Sea clutter K distribution shape parameter neural network modeling method

Abstract

The invention discloses a sea clutter K distribution shape parameter neural network modeling method. The sea clutter K distribution shape parameter neural network modeling method comprises the following steps of 1, obtaining N groups of sea clutter data and marine environment parameter data under different measurement parameter conditions; 2, initializing a sea clutter K distribution shape parameter modeling data matrix into an empty matrix; step 3, calculating a mopping angle value corresponding to each range gate based on the nth group of sea clutter data; step 4, dividing the sea clutter data into a plurality of sub-data blocks according to the grazing angle value; the method further comprises other calculation steps. According to the modeling method disclosed by the invention, modeling is carried out by utilizing five parameters including the grazing angle, the radar spatial resolution unit area, the effective wave height, the average wave period and the wave direction, and the obtained model has better universality compared with an existing method; an empirical model nonlinear function term is combined with a BP neural network, and the obtained model is higher in precision compared with an existing method.

Description

Technical Field

[0001] The invention belongs to the field of radar clutter characteristics research, and particularly relates to a sea clutter K distribution shape parameter neural network modeling method in the field, which can be used for radar sea clutter characteristics analysis and modeling research. Background Art

[0002] Accurately understanding the characteristics of sea clutter is an important basis for sea radar system design, performance evaluation and target detection. Among them, the amplitude distribution characteristics are particularly important, which determines the design of the target detection algorithm and the setting of the constant false alarm detection threshold. Currently, the commonly used sea clutter amplitude distribution models include Rayleigh distribution, Weibull distribution, lognormal distribution and K distribution. With the improvement of radar resolution and the introduction of the composite Gaussian model, the K distribution model has been widely used. Relevant studies have shown that the K distribution model is more effective in describing sea clutter in high sea conditions. The effectiveness of the K distribution model is not only supported by experimental data and observational data, but also can be reasonably explained from the scattering mechanism of sea clutter. Therefore, the K distribution model is considered to be one of the most effective sea clutter amplitude distribution models.

[0003] In order to achieve the effective application of sea clutter characteristics, in addition to directly using the measured sea clutter data to estimate the sea clutter characteristic parameters, another method is to predict the sea clutter characteristic parameters based on radar parameters, wave parameters and radar observation geometric parameters, so as to determine the radar design scheme and evaluate the radar performance based on the radar working condition parameters. At present, the existing sea clutter K distribution shape parameter modeling method only considers the ground grazing angle, radar resolution unit irradiation area and wave direction angle, and uses the least squares fitting method to build the model. The model accuracy is low and it is difficult to meet engineering applications. Summary of the invention

[0004] The technical problem to be solved by the present invention is to study a method for modeling sea clutter K distribution shape parameters based on radar parameters, wave parameters and radar observation geometric parameters, and provide a method for modeling sea clutter K distribution shape parameters based on a neural network, which can realize sea clutter K distribution shape parameter modeling based on measured data.

[0005] The present invention adopts the following technical solution:

[0006] A sea clutter K distribution shape parameter neural network modeling method, the improvement of which is that it comprises the following steps:

[0007] Step 1: Obtain N groups of sea clutter data and ocean environment parameter data under different measurement parameter conditions:

[0008] The nth set of sea clutter data is represented by a complex matrix G represents the number of range gates, L represents the number of pulses, and the ocean environment parameter data corresponding to the nth group of sea clutter data includes the effective wave height Average wave cycle and wave angle n=1,2,...,N;

[0009] Step 2, initializing the sea clutter K distribution shape parameter modeling data matrix S to an empty matrix;

[0010] Step 3, based on the nth group of sea clutter data The radar slant range corresponding to each range gate is Calculate the ground-rubbing angle value corresponding to each range gate H represents the relative height between the radar and the sea surface, g = 1, 2, ..., G;

[0011] Step 4: According to the floor rubbing angle value Sea clutter data Divide into several sub-data blocks;

[0012] Step 5, based on the sea clutter sub-data block, calculate the corresponding sea clutter K distribution shape parameter, and construct a sea clutter K distribution shape parameter modeling sample;

