K-distribution sea clutter shape parameter estimation method based on neural network and sample quantile ratio

Abstract

The present invention relates to the field of radar signal processing technology, and in particular to a K-distribution sea clutter shape parameter estimation method based on a neural network and a sample quantile ratio, wherein a neural network model is established for training, and then the logarithm of the two sample quantile ratios is input into the neural network model, thereby reducing the computational complexity while ensuring the estimation accuracy of the K-distribution sea clutter shape parameter, and not relying on table lookup. Furthermore, the estimated K-distribution sea clutter shape parameter can be applied to a radar target detection method under a sea clutter background, thereby improving the radar detection performance of targets in the K-distribution sea clutter.

Description

Technical Field

[0001] The invention relates to the technical field of radar signal processing, and in particular to a K-distribution sea clutter shape parameter estimation method based on a neural network and a sample quantile ratio. Background Art

[0002] Radar is a powerful weapon for early warning and detection of sea surface targets. However, electromagnetic backscattering (called sea clutter) on the sea surface seriously affects the detection performance of radar for sea surface targets. The improvement of sea conditions and the improvement of radar resolution will cause obvious non-Gaussian characteristics of sea clutter. The radar target detection algorithm under the Gaussian model will suffer serious performance loss at this time. In order to accurately model non-Gaussian sea clutter, some scholars have proposed a composite Gaussian model, which mathematically represents sea clutter as the product of texture component and speckle component. The earliest and most widely used distribution proposed under the composite Gaussian model is the K distribution in which the texture component obeys the gamma distribution. The K distribution model contains shape parameters that control the non-Gaussianity of sea clutter, which determines the working performance of the radar target detection algorithm under the background of K-distributed sea clutter. Therefore, it is of great significance to study the estimation method of shape parameters of K-distributed sea clutter.

[0003] The literature “IRJoughin, DBPercival, and DPWinebrenner, “Maximum likelihood estimation of K distribution parameters for SAR data,” IEEE Transactions on Geoscience and Remote Sensing, vol.31, no.5. pp.989–999, Sep.1993. doi:10.1109 / 36.263769.” proposed a maximum likelihood estimation method for K distribution parameters. The literature “DRIskander, AMZoubir, and B.Boashash, “A Method for estimating the parameters of the K distribution,” IEEE Transactions on Signal Processing, vol.47, no.4, pp.1147–1151, 1999.” studied the moment estimation method for K distribution parameters. However, when abnormal sample echoes such as targets and islands and reefs appear in sea clutter, the estimation accuracy of the maximum likelihood estimation method and the moment estimation method will drop rapidly.

[0004] In order to solve the problem of estimating the shape parameters of K-distributed sea clutter in the presence of abnormal sample echoes, the patent "K-distributed sea clutter shape parameter estimation method based on neural network, CN107247259A" proposes to use 25 quantile samples and neural network to estimate parameters, but the estimation accuracy of the method proposed in this patent is insufficient in the absence of abnormal samples. In addition, the patent "Three-point point parameter estimation method based on K-distributed sea clutter amplitude model, CN109143196A" proposes the estimation of K-distributed sea clutter shape parameters based on sample quantile ratios, but this patent relies on table lookup, and the computational cost will gradually increase as the table size increases.

[0005] None of the methods proposed in the above literature can guarantee accurate estimation of K-distributed sea clutter shape parameters in pure environments and environments with abnormal samples while also having low computational complexity. Summary of the invention

[0006] The present invention overcomes the shortcomings of the prior art and provides a K-distribution sea clutter shape parameter estimation method based on a neural network and a sample quantile ratio. The present invention reduces the complexity of the K-distribution sea clutter shape parameter estimation method while ensuring the estimation accuracy of the K-distribution sea clutter shape parameter. At the same time, the estimated values ​​of the K-distribution sea clutter shape parameters obtained by the present invention are applied to a radar target detection method under a sea clutter background, thereby improving the detection performance of the radar for the K-distribution sea clutter.

