Underwater target positioning method and device

Abstract

The present invention discloses a method and device for locating underwater targets, including: receiving measurement data collected by multiple sensors arranged on a detector at multiple snapshot moments, constructing a measurement signal matrix; using a multi-target current signal extraction algorithm based on multi-measurement vector correlation to iteratively estimate the current signal amplitude matrix of all received targets at corresponding frequencies; sequentially inputting the estimated amplitude matrices of multiple targets into a spatial perception model based on multi-size one-dimensional convolution kernels for nonlinear mapping and feature representation, obtaining the distance value from the geometric center of the detector to the geometric center of each target, and the angle value between the line connecting the two geometric centers and the bow direction of the detector, so as to realize the positioning of underwater targets. The present invention can effectively work under various water quality conditions through an optimized electrode arrangement scheme and advanced signal processing algorithms, thereby improving the robustness and accuracy of underwater detection.

Description

Technical Field

[0001] The present invention relates to the field of underwater detection, and in particular to an underwater target positioning method and device. Background Art

[0002] At present, underwater target detection methods can be roughly divided into two categories: optical detection and acoustic detection. Optical detection methods rely on the propagation of light and are suitable for underwater environments with clear water quality. This method can provide high-resolution visual images. Therefore, this also leads to its very demanding requirements on water quality. Turbid water or insufficient and uneven lighting environments will greatly reduce its performance. In contrast, acoustic-based detection technology identifies underwater targets by emitting sound waves and receiving their echoes. It has the advantages of long detection distance and high long-distance accuracy, and is not affected by the transparency of water. However, in limited waters or close-range detection, the acoustic method has a low resolution and is easily interfered by the multipath effect of complex underwater terrain, resulting in insufficient accuracy.

[0003] Active electric field detection technology originates from the study of the electro-sensing mechanism of weak electric fish. These fish can generate low-frequency, low-amplitude bioelectric fields in their bodies. When there are metal objects nearby, although these objects do not emit electrical signals themselves, they will generate secondary equivalent excitation electric fields under the stimulation of active electric fields, causing changes in the potential and electric field distribution of the surrounding waters. The electrosensory organs on the surface of their skin can sense these distortions and thus identify the surrounding environment. These changes in bioelectric fields can not only be captured by receiving electrodes, but also recorded and analyzed by data acquisition equipment, so as to accurately obtain key information such as the position, volume, and shape of underwater targets. Therefore, this perception mechanism of weak electric fish provides a solid bionic foundation for the development of underwater target positioning and recognition technology based on active electric field changes. Compared with traditional passive electric field detection methods, active electric field detection technology does not rely on the electric field in the natural environment, but obtains information by emitting electric fields and analyzing their changes. Unlike optical and acoustic methods that rely on water clarity and distance, active electric field detection has shown results in various water quality conditions, especially turbid and insufficiently illuminated water environments, as well as in close-range scenes. These characteristics make up for the limitations of traditional acoustic and optical methods.

[0004] However, previous studies have often focused on system design, ignoring the spatial geometric information provided by the sensor electrode array. At the same time, it is crucial to detect targets of different materials and shapes under different water quality conditions. This is often the key to whether active electric field technology can be widely used in actual sea areas. Summary of the invention

[0005] In view of the shortcomings of the prior art, the present invention proposes an underwater target positioning method and device to solve the problem that traditional acoustic and optical detection methods cannot locate at close range and in turbid water environments, and the problem that active electric field detection methods cannot fully utilize spatial information to extract, process and locate current signals in complex environments. Since the performance of signals arriving from the same or similar directions on spatially distributed sensors will have a high degree of coherence, the present invention addresses the problem of insufficient utilization of spatial information by previous active electric field detection systems. It jointly considers the electrode sensor array arrangement, current signal extraction algorithm, and spatial perception model, optimizes the electrode arrangement to provide maximum water area electric field spatial information, and fully considers the correlation of data collected by multiple sensors at the same time when capturing and processing electrode signals. This idea originates from the Direction Of Arrival (DOA) estimation problem, and the core idea is to process MMVs and effectively utilize this spatial information.