[0013] Step 6: let n = 1, 2, ..., N, repeat steps 3 to 5 to form the sea clutter K distribution shape parameter modeling data matrix S F×6 ;

[0014] Step 7: Model the data matrix S based on the shape parameters of the sea clutter K distribution F×6 , obtain the input data X for modeling the shape parameters of sea clutter K distribution F×5 and the expected output y F×1 :

[0015] X F×5 (:,1:4)=log 10 (S F×6 (:,1:4)),

[0016] X F×5 (:,5)=cos(2×S F×6 (:,5)),

[0017] y F×1 =log 10 (S F×6 (:,6))

[0018] In the above formula, the ":" before the matrix "," means taking all rows, and the "1:4" after the "," means taking the 1st to 4th columns, 5 means taking the 5th column, and 6 means taking the 6th column;

[0019] Step 8: Construct a BP neural network using the input data X obtained in step 7 F×5 and the expected output y F×1 As training samples, the neural network model net of shape parameters of sea clutter K distribution is trained and generated;

[0020] Step 9, input the sea clutter K distribution shape parameter model input parameter vector into the neural network model net, and obtain the sea clutter K distribution shape parameter based on the network model output:

[0021] Let x represent the input parameter vector of the sea clutter K distribution shape parameter model, x = [log 10 (φ),log 10 (A),log 10 (h sw ),log 10 (p wa ),cos(2θ wa )], where 0<φ≤90, h sw >0, p wa >0,0≤θ wa ≤180, x is used as the input of the sea clutter K distribution shape parameter neural network model net, and the network output is y. The sea clutter K distribution shape parameter obtained based on the model is 10 y .

[0022] Furthermore, step 4 specifically includes the following steps:

[0023] Step 41, setting the floor-rubbing angle grouping interval, assuming that the floor-rubbing angle value interval is [0.1, 90], dividing it into Q = 899 sub-intervals at intervals of 0.1, and the value range of the qth sub-interval is [0.1q, 0.1(q+1)];

[0024] Step 42: According to the floor rubbing angle value Calculate the corresponding floor-rubbing angle subinterval [0.1q1,0.1(q2+1)]:

[0025]

[0026] Among them, ceil(·) means rounding up, and floor(·) means rounding down;

[0027] Step 43: Initialize the sub-data block q=q1,q1+1,...,q2;

[0028] Step 44: According to the floor rubbing angle value Let g=1,2,...,G in turn. If If it falls into the qth ground-rubbing angle sub-interval, the data of the range gate will be merged into the sub-data block that is in Representation Matrix The data in row g.

[0029] Furthermore, step 5 specifically includes the following steps:

[0030] Step 51: Sea clutter sub-data block The D elements in are denoted as {x1,...,x d ,...,x D}, calculate the shape parameter v of the sea clutter K distribution:

[0031]

[0032] in, i=1 or 2

[0033] Step 52, the sea clutter K distribution shape parameter modeling sample is represented as a vector

[0034] φ=0.1q+0.05,

[0035] A=θπHΔR / sin(φ) / cos(φ)

[0036] In the above formula, θ is the radar azimuth beamwidth, and ΔR is the radar range resolution;

[0037] Step 53, counting the number of rows F of the current matrix S, and taking the vector s as the F+1th row element of the matrix S;

[0038] Step 54, let q=q1, q1+1, ..., q2 in sequence, and repeat steps 51 to 53.

[0039] Furthermore, step 8 specifically includes the following steps:

[0040] Step 81, the neural network input layer includes five inputs, the output layer includes one output, the network structure is double hidden layer, the hidden layer uses the hyperbolic tangent function as the activation function, the output layer uses the linear function, the learning method uses the weight or threshold learning function of the gradient descent with momentum, the training function uses the Levenberg-Marquardt optimization algorithm, and the performance function is the mean square error function;

[0041] Step 82, set the learning rate, the maximum number of training times is not less than 1000 times, the expected error value is 0.001 to 0.00001, the number of neurons in each hidden layer is an exponential power of 2, and use the input data X F×5 and the expected output y F×1 The sea clutter K distribution shape parameter neural network model net is generated as training samples.

