Underwater target intensity prediction method based on neural network

Through the BP neural network method, the target intensity database is constructed and trained using the plate element method, which solves the problem of slow prediction of complex underwater target intensity, and achieves fast and efficient underwater target intensity forecasting, which is suitable for real-time application scenarios .

CN119046637APending Publication Date: 2024-11-29SHANGHAI JIAOTONG UNIV
View PDF 3 Cites 0 Cited by

Patent Information

Application Number
CN202411176423.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-26
Publication Date
2024-11-29

AI Technical Summary

Technical Problem

The prior art is slow in the calculation speed and low efficiency when dealing with the intensity prediction of complex underwater targets. It is difficult to choose a suitable interpolation model especially in high-dimensional data, and artificial intelligence is less used in the prediction of physical characteristics of underwater targets.

Method used

Using a BP neural network-based method, a target intensity database is constructed by designing the network layer number and activation function, and a plate element method is used to construct a target intensity database, and train and verify it to achieve a rapid forecast of the underwater target intensity.

Benefits of technology

It realizes rapid forecasting of the physical attribute characteristics of underwater complex structure targets, with the characteristics of high efficiency, strong adaptability and convenient use, and can flexibly respond to different environmental conditions and target characteristics, improving the robustness and real-timeness of the prediction model.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119046637A_ABST
    Figure CN119046637A_ABST
Patent Text Reader

Abstract

The invention discloses an underwater target strength prediction method based on a neural network, and the method comprises the steps: building a training data set of underwater target strength through plate element calculation, and learning the strength characteristics and change rules of an underwater complex target through the highly nonlinear mapping capability of a BP neural network. And the transmitting frequency and the azimuth angle and pitch angle parameters of the to-be-measured target are input into the trained BP neural network model, so that the target strength can be quickly predicted. According to the method, the neural network is adopted to construct the prediction model reflecting the change rule of the target strength data, in practical engineering application, after network training is completed, data features can be generalized, field simulation calculation is not needed, and compared with a traditional prediction method, the method has the advantages of being high in calculation speed, high in efficiency and convenient to use, and the method is suitable for popularization and application. And the target intensity can be quickly predicted in a complex underwater environment.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention belongs to the field of underwater target positioning, detection and identification, and specifically relates to an underwater target intensity prediction method based on a neural network. Background Art

[0002] Active sonar target echo modulates the information of the transmitting signal, target and acoustic wave transmission channel. The acquisition of target echo physical characteristics is of great significance for active sonar target detection, classification and recognition, and acoustic feature control stealth design. In addition to direct measurement of target echo characteristics, target echo simulation modeling is an important means to obtain target echo characteristics. Based on the concepts of radar cross section (RCS) and plate element, the underwater target plate element analysis method established by Krichhoff approximation converts the target complex scattering surface integral into a superposition of a set of plane scattering, simplifies the calculation of complex target scattering sound field, and can accurately predict the geometric backscattering waves of underwater complex targets. However, the number of target plate divisions is huge, the calculation speed is slow, and the efficiency is low. In early engineering, in order to quickly calculate the target intensity, the table lookup method or interpolation method is often used. However, this method is difficult to select an interpolation model that meets the target intensity feature mapping in the case of high-dimensional data with many variables and a large range of parameter changes. At present, artificial intelligence has been widely used in the field of underwater acoustic processing, mainly focusing on the extraction and identification of underwater target features, which belongs to the application scope of signal processing. However, it is relatively rare to use artificial intelligence methods to predict the physical properties of underwater targets. It is expected to explore new forecasting models, which is simple and efficient in predicting target intensity. Summary of the invention

[0003] In order to overcome the shortcomings of the above-mentioned prior art, the present invention provides a method for quickly predicting the strength of underwater targets based on a neural network. The method adopts a neural network to predict the strength of underwater targets. In actual engineering applications, after the network training is completed, the corresponding target strength can be output given the input parameters. There is no need to perform on-site simulation calculations. The physical property characteristics of underwater complex structure targets can be quickly predicted. The method has the characteristics of strong adaptability, efficient calculation and easy use.