[0006] The specific technical solutions are as follows:

[0007] A method for locating an underwater target comprises the following steps:

[0008] S1: Receive measurement data collected by multiple sensors arranged on the detector at multiple snapshot times and construct a measurement signal matrix ;

[0009] S2: Using the multi-target current signal extraction algorithm with multi-measurement vector correlation, the current signal amplitude matrix of all received targets at the corresponding frequency is iteratively estimated, which is specifically achieved through the following sub-steps:

[0010] (2.1) Using the measurement signal matrix , a preliminary frequency estimate is obtained by maximizing the generalized likelihood ratio cost function on the discrete Fourier transform grid ,and Within the frequency range obtained by a priori estimation of the current target based on the seabed environment and target characteristics; the corresponding preliminary amplitude estimate is calculated based on the preliminary frequency estimate;

[0011] (2.2) The frequency estimate is updated by the first and second derivatives of the generalized likelihood ratio cost function: ,according to Calculate the corresponding amplitude at each snapshot moment, and collect the amplitude matrix corresponding to the current target ;

[0012] (2.3) Subtract the signal components of all detected targets from The contribution in is obtained, and the residual matrix is ​​obtained;

[0013] (2.4) Determine whether the detection of all targets is completed. If so, stop the detection and output the amplitude matrix corresponding to each target; if not, determine whether the residual matrix is ​​less than the set threshold. If it is less than the threshold, stop the detection. Otherwise, update the measurement signal matrix. is the residual matrix , and return to step (2.1) to start the detection of the next target;

[0014] S3: The amplitude matrices of multiple targets estimated by S2 are sequentially input into the spatial perception model based on multi-size one-dimensional convolution kernels for nonlinear mapping and feature representation, and the distance value from the geometric center of the detector to the geometric center of each target object and the angle value between the line connecting the two geometric centers and the bow direction of the detector are obtained to realize the positioning of underwater targets.

[0015] Furthermore, in (2.1), the preliminary frequency estimate The calculation expression is as follows:

[0016] ;

[0017] ;

[0018] In the formula, The measurement signal matrix The elements in represent the measurement vector at the t-th snapshot time, t=1,2,3,…,T, T is the total number of snapshot times; In a finite set of discrete candidate frequency points Above; γ represents the oversampling factor relative to the discrete Fourier transform grid; It is a column vector related to the frequency ω, representing the signal response of each sensor to the frequency ω.

[0019] Furthermore, the expression for calculating the corresponding preliminary amplitude estimate based on the preliminary frequency estimate is as follows:

[0020] ;

[0021] In the formula, represents a preliminary amplitude estimate, The measurement signal matrix The elements in represent the measurement vector at the t-th snapshot time, t=1,2,3,…,T, T is the total number of snapshot times; is related to frequency The relevant column vector represents the frequency of each sensor signal response;

[0022] according to Calculate the corresponding amplitude at the tth snapshot time The expression is as follows:

[0023] ;

[0024] In the formula, is related to frequency The relevant column vector represents the frequency of each sensor signal response.

[0025] Furthermore, the input of the spatial perception model based on multi-size one-dimensional convolution kernels is an n-dimensional vector, including the amplitude values ​​of n-1 sensors, that is, the amplitude matrix of a target, and the distance of the current detector from the bottom of the water; the output is a one-dimensional vector, which contains two components, including the distance value from the geometric center of the detector to the geometric center of the target, and the angle value between the line from the geometric center of the detector to the geometric center of the target and its own bow direction.

[0026] Furthermore, in S3, the spatial perception model based on the multi-size one-dimensional convolution kernel includes a plurality of residual modules of multi-size one-dimensional convolution connected in series and a multi-layer perceptron alternately, that is, a multi-layer perceptron is arranged behind each residual module, and the output of the current multi-layer perceptron is used as the input of the next residual module; the output of the last multi-layer perceptron is a one-dimensional vector including two components: the distance value from the geometric center of the detector to the geometric center of the target object, and the angle value between the line connecting the geometric center of the detector to the geometric center of the target object and the heading of the detector itself.

