Target autonomous discrimination method driven by multi-category features

Abstract

The invention provides a multi-category feature-driven target autonomous discrimination method, which belongs to the technical field of ocean intelligent detection, and comprises the following steps of: by taking multi-category feature parameters with clear physical significance, such as a target motion feature, an energy feature, a depth feature and a frequency spectrum feature, as input, respectively endowing a precision factor to each dimension of feature, the method can effectively solve the problem of autonomous discrimination of a weak acoustic feature target, achieves the autonomous discrimination of the target under the condition that people are not in a loop, and can improve the discrimination accuracy and robustness of the type of the underwater acoustic target.

Description

Technical Field

[0001] The present invention belongs to the technical field of marine intelligent detection, and in particular relates to a multi-category feature-driven target autonomous discrimination method. Background Art

[0002] Underwater unmanned mobile detection equipment is a typical dual-use military and civilian technical equipment. With its advantages of low noise level, high concealment performance, strong environmental adaptability, low cost, and easy operation, it has obvious application value in marine environment monitoring and underwater target detection. For underwater unmanned platform target detection applications, the most important feature is autonomous target recognition. The realization of this technology and its performance directly affect the operation of the entire system, and it is also one of the key points and difficulties in the field of underwater unmanned detection. Traditional hydroacoustic target recognition requires people with rich knowledge and professional experience in hydroacoustic signal processing to extract target data features, while unmanned platforms need to realize autonomous discrimination of target types, which puts higher requirements on target recognition methods. Due to the autonomous recognition of target types by underwater unmanned detection equipment, large sample target data and highly robust target recognition algorithms are required. In addition, the unmanned platform has a complex marine application environment, and the detection target is easily interfered by fishing boats, merchant ships, engineering ships, etc., resulting in low recognition accuracy at present. Therefore, passive recognition of hydroacoustic signal target types has always been a difficult problem faced by countries around the world. Summary of the invention

[0003] The present invention proposes a multi-category feature-driven autonomous target identification method. By taking multi-category feature parameters with clear physical meanings such as target motion features, energy features, depth features, and spectrum features as input, a precision factor (recognition probability) is assigned to each dimensional feature, and combined with the DS evidence theory, a comprehensive recognition result of the underwater acoustic target type is obtained. The method can effectively solve the problem of autonomous identification of targets with weak acoustic features, realize autonomous identification of targets under the condition of "man is not in the loop", and improve the accuracy and robustness of underwater acoustic target type identification.

[0004] To achieve the above object, the technical solution of the present invention is:

[0005] A multi-category feature-driven target autonomous identification method comprises the following steps:

[0006] Step S1: performing attitude correction preprocessing on the signals collected by the vector acoustic system of the underwater unmanned platform;

[0007] Step S2: performing azimuth estimation on the preprocessed acoustic signal obtained in step S1, and extracting motion characteristic parameters of the target according to the azimuth history information;

[0008] Step S3: Calculate the coherence function of the acoustic signal and extract the energy spectrum characteristic parameters of the target according to the coherence function;

[0009] Step S4: performing target delay analysis on the acoustic signal, and extracting depth feature parameters of the target according to the delay information;

[0010] Step S5: performing spectrum analysis on the acoustic signal, and extracting the line spectrum characteristic parameters of the target according to the spectrum information;

[0011] Step S6: Assign an underwater target recognition probability to each feature parameter obtained in steps S2 to S5, and use it as the input of the DS classifier to obtain a comprehensive recognition result of the target type.