[0004] The technical solution of the present invention is as follows:

[0005] A method for predicting underwater target strength based on a neural network, characterized in that the method comprises the following steps:

[0006] 1) Target intensity is a function of target geometry, frequency, azimuth and elevation. For the prediction of the acoustic scattering field of a target with a specific geometry, the frequency f, azimuth θ, and elevation angle are used. Construct a three-dimensional vector as the input parameter of the BP neural network, and use the scattering cross-sectional area corresponding to the input parameter as the output parameter of the neural network;

[0007] 2) Establish BP neural network, including designing the number of network layers, hidden layer and output layer activation functions, and training functions;

[0008] 3) Build a target strength database and use the plate element method to calculate different frequencies f, azimuth angles θ, and pitch angles Target strength TS value in the case;

[0009] 4) Normalize the input and output parameters of the target strength database samples, randomly select 70% of the data in the database as training samples, 15% as verification samples, and 15% as test samples;

[0010] 5) Use training samples to train the network, and then use verification samples to verify the trained network. According to the verification results, adjust the initial parameters and network structure of the network, and repeat the network training until the error between the predicted value and the actual value meets the accuracy requirements;

[0011] 6) Use the test samples to test the trained network model, and give the probability density statistical distribution results of the target intensity TS value of the test samples and the model prediction results respectively.

[0012] 7) The target intensity TS value of the database described above is calculated by using graphic software such as AutoCAD to geometrically model the target; the target intensity TS value is calculated using the plate element method. For high-frequency conditions, it is usually assumed that: a. the contribution of the geometric shadow area to the sound field can be ignored; b. the surface of the object satisfies the rigid boundary condition.

[0013] In the far field, k0r>>1 (k0 is the wave number, λ is the wavelength of the transmitted signal, c is the speed of sound wave propagation in water, and r is the distance between the receiving point and the target geometry), the sonar target strength of the combined transmission and reception situation is:

[0014]

[0015] In the formula, The key to finding the far-field target intensity is to calculate the surface integral I. is the vector from the point where the surface element is located to the reference point, is the unit normal vector of the surface element, is the unit vector from the receiving point to the reference point.

[0016] The plate element method calculates the target strength of complex targets. First, the target surface is divided into N×M polygonal small plate networks to obtain a series of plate elements S i,j Since the plate elements of direct construction have different tendencies, it is necessary to transform the plate elements S i,j By transforming the coordinates to a certain plane, the surface integral can be expressed as,

[0017]

[0018] Where K is the number of polygon vertices, (x n ,y n ) are the coordinates of the polygon vertices, And order x' n ,y' n is the vertex coordinates of the (i,j)th plate after transformation to the two-dimensional plane, V(α' ij ) is the local surface reflection coefficient of the (i,j)th plate, α i ' j is the angle between the normal of the (i,j)th plate and the incident sound ray.

[0019] In the near field, the target surface integral I is calculated as follows:

[0020]

[0021] In the formula, u = 2ku', v = 2kv', w = w', and then the obtained acoustic fields of each plate are synthesized into the scattered field of the entire target where A is the acoustic scattering amplitude, is the distance from the geometric center of the plate element, and the normal of the plane is N is the number of plates on the target that can be illuminated by the incident wave, satisfying is the unit normal vector of the surface element.

[0022] 8) The BP neural network design described above assumes that x i is the input vector, y j is the hidden layer node parameter, o l is the output node parameter; w ij is the network weight between the input node and the hidden layer node, and its threshold is θ i ; T li is the network weight between the output layer nodes and the hidden layer nodes, and its threshold θ l When t l As the expected output of the network, the output of the hidden layer node is,

[0023] y j =f(Σ j w ij x i -θ i ) (5)

[0024] The output of the output layer node is,

[0025] o l=f(Σ j T li y j -θ l ) (6)

[0026] The error formula is,

[0027]

[0028] The weight update process uses momentum and adaptive gradient decreasing training function.