[0027] Furthermore, there are multiple residual modules for extracting spatial features of different scales, and each residual module includes: an input layer, multiple branches, and a fully connected layer; the number of the branches is the number of even powers of 2 that are smaller than the data size of the input layer of the current residual module plus 1; the convolution kernel sizes of the one-dimensional residual blocks of each branch are, in sequence, even powers of 2 starting from 1 and increasing in size, and if the next even power is larger than the data size of the input layer, the data size of the input layer is taken as the convolution kernel size of the last branch;

[0028] Each branch fills the input according to the different sizes of the one-dimensional convolution kernel and the step length is 1, so that the output size obtained in each branch is consistent and the length is the same as the input of the current residual module; the output of each branch is expanded by channel, and in the fully connected layer, it is spliced ​​into a new one-dimensional vector in the order of each channel, and then the one-dimensional vectors obtained by each branch are spliced ​​to obtain the final output of the current residual module;

[0029] The one-dimensional vector output by the current residual module is sent to the multi-layer perceptron. After the multi-layer perceptron performs feature transformation and fully connected layer processing on the one-dimensional vector, a new feature representation is generated and a new one-dimensional vector is output. This one-dimensional vector is then sent to the new residual module. The operation is repeated until it passes through the last residual module and multi-layer perceptron and the final one-dimensional vector is output.

[0030] An underwater target positioning device, used to implement the underwater target positioning method, comprising: a detector with a mobile function, a multi-channel transceiver electrode sensor system mounted on the detector, and a host computer;

[0031] The multi-channel transceiver electrode sensor system includes a transmitting electrode and multiple receiving electrodes; the transmitting electrode is located on one end face of the detector, and the receiving electrodes are arranged on the other end faces of the detector in the form of maximum spacing; the transmitting electrode and a receiving electrode constitute an electrode pair, each electrode pair works in turn, and the non-working electrode pair is switched to a floating potential without affecting the electric field distribution; one electrode pair is equivalent to a sensor;

[0032] The host computer includes a current signal extraction module and a feature extraction module. The current signal extraction module is used to estimate the amplitude matrix of the signal received by each receiving electrode at a specific frequency based on the spatial information between the measurement sensors in various directions; the feature extraction module is used to extract features of the current signal output by the current signal extraction module to obtain the distance from the geometric center of the detector to the geometric center of the target object, and the angle value between the line connecting the two geometric centers and the bow direction of the detector.

[0033] Furthermore, the detector is a remote-controlled unmanned submersible.

[0034] Furthermore, the normal of the left and right end faces of the detector is horizontal, and the direction parallel to the upper end face and the left end face is longitudinal, and the normal of the upper and lower end faces is vertical; the transmitting electrode is arranged in the middle of the left end face of the detector, and the receiving electrodes are arranged at the maximum interval on the periphery of the detector body except the left end face, wherein a receiving electrode is set in the middle of the right end face, and the remaining receiving electrodes are arranged in the same form on the front and rear end faces of the detector, respectively. Taking the front end face as an example, the receiving electrodes are arranged symmetrically on both sides of the vertical direction, and the receiving electrodes on one side are evenly distributed in the horizontal direction with the maximum spacing; if one receiving electrode does not work, it is switched to a floating potential without affecting the electric field distribution.

[0035] Furthermore, a row of spare electrodes is provided at the center of the lower end surface of the detector body as spare receiving electrodes. When the target object is located directly below the detector, the spare electrode is turned on to enhance the perception and confirm whether the target object is located directly below. If one of the main working receiving electrodes is turned off, a corresponding spare electrode is turned on, and the non-working receiving electrode is switched to a floating potential and does not participate in the work.

[0036] The beneficial effects of the present invention are:

[0037] (1) The present invention uses a current signal extraction algorithm that considers multiple measurement vectors, integrates the spatial information between sensors in various directions for estimation, and fully captures the correlation of the spatial information of the surrounding electric field environment to improve the sensitivity of the detection electrode.

[0038] (2) The underwater target positioning device proposed in the present invention has an optimized sensor electrode arrangement and configuration, which maximizes the detection range and resolution. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 1 is a schematic diagram of the structure of an underwater target positioning device in an embodiment of the present invention, wherein (a) is a schematic diagram of the electrode arrangement structure, and (b) is a schematic diagram of the electrode type and numbering.

[0040] Figure 2 4 is a flow chart of an underwater target positioning method in an embodiment of the present invention.

