schematic diagram of the multi-beam LOFAR spectrum.

[0046] Figure 3 A schematic diagram of azimuthal energy distribution

[0047] Figure 4 This is a schematic diagram of the target LOFAR spectrum.

[0048] Figure 5 This is a diagram of the sample storage array in the data set. DETAILED DESCRIPTION

[0049] The method provided by the present invention relates to a method and device for generating a water acoustic passive tracking target data set in real time. Various exemplary embodiments of the present invention will now be described in detail with reference to the accompanying drawings. It should be noted that unless otherwise specifically stated, the relative arrangement of the components and steps, numerical expressions and numerical values ​​described in these embodiments do not limit the scope of the present invention.

[0050] The following description of at least one exemplary embodiment is merely illustrative in nature and is in no way intended to limit the invention, its application, or uses.

[0051] Technologies, methods, and equipment known to ordinary technicians in the relevant art may not be discussed in detail, but where appropriate, the technologies, methods, and equipment should be considered as part of the specification.

[0052] In all examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not limiting. Therefore, other examples of the exemplary embodiments may have different values.

[0053] It should be noted that like reference numerals and letters refer to similar items in the following figures, and therefore, once an item is defined in one figure, it need not be further discussed in subsequent figures.

[0054] Example 1

[0055] Refer to the following Figures 1 to 5 To describe the data set generation method provided according to an embodiment of the present invention.

[0056] The present invention provides a method for generating a data set of underwater acoustic passive tracking targets in real time, and the processing flow is as follows: Figure 1 shown.

[0057] A method for generating underwater acoustic passive tracking target dataset in real time, such as Figure 1 As shown, the specific steps are as follows:

[0058] S1, the passive sonar array receives the signal, and after filtering, the signal is processed using beamforming. A multi-beam LOFAR spectrogram is generated, such as Figure 2 Generate azimuth energy distribution diagram, as shown Figure 3 shown.

[0059] Wherein, the beam forming method is:

[0060] A passive sonar array consisting of M hydrophones is used to receive water acoustic signals. The sonar array is in the form of a uniform linear array with an array element spacing of d.

[0061] S11, divide the azimuth range of 0°-180° into 128 beams. Use the sine equal value division method to divide the mean value from -1 to 1 into 128 parts, and then calculate the arc sine value to obtain the beam pointing angle sequence [θ1,θ2,…,θ s ];

[0062] S12 uses frequency domain broadband beamforming to transform the accumulated 1s time domain signal x from multiple hydrophone channels into i (t) Convert each channel into frequency domain signal x i (f);

[0063] S13, taking the first array element as the reference array element, according to △T(i,θ s )=D i sinθ s / υ c , traverse all array elements and beam pointing angles, and calculate the time difference △T(i,θ s ), where D i is the distance from the rest of the array elements to the reference array element, υ c is the speed of sound, θ s is the beam pointing angle.

[0064] S14, based on the delay calculation of S13, expands to the frequency domain and calculates the frequency domain broadband beamforming weight vector matrix w(i,θ s ,f)=exp(j2πf△T(i,θ s ), and the frequency domain signal x i (f) are multiplied to generate the LOFAR data B(θ s ), as shown in formula (1):

[0065] B(θ s)=w(i,θ s ,f)x i (f) / M (1)

[0066] Then the multi-beam LOFAR diagram is Y = [y(θ1), y(θ2), …, y(θ s )], and obtain the frequency distribution change corresponding to the direction;

[0067] S15, based on the beam LOFAR data B(θ obtained in S14 s ), calculate the beam pointing θ s The azimuthal energy y(θ s ), as shown in formula (2):

[0068] y(θ s )=20lg∑B(θ s ) (2)

[0069] Then the azimuth energy distribution diagram during this period of time is Y = [y(θ1), y(θ2), …, y(θ s )], and obtain the energy distribution changes in different directions within the entire frequency band;

[0070] S2, joint confirmation of the target to be tracked. By comparing the peak (maximum) in the azimuth energy distribution diagram and the spectrum energy distribution in the multi-beam LOFAR spectrum diagram, the target to be tracked and its azimuth angle are confirmed, and the frequency sub-band interval of the target's frequency band range is confirmed.

