Method and device for selecting training samples based on prior knowledge for doppler post-stap

Abstract

The application discloses a prior knowledge-based training sample selection method for Doppler post-STAP, which comprises the following steps: mapping each information point in a geographic database in a radar beam irradiation range to a corresponding distance unit; re-dividing the information points mapped to the distance unit according to a Doppler frequency, indexing into a corresponding distance-Doppler unit, and forming a normalized vector of each distance-Doppler unit according to a landform type; constructing a weighted normalized vector of each distance-Doppler unit, calculating the Euclidean distance between a training unit and a normalized vector of a to-be-detected distance-Doppler unit, and selecting the first L samples with the smallest distance as uniform training samples; performing power difference correction on the selected uniform training samples; estimating a clutter noise covariance matrix based on the corrected uniform training samples, and calculating a space-time filtering weight vector. The application also provides a corresponding prior knowledge-based training sample selection device for Doppler post-STAP.

Description

Technical Field

[0001] This invention belongs to the field of signal processing technology, and more specifically, relates to a method and apparatus for selecting training samples based on prior knowledge for post-Doppler STAP. Background Technology

[0002] Space-Time Adaptive Processing (STAP) requires accurate estimation of clutter distribution characteristics in airborne radar echo signals during weight formation to effectively suppress strong surface clutter. However, in real-world environments, clutter exhibits a severely non-uniform distribution in the range direction due to factors such as terrain variations, internal clutter motion, and strong isolated clutter caused by man-made objects. The severe lack of uniform training samples leads to a large deviation between the estimated clutter covariance matrix and the true covariance matrix, causing a sharp decline in the performance of the statistically based STAP method.

[0003] Indirect application of KA-STAP (knowledge-aided space-time adaptive processing) is a typical method that utilizes prior knowledge to improve the performance of non-uniform clutter suppression. The method of selecting samples based on prior terrain data is a typical example of this type of method. This method first divides a single range cell into several range-Doppler (RD) cells according to Doppler frequency. Then, it quantizes the geographic information of each RD cell into a terrain vector, classifying the proportion of each terrain type within the terrain vector. The Euclidean distance between the training samples and the terrain vector of the cell to be detected is calculated, and finally, samples are selected using the distance metric. Because this method selects samples for each range-Doppler cell, it is only applicable to the 1DT method. However, the 1DT method, due to its limited system degrees of freedom, can only form adaptive notches in the spatial domain. Compared to the STAP method, which combines multiple Doppler channels, its clutter suppression performance is relatively poor. Therefore, in practical engineering, the STAP method, which combines multiple Doppler channels for stronger clutter suppression, is usually used in strong clutter areas.

[0004] Therefore, there is an urgent need to invent an effective method for selecting training samples based on prior knowledge for post-Doppler STAP, so as to meet the requirement of airborne radar to effectively improve the suppression capability of strong clutter regions in practical engineering. Summary of the Invention

[0005] Therefore, this invention provides a prior knowledge-based training sample selection method for post-Doppler STAP, in order to overcome the problems existing in the prior art.

[0006] To achieve the above objectives, according to one aspect of the present invention, a method for selecting training samples based on prior knowledge for post-Doppler STAP is provided, comprising the following steps:

[0007] Step 1: Based on prior knowledge and aircraft parameters and radar system parameters, map each information point in the geographic database within the radar beam illumination range to the corresponding range cell;

[0008] Step 2: Re-divide the information points mapped to the range cells in Step 1 according to the Doppler frequency, index them into the corresponding range-Doppler cells, and form normalized vectors for each range-Doppler cell according to the landform type.

[0009] Step 3: Construct the weighted normalized vector of each distance-Doppler unit, calculate the Euclidean distance between the training unit and the normalized vector of the distance-Doppler unit to be detected, and select the L samples with the smallest distance as uniform training samples.

[0010] Step 4: Perform power difference correction on the uniform training samples selected in Step 3;

[0011] Step 5: Estimate the clutter noise covariance matrix based on the corrected uniform training samples from Step 4, and calculate the space-time filtering weight vector based on the clutter noise covariance matrix.