[0007] The technical problem solved by the present invention can be achieved by adopting the following technical solutions:

[0008] The K-distribution sea clutter shape parameter estimation method based on neural network and sample quantile ratio includes the following methods

[0009] S1: Calculate the K distribution sea clutter amplitude value r at the αth and βth quantiles when the K distribution shape parameter is ν. ν,α and r ν,β , where r is the amplitude value of K-distributed sea clutter, and ν is a real number greater than zero;

[0010] S2: The sea clutter amplitude value obtained from K distribution in S1 is r ν,α The amplitude of K-distributed sea clutter is r ν,β The logarithm of the ratio between the two values ​​is calculated as

[0011] S3: Create a logarithm based on the ratio of quantiles of sea clutter samples The neural network model with shape parameter ν as input and output is based on the logarithm of the ratio of the quantiles of the sea clutter samples in S2. For training data Using this training data and the value of shape parameter ν to train the neural network model;

[0012] S4: Extract the actual K-distributed sea clutter amplitude values ​​at the αth and βth quantiles from the actual radar sea clutter data and the actual K-distributed sea clutter amplitude value Among them is is the actual K-distributed sea clutter amplitude value, calculate the actual K-distributed sea clutter amplitude value and the actual K-distributed sea clutter amplitude value The logarithm of the ratio between two values ​​of

[0013] S5: Logarithm of the actual sea clutter sample quantile ratio obtained in S4 Input the trained neural network model in S3 to obtain the estimated value of the shape parameter of K-distributed sea clutter

[0014] S6: Estimated values ​​of the shape parameters of the K-distributed sea clutter obtained in S5 This estimate It is applied to radar target detection methods under sea clutter background to improve the radar detection performance of targets in K-distributed sea clutter.

[0015] Furthermore, in S1, the K distribution sea clutter amplitude values ​​r at the αth and βth quantiles are calculated according to the cumulative probability distribution function of the K distribution when the K distribution shape parameter is ν. ν,α and r ν,β , where the value range of α is: 0<α<100, the value range of β is: 0<β<100, where β>α.

[0016] Furthermore, the K distribution sea clutter amplitude value r at the αth and βth quantiles is calculated based on the cumulative probability distribution function of the K distribution when the K distribution shape parameter is ν. ν,α and r ν,β Also includes the following methods:

[0017] S01: It is known that the value of the shape parameter ν of the K distribution is a specific real value, and r is set respectively ν,α and r ν,β An initial known value and Let the cumulative probability values ​​of the cumulative probability distribution function of K distribution be α% and β% respectively;

[0018] S02: The known value of the K distribution shape parameter ν in S01 and the known value and The value of is brought into the cumulative probability distribution function of the K distribution, and the cumulative probability values ​​of the cumulative probability distribution function of the K distribution at this time are obtained as α(0) % and β (0) %;

[0019] S03: The cumulative probability value α of the cumulative probability distribution function of the K distribution obtained in S02 (0) % and β (0) α in % (0) and β (0) Compared with α and β in S1, according to α (0) With α, β (0) The size between β and and The size of the value, and then adjust the new and new Substitute the value into the cumulative probability distribution function of the K distribution to obtain the cumulative probability value of the new cumulative probability distribution function of the K distribution, the new α (0) % and new β (0) %, until the new α (0) =α, new β (0) =β, the new and new That is, when the shape parameter of the K distribution in S1 is ν, the K distribution sea clutter amplitude value r at the αth and βth quantiles ν,α and r ν,β .