[0012] Preferably, in step S1: the underwater unmanned platform vector acoustic system is a single vector acoustic system, and its signal receiving end is a co-oscillating three-axis single vector hydrophone;

[0013] The method of attitude correction preprocessing is: let the Cartesian coordinate system of the underwater unmanned platform be Ox b y b z b , the geographic coordinate system used by the attitude sensor is Ox n y n z n , Ox in geographic coordinate system n Axis, Oy n Axes and Oz n The axes are along the north, east and sky directions respectively; when the coordinate system Ox b y b z b Relative to coordinate system Ox n y n z n When rotating, around Oz n Axis, Ox n Axis and Oy n The axis rotations respectively obtain the heading angle α, pitch angle β and roll angle γ of the platform;

[0014] If (x b ,y b ,z b ) is in the coordinate system Ox b y b z b The coordinate value under (x n ,y n ,z n ) is the coordinate system Ox in the ground n y n z n The coordinate value under the attitude transformation matrix is:

[0015]

[0016] In formula (1), “T” represents matrix transpose, is the attitude transformation matrix, and its expression is:

[0017]

[0018] Preferably, the step S2 comprises the following specific steps:

[0019] The sound pressure signal and particle velocity signal in the preprocessed acoustic signal are Fourier transformed and conjugate multiplied to obtain the complex sound intensity S pv (f):

[0020]

[0021] In formula (3), S p (f) and S v (f) represent the frequency domain results of FFT of sound pressure and particle velocity, respectively. The symbol * indicates conjugation.

[0022] Complex sound intensity is also expressed in the form of active sound intensity and reactive sound intensity:

[0023] S pv (f) = S a,pv (f)+i·S r,pv (f) (4)

[0024] In formula (4), the active sound intensity S a,pv (f) is the real part of the complex sound intensity, the reactive sound intensity S r,pv (f) is the imaginary part of the complex sound intensity; using the active sound intensity S a,pv (f) Orthogonal projection in the x and y directions and Calculate the target azimuth θ(f), the expression is:

[0025]

[0026] Formula (5) is used to achieve unambiguous direction finding of the target in the entire space, and there is no ambiguity problem of the left and right sides of the uniform sound pressure linear array. It can be seen from Formula (5) that the target azimuth calculated by it is inconsistent at different frequency points f, and the frequency-weighted histogram azimuth statistics are used for processing, and the expression is as follows:

[0027] k=[θ(f)*180 / π] (6)

[0028] N k =N k +1 (7)

[0029] In formula (6), [] represents the rounding operation, k∈[1,360]; formula (7) is the frequency weighted histogram expression, N k is the frequency of azimuth k;

[0030] For underwater unmanned platforms, a single-station pure azimuth target motion model is constructed. T Under certain conditions, the azimuth course θ t The change of is only a function of the velocity-to-initial ratio ε, that is:

[0031]

[0032] The target motion characteristic parameters are accurately calculated by combining the weighted azimuth history fitting method. The target motion characteristic parameters include the change of the target azimuth angle relative to the detection platform, and the target speed initial distance ratio, target motion heading angle, target crossing platform positive transverse time and crossing positive transverse azimuth are obtained.

[0033] Preferably, the step S3 comprises the following specific steps:

[0034] Using complex sound intensity S pv (f) Orthogonal projection in the x, y, and z directions The overall coherence function of the sound field is Expression:

[0035]

[0036] The coherent energy spectrum S in the sound field is obtained by using the total coherence function coh (f) and diffusion energy spectrum S dif (f) as follows:

[0037]

[0038] Coherent energy spectrum S coh (f) and diffusion energy spectrum S dif (f) is the characteristic parameter of the energy spectrum.

[0039] Preferably, step S4 includes the following specific steps:

[0040] The delay information presents interference fringes on the time-frequency image. The target delay characteristics are extracted by image processing, and then the target depth is estimated. The relationship between the delay of the target direct wave and the sea surface reflected wave and the horizontal distance of the target is:

[0041]

[0042] In formula (12), r1 and r2 are the acoustic paths of the direct signal and the sea surface reflected signal, r0 is the horizontal distance of the target relative to the receiver, and z r and z sare the receiving depth and the sound source depth respectively; therefore, the delay information is related to the target distance, target depth, and receiving depth. In formula (12), the target depth z s As an unknown quantity, and the other three pieces of information as known quantities, then this is a s Solve the quartic equation, which has two imaginary roots, a negative real root and a positive real root. Take the positive real root to get the expression of the target depth parameter:

[0043]

[0044] In formula (13), c and z r It is known that the distance r0 is obtained through pure bearing motion analysis. Therefore, the estimation problem of the time delay t0 is the estimation problem of the target depth parameter.