[0041] Figure 3 4 is a flow chart of a current signal extraction algorithm in an embodiment of the present invention.

[0042] Figure 4 It is a spatial perception model based on multi-sized one-dimensional convolution kernels in an embodiment of the present invention.

[0043] Figure 5 : is a schematic diagram of the relationship between the current change of the receiving electrodes at different positions and the forward distance when the target is placed 2 meters away from the ROV in an embodiment of the present invention, where (a) is Figure 1 (a) is a schematic diagram of the relationship between the current change and the distance advanced for the electrodes numbered 7, 10, 8, and 9; (b) is a schematic diagram of the relationship between the current change and the distance advanced for the electrodes numbered 4, 5, 6, and 7; (c) is a schematic diagram of the relationship between the current change and the distance advanced for the electrodes numbered 7, 11, 12, and 13; and (d) is a schematic diagram of the relationship between the current change and the distance advanced for the electrodes numbered 1, 2, 3, and 7. DETAILED DESCRIPTION

[0044] The present invention will be described in detail below according to the accompanying drawings and preferred embodiments, and the purpose and effect of the present invention will become more clear. The present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0045] like Figure 1As shown, an underwater target positioning device includes: a detector with a mobile function, a multi-channel transceiver electrode sensor system mounted on the detector, and a host computer. The multi-channel transceiver electrode sensor system includes a single positive electrode as a transmitting electrode and multiple single negative electrodes as receiving electrodes.

[0046] The power supply system uses DC voltage excitation-current measurement as the power detection excitation method. The transmitting electrode is located on one end face of the detector, and the receiving electrode is arranged on the other end face of the detector in the form of maximum spacing; the transmitting electrode and a receiving electrode form an electrode pair, and each electrode pair works in turn. The inoperative electrode pair is switched to a floating potential without affecting the electric field distribution. The electrode pair is equivalent to integrating measurement sensors in multiple directions on the detector, replacing the process that requires the detector to move and turn for measurement.

[0047] The host computer includes a current signal extraction module and a feature extraction module. The current signal extraction module is used to estimate the amplitude matrix of the signal received by each receiving electrode at a specific frequency based on the spatial information between the measurement sensors in each direction. The feature extraction module is used to extract the features of the current signal output by the current signal extraction module to obtain the distance from the geometric center of the detector to the geometric center of the target, as well as the angle between the line connecting the two geometric centers and the bow direction of the detector.

[0048] In this embodiment, the detector uses a remotely operated vehicle (hereinafter referred to as ROV), and 13 receiving electrodes are set. The ROV is a rectangular parallelepiped, 1.5m long × 0.85m wide × 0.77m high, with the normals of the left and right end faces as the horizontal direction, the direction parallel to the upper end face and the left end face as the longitudinal direction, and the normals of the upper and lower end faces as the vertical direction. The transmitting electrode is arranged in the middle of the left end face of the ROV, and the 13 receiving electrodes are arranged at the maximum interval on the outer periphery of the ROV except the left end face, among which a receiving electrode is set in the middle of the right end face ( Figure 1 The green number 0 in (b) is used to stretch the electric field lines as much as possible, enhance the ductility of the electric field lines and the sensing accuracy; the remaining receiving electrodes are arranged in the same form on the front and rear ends of the ROV. Taking the front end as an example, the receiving electrodes are symmetrically arranged on both sides of the vertical direction, and the receiving electrodes on one side (numbered 8, 9, 10 or 11, 12, 13) are evenly distributed in the horizontal direction with the maximum spacing. The receiving electrodes are arranged symmetrically on both sides of the vertical direction, which can be used for differential calculation to identify the direction. The reception on the front and rear ends is used to measure the differential current. If a receiving electrode does not work, it will be switched to a floating potential without affecting the electric field distribution.