[0042] The beneficial effects of the present invention are:

[0043] The modeling method disclosed in the present invention uses five parameters, namely, ground scraping angle, radar spatial resolution unit area, effective wave height, average wave period and wave direction, to perform modeling. The obtained model has better universality than the existing method. The nonlinear function term of the empirical model is combined with the BP neural network. The obtained model has higher accuracy than the existing method. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 It is a schematic diagram of the flow chart of the modeling method disclosed in the present invention;

[0045] Figure 2 This is a diagram of the prediction results of the sea clutter K distribution shape parameter model obtained based on the existing method;

[0046] Figure 3 It is a diagram of the prediction results of the sea clutter K distribution shape parameter model obtained based on the modeling method of the present invention. DETAILED DESCRIPTION

[0047] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is 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.

[0048] Embodiment 1, this embodiment discloses a neural network modeling method for sea clutter K distribution shape parameters, such as Figure 1 As shown, the following steps are included:

[0049] Step 1: Obtain N groups of sea clutter data and ocean environment parameter data under different measurement parameter conditions:

[0050] The nth set of sea clutter data is represented by a complex matrix G represents the number of range gates, L represents the number of pulses, and the ocean environment parameter data corresponding to the nth group of sea clutter data includes the effective wave height Average wave cycle and wave angle n=1,2,...,N;

[0051] Step 2, initializing the sea clutter K distribution shape parameter modeling data matrix S to an empty matrix;

[0052] Step 3, based on the nth group of sea clutter data The radar slant range corresponding to each range gate is Calculate the ground-rubbing angle value corresponding to each range gate H represents the relative height between the radar and the sea surface, g = 1, 2, ..., G;

[0053] Step 4: According to the floor rubbing angle value Sea clutter data Divide into several sub-data blocks;

[0054] Step 4 specifically includes the following steps:

[0055] Step 41, setting the floor-rubbing angle grouping interval, assuming that the floor-rubbing angle value interval is [0.1, 90], dividing it into Q = 899 sub-intervals at intervals of 0.1, and the value range of the qth sub-interval is [0.1q, 0.1(q+1)];

[0056] Step 42: According to the floor rubbing angle value Calculate the corresponding floor-rubbing angle subinterval [0.1q1,0.1(q2+1)]:

[0057]

[0058] Among them, ceil(·) means rounding up, and floor(·) means rounding down;

[0059] Step 43: Initialize the sub-data block q=q1,q1+1,...,q2;

[0060] Step 44: According to the floor rubbing angle value Let g=1,2,...,G in turn. If If it falls into the qth ground-rubbing angle sub-interval, the data of the range gate will be merged into the sub-data block that is in Representation Matrix The data in row g.

[0061] Step 5: Based on the sea clutter sub-data block Calculate the corresponding sea clutter K distribution shape parameters and construct a sea clutter K distribution shape parameter modeling sample;

[0062] Step 5 specifically includes the following steps:

[0063] Step 51: Sea clutter sub-data block The D elements in are denoted as {x1,...,x d ,...,x D}, calculate the shape parameter v of the sea clutter K distribution:

[0064]

[0065] in, i=1 or 2

[0066] Step 52, the sea clutter K distribution shape parameter modeling sample is represented as a vector

[0067] φ=0.1q+0.05,

[0068] A=θπHΔR / sin(φ) / cos(φ)

[0069] In the above formula, θ is the radar azimuth beamwidth, and ΔR is the radar range resolution;

[0070] Step 53, counting the number of rows F of the current matrix S, and taking the vector s as the F+1th row element of the matrix S;

[0071] Step 54, let q=q1, q1+1, ..., q2 in sequence, and repeat steps 51 to 53.