[0045] Preferably, the step S5 comprises the following specific steps:

[0046] Adaptive line spectrum enhancement (ALE) utilizes the fact that narrowband signals have a larger coherence radius than broadband noise, and uses the least mean square error (LMS) algorithm to perform adaptive cancellation processing to eliminate part of the broadband noise, improve the signal-to-noise ratio of the line spectrum signal, and enhance the line spectrum signal in the time domain; after adaptive line spectrum enhancement, the characteristic line spectrum parameters of the weak target signal are extracted.

[0047] Preferably, the method of step S6 is:

[0048] The target motion characteristic parameters, energy characteristic parameters, depth characteristic parameters, and spectrum characteristic parameters obtained in steps S2 to S5 are used as input, and the recognition probability of various characteristic targets is obtained using a support vector machine (SVM) classifier. Combined with the DS evidence theory, a comprehensive recognition result of the underwater acoustic target type is obtained.

[0049] Preferably, the step S6 comprises the following specific steps:

[0050] Let θ = {θ1,θ2,......,θ n} is the identification framework of the current research problem, which contains n complete and mutually exclusive independent propositions θ i According to the DS fusion theory, under the current recognition framework, Θ represents the set of all elements in Θ and the union of the elements, which corresponds to the discrimination of all target types; when identifying and classifying surface and underwater targets, Θ = {θ1, θ2}, where θ1 represents "surface target type" and θ2 represents "underwater target type"; and there are Among them, θ1∪θ2 indicates that it may be both on the water surface and underwater. According to DS theory, the mass function in the current recognition framework can be expressed as:

[0051] and

[0052] for The Dempster synthesis rule on the underwater acoustic target noise recognition framework Θ is expressed as:

[0053]

[0054] In formula (15):

[0055]

[0056] The beneficial effects of the multi-category feature-driven target autonomous discrimination method of the present invention are:

[0057] The present invention can make full use of the characteristic information related to the marine detection targets, and carry out combined reasoning on various characteristics, taking advantage of their strengths and making up for their weaknesses, so that the underwater unmanned detection equipment can obtain more comprehensive and accurate discrimination of the target types, and can significantly improve the accuracy, robustness and anti-false target interference performance of target discrimination. The underwater unmanned detection equipment can autonomously discriminate the types of targets detected by the underwater unmanned platform, and realize autonomous discrimination of targets under the condition of "no one is in the loop".

[0058] Instruction Manual

[0059] Figure 1 It is a technical flow chart of the present invention.

[0060] Figure 2 Schematic diagram for defining the attitude angle of an underwater unmanned platform.

[0061] Figure 3 Schematic diagram of the relative position of the target sound source and the receiver.

[0062] Figure 4 Block diagram of adaptive line spectrum enhancement. DETAILED DESCRIPTION

[0063] The following description is only a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

[0064] The following embodiments may be understood as individually expressing a part of a local structure or method of the present invention, or may be understood as a combination of the embodiments to explain the connotation of a larger structure or method of the present invention.

[0065] Example 1

[0066] A multi-category feature-driven target autonomous identification method, such as Figure 1 As shown, the following steps are included:

[0067] Step S1: performing attitude correction preprocessing on the signals collected by the vector acoustic system of the underwater unmanned platform;

[0068] Step S2: performing azimuth estimation on the preprocessed acoustic signal obtained in step S1, and extracting motion characteristic parameters of the target according to the azimuth history information;

[0069] Step S3: Calculate the coherence function of the acoustic signal and extract the energy spectrum characteristic parameters of the target according to the coherence function;

[0070] Step S4: performing target delay analysis on the acoustic signal, and extracting depth feature parameters of the target according to the delay information;

[0071] Step S5: performing spectrum analysis on the acoustic signal, and extracting the line spectrum characteristic parameters of the target according to the spectrum information;

[0072] Step S6: Assign an underwater target recognition probability to each feature parameter obtained in steps S2 to S5, and use it as the input of the DS classifier to obtain a comprehensive recognition result of the target type.