[0020] Furthermore, the cumulative probability distribution function formula of the K distribution is:

[0021]

[0022] Where r represents the sea clutter amplitude, ν represents the shape parameter of the K distribution, μ represents the scale parameter of the K distribution. Since the solution of the shape parameter is independent of the scale parameter, the scale parameter is set to 1, that is, μ = 1, Γ(·) represents the gamma function, and K(·) represents the second kind of modified Bessel function;

[0023] In S01, it is known that the shape parameter of the K distribution is ν. Let the cumulative probability values ​​of the cumulative probability distribution function of the K distribution be α% and β% respectively. Substitute them into the cumulative probability distribution function formula (1) of the K distribution to obtain the cumulative probability distribution function formula (2) of the K distribution as follows:

[0024]

[0025] where ν represents the shape parameter of the K distribution, Γ(·) represents the gamma function, and r ν,α represents the amplitude value corresponding to the αth quantile of the K-distributed sea clutter when the shape parameter is ν, r ν,β It represents the amplitude value corresponding to the βth quantile of K-distributed sea clutter when the shape parameter is ν;

[0026] In S02, the value of the known K distribution shape parameter ν in S01 and the known value and Substitute the value of into the cumulative probability distribution function formula (2) of the K distribution, and obtain formula (3) as follows:

[0027]

[0028] The new and Substitute the value into formula (3) to obtain the cumulative probability value of the new K distribution cumulative probability distribution function new α (0) % and new β (0) %.

[0029] Further, in the S03, according to α (0) With α, β (0) The size between β and and The value size includes the following methods:

[0030] S001: When α (0) When it is greater than α, it decreases Lingxin

[0031] S002: When α (0) When it is less than α, it increases Lingxin

[0032] S003: When β (0) When it is greater than β, it decreases Lingxin

[0033] S004: When β (0) When it is less than β, it increases Lingxin

[0034] Furthermore, the neural network model includes an input layer, a hidden layer and an output layer, the hidden layer is not less than two layers, and the number of neurons in the input layer and the output layer is 1.

[0035] Furthermore, the logarithm of the ratio of the quantiles of the sea clutter samples in S2 is For training data Also includes the following methods:

[0036] When ν takes a number of real numbers greater than zero, the logarithm of the ratio of the quantiles of the sea clutter samples obtained in S2 is a number of In S3, there are several As a data set, then several The data is arranged in random order, and finally several The data are randomly divided into neural network training data for neural network model training and neural network verification data for neural network model verification, wherein the number of neural network training data is greater than the number of neural network verification data.

[0037] Furthermore, the K-distributed sea clutter amplitude value obtained in S1 is calculated in S2 as r ν,α The amplitude of K-distributed sea clutter is r ν,β The logarithm of the ratio is then taken to obtain the logarithm of the sea clutter sample quantile ratio: in

[0038] Furthermore, the actual K-distributed sea clutter amplitude value is calculated in S4 and the actual K-distributed sea clutter amplitude value The ratio between the two values ​​of , and then take the logarithm of the ratio, and get the logarithm of the actual sea clutter sample quantile ratio: in

[0039] The beneficial effects of the present invention are:

[0040] Compared with the prior art, the present invention trains by establishing a neural network model and then inputting the logarithm of the ratio of two sample quantiles into the neural network model, thereby reducing the computational complexity while ensuring the estimation accuracy of the shape parameters of K-distributed sea clutter, and does not rely on table lookup. Furthermore, the estimated shape parameters of K-distributed sea clutter can be applied to the radar target detection method under the background of sea clutter, so as to improve the detection performance of the radar for targets in K-distributed sea clutter. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] The present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0042] Figure 1 It is a schematic diagram of the implementation process of the present invention.

[0043] Figure 2 The figure is a schematic diagram of shape parameter estimation error curves for K-distributed sea clutter data using the method of the present invention and the existing three-point point estimation method, first, second and third order moment estimation method and second and fourth order moment estimation method.

[0044] Figure 3 It is a schematic diagram of shape parameter estimation errors of K-distributed sea clutter data containing 2% outliers by the method of the present invention and the existing second and fourth order moment estimation methods, first, second and third order moment estimation methods and three-point point estimation methods. DETAILED DESCRIPTION

[0045] The following is a detailed description of the K-distribution sea clutter shape parameter estimation method based on a neural network and sample quantile ratios provided by the embodiment of the present invention through several specific embodiments.