[0012] Further, step 1 specifically involves: accurately mapping each information point in the geographic information database to the range-Doppler cells divided by the clutter signal model; firstly, converting the geodetic coordinate coefficients of each information point in the database and the carrier aircraft to the Cartesian coordinate system using Gauss-Kruger projection; and then uniformly mapping each information point to different range cells in the Cartesian coordinate system.

[0013] Further, step 1 specifically includes: Let the equivalent radius of the Earth be R. e The coordinates of the aircraft after Gauss-Kruger projection are (x a y a H a The coordinates of the information point are (x c y c H c ),but

[0014]

[0015] The distance and pitch angle of the information point relative to the carrier aircraft can be obtained by using the above set of equations;

[0016] The azimuth angle corresponding to the information point is

[0017]

[0018] Based on the distance and azimuth parameters of each information point, the location is... The information points in the interval are mapped to a distance cell with a distance of R, where ΔR is the aircraft distance resolution; this step is repeated until all information points in the database are mapped to the corresponding distance cells.

[0019] Further, step 2 specifically includes: re-dividing the information points within the same distance ring according to the Doppler frequency and indexing them into the corresponding range-Doppler units;

[0020] Let R be the distance corresponding to the l-th distance unit. l The pitch angle is but

[0021]

[0022] The maximum normalized Doppler frequency of the clutter block in this range cell is

[0023]

[0024] Therefore, the normalized Doppler frequency range of the l-th distance cell is [-f l f l ];

[0025] Let J be the number of Doppler units, then the frequency formula for the j-th Doppler frequency point is:

[0026]

[0027] The corresponding azimuth angle is

[0028]

[0029] Then the set of azimuth angles for the l-th distance unit is:

[0030] θ l =[θ l,1 ,θ l,2 ,…,θ l,J+1 (7)

[0031] J+1 azimuth angles are used to form boundary points, forming J Doppler units; based on the azimuth angle information corresponding to the information points, the information points are mapped to the corresponding Doppler units;

[0032] The geographic information database used is divided into 8 landform types. The landform type attribute of each range-Doppler cell is represented by a vector containing 8 elements, where each element represents a landform type. The proportion of that landform type in the range-Doppler cell is obtained by dividing the number of information points for that landform type by the total number of information points in the range-Doppler cell, thus forming a normalized vector containing 8 elements. Therefore, the normalized vector of the landform type in the l-th range-Doppler cell of the k-th Doppler channel is expressed as follows:

[0033] I k,l =[i k,l,1 ,…,i k,l,8 ] T (8).

[0034] Further, step 3 specifically includes: if the l-th distance-Doppler cell of the k-th Doppler channel is the distance-Doppler cell to be detected, then the landform type normalized vectors of its two adjacent distance-Doppler cells are respectively...

[0035] I k-1,l =[i k-1,l,1 ,…,i k-1,l,8 ] T (9)

[0036] I k+1,l =[i k+1,l,1 ,…,i k+1,l,8 ] T (10)

[0037] The weighted normalized vector is then...

[0038]

[0039] The weighting coefficients represent the influence of the Doppler channel on the sample selection decision during the sample selection process;

[0040] The selection criteria for the weighting coefficients of adjacent range-Doppler cells are given; assuming the weighted scattering coefficient of the l-th range-Doppler cell in the k-th Doppler channel is σ. k，l Its calculation formula is

[0041]

[0042] Where i k，l，t σ represents the proportion of the t-th landform type in that distance-Doppler cell. k，l，t This represents the empirical backscattering coefficient for the t-th landform type; similarly, the weighted scattering coefficient of two adjacent distance-Doppler cells is calculated as σ. k-1，l and σ k+1，l The formula for calculating the optimal weight coefficient is:

[0043] α:β:γ=σ k-1,l :σ k,l :σ k+1,l (13)

[0044] Combining equations (11), we get:

[0045]

[0046] The normalized vector after further considering the influence of different landform types is:

[0047]

[0048] Where e represents the Hardmard product.

[0049] A = [a1 a2 … a8] T (16)

[0050] a i This represents the normalized scattering intensity of the i-th landform type relative to all eight landform types; the proportion of landform types with larger scattering coefficients is increased through weighting.