[0049] The above-mentioned receiving electrode is used as the main working electrode, and a row of spare electrodes is set in the center of the lower end surface of the detector as a spare receiving electrode as an additional sensing tool. When the target is directly below the ROV, the spare electrode can be turned on for signal enhancement to enhance the perception and confirm whether the target is directly below; the rest of the time, the spare electrode is turned off, maintains a suspended potential, and does not participate in the work; in this embodiment, the spare electrodes are evenly distributed in the horizontal direction with the maximum spacing. Insulating materials are used on the surface of the detector to avoid leakage, the formation of additional current loops, and the interference of metal equipotentials on the reaction field. The electrode area of ​​the receiving electrode and the spare electrode is 40mm×40mm square, and a replaceable carbon block is used as the exposed electrode, which is easy to replace after electrochemical corrosion. When the total number of single negative electrodes remains unchanged, the main working electrode and the spare electrode work in turn (one spare electrode is turned on, and one main working electrode is turned off accordingly), and the non-working receiving electrode is switched to a suspended potential without affecting the electric field distribution.

[0050] In this device, multi-channel electrode detection is regarded as a multiple measurement vector (MMVs) problem to receive current signals.

[0051] Based on the above underwater target positioning device, an embodiment of the present invention proposes an underwater target positioning method, such as Figure 2 As shown, the specific steps include:

[0052] S1: Receive measurement data collected by multiple sensors at multiple snapshot moments, where each time point is called a snapshot moment, and construct a measurement signal matrix ,in Represents the measurement data of N sensors at T snapshot times, represents a dictionary matrix containing K frequency information, Represents the amplitude matrix of the spectrum value at the corresponding frequency of the transmitted signal, represents an independent and identically distributed Gaussian noise matrix. K is numerically equal to the number of discrete Fourier transform (DFT) points set in the subsequent multi-target current signal extraction algorithm, which is generally greater than the number of sensors N. The obtained measurement signal matrix This will be passed as input to the next step, the detection stage.

[0053] S2: Using the multi-target current signal extraction algorithm based on multi-measurement vector correlation, the current signal amplitude matrix of all received targets at the corresponding frequency is estimated. The multi-target current signal extraction algorithm achieves estimation optimization through multiple iterations. The initial iteration targets one target, and each subsequent iteration adds a new target. This is achieved through the following sub-steps:

[0054] (2.1) Using the measurement signal matrix , on the DFT grid, a preliminary frequency estimate is obtained by maximizing the generalized likelihood ratio cost function :

[0055]

[0056]

[0057] In the formula, The measurement signal matrix The elements in represent the measurement vector at the t-th snapshot time, t=1,2,3,…,T; In a finite set of discrete candidate frequency points superior, The upper and lower frequency ranges (i.e. and The value of ) is obtained by prior estimation and is determined according to the specific seabed environment and different target characteristics, so that there is a priori estimation of the frequency range of each specific target; γ represents the oversampling factor relative to the DFT grid; It is a column vector related to the frequency ω, which represents the signal response of each electrode to the frequency ω, and further the spatial distribution on the device structure.

[0058] Further select the frequency that maximizes the cost function as a preliminary frequency estimate, and from it obtain the corresponding preliminary amplitude estimate .

[0059] (2.2) Using the preliminary frequency estimate obtained in the detection phase and preliminary amplitude estimates , further refine the frequency and update the frequency estimate using the following formula:

[0060]

[0061] In the formula, is the generalized likelihood ratio cost function The first derivative of is the generalized likelihood ratio cost function The second-order derivatives of are calculated as:

[0062]

[0063]

[0064] In the formula, represents the real part, and They are The first and second order derivatives of .

[0065] The output of this step is a refined frequency estimate , and use the method in (2.1) to further estimate the amplitude of each snapshot moment , and then the amplitudes of all these snapshots are combined into the amplitude matrix corresponding to the total frequency .

[0066] (2.3) Use the refined frequency obtained from this round of target and its corresponding amplitude matrix, and the refined frequency and amplitude corresponding to other targets estimated previously are combined to calculate the residual matrix corresponding to the next target, that is, the total measurement signal matrix received Subtract the signal components of all previously detected targets from the total measured signal matrix The contribution of is expressed as follows:

[0067]

[0068] In the formula, Indicates the current The signal components of the target are related to the total measurement signal matrix received contribution.

[0069] (2.4) Determine whether the detection of all targets is completed. If so, stop the detection and output the amplitude matrix corresponding to each target; if not, continue to determine whether the residual matrix is ​​less than the set threshold. If it is less than, stop the detection, otherwise, update the measurement signal matrix is the residual matrix (That is, using the residual matrix Replace the original measurement signal matrix ), and return to step (2.1) to start the detection of the next target.