[0046] Reference Figure 1 As shown, the K-distribution sea clutter shape parameter estimation method based on neural network and sample quantile ratio includes the following methods

[0047] S1: Calculate the K distribution sea clutter amplitude value r at the αth and βth quantiles when the K distribution shape parameter is ν. ν,α and r ν,β , where r is the amplitude value of K-distributed sea clutter, and ν is a real number greater than zero;

[0048] In the above method S1 of the present invention, the K distribution sea clutter amplitude value r at the αth and βth quantiles is calculated respectively according to the cumulative probability distribution function of the K distribution when the K distribution shape parameter is ν. ν,α and r ν,β , where the value range of α is: 0<α<100, the value range of β is: 0<β<100, where β>α;

[0049] Among them, the K distribution sea clutter amplitude value r at the αth and βth quantiles is calculated according to the cumulative probability distribution function of the K distribution when the K distribution shape parameter is ν. ν,α and r ν,β Also includes the following methods:

[0050] S01: It is known that the value of the shape parameter ν of the K distribution is a specific real value, and r is set respectively ν,α and r ν,β An initial known value and Let the cumulative probability values ​​of the cumulative probability distribution function of K distribution be α% and β% respectively;

[0051] Among them, the cumulative probability distribution function formula of K distribution is:

[0052]

[0053] Where r represents the sea clutter amplitude, ν represents the shape parameter of the K distribution, μ represents the scale parameter of the K distribution. Since the solution of the shape parameter is independent of the scale parameter, the scale parameter is set to 1, that is, μ = 1, Γ(·) represents the gamma function, and K(·) represents the second kind of modified Bessel function;

[0054] In S01, the shape parameter of K distribution is known to be ν. Let the cumulative probability values ​​of the cumulative probability distribution function of K distribution be α% and β% respectively. Substitute them into the cumulative probability distribution function formula (1) of K distribution, and the cumulative probability distribution function formula (2) of K distribution is obtained as follows:

[0055]

[0056] where ν represents the shape parameter of the K distribution, Γ(·) represents the gamma function, and r ν,α represents the amplitude value corresponding to the αth quantile of the K-distributed sea clutter when the shape parameter is ν, r ν,β It represents the amplitude value corresponding to the βth quantile of K-distributed sea clutter when the shape parameter is ν;

[0057] S02: The known value of the K distribution shape parameter ν in S01 and the known value and The value of is brought into the cumulative probability distribution function of the K distribution, and the cumulative probability values ​​of the cumulative probability distribution function of the K distribution at this time are obtained as α (0) % and β (0) %;

[0058] In S02, the value of the known K distribution shape parameter ν in S01 and the known value and Substitute the value of into the cumulative probability distribution function formula (2) of the K distribution, and obtain formula (3) as follows:

[0059]

[0060] S03: The cumulative probability value α of the cumulative probability distribution function of the K distribution obtained in S02 (0) % and β (0) α in % (0) and β (0) Compared with α and β in S1, according to α (0) With α, β (0) The size between β and and The size of the value, and then adjust the new and new Substitute the value into the cumulative probability distribution function of the K distribution to obtain the cumulative probability value of the new cumulative probability distribution function of the K distribution, the new α (0) % and new β (0) %, specifically the adjusted new and Substitute the value into formula (3) to obtain the cumulative probability value of the new K distribution cumulative probability distribution function new α (0) % and new β (0) %; until the new α (0) =α, new β (0) =β, the new and new That is, when the shape parameter of the K distribution in S1 is ν, the K distribution sea clutter amplitude value r at the αth and βth quantiles ν,α and rν,β .