[0051] The Euclidean distance between the l′-th training unit and the l′-th distance-Doppler unit to be detected is:

[0052]

[0053] The Euclidean distance of each training unit is obtained from the above formula. The units are sorted from smallest to largest, and the top L samples with the smallest distance are selected as training samples, where L must satisfy the RMB criterion.

[0054] Further, step 4 specifically includes: setting the echo power of the distance-Doppler unit to be detected as P. j Then, the power values ​​of all J Doppler frequency points contained in the detection range unit are averaged, i.e.

[0055]

[0056] Similarly, the average power of the l-th training unit is obtained. The power difference correction factor due to range and antenna pattern modulation is obtained by normalizing the echo power of the range cell to be detected for the l-th training unit.

[0057]

[0058] After power difference correction, the l-th training sample corresponding to the k-th Doppler channel to be detected is...

[0059]

[0060] Where X k，l This represents the original training samples.

[0061] Further, step 5 specifically includes: estimating the clutter noise covariance matrix of the k-th Doppler channel to be detected as follows:

[0062]

[0063] At this time, its corresponding space-time filtering weight vector is

[0064]

[0065] Where S is the space-time steering vector of the target signal.

[0066] According to another aspect of the present invention, a prior knowledge-based training sample selection device for post-Doppler STAP is also provided, comprising at least one processor and a memory, wherein the at least one processor and the memory are connected via a data bus, and the memory stores instructions executable by the at least one processor, wherein the instructions, after being executed by the processor, are used to complete the prior knowledge-based training sample selection method for post-Doppler STAP.

[0067] In summary, the technical solutions conceived by this invention have the following beneficial effects compared with the prior art:

[0068] This invention provides a prior knowledge-based training sample selection method for post-Doppler STAP. It treats multiple adjacent Doppler channels participating in adaptive processing as a whole for sample selection, forms weights for each range-Doppler unit based on clutter scattering coefficients, and constructs Euclidean distance measures between each training sample and the unit to be detected by weighted normalized vectors. Finally, it achieves effective selection of uniform training samples. Attached Figure Description

[0069] Figure 1 This is a flowchart of a training sample selection method based on prior knowledge for post-Doppler STAP in an embodiment of the present invention. Detailed Implementation

[0070] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0071] like Figure 1As shown, this invention provides a method for selecting training samples based on prior knowledge for post-Doppler STAP, comprising the following steps:

[0072] Step 1: Based on prior knowledge and aircraft parameters and radar system parameters, map each information point in the geographic database within the radar beam illumination range to the corresponding range cell;

[0073] Specifically, each information point in the geographic information database is precisely mapped to a range-Doppler cell in the clutter signal model. First, the geodetic coordinate coefficients of each information point and the aircraft in the database are converted to a Cartesian coordinate system using Gauss-Kruger projection. Then, each information point is mapped to a different range cell in the Cartesian coordinate system. The specific mapping process is described below.

[0074] Let the equivalent radius of the Earth be R. e The coordinates of the aircraft after Gauss-Kruger projection are (x a y a H a The coordinates of the information point are (x c y c H c ),but

[0075]

[0076] The distance and pitch angle of the information point relative to the carrier aircraft can be obtained by using the above set of equations.

[0077] The azimuth angle corresponding to the information point is

[0078]

[0079] Based on the distance and azimuth parameters of each information point, the location is... Information points within an interval are mapped to distance cells of distance R, where ΔR is the aircraft's distance resolution. This step is repeated until all information points in the database are mapped to their corresponding distance cells.

[0080] Step 2: Re-divide the information points mapped to the range cells in Step 1 according to the Doppler frequency, index them into the corresponding range-Doppler cells, and form normalized vectors for each range-Doppler cell according to the landform type.

[0081] Specifically, information points within the same distance ring are re-divided according to Doppler frequency to form a range-Doppler unit composed of information points.

[0082] Let R be the distance corresponding to the l-th distance unit. l The pitch angle is but

[0083]

[0084] The maximum normalized Doppler frequency of the clutter block in this range cell is

[0085]

[0086] Therefore, the normalized Doppler frequency range of the l-th distance cell is [-f l f l ].