[0070] S3: Input the amplitude matrix of multiple targets estimated by S2 into the spatial perception model based on multi-size one-dimensional convolution kernels for nonlinear mapping and feature representation, obtain the distance value from the geometric center of the detector to the geometric center of the target, and the angle value between the line connecting the two geometric centers and the bow of the detector, so as to realize the positioning of underwater targets. This is achieved through the following sub-steps:

[0071] The spatial perception model based on multi-size one-dimensional convolution kernels is deployed in the host computer. For the detection of a single target, the input of the model is an n-dimensional vector, which contains the amplitude of the current of n-1 receiving electrodes (i.e., the amplitude matrix of a certain target) and the distance of the current ROV from the bottom of the water. The output is a one-dimensional vector, which contains two components: the distance value from the geometric center of the ROV to the geometric center of the target, and the angle value between the line connecting the geometric center of the ROV to the geometric center of the target and its own bow direction.

[0072] The spatial perception model based on multi-size one-dimensional convolution kernels is composed of multiple multi-size one-dimensional convolution residual modules and multilayer perceptrons (MLP) alternately connected in series; that is, each residual module is followed by a multilayer perceptron, and the output of the current multilayer perceptron is used as the input of the next residual module.

[0073] like Figure 4 As shown, there are multiple residual modules for extracting spatial features of different scales. Each residual module includes: an input layer, multiple branches, and a fully connected layer. The number of branches and the convolution kernel size of the one-dimensional residual block of each branch are determined by the data size of the input layer input of the current residual module. Specifically, the number of branches is the number of even powers of 2 that are smaller than the data size of the input layer input of the current residual module plus 1; the convolution kernel size of the one-dimensional residual block of each branch is an even power of 2 from 1 to large. If the next even power is larger than the current input data size, the current input data size is taken as the convolution kernel size of the last branch. An example is provided below for easy understanding: the data size of the input layer input of the current residual module is 14 (including the amplitude matrix of 13 receiving electrode currents and the current distance of the ROV from the bottom of the water). Since there are 2 even powers of 2 smaller than 14, 0 and 2 2 , then the number of branches is 3, and the convolution kernel sizes of the one-dimensional residual blocks of each branch are 1×1, 1×4, and 1×14 respectively.

[0074] At the same time, each branch fills the input according to the different sizes of the one-dimensional convolution kernel and the step length is 1, so that the output size obtained in each branch is consistent and the length is the same as the input of the residual module (that is, they are all converted into the form of 1×M, M represents the size length of the feature representation of the previous multi-layer perceptron output). The output of each branch is expanded by channel, and spliced ​​into a new one-dimensional vector in the order of each channel in the fully connected layer, and then the one-dimensional vectors obtained by each branch are spliced ​​again to obtain the final output of the residual module.

[0075] The one-dimensional vector output by the residual module is sent to a multi-layer perceptron. After the one-dimensional vector is transformed in features, it is processed through a series of fully connected layers to generate a new feature representation and output a new one-dimensional vector. This one-dimensional vector is then sent to a new residual module and the operation is repeated until it passes through the last residual module and multi-layer perceptron and outputs the final one-dimensional vector.

[0076] During underwater detection by ROV, the electric field signal generated by the electric field emission electrode will propagate in the water and interact with different targets in the water, such as Figure 5 The figure shows the relationship between the current change of the receiving electrodes in the four main directions and the distance the ROV moves toward the target when the distance between the target and the ROV is 2 meters. It can be observed that the current of most electrodes decreases significantly when approaching the target, and the current increases rapidly after passing the target. This experimental phenomenon proves the high sensitivity of the system at close distances.

[0077] Through the above specific implementation methods, the present invention provides an underwater target positioning method and device that can perform high-precision target positioning in a complex underwater environment, and uses a neural network model with a multi-scale one-dimensional convolution kernel to extract features of the current signal extracted by the device and further locate the target, wherein the number of branches and the size of the kernel of each convolution module are adaptively changed with the size of the one-dimensional vector input of the corresponding layer, thereby better extracting the features of 14 special physical quantities in the actual experiment, and at the level of the positioning algorithm, the current value correlation at each scale between the 13 electrodes is still considered, corresponding to the actual physical model. Thereby improving the final positioning accuracy of the entire system for buried / non-buried objects. Through the optimized electrode arrangement scheme and advanced signal processing algorithm, the present invention can work effectively under various water quality conditions, thereby improving the robustness and accuracy of underwater detection.