[0061] In S03, according to α (0) With α, β (0) The size between β and and The value size includes the following methods:

[0062] S001: When α (0) When it is greater than α, it decreases Lingxin

[0063] S002: When α (0) When it is less than α, it increases Lingxin

[0064] S003: When β (0) When it is greater than β, it decreases Lingxin

[0065] S004: When β (0) When it is less than β, it increases Lingxin

[0066] It should be noted that the brackets in the formulas from S001 to S004 above and These are the values ​​after each adjustment.

[0067] Furthermore, in S03, the cumulative probability value α of the cumulative probability distribution function of the K distribution obtained in S02 is (0) % and β (0) α in % (0) and β (0) Compared with α and β in S1, the following methods are also included:

[0068] Set when α (0) With α, β (0) The absolute value of the difference between α and β (0) -α| and |β (0) -β| are both less than m, m is taken as the minimum value set, and the new and new That is, when the shape parameter of the K distribution in S1 is ν, the K distribution sea clutter amplitude value r at the αth and βth quantiles ν,α and r ν,β ;

[0069] When the actual ν value range is 0.1, 0.2, ..., 20, and the interval of each number is 0.01, m is set to 10 -6 ; Therefore, when α(0) The absolute value of the difference with α|α (0) -α|less than 10 -6 , then new

[0070] When α (0) The difference α from α (0) -α is greater than 10 -6 , then reduce the new

[0071] Lingxin Then the adjusted new Substitute into formula (3) and obtain the new α (0) %;

[0072] When α (0) The difference α from α (0) -α is less than -10 -6 , then increase the new Lingxin Then the adjusted new Substitute into formula (3) and obtain the new α (0) %;

[0073] When β (0) The absolute value of the difference with β|β (0) -β|less than 10 -6 , then new

[0074] When β (0) The difference between β and (0) -β greater than 10 -6 , then reduce the new Lingxin Then the adjusted new Substitute into formula (3) and obtain the new β (0) %;

[0075] When β (0) The difference between β and (0) -β is less than 10 -6 , then increase the new Lingxin Then the adjusted new Substitute into formula (3) and obtain the new β (0) %.

[0076] S2: The sea clutter amplitude value obtained from K distribution in S1 is r ν,α The amplitude of K-distributed sea clutter is r ν,β Calculate the ratio between the two values, and then take the logarithm of the ratio to get the logarithm of the sea clutter sample quantile ratio: in

[0077]

[0078] S3: Create a logarithm based on the ratio of quantiles of sea clutter samples The neural network model with shape parameter ν as input and output is based on the logarithm of the ratio of the quantiles of the sea clutter samples in S2. For training data Using this training data and the value of shape parameter ν to train the neural network model;

[0079] The neural network model includes an input layer, a hidden layer and an output layer, wherein the hidden layer is not less than two layers, and the number of neurons in the input layer and the output layer is 1; when the present invention performs actual K-distributed sea clutter shape parameter estimation, the hidden layer is two layers, the first layer uses 32 neurons, and the second layer uses 16 neurons, which can meet the estimation of the K-distributed sea clutter shape parameters of the present invention, so that the estimated value is accurate; the number of neurons in the input layer is 1, which is used to input the logarithm of the quantile ratio of the sea clutter samples The number of neurons in the output layer is 1, which is used to output the estimated value of the shape parameters of the K-distributed sea clutter.

[0080] The logarithm of the ratio of the sea clutter sample quantiles in S2 in S3 is For training data Also includes the following methods:

[0081] When ν takes a number of real numbers greater than zero, the logarithm of the ratio of the quantiles of the sea clutter samples obtained in S2 is a number of In S3, there are several As a data set, then several The data is arranged in random order, and finally several The data is divided into neural network training data for neural network model training and neural network verification data for neural network model data verification, wherein the number of neural network training data is greater than the number of neural network verification data. In the actual application of the present invention, 80% of the data set is selected as neural network training data, and 20% is selected as neural network verification data, thereby realizing a training and verification mode, avoiding excessive fitting of the neural network model, and causing inaccurate final results.