[0087] Let J be the number of Doppler units, then the frequency formula for the j-th Doppler frequency point is:

[0088]

[0089] The corresponding azimuth angle is

[0090]

[0091] Then the set of azimuth angles for the l-th distance unit is:

[0092] θ l =[θ l,1 ,θ l,2 ,…,θ l,J+1 (7)

[0093] Boundary points are formed using J+1 azimuth angles, creating J Doppler cells. Based on the azimuth information corresponding to each information point, the information points are mapped to their respective Doppler cells.

[0094] The geographic information database used is divided into 8 landform types. The landform type attribute of each range-Doppler cell is represented by a vector containing 8 elements, where each element represents a landform type. Dividing the number of information points for that landform type by the total number of information points within the range-Doppler cell yields the proportion of that landform type in the cell, thus forming a normalized vector containing 8 elements. Therefore, the normalized vector of the landform type for the l-th range-Doppler cell in the k-th Doppler channel is expressed as:

[0095] I k,l =[i k,l,1 ,…,i k,l,8 ] T (8)

[0096] Step 3: Construct the weighted normalized vector of each distance-Doppler unit, calculate the Euclidean distance between the training unit and the normalized vector of the distance-Doppler unit to be detected, and select the L samples with the smallest distance as uniform training samples.

[0097] Specifically, if the l-th distance-Doppler cell in the k-th Doppler channel is the distance-Doppler cell to be detected, then the normalized vectors of the landform types of its two adjacent distance-Doppler cells are respectively...

[0098] I k-1,l =[i k-1,l,1 ,…,i k-1,l,8 ] T (9)

[0099] I k+1,l =[i k+1,l,1 ,…,i k+1,l,8 ] T (10)

[0100] The weighted normalized vector is then...

[0101]

[0102] The weighting coefficients represent the influence of the Doppler channel on the sample selection decision during the sample selection process.

[0103] The selection criteria for the weighting coefficients of adjacent range-Doppler cells are given. Assume the weighted scattering coefficient of the l-th range-Doppler cell in the k-th Doppler channel is σ. k，l Its calculation formula is

[0104]

[0105] Where i k，l，t σ represents the proportion of the t-th landform type in that distance-Doppler cell. k，l，t Let represent the empirical backscattering coefficient for the t-th landform type. Similarly, the weighted scattering coefficient for two adjacent distance-Doppler cells is calculated as σ. k-1，l and σ k+1，l The formula for calculating the optimal weight coefficient is:

[0106] α:β:γ=σ k-1,l :σ k,l :σ k+1,l (13)

[0107] Combining equations (11), we get:

[0108]

[0109] The normalized vector after further considering the influence of different landform types is:

[0110]

[0111] Where e represents the Hardmard product.

[0112] A = [a1 a2 … a8] T (16)

[0113] a i This represents the normalized scattering intensity of the i-th landform type relative to all eight landform types. Weighting was used to increase the proportion of landform types with larger scattering coefficients.

[0114] The Euclidean distance between the l′-th training unit and the l′-th distance-Doppler unit to be detected is:

[0115]

[0116] The Euclidean distance of each training unit is obtained from the above formula. The units are sorted from smallest to largest, and the top L samples with the smallest distance are selected as training samples, where L must satisfy the RMB criterion.

[0117] Step 4: Perform power difference correction on the uniform training samples selected in Step 3;

[0118] Specifically, let the echo power of the distance-Doppler unit to be detected be P. j Then, the power values ​​of all J Doppler frequency points contained in the detection range unit are averaged, i.e.

[0119]

[0120] Similarly, the average power of the l-th training unit is obtained. The power difference correction factor due to range and antenna pattern modulation is obtained by normalizing the echo power of the range cell to be detected for the l-th training unit.

[0121]

[0122] After power difference correction, the l-th training sample corresponding to the k-th Doppler channel to be detected is...

[0123]

[0124] Where X k，l This represents the original training samples.

[0125] Step 5: Estimate the clutter noise covariance matrix based on the corrected uniform training samples from Step 4, and calculate the space-time filtering weight vector based on the clutter noise covariance matrix.