[0078] Those skilled in the art can understand that the above are only preferred examples of the invention and are not intended to limit the invention. Although the invention is described in detail with reference to the above examples, those skilled in the art can still modify the technical solutions recorded in the above examples or replace some of the technical features therein with equivalents. Any modification, equivalent replacement, etc. made within the spirit and principle of the invention shall be included in the protection scope of the invention.

Claims

1. A method for locating an underwater target, characterized in that: The following steps are involved: S1: Receive measurement data collected by multiple sensors arranged on the detector at multiple snapshot times and construct a measurement signal matrix ; S2: Using the multi-target current signal extraction algorithm with multi-measurement vector correlation, the current signal amplitude matrix of all received targets at the corresponding frequency is iteratively estimated, which is specifically achieved through the following sub-steps: (2.1) Using the measurement signal matrix , a preliminary frequency estimate is obtained by maximizing the generalized likelihood ratio cost function on the discrete Fourier transform grid ,and Within the frequency range obtained by a priori estimation of the current target based on the seabed environment and target characteristics; the corresponding preliminary amplitude estimate is calculated based on the preliminary frequency estimate; (2.2) The frequency estimate is updated by the first and second derivatives of the generalized likelihood ratio cost function: ,according to Calculate the corresponding amplitude at each snapshot moment, and collect the amplitude matrix corresponding to the current target ; (2.3) Subtract the signal components of all detected targets from The contribution in is obtained, and the residual matrix is ​​obtained; (2.4) Determine whether the detection of all targets is completed. If so, stop the detection and output the amplitude matrix corresponding to each target; if not, determine whether the residual matrix is ​​less than the set threshold. If it is less than the threshold, stop the detection. Otherwise, update the measurement signal matrix. is the residual matrix , and return to step (2.1) to start the detection of the next target; S3: The amplitude matrices of multiple targets estimated by S2 are sequentially input into the spatial perception model based on multi-size one-dimensional convolution kernels for nonlinear mapping and feature representation, and the distance value from the geometric center of the detector to the geometric center of each target object and the angle value between the line connecting the two geometric centers and the bow direction of the detector are obtained to realize the positioning of underwater targets.

2. The underwater target positioning method according to claim 1, characterized in that: In (2.1), the initial frequency estimate The calculation expression is as follows: ； ； In the formula, The measurement signal matrix The elements in represent the measurement vector at the t-th snapshot time, t=1,2,3,…,T, T is the total number of snapshot times; In a finite set of discrete candidate frequency points Above; γ represents the oversampling factor relative to the discrete Fourier transform grid; It is a column vector related to the frequency ω, representing the signal response of each sensor to the frequency ω.

3. The underwater target positioning method according to claim 1, characterized in that: The expression for calculating the corresponding preliminary amplitude estimate based on the preliminary frequency estimate is as follows: ； In the formula, represents a preliminary amplitude estimate, The measurement signal matrix The elements in represent the measurement vector at the t-th snapshot time, t=1,2,3,…,T, T is the total number of snapshot times; is related to frequency The relevant column vector represents the frequency of each sensor signal response; according to Calculate the corresponding amplitude at the tth snapshot time The expression is as follows: ； In the formula, is related to frequency The relevant column vector represents the frequency of each sensor signal response.

4. The underwater target positioning method according to claim 1, characterized in that: The input of the spatial perception model based on multi-size one-dimensional convolution kernels is an n-dimensional vector, including the amplitude values ​​of n-1 sensors, that is, the amplitude matrix of a certain target, and the distance between the current detector and the bottom of the water; The output is a one-dimensional vector, which contains two components: the distance from the geometric center of the detector to the geometric center of the target, and the angle between the line from the geometric center of the detector to the geometric center of the target and its own heading.