[0082] The transfer function used in the hidden layer of the neural network model in the present invention is a linear rectification function, the transfer function used in the output layer is a linear function, the loss function is also provided in the neural network model as a relative root mean square error function, and the optimization algorithm is also provided in the neural network model as an adaptive moment estimation method. The linear rectification function, linear function, relative root mean square error function and adaptive moment estimation method mentioned in the present invention are all prior arts.

[0083] S4: Extract the actual K-distributed sea clutter amplitude values ​​at the αth and βth quantiles from the actual radar sea clutter data and the actual K-distributed sea clutter amplitude value Among them is is the actual K-distributed sea clutter amplitude value, calculate the actual K-distributed sea clutter amplitude value and the actual K-distributed sea clutter amplitude value The ratio between the two values ​​of , and then take the logarithm of the ratio to get the logarithm of the actual sea clutter sample quantile ratio: in

[0084] S5: Logarithm of the actual sea clutter sample quantile ratio obtained in S4 Input the trained neural network model in S3 to obtain the estimated value of the shape parameter of K-distributed sea clutter

[0085] S6: Estimated values ​​of the shape parameters of the K-distributed sea clutter obtained in S5 The estimated value It is applied to radar target detection methods under sea clutter background, and the test statistics of radar target detection methods are calculated to improve the detection performance of radar for targets in K-distributed sea clutter.

[0086] The radar target detection method described in S6 is as follows:

[0087] Where α represents the test statistic of the radar target detection method, M represents the number of accumulated coherent pulses, z represents the radar echo data within a radar coherent processing interval, R represents the sea clutter covariance matrix, p represents the normalized steering vector of the radar target, and ν represents the shape parameter of the K-distributed sea clutter;

[0088] The obtained K-distribution sea clutter shape parameters based on neural network and sample quantile ratio are Substitute into the equation to replace ν, and obtain the test statistic of the radar target detection method. The estimated value of the K-distributed sea clutter shape parameter obtained by the method of the present invention is The test statistic obtained by applying it to the radar target detection method is more accurate, thus further improving the radar's detection performance for targets in sea clutter.

[0089] The effect of the present invention is further described below in conjunction with simulation experiments.

[0090] 1. Simulation parameters

[0091] Matlab software is used to simulate K-distributed sea clutter data. The length of sea clutter data is N=5000, the scale parameter is 1, and the shape parameter ν increases from 0.01 to 20, with a step size of 0.01. Each shape parameter is estimated 10,000 times. The quantile order β=95, the quantile order α=-0.22 (β / 100) 2 +0.42(β / 100)-0.03.

[0092] First, the generated pure sea clutter data is used to perform shape parameter estimation experiments. Then, 2% abnormal samples with uniform intensity distribution in the interval [10,100] are added to the generated sea clutter data, and the shape parameter estimation experiment is performed again.

[0093] 2. Simulation experiment content

[0094] In the simulation experiment, the present invention uses the first, second and third order moment estimation method, the second and fourth order moment estimation method and the three-point point estimation method to estimate the shape parameters of the sea clutter data. Figure 2 The relative root mean square error curve of the estimation results is drawn in .

[0095] The experimental results are as follows Figure 2 and Figure 3 As shown, the horizontal axis represents the true value of the shape parameter, and the vertical axis represents the relative root mean square error corresponding to the estimation method.

[0096] from Figure 2 It can be seen that under the pure sea clutter background, the relative root mean square errors of the method of the present invention and the three-point estimation method are close, and the method of the present invention has good estimation performance.

[0097] from Figure 3 It can be seen that in the presence of abnormal samples, the estimation error of the method of the present invention is better than that of the three-point estimation method and other methods. The three-point estimation method needs to rely on table lookup, and when the size of the table increases, its computational cost will increase rapidly. However, the method of the present invention does not rely on table lookup and has low computational complexity.