[0073] Example 2

[0074] In the step S1: the underwater unmanned platform vector acoustic system is a single vector acoustic system, and its signal receiving end is a co-oscillating three-axis single vector hydrophone; when the vector acoustic system is applied to the underwater unmanned platform, due to the influence of internal waves, tides, circulations, etc. in the complex ocean environment, the platform has attitude changes. The target orientation detected by the vector acoustic system on the platform is the orientation relative to the coordinate system of its own carrier. In order to obtain the accurate orientation information of the target, an attitude sensor is also required to measure the real-time attitude information of the vector acoustic system relative to the geographic coordinate system.

[0075] The method of attitude correction preprocessing is: let the Cartesian coordinate system of the underwater unmanned platform be Ox b y b z b , the geographic coordinate system used by the attitude sensor is Ox n y n z n , Ox in geographic coordinate system n Axis, Oy n Axes and Oz n The axes are along the north, east and sky directions respectively; when the coordinate system Ox b y b z b Relative to coordinate system Ox ny n z n When rotating, around Oz n Axis, Ox n Axis and Oy n The platform's heading angle α, pitch angle β and roll angle γ are obtained by rotating the axis respectively; Figure 2 shown.

[0076] If (x b ,y b ,z b ) is in the coordinate system Ox b y b z b The coordinate value under (x n ,y n ,z n ) is the coordinate system Ox in the ground n y n z n The coordinate value under the attitude transformation matrix is:

[0077]

[0078] In formula (1), “T” represents matrix transpose, is the attitude transformation matrix, and its expression is:

[0079]

[0080] Example 3

[0081] Preferably, step S2 includes the following specific steps:

[0082] The sound pressure signal and particle velocity signal in the preprocessed acoustic signal are Fourier transformed and conjugate multiplied to obtain the complex sound intensity (cross spectrum) S pv (f):

[0083]

[0084] In formula (3), S p (f) and S v (f) represent the frequency domain results of FFT of sound pressure and particle velocity, respectively. The symbol * indicates conjugation.

[0085] Complex sound intensity is also expressed in the form of active sound intensity and reactive sound intensity:

[0086] S pv (f) = S a,pv (f)+i·S r,pv (f) (20)

[0087] In formula (4), the active sound intensity S a,pv(f) is the real part of the complex sound intensity, the reactive sound intensity S r,pv (f) is the imaginary part of the complex sound intensity; using the active sound intensity S a,pv (f) Orthogonal projection in the x and y directions and Calculate the target azimuth θ(f), the expression is:

[0088]

[0089] Formula (5) is used to achieve unambiguous direction finding of the target in the entire space, and there is no ambiguity problem of the left and right sides of the uniform sound pressure linear array. It can be seen from Formula (5) that the target azimuth calculated by it is inconsistent at different frequency points f, and the frequency-weighted histogram azimuth statistics are used for processing, and the expression is as follows:

[0090] k=[θ(f)*180 / π] (22)

[0091] N k =N k +1 (23)

[0092] In formula (6), [] represents the rounding operation, k∈[1,360]; formula (7) is the frequency weighted histogram expression, N k is the frequency of azimuth k;

[0093] The target motion feature difference is used to classify the target type. It is a target type identification feature with good robustness. For the underwater unmanned detection platform, a single-station pure azimuth target motion model is constructed. The total detection time T, the initial azimuth angle θ0, the terminal azimuth angle θ T Under certain conditions, the azimuth course θ t The change of is only a function of the velocity-to-initial ratio ε, that is:

[0094]

[0095] The target motion characteristic parameters are accurately calculated by combining the weighted azimuth history fitting method. The target motion characteristic parameters include the change of the target azimuth angle relative to the detection platform, and the target speed initial distance ratio, target motion heading angle, target crossing platform positive transverse time and crossing positive transverse azimuth are obtained.