[0029] dw t =m c ×dw t-1 +lr×m c ×dE / dw t (8)

[0030] In the formula, m c is the momentum parameter, lr is the learning rate.

[0031] Furthermore, to predict the acoustic physical characteristics of multiple target types, the input vector needs to contain the geometric structure parameters of different types of targets.

[0032] The present invention has the following advantages:

[0033] (1) The present invention uses a neural network to capture the complex nonlinear relationship between the physical characteristics of the transmitted signal and the target echo it stimulates, and can directly learn and establish a prediction model for target intensity from the original data, thereby simplifying the entire prediction process and improving efficiency;

[0034] (2) The neural network has good adaptability and generalization ability and can process data sets of different types and sizes, which enables the neural network to flexibly respond to different underwater environmental conditions and target characteristics (such as target geometry, shape, material, etc.), thereby improving the robustness of the prediction model;

[0035] (3) When dealing with complex underwater targets, the plate element method simplifies the calculation process of the integral equation by dividing the target surface into several planar plates, making the calculation of underwater target strength more efficient. After the BP neural network is fully trained with the training data set established by plate element calculation, the model can quickly respond to changes in input parameters and output the corresponding target strength prediction value. This fast prediction speed is of great significance for application scenarios with high real-time requirements (such as underwater detection, target tracking, etc.).

[0036] (3) Compared with the calculation method of physical acoustics, the network model is trained in advance. In actual engineering applications, the corresponding target intensity can be output given the input parameters, which has the advantage of being easy to use. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 This is a schematic diagram of rapid prediction of underwater target intensity based on neural network;

[0038] Figure 2 This is a schematic diagram of the benchmark submarine surface element division;

[0039] Figure 3 is a neural network graph;

[0040] Figure 4 is a graph of neural network training performance;

[0041] Figure 5 is the neural network regression graph;

[0042] Figure 6 It is a comparison chart between target strength prediction and simulation results;

[0043] Figure 7 It is a comparison chart between target intensity probability distribution prediction and simulation results;

[0044] Numbers in the figure: 1, target; 2, transmitting and receiving combined underwater acoustic transducer. DETAILED DESCRIPTION

[0045] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0046] The schematic diagram of the embodiment of the method for rapid prediction of underwater target intensity based on neural network of the present invention is as follows: Figure 1 The method for rapid prediction of underwater target intensity based on neural network of the present invention comprises the following steps:

[0047] 1) Target intensity is a function of target geometry, frequency, azimuth and elevation. For the prediction of acoustic scattering field of a target with a specific geometry, the frequency f = [10000:50:30000], the azimuth θ = [(0:5:180)×π / 180], the elevation Construct a three-dimensional vector as the input parameter of the BP neural network, and use the scattering cross-sectional area corresponding to the input parameter as the output parameter of the neural network;

[0048] 2) Establish BP neural network, Figure 2 Given a network model diagram, design a 3-layer network, including input layer, hidden layer and output layer, where the hidden layer has 198 neurons. The hidden layer uses the hyperbolic tangent activation function, the output layer uses the linear activation function, and the network is trained using the gradient descent method;

[0049] 3) Build a target intensity database, including 14837 samples. The target is 62m long and 6m in diameter. The plate element method is used to calculate different frequencies f, azimuth angles θ, and pitch angles. Target strength TS value in the case;

[0050] 4) Normalize the input and output parameters of the target intensity database samples, randomly select 70% of the data in the database as training samples, 15% as validation samples, and 15% as test samples, where the training samples, validation samples, and test samples do not overlap;

[0051] 5) Use training samples to train the network, and then use verification samples to verify the trained network. According to the verification results, adjust the network initial parameters and network structure, and repeat the network training until the error between the predicted value and the actual value meets the set accuracy. The target error is set to 0.01. Figure 3 It is a training performance graph. The training continues and approaches this error target before reaching the maximum training cycle, and the training will end early. Figure 4 It is a prediction regression graph, which reflects the degree of fit to the target intensity.