[0098] The above is a detailed description of the embodiments of the present invention in conjunction with the accompanying drawings, but the present invention is not limited to the above embodiments. Within the knowledge scope of ordinary technicians in the field, various changes can be made without departing from the purpose of the present invention, which are all within the protection scope of this technology.

[0099] The technical solutions between the various embodiments can be combined with each other, but they must be based on the fact that they can be implemented by ordinary technicians in the field. When the combination of technical solutions is contradictory or cannot be implemented, it should be deemed that such combination of technical solutions does not exist and is not within the scope of protection required by the present invention.

Claims

1. A K-distribution sea clutter shape parameter estimation method based on neural network and sample quantile ratio, characterized by: Includes the following methods S1: Calculate the K distribution sea clutter amplitude value r at the αth and βth quantiles when the K distribution shape parameter is ν. ν,α and r ν,β , where r is the amplitude value of K-distributed sea clutter, and ν is a real number greater than zero; S2: The sea clutter amplitude value obtained from K distribution in S1 is r ν,α The amplitude of K-distributed sea clutter is r ν,β The logarithm of the ratio between the two values ​​of S3: Create a logarithm based on the ratio of quantiles of sea clutter samples The neural network model with shape parameter ν as input and output is based on the logarithm of the ratio of the quantiles of the sea clutter samples in S2. For training data Using this training data and the value of shape parameter ν to train the neural network model; S4: Extract the actual K-distributed sea clutter amplitude values ​​at the αth and βth quantiles from the actual radar sea clutter data and the actual K-distributed sea clutter amplitude value Among them is is the actual K-distributed sea clutter amplitude value, calculate the actual K-distributed sea clutter amplitude value and the actual K-distributed sea clutter amplitude value The logarithm of the ratio between two values ​​of is S5: Logarithm of the actual sea clutter sample quantile ratio obtained in S4 Input the trained neural network model in S3 to obtain the estimated value of the shape parameter of K-distributed sea clutter S6: Estimated values ​​of the shape parameters of the K-distributed sea clutter obtained in S5 The estimated value It is applied to radar target detection methods under sea clutter background to improve the radar detection performance of targets in K-distributed sea clutter.

2. The K-distribution sea clutter shape parameter estimation method based on neural network and sample quantile ratio according to claim 1 is characterized by: In S1, the K distribution sea clutter amplitude value r at the αth and βth quantiles is calculated according to the cumulative probability distribution function of the K distribution when the K distribution shape parameter is ν. ν,α and r ν,β , where the value range of α is: 0<α<100, the value range of β is: 0<β<100, where β>α.

3. The K-distribution sea clutter shape parameter estimation method based on neural network and sample quantile ratio according to claim 2 is characterized by: The K distribution sea clutter amplitude value r at the αth and βth quantiles is calculated based on the cumulative probability distribution function of the K distribution when the K distribution shape parameter is ν. ν,α and r ν,β Also includes the following methods: S01: It is known that the value of the shape parameter ν of the K distribution is a specific real value, and r is set respectively ν,α and r ν,β An initial known value and Let the cumulative probability values ​​of the cumulative probability distribution function of K distribution be α% and β% respectively; S02: The known value of the K distribution shape parameter ν in S01 and the known value and The value of is brought into the cumulative probability distribution function of the K distribution, and the cumulative probability values ​​of the cumulative probability distribution function of the K distribution at this time are obtained as α (0) % and β (0) %; S03: The cumulative probability value α of the cumulative probability distribution function of the K distribution obtained in S02 (0) % and β (0) α in % (0) and β (0) Compared with α and β in S1, according to α (0) With α, β (0) The size between β and and The size of the value, and then adjust the new and new Substitute the value into the cumulative probability distribution function of the K distribution to obtain the cumulative probability value of the new cumulative probability distribution function of the K distribution, the new α (0) % and new β (0) %, until the new α (0) =α, new β (0) =β, the new and new That is, when the shape parameter of the K distribution in S1 is ν, the K distribution sea clutter amplitude value r at the αth and βth quantiles ν,α and r ν,β .