[0072] Step 6: let n = 1, 2, ..., N, repeat steps 3 to 5 to form the sea clutter K distribution shape parameter modeling data matrix S F×6 ;

[0073] Step 7: Model the data matrix S based on the shape parameters of the sea clutter K distribution F×6 , obtain the input data X for modeling the shape parameters of sea clutter K distribution F×5 and the expected output y F×1 :

[0074] X F×5 (:,1:4)=log 10 (S F×6 (:,1:4)),

[0075] X F×5 (:,5)=cos(2×S F×6 (:,5)),

[0076] y F×1 =log 10 (S F×6 (:,6))

[0077] In the above formula, the ":" before the matrix "," means taking all rows, and the "1:4" after the "," means taking the 1st to 4th columns, 5 means taking the 5th column, and 6 means taking the 6th column;

[0078] Step 8: Construct a BP neural network using the input data X obtained in step 7 F×5 and the expected output y F×1 As training samples, the neural network model net of shape parameters of sea clutter K distribution is trained and generated;

[0079] Step 8 specifically includes the following steps:

[0080] Step 81, the neural network input layer includes five inputs, the output layer includes one output, the network structure is double hidden layers, the hidden layer uses the hyperbolic tangent function (tansig) as the activation function, the output layer uses the linear function (purelin), the learning method uses the weight / threshold learning function (learngdm) with gradient descent with momentum, the training function uses the Levenberg-Marquardt optimization algorithm (trainlm), and the performance function is the mean square error function;

[0081] Step 82, set the learning rate, the maximum number of training times is not less than 1000 times, the expected error value is 0.001 to 0.00001, the number of neurons in each hidden layer is generally an exponential power of 2 (for example, 16, 32, 64, etc.), and use the input data X F×5 and the expected output y F×1 The sea clutter K distribution shape parameter neural network model net is generated as training samples.

[0082] Step 9, input the sea clutter K distribution shape parameter model input parameter vector into the neural network model net, and obtain the sea clutter K distribution shape parameter based on the network model output:

[0083] Let x represent the input parameter vector of the sea clutter K distribution shape parameter model, x = [log 10 (φ),log 10 (A),log 10 (h sw ),log 10 (p wa ),cos(2θ wa )], where 0<φ≤90, h sw >0, p wa >0,0≤θ wa ≤180, x is used as the input of the sea clutter K distribution shape parameter neural network model net, and the network output is y. The sea clutter K distribution shape parameter obtained based on the model is 10 y .

[0084] The effect of the modeling method disclosed in the present invention is further illustrated by the following experimental results:

[0085] Using the sea clutter data measured by a shore-based X-band radar and the ocean environment parameter data measured by a wave buoy, according to steps 2 to 6 of the modeling method of the present invention, a data matrix S for modeling the shape parameters of the sea clutter K distribution is formed. F×6 , where F = 47772, further based on the data matrix S F×6 , according to steps 7 to 9, the sea clutter K distribution shape parameter neural network model is obtained.

[0086] Compared with the existing method, the average relative error of the sea clutter K distribution shape parameter model obtained by the modeling method of the present invention is reduced by 41%. Figure 2 This is a diagram showing the prediction results of the sea clutter K distribution shape parameter model obtained based on the existing method. Figure 3 The figure is a prediction result diagram of the sea clutter K distribution shape parameter model obtained based on the modeling method of the present invention. Figure 2 and Figure 3 It can be seen that Figure 3 The error between the predicted value and the true value is significantly smaller than Figure 2 , which shows the effectiveness of the model obtained by the modeling method of the present invention.