[0096] Example 4

[0097] The step S3 comprises the following specific steps:

[0098] Using complex sound intensity S pv (f) Orthogonal projection in the x, y, and z directions The overall coherence function of the sound field is Expression:

[0099]

[0100] The coherent energy spectrum S in the sound field is obtained by using the total coherence function coh (f) and diffusion energy spectrum S dif (f) as follows:

[0101]

[0102] Coherent energy spectrum S coh (f) and diffusion energy spectrum S dif (f) is the characteristic parameter of the energy spectrum.

[0103] Example 5

[0104] The step S4 comprises the following specific steps:

[0105] The target depth information is the most essential feature in the classification and discrimination of targets. The time delay information between the target direct wave and the sea surface reflected wave received by the acoustic system is sensitive to the target depth. The time delay information shows obvious interference fringes on the time-frequency image. The target time delay features are extracted by image processing, and then the target depth is estimated. Figure 3 , the relationship between the time delay of the target direct wave and the sea surface reflected wave and the horizontal distance of the target is:

[0106]

[0107] In formula (12), r1 and r2 are the acoustic paths of the direct signal and the sea surface reflected signal, r0 is the horizontal distance of the target relative to the receiver, and z r and z s are the receiving depth and the sound source depth respectively; therefore, the delay information is related to the target distance, target depth, and receiving depth. In formula (12), the target depth z s As an unknown quantity, and the other three pieces of information as known quantities, then this is a s Solve the quartic equation, which has two imaginary roots, a negative real root and a positive real root. Take the positive real root to get the expression of the target depth parameter:

[0108]

[0109] In formula (13), c and z r The distance r0 is known, and is obtained through pure bearing motion analysis. The problem of estimating the time delay t0 is the problem of estimating the target depth parameter.

[0110] Example 6

[0111] The step S5 comprises the following specific steps:

[0112] The low-frequency line spectrum in the underwater target radiation noise has the characteristics of high intensity, relatively stable amplitude frequency, and small propagation loss, making it the main feature of passive sonar detection and identification of targets. Adaptive line spectrum enhancement (ALE) uses the characteristic that narrowband signals have a larger coherence radius than broadband noise, and uses the minimum mean square error algorithm (LMS) for adaptive cancellation processing to eliminate part of the broadband noise, improve the signal-to-noise ratio of the line spectrum signal, and enhance the line spectrum signal in the time domain;

[0113] The structural block diagram of ALE is as follows Figure 4 As shown, the input signal is input into the tapped delay line with a length of L after a decorrelation delay of Δ, and is multiplied by the weighting coefficient to obtain the output signal. The output signal is subtracted from the reference signal to obtain the error signal, and the weighting coefficient is updated using the LMS algorithm.

[0114] In essence, ALE is an adaptive FIR filter. When the filter reaches a steady state through continuous iteration, its frequency response consists of one or more peaks, which correspond to the frequency of the unknown line spectrum signal. When a signal containing additive noise passes through the filter, the narrowband line spectrum signal passes through and the additive noise is suppressed, thereby improving the output signal-to-noise ratio.

[0115] After adaptive line spectrum enhancement, the target line spectrum can be changed more obviously, so that the characteristic line spectrum parameters of weak target signals can be effectively extracted.

[0116] Example 7

[0117] The method of step S6 is:

[0118] The target motion characteristic parameters, energy characteristic parameters, depth characteristic parameters, and spectrum characteristic parameters obtained in steps S2 to S5 are used as input, and the recognition probability of various characteristic targets is obtained using a support vector machine (SVM) classifier. Combined with the DS evidence theory, a comprehensive recognition result of the underwater acoustic target type is obtained.