[0052] 6) Use test samples to test the trained network model. Figure 5 is the comparison result between the target intensity prediction and the verification sample, Figure 6 The probability density statistical distribution results of the target intensity TS values ​​of the test samples and the model prediction results are given.

[0053] 7) The target intensity TS value of the database described above is calculated by using graphic software such as AutoCAD to geometrically model the target; the target intensity TS value is calculated using the plate element method. For high-frequency conditions, it is usually assumed that: a. the contribution of the geometric shadow area to the sound field can be ignored; b. the surface of the object satisfies the rigid boundary condition.

[0054] The distance r between the transmitting and receiving sonar and the target center is 500m, and the target strength is calculated according to the far-field plate element formula. λ is the wavelength of the transmitted signal, c is the speed of sound wave propagation in water, c = 1500m / s,), sonar target intensity,

[0055]

[0056] In the formula, The key to finding the far-field target intensity is to calculate the surface integral I. is the vector from the point where the surface element is located to the reference point, is the unit normal vector of the surface element, is the unit vector from the receiving point to the reference point.

[0057] The plate element method calculates the target intensity of complex targets and divides the target surface into triangular face elements with 1 / 4 wavelength. Figure 7 Given a series of triangular plate elements S i,j Since the plate elements of direct construction have different tendencies, it is necessary to transform the plate elements S i,j , by transforming the coordinates to a certain plane, the surface integral can be expressed as,

[0058]

[0059] Where K is the number of polygon vertices, here K = 3, (x n ,y n ) are the coordinates of the triangle vertices, And order x' n ,y' n is the vertex coordinates of the (i,j)th plate after transformation to the two-dimensional plane, V(α' ij ) is the local surface reflection coefficient of the (i, j)th plate, which takes the value 1 assuming the target is rigid, and α i ' j is the angle between the normal of the (i,j)th plate and the incident sound ray.

[0060] 8) The BP neural network design described above, the input vector y j is the hidden layer node parameter, o l is the output node parameter; w ij is the network weight between the input node and the hidden layer node, and its threshold is θ i ; T li is the network weight between the output layer nodes and the hidden layer nodes, and its threshold θ l When t l As the expected output of the network, the output of the hidden layer node is,

[0061] y j =f(Σ j w ij x i -θ i ) (5)

[0062] The output of the output layer node is,

[0063] o l =f(∑ j T li y j -θ l ) (6)

[0064] The error formula is,

[0065]

[0066] The weight update process uses momentum and adaptive gradient decreasing training function.

[0067] dw t =m c ×dw t-1 +lr×m c ×dE / dw t (8)

[0068] In the formula, m c is the momentum parameter, the initial value is set to 0.5; lr is the learning rate, the initial value is set to 0.01.

[0069] The present invention predicts the acoustic physical characteristics of various target types, and the input vector needs to include geometric structure parameters of different types of targets.

[0070] Experiments show that the method of the present invention uses the highly nonlinear mapping ability of the BP neural network to achieve rapid prediction of target intensity. After the network training is completed, the data features can be generalized without the need for on-site simulation calculations. Compared with traditional forecasting methods, it has the characteristics of fast calculation speed, high efficiency and convenient use.

[0071] The above product can execute the method provided by the embodiment of the present invention, and has the functional modules and beneficial effects corresponding to the execution method. For technical details not described in detail in this embodiment, please refer to the method provided by the embodiment of the present invention.