Claims

1. A sea clutter K distribution shape parameter neural network modeling method, characterized in that: The steps include: Step 1: Obtain N groups of sea clutter data and ocean environment parameter data under different measurement parameter conditions: The nth set of sea clutter data is represented by a complex matrix G represents the number of range gates, L represents the number of pulses, and the ocean environment parameter data corresponding to the nth group of sea clutter data includes the effective wave height Average wave cycle and wave angle Step 2, initializing the sea clutter K distribution shape parameter modeling data matrix S to an empty matrix; Step 3, based on the nth group of sea clutter data The radar slant range corresponding to each range gate is Calculate the ground-rubbing angle value corresponding to each range gate H represents the relative height between the radar and the sea surface, g = 1, 2, ..., G; Step 4: According to the floor rubbing angle value Sea clutter data Divide into several sub-data blocks; Step 5, based on the sea clutter sub-data block, calculate the corresponding sea clutter K distribution shape parameter, and construct a sea clutter K distribution shape parameter modeling sample; Step 6: let n = 1, 2, ..., N, repeat steps 3 to 5 to form the sea clutter K distribution shape parameter modeling data matrix S F×6 ; Step 7: Model the data matrix S based on the shape parameters of the sea clutter K distribution F×6 , obtain the input data X for modeling the shape parameters of sea clutter K distribution F×5 and the expected output y F×1 : X F×5 (:,1:4)=log 10 (S F×6 (:,1:4)), X F×5 (:,5)=cos(2×S F×6 (:,5)), y F×1 =log 10 (S F×6 (:,6)) In the above formula, the ":" before the "" in the matrix means taking all rows, the 1:4 after the "" means taking the 1st to 4th columns, the 5 means taking the 5th column, and the 6 means taking the 6th column; Step 8: Construct a BP neural network using the input data X obtained in step 7 F×5 and the expected output y F×1 As training samples, the neural network model net of shape parameters of sea clutter K distribution is trained and generated; Step 9, input the sea clutter K distribution shape parameter model input parameter vector into the neural network model net, and obtain the sea clutter K distribution shape parameter based on the network model output: Let x represent the input parameter vector of the sea clutter K distribution shape parameter model, x = [log 10 (φ),log 10 (A),log 10 (h sw ),log 10 (p wa ),cos(2θ wa )], where 0<φ≤90, h sw >0, p wa >0,0≤θ wa ≤180, x is used as the input of the sea clutter K distribution shape parameter neural network model net, and the network output is y. The sea clutter K distribution shape parameter obtained based on the model is 10 y .

2. The sea clutter K distribution shape parameter neural network modeling method according to claim 1, characterized in that: Step 4 specifically includes the following steps: Step 41, setting the floor-rubbing angle grouping interval, assuming that the floor-rubbing angle value interval is [0.1, 90], dividing it into Q = 899 sub-intervals at intervals of 0.1, and the value range of the qth sub-interval is [0.1q, 0.1(q+1)]; Step 42: According to the floor rubbing angle value Calculate the corresponding floor-rubbing angle subinterval [0.1q1,0.1(q2+1)]: Among them, ceil(·) means rounding up, and floor(·) means rounding down; Step 43: Initialize the sub-data block Step 44: According to the floor rubbing angle value Let g=1,2,...,G in turn. If If it falls into the qth ground-rubbing angle sub-interval, the data of the range gate will be merged into the sub-data block that is in Representation Matrix The data in row g.

3. The sea clutter K distribution shape parameter neural network modeling method according to claim 2, characterized in that: Step 5 specifically includes the following steps: Step 51: Sea clutter sub-data block The D elements in are denoted as {x1,...,x d ,...,x D }, calculate the shape parameter v of the sea clutter K distribution: in, Step 52, the sea clutter K distribution shape parameter modeling sample is represented as a vector φ=0.1q+0.05, A=θπHΔR / sin(φ) / cos(φ) In the above formula, θ is the radar azimuth beamwidth, and ΔR is the radar range resolution; Step 53, counting the number of rows F of the current matrix S, and taking the vector s as the F+1th row element of the matrix S; Step 54, let q=q1, q1+1, ..., q2 in sequence, and repeat steps 51 to 53.

4. The sea clutter K distribution shape parameter neural network modeling method according to claim 3, characterized in that: Step 8 specifically includes the following steps: Step 81, the neural network input layer includes five inputs, the output layer includes one output, the network structure is double hidden layer, the hidden layer uses the hyperbolic tangent function as the activation function, the output layer uses the linear function, the learning method uses the weight or threshold learning function of the gradient descent with momentum, the training function uses the Levenberg-Marquardt optimization algorithm, and the performance function is the mean square error function; Step 82, set the learning rate, the maximum number of training times is not less than 1000 times, the expected error value is 0.001 to 0.00001, the number of neurons in each hidden layer is an exponential power of 2, and use the input data X F×5 and the expected output y F×1 The sea clutter K distribution shape parameter neural network model net is generated as training samples.

Images