[0071] Preferably, the frequency sub-band interval range is:

[0072] Divided according to the frequency range of 0-100Hz, 100-200Hz, 200-400Hz, 400-800Hz, 800-1600Hz, 1600-3200Hz.

[0073] S21, screening all the peak (maximum) positions θ in the azimuth energy graph that exceed the threshold s , find the corresponding single-beam LOFAR spectrum in the multi-beam LOFAR spectrum, confirm its frequency distribution interval, and divide it into the corresponding frequency sub-band interval. Determine a potential target based on the three characteristics of energy threshold, azimuth and frequency distribution;

[0074] S22, after the potential target is determined, if a target with similar characteristics appears more than three times in a row, the potential target is confirmed as the tracked target;

[0075] S3, for the target to be tracked confirmed in S2, select the single-beam LOFAR data of the corresponding azimuth according to its azimuth and frequency sub-band interval, and intercept the corresponding single-beam sub-band LOFAR data B′(θ s).

[0076] S4, such as Figure 4 As shown in S3, the single-beam sub-band LOFAR data B′(θ s ), use inverse Fourier transform to obtain time domain data, accumulate three time domain data, and then use short-time Fourier transform to generate a two-dimensional time-frequency matrix containing target feature information and orientation information, namely the target LOFAR spectrogram.

[0077] The time-frequency two-dimensional matrix is ​​generated as follows:

[0078] S41, single beam sub-band LOFAR data B′(θ s ), using inverse Fourier transform, we can obtain the target time domain data X′(θ s ). As shown in formula (3):

[0079] X′(θ s )=IFFT(B′(θ s )) (3)

[0080] S42, target time domain data X′(θ s ) is accumulated three times to obtain the target time domain data X(θ s ).

[0081] S43, the accumulated target time domain data X(θ s ) uses short-time Fourier transform to generate a two-dimensional time-frequency matrix Z(θ s ).

[0082] Z(θ s ) = STFT(X(θ s )) (4)

[0083] S5, such as Figure 5 As shown, the time-frequency two-dimensional matrix of the target obtained in S4 is added with the target label information to generate a multidimensional array and stored in the data set.

[0084] S6, repeats S1-S5, and generates data sets of all tracked targets during the tracking process.

[0085] In another embodiment, a device for generating a water acoustic passive tracking target data set in real time is also proposed, comprising:

[0086] Processor and memory;

[0087] The memory is used to store programs;

[0088] The processor is used to execute the program in the memory, so that the device for generating a target data set for passive underwater acoustic tracking in real time executes the above-mentioned method for generating a target data set for passive underwater acoustic tracking in real time.

[0089] Although the present invention has been specifically shown and described in conjunction with the preferred embodiments, it should be understood by those skilled in the art that various changes may be made to the present invention in form and details without departing from the spirit and scope of the present invention as defined by the appended claims, all of which are within the scope of protection of the present invention.