[0072] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for predicting underwater target strength based on neural network, characterized in that: The steps include: S1. Identify the target intensity as a function of the target geometry, frequency, azimuth and elevation, construct an input parameter set, wherein the input parameter includes a three-dimensional vector formed by the frequency, azimuth and elevation of the incident wave; use a BP neural network, with the input parameter set as input, to predict the acoustic scattering characteristics of the target geometry under different detection conditions; set the scattering cross-sectional area corresponding to the input parameter, i.e., the target intensity TS value, as an output parameter of the BP neural network; S2. Establish a BP neural network model, which includes an appropriate number of network layers, hidden layers and output layers, and set activation functions and training functions for each layer; S3. Build a target strength database, and use the plate element method to calculate the target strength TS values ​​under different frequencies, azimuths, and elevation angles to build a comprehensive target strength database; S4. Data preprocessing and division: normalizing the input parameters and output parameters in the target intensity database, and randomly dividing the processed data set into training samples, verification samples and test samples, wherein 70% of the data is used as training samples, 15% of the data is used as verification samples, and 15% of the data is used as test samples; S5. Train the neural network, use the training samples to train the BP neural network model, use the verification samples to verify the trained network, and adjust the network initial parameters and network structure according to the verification results, and repeat the training process until the error between the predicted value and the actual value meets the preset accuracy requirements; S6. Testing and evaluation: Use test samples to test the trained and verified neural network model. By comparing the target intensity TS value of the test sample with the model prediction result, the probability density statistical distribution result of the target intensity TS value of the test sample and the model prediction result is given to evaluate the prediction performance of the model.

2. The underwater target intensity prediction method based on neural network according to claim 1 is characterized in that: The calculation of the target strength TS value comprises the following steps: (1) Use the graphics software AutoCAD to geometrically model the target; (2) The target strength TS is calculated using the plate element method, and the formula is as follows: In the far field, that is, k0r>>1, k0 is the wave number, λ is the wavelength of the transmitted signal, c is the speed of sound wave propagation in water, f is the frequency or frequency band of the transmitted signal, r is the distance between the receiving point and the geometric acoustic center of the target, and the target surface integral I is in, is the vector from the point where the surface element is located to the reference point, is the unit normal vector of the surface element, is the unit vector from the receiving point to the reference point; To determine the target strength of a complex target, we first need to divide the target surface into a network of N×M polygonal small plates to obtain a series of plate elements S i,j , and then calculate the target surface integral I, the formula is as follows: In formula 2, K is the number of polygon vertices, (x n ,y n ) are the coordinates of the polygon vertices, And order x' n ,y' n is the vertex coordinates of the (i,j)th plate after transformation to the two-dimensional plane, V(α' ij ) is the local surface reflection coefficient of the (i,j)th plate, α' ij is the angle between the normal of the (i,j)th plate and the incident sound ray; In the near field, the target surface integral I is calculated as follows: In the formula, u = 2ku', v = 2kv', w = w', and then the obtained acoustic fields of each plate are synthesized into the scattered field of the entire target where A is the acoustic scattering amplitude, is the distance from the geometric center of the plate element, and the normal of the plane is N is the number of plates on the target that can be illuminated by the incident wave, satisfying is the unit normal vector of the surface element.

3. The underwater target intensity prediction method based on neural network according to claim 1 is characterized in that: It also includes using the back-propagation algorithm to adjust the network weights and biases to minimize the prediction error.

4. The method for predicting underwater target strength based on a neural network according to any one of claims 1 to 3, characterized in that: The step of evaluating the prediction performance of the model also includes calculating the correlation coefficient, mean square error or absolute error between the predicted result and the actual result to quantify the prediction accuracy of the model.

Citation Information

Patent Citations

  • Target tracking systems and methods for uuv

    CA3067573A1

  • Target electromagnetic scattering characteristic rapid prediction method based on deep learning

    CN113177356A

  • Method for operational identification of marine targets by their information fields based on neuro-fuzzy models

    RU2763125C1