[0119] Example 8

[0120] The step S6 comprises the following specific steps:

[0121] Let θ = {θ1,θ2,......,θ n} is the identification framework of the current research problem, which contains n complete and mutually exclusive independent propositions θ i According to the DS fusion theory, under the current recognition framework, Θ represents the set of all elements in Θ and the union of the elements, which corresponds to the discrimination of all target types; when identifying and classifying surface and underwater targets, Θ = {θ1, θ2}, where θ1 represents "surface target type" and θ2 represents "underwater target type"; and there are Among them, θ1∪θ2 indicates that it may be both on the water surface and underwater. According to DS theory, the mass function in the current recognition framework can be expressed as:

[0122] and

[0123] for The Dempster synthesis rule on the underwater acoustic target noise recognition framework Θ is expressed as:

[0124]

[0125] In formula (15):

[0126]

Claims

1. A multi-category feature-driven target autonomous discrimination method, characterized by: The steps include: Step S1: performing attitude correction preprocessing on the signals collected by the underwater unmanned platform vector acoustic system; Step S2: performing position estimation on the pre-processed acoustic signal obtained in step S1, and extracting motion characteristic parameters of the target based on the position history information; Step S3: Calculate the coherence function of the acoustic signal and extract the energy spectrum characteristic parameters of the target according to the coherence function; Step S4: performing target delay analysis on the acoustic signal and extracting the depth feature parameters of the target based on the delay information; Step S5: performing spectrum analysis on the acoustic signal and extracting the line spectrum characteristic parameters of the target based on the spectrum information; Step S6: Assign an underwater target recognition probability to each feature parameter obtained from step S2 to step S5, and use it as the input of the DS classifier to obtain a comprehensive recognition result of the target type.

2. The multi-category feature-driven autonomous target identification method according to claim 1, characterized in that: In step S1, the underwater unmanned platform vector acoustic system is a single vector acoustic system, and its signal receiving end is a co-vibration three-axis single vector hydrophone; The method of attitude correction preprocessing is: let the Cartesian coordinate system of the underwater unmanned platform be Ox b y b z b , the geographic coordinate system used by the attitude sensor is Ox n y n z n , Ox in geographic coordinate system n Axis, Oy n Axes and Oz n The axes are along the north, east and sky directions respectively; when the coordinate system Ox b y b z b Relative to coordinate system Ox n y n z n When rotating, around Oz n Axis, Ox n Axis and Oy n The axis rotations respectively obtain the heading angle α, pitch angle β and roll angle γ of the platform; If (x b ,y b ,z b ) is in the coordinate system Ox b y b z b The coordinate value under (x n ,y n ,z n ) is the ground coordinate system Ox n y n z n The coordinate value under the attitude transformation matrix is: In formula (1), "T" represents matrix transpose, is the posture transformation matrix, which is expressed as:

3. The multi-category feature-driven autonomous target identification method according to claim 2, characterized in that: The step S2 includes the following specific steps: The sound pressure signal and particle velocity signal in the preprocessed acoustic signal are Fourier transformed and then conjugate multiplied to obtain the complex sound intensity S pv (f): In formula (3), S p (f) and S v (f) represent the frequency domain results of FFT of sound pressure and particle velocity, respectively. The symbol * indicates conjugation. Complex sound intensity can also be expressed in the form of active sound intensity and reactive sound intensity: S pv (f)=S a,pv (f)+i·S r,pv (f) (4) In formula (4), the active sound intensity S a,pv (f) is the real part of the complex sound intensity, the reactive sound intensity S r,pv (f) is the imaginary part of the complex sound intensity; using the active sound intensity S a,pv (f) Orthogonal projection in the x and y directions and Calculate the target azimuth θ(f), the expression is: Equation (5) is used to achieve unambiguous direction finding of the target in the entire space, and there is no ambiguity problem of the left and right sides of the uniform sound pressure linear array. As can be seen from Equation (5), the target direction calculated by it is inconsistent at different frequency points f. The frequency-weighted histogram direction statistics are used for processing, and the expression is as follows: k=[θ(f)*180 / π] (6) N k =N k +1 (7) In formula (6), [] represents the rounding operation, k∈[1,360]; formula (7) is the frequency weighted histogram expression, N k is the frequency of azimuth k; For underwater unmanned platforms, a single-station pure bearing target motion model is constructed. Under the conditions of the total detection time T, the initial azimuth angle θ0, and the final azimuth angle θT, the azimuth course θ t The change of is only a function of the velocity-to-initial distance ratio ε, that is: The target motion characteristic parameters are accurately calculated by combining the weighted azimuth history fitting method. The target motion characteristic parameters include the azimuth angle change of the target relative to the detection platform, and the target velocity initial distance ratio, target motion heading angle, target crossing platform positive time and passing positive azimuth are obtained.