4. The method for estimating shape parameters of K-distributed sea clutter based on neural network and sample quantile ratio according to claim 3, characterized in that: The cumulative probability distribution function formula of the K distribution is: Where r represents the sea clutter amplitude, ν represents the shape parameter of the K distribution, μ represents the scale parameter of the K distribution. Since the solution of the shape parameter is independent of the scale parameter, the scale parameter is set to 1, that is, μ = 1, Γ(·) represents the gamma function, and K(·) represents the second kind of modified Bessel function; In S01, it is known that the shape parameter of the K distribution is ν. Let the cumulative probability values ​​of the cumulative probability distribution function of the K distribution be α% and β% respectively. Substitute them into the cumulative probability distribution function formula (1) of the K distribution to obtain the cumulative probability distribution function formula (2) of the K distribution as follows: where ν represents the shape parameter of the K distribution, Γ(·) represents the gamma function, and r ν,α represents the amplitude value corresponding to the αth quantile of the K-distributed sea clutter when the shape parameter is ν, r ν,β It represents the amplitude value corresponding to the βth quantile of K-distributed sea clutter when the shape parameter is ν; In S02, the value of the known K distribution shape parameter ν in S01 and the known value and Substitute the value of into the cumulative probability distribution function formula (2) of the K distribution, and obtain formula (3) as follows: The adjusted new and Substitute the value into formula (3) to obtain the cumulative probability value of the new K distribution cumulative probability distribution function new α (0) % and new β (0) %.

5. The method for estimating shape parameters of K-distributed sea clutter based on neural network and sample quantile ratio according to claim 3 or 4, characterized in that: In the S03 described above, according to α (0) With α, β (0) The size between β and and The size of the value includes the following methods: S001: When α (0) When it is greater than α, it decreases Lingxin S002: When α (0) When it is less than α, it increases Lingxin S003: When β (0) When it is greater than β, it decreases Lingxin S004: When β (0) When it is less than β, it increases Lingxin 6. The method for estimating shape parameters of K-distributed sea clutter based on neural network and sample quantile ratio according to claim 1, characterized in that: The neural network model comprises an input layer, a hidden layer and an output layer, the hidden layer has no less than two layers, and the number of neurons in the input layer and the output layer is 1.

7. The method for estimating shape parameters of K-distributed sea clutter based on neural network and sample quantile ratio according to claim 1, characterized in that: The logarithm of the ratio of the sea clutter sample quantiles in S2 in S3 is For training data Also includes the following methods: When ν takes a number of real numbers greater than zero, the logarithm of the ratio of the quantiles of the sea clutter samples obtained in S2 is a number of In S3, there are several As a data set, then several The data is arranged in random order, and finally several The data are randomly divided into neural network training data for neural network model training and neural network verification data for neural network model verification, wherein the number of neural network training data is greater than the number of neural network verification data.

8. The method for estimating shape parameters of K-distributed sea clutter based on neural network and sample quantile ratio according to claim 1, characterized in that: The K-distributed sea clutter amplitude value obtained in S1 is calculated in S2 as r ν,α The amplitude of K-distributed sea clutter is r ν,β The logarithm of the ratio is then taken to obtain the logarithm of the sea clutter sample quantile ratio: in 9. The method for estimating shape parameters of K-distributed sea clutter based on neural network and sample quantile ratio according to claim 1, characterized in that: The actual K distribution sea clutter amplitude value is calculated in S4 and the actual K-distributed sea clutter amplitude value The ratio between the two values ​​of , and then take the logarithm of the ratio, and get the logarithm of the actual sea clutter sample quantile ratio: in

Images