5. The underwater target positioning method according to claim 1, characterized in that: In S3, the spatial perception model based on multi-size one-dimensional convolution kernels includes a plurality of residual modules of multi-size one-dimensional convolutions connected in series and a multi-layer perceptron alternately, that is, a multi-layer perceptron is arranged behind each residual module, and the output of the current multi-layer perceptron is used as the input of the next residual module; the output of the last multi-layer perceptron is a one-dimensional vector including two components: the distance value from the geometric center of the detector to the geometric center of the target object, and the angle value between the line connecting the geometric center of the detector to the geometric center of the target object and the heading of the detector itself.

6. The underwater target positioning method according to claim 5, characterized in that: There are multiple residual modules for extracting spatial features of different scales, and each residual module includes: an input layer, multiple branches, and a fully connected layer; the number of the branches is the number of even powers of 2 that are smaller than the data size of the input layer of the current residual module plus 1; the convolution kernel sizes of the one-dimensional residual blocks of each branch are, in order, even powers of 2 from 1 to large, and if the next even power is larger than the data size of the input layer, the data size of the input layer is taken as the convolution kernel size of the last branch; Each branch fills the input according to the different sizes of the one-dimensional convolution kernel and the step length is 1, so that the output size obtained in each branch is consistent and the length is the same as the input of the current residual module; the output of each branch is expanded by channel, and in the fully connected layer, it is spliced ​​into a new one-dimensional vector in the order of each channel, and then the one-dimensional vectors obtained by each branch are spliced ​​to obtain the final output of the current residual module; The one-dimensional vector output by the current residual module is sent to the multi-layer perceptron. After the multi-layer perceptron performs feature transformation and fully connected layer processing on the one-dimensional vector, a new feature representation is generated and a new one-dimensional vector is output. This one-dimensional vector is then sent to the new residual module. The operation is repeated until it passes through the last residual module and multi-layer perceptron and the final one-dimensional vector is output.

7. An underwater target positioning device, used to implement the underwater target positioning method according to any one of claims 1 to 6, characterized in that: include: A detector with mobile function, a multi-channel transceiver electrode sensor system mounted on the detector, and a host computer; The multi-channel transceiver electrode sensor system includes a transmitting electrode and multiple receiving electrodes; the transmitting electrode is located on one end face of the detector, and the receiving electrodes are arranged on the other end faces of the detector in the form of maximum spacing; the transmitting electrode and a receiving electrode constitute an electrode pair, each electrode pair works in turn, and the non-working electrode pair is switched to a floating potential without affecting the electric field distribution; one electrode pair is equivalent to a sensor; The host computer includes a current signal extraction module and a feature extraction module. The current signal extraction module is used to estimate the amplitude matrix of the signal received by each receiving electrode at a specific frequency based on the spatial information between the measurement sensors in various directions; the feature extraction module is used to extract features of the current signal output by the current signal extraction module to obtain the distance from the geometric center of the detector to the geometric center of the target object, and the angle value between the line connecting the two geometric centers and the bow direction of the detector.

8. The underwater target positioning device according to claim 7, characterized in that: The detector is a remote-controlled unmanned submersible.

9. The underwater target positioning device according to claim 7, characterized in that: The normals of the left and right end faces of the detector are horizontal, and the directions parallel to the upper and left end faces are longitudinal, and the normals of the upper and lower end faces are vertical; the transmitting electrode is arranged in the middle of the left end face of the detector, and the receiving electrodes are arranged at the maximum interval on the periphery of the detector body except the left end face, wherein a receiving electrode is set in the middle of the right end face, and the remaining receiving electrodes are arranged in the same form on the front and rear end faces of the detector, respectively. Taking the front end face as an example, the receiving electrodes are symmetrically arranged on both sides of the vertical direction, and the receiving electrodes on one side are evenly distributed in the horizontal direction with the maximum spacing; if a receiving electrode does not work, it will be switched to a floating potential without affecting the electric field distribution.

10. The underwater target positioning device according to claim 7, characterized in that: A row of spare electrodes is also provided in the center of the lower end surface of the detector body as spare receiving electrodes. When the target object is located directly below the detector, the spare electrode is turned on to enhance the perception and confirm whether the target object is located directly below. If one of the main working receiving electrodes is turned off, a corresponding spare electrode is turned on, and the non-working receiving electrode is switched to a floating potential and does not participate in the work.

Images