Claims

1. A method for generating a data set of underwater acoustic passive tracking targets in real time, characterized in that: The specific steps include: S1, the passive sonar array receives the signal, and after filtering, the signal is processed using beamforming to generate a multi-beam LOFAR spectrogram and azimuth energy distribution map; S11, divide the azimuth range of 0°-180° into 128 beams; use the sine equal value division method to divide the mean value from -1 to 1 into 128 parts, and then calculate the arc sine value to obtain the beam pointing angle sequence [θ1,θ2,…,θ s ]; S12 uses frequency domain broadband beamforming to transform the accumulated 1s time domain signal x from multiple hydrophone channels into i (t) Convert each channel into frequency domain signal x i (f); S13, taking the first array element as the reference array element, according to △T(i,θ s )=D i sinθ s / υ c , traverse all array elements and beam pointing angles, and calculate the time difference △T(i,θs) of all array elements relative to the reference array element under all beam pointing angles; where D i is the distance from the rest of the array elements to the reference array element, υ c is the speed of sound, θ s is the beam pointing angle; S14, based on the delay calculation of S13, expands to the frequency domain and calculates the frequency domain broadband beamforming weighting vector matrix w(i,θ s ,f)=exp(j2πf△T(i,θ s ), and the frequency domain signal x i (f) are multiplied to generate the LOFAR data B(θ s ), as shown in formula (1): B(θ s )=w(i,θ s ,f)x i (f) / M (1) Then the multi-beam LOFAR diagram is Y = [y(θ1), y(θ2), …, y(θ s )], and obtain the frequency distribution change corresponding to the azimuth; wherein M represents the number of hydrophones in the passive sonar array composed of hydrophones; S15, calculate the beam pointing θ according to the beam LOFAR data B(θs) obtained in S14 s The azimuthal energy y(θ s ), as shown in formula (2): y(θ s )=20lg∑B(θ s ) (2) Then the azimuth energy distribution diagram during this period of time is Y = [y(θ1), y(θ2), …, y(θ s )], and obtain the energy distribution changes in different directions within the entire frequency band; S2, divide the frequency sub-bands and confirm the target to be tracked; compare the peaks in the azimuth energy distribution diagram and the spectrum energy distribution in the multi-beam LOFAR spectrum diagram to confirm the target to be tracked and its azimuth, and confirm the frequency sub-band interval in which the frequency band range of the target is located; S3, for the target to be tracked confirmed in S2, according to its azimuth and frequency sub-band interval, selecting the single-beam LOFAR data of the corresponding azimuth, and intercepting the corresponding single-beam sub-band LOFAR data; S3 obtains the single-beam LOFAR data of the frequency sub-band interval where the target is located, uses inverse Fourier transform to convert it into time domain data, accumulates the time domain data three times, and then uses short-time Fourier transform to generate the target LOFAR spectrogram in the form of a two-dimensional time-frequency matrix containing target feature information and azimuth information, which is located in a certain frequency sub-band; S4, the single-beam sub-band LOFAR data of the target obtained in S3 is subjected to inverse Fourier transform to obtain time domain data, the time domain data is accumulated three times, and short-time Fourier transform is then used to generate a time-frequency two-dimensional matrix containing target feature information and orientation information, i.e., the target LOFAR spectrogram; S5, the time frequency two-dimensional matrix of the target obtained in S4 is added with the target label information to generate a multidimensional array and stored in the data set; S6, repeat S1-S5, and generate a data set of all tracked targets during the tracking process; In the process of real-time data set generation, the azimuth frequency distribution and azimuth energy distribution are combined to confirm the tracked target; the target is assigned to different frequency sub-bands to distinguish different targets in the same azimuth; the target single-beam sub-band LOFAR data is converted into time domain data and accumulated, and then the accumulated data is short-time Fourier transformed to generate the target LOFAR spectrum; the target LOFAR spectrum is combined with the target label information in the form of a two-dimensional matrix to form a multi-dimensional array and stored in the data set.

2. The method for generating a real-time underwater acoustic passive tracking target data set according to claim 1, characterized in that: In S2, the single-beam LOFAR data of the target is azimuthally weighted; in S3, only the single-beam LOFAR data B′(θ s ); The frequency sub-band interval ranges are: 0-100Hz, 100-200Hz, 200-400Hz, 400-800Hz, 800-1600Hz, 1600-3200Hz.

3. A method for generating a data set of underwater acoustic passive tracking targets in real time according to claim 1 or 2, characterized in that: In S5, samples of the tracked target generated in real time are stored in the data set in the form of a multidimensional array, and the multidimensional array includes the time-frequency two-dimensional matrix containing the target feature information and orientation information and the target label information described in S3.

4. A device for real-time generation of underwater acoustic passive tracking target data set is applicable to the method for real-time generation of underwater acoustic passive tracking target data set according to claim 1, characterized in that: include: Processor and memory; The memory is used to store programs; The processor is configured to execute the program in the memory.

Images