4. The multi-category feature-driven autonomous target identification method according to claim 3, characterized in that: The step S3 includes the following specific steps: Using complex sound intensity S pv (f) Orthogonal projection in the x, y, and z directions The overall coherence function of the sound field Expression: The coherent energy spectrum S in the sound field is obtained using the total coherence function coh (f) and diffusion energy spectrum S dif (f), as follows: Coherent energy spectrum S coh (f) and diffusion energy spectrum S dif (f) is the energy spectrum characteristic parameter.

5. The multi-category feature-driven autonomous target identification method according to claim 4, characterized in that: Step S4 includes the following specific steps: The time delay information appears as interference fringes on the time-frequency image. Image processing is used to extract the target delay characteristics and then estimate the target depth. The relationship between the time delay of the target direct wave and the sea surface reflected wave and the horizontal distance to the target is: In formula (12), r1 and r2 are the acoustic ranges of the direct signal and the sea surface reflected signal, r0 is the horizontal distance of the target relative to the receiver, and z r and z s are the receiving depth and the sound source depth respectively; therefore, the delay information is related to the target distance, target depth, and receiving depth. In formula (12), the target depth z s As an unknown quantity, and the other three pieces of information as known quantities, then this is a s Solve the quartic equation, which has two imaginary roots, a negative real root and a positive real root. Take the positive real root to get the expression of the target depth parameter: In formula (13), c and z r is known, and the distance r0 is obtained through pure bearing motion analysis. Therefore, the problem of estimating the time delay t0 is the problem of estimating the target depth parameter.

6. The multi-category feature-driven autonomous target identification method according to claim 5, characterized in that: The step S5 includes the following specific steps: Adaptive Line Spectrum Enhancement (ALE) utilizes the fact that narrowband signals have a larger coherence radius than broadband noise, and uses the Least Mean Square Error (LMS) algorithm for adaptive cancellation processing to eliminate some broadband noise, improve the signal-to-noise ratio of the line spectrum signal, and enhance the line spectrum signal in the time domain. After adaptive line spectrum enhancement, the characteristic line spectrum parameters of weak target signals are extracted.

7. The multi-category feature-driven autonomous target identification method according to claim 6, characterized in that: The method of step S6 is: The target motion characteristic parameters, energy characteristic parameters, depth characteristic parameters, and spectrum characteristic parameters obtained in steps S2 to S5 are used as input, and the support vector machine (SVM) classifier is used to obtain the recognition probability of various characteristic targets, and the DS evidence theory is combined to obtain the comprehensive recognition result of the underwater acoustic target type.

8. The multi-category feature-driven autonomous target identification method according to claim 7, characterized in that: The step S6 includes the following specific steps: Let Θ={θ1,θ2,......,θ n } is the identification framework of the current research problem, which contains n complete and mutually exclusive independent propositions θ i According to DS fusion theory, under the current recognition framework, 2 Θ Represents the set of all elements in Θ and the union of the elements, which corresponds to the discrimination of all target types; when identifying and classifying surface and underwater targets, Θ = {θ1,θ2}, where θ1 represents "surface target type" and θ2 represents "underwater target type"; and there is Among them, θ1∪θ2 represents the situation that it may be both on the water surface and underwater. According to DS theory, the mass function in the current recognition framework can be expressed as: and for The Dempster synthesis rule on the underwater acoustic target noise recognition framework Θ is expressed as: In formula (15):

Images