[0126] Specifically, the clutter noise covariance matrix of the k-th Doppler channel to be detected, estimated accordingly, is:

[0127]

[0128] At this time, its corresponding space-time filtering weight vector is

[0129]

[0130] Where S is the space-time steering vector of the target signal.

[0131] Furthermore, the present invention also provides a prior knowledge-based training sample selection device for post-Doppler STAP, comprising at least one processor and a memory, wherein the at least one processor and the memory are connected via a data bus, and the memory stores instructions executable by the at least one processor, wherein the instructions, after being executed by the processor, are used to complete the prior knowledge-based training sample selection method for post-Doppler STAP.

[0132] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for selecting training samples based on prior knowledge for post-Doppler STAP, characterized in that, Includes the following steps, Step 1: Based on prior knowledge and aircraft parameters and radar system parameters, map each information point in the geographic database within the radar beam illumination range to the corresponding range cell; Step 2: Re-divide the information points mapped to the range cells in Step 1 according to Doppler frequency, index them into the corresponding range-Doppler cells, and form normalized vectors for each range-Doppler cell based on the terrain type; Step 2 specifically includes: Information points within the same range cell are re-divided according to Doppler frequency and indexed into the corresponding range-Doppler cell. Let the first cell be... l The distance corresponding to each distance unit is R l The pitch angle is The Earth's equivalent radius is R e The coordinates of the aircraft after Gauss-Kruger projection are ( x a , y a , H a ),but (3) The maximum normalized Doppler frequency of the clutter block in this range cell is (4) Therefore, the first l The normalized Doppler frequency range for each distance cell is [ -f l ， f l ]; Let the number of Doppler units be . J Then the first j The formula for the frequency of each Doppler frequency point is: (5) The corresponding azimuth angle is (6) Then the first l The set of azimuth angles for each distance unit is (7) by J+1 The boundary points are formed by the azimuth angles, forming J Each Doppler unit maps the information point to the corresponding Doppler unit based on the azimuth information corresponding to the information point. The geographic information database used is divided into 8 landform types. The landform type attribute of each distance-Doppler cell is represented by a vector containing 8 elements, where each element represents a landform type. The number of information points for that landform type is divided by the total number of information points within the distance-Doppler cell to obtain the proportion of that landform type in the distance-Doppler cell, thus forming a normalized vector containing 8 elements. k The first Doppler channel l The landform type normalized vector of each distance-Doppler cell is represented as follows: (8)； Step 3: Construct the weighted normalized vectors of each distance-Doppler unit, calculate the Euclidean distance between the normalized vectors of the training unit and the distance-Doppler unit to be detected, and select the unit with the smallest distance. L Each sample is used as a uniform training sample; Step 4: Perform power difference correction on the uniform training samples selected in Step 3; Step 5: Estimate the clutter noise covariance matrix based on the corrected uniform training samples from Step 4, and calculate the space-time filtering weight vector based on the clutter noise covariance matrix.

2. The method for selecting training samples based on prior knowledge for post-Doppler STAP as described in claim 1, characterized in that, Step 1 specifically involves: Each information point in the geographic information database is precisely mapped to the range-Doppler cell of the clutter signal model. First, the geodetic coordinate coefficients of each information point and the carrier aircraft in the database are converted to the Cartesian coordinate system through Gauss-Kruger projection. Then, each information point is mapped to a different range cell in the Cartesian coordinate system.

3. The method for selecting training samples based on prior knowledge for post-Doppler STAP according to claim 1 or 2, characterized in that, Step 1 specifically includes: Let the coordinates of the information point be ( x c , y c , H c ),but (1) The distance and pitch angle of the information point relative to the carrier aircraft can be obtained by using the above set of equations; The azimuth angle corresponding to the information point is (2) Based on the distance and azimuth parameters of each information point, the location is... Information points in an interval are mapped to distances of R In the distance cell, where Δ R This is the range resolution for the aircraft; repeat this step until all information points in the database are mapped to the corresponding range cells.

4. The method for selecting training samples based on prior knowledge for post-Doppler STAP as described in claim 1, characterized in that, Step 3 specifically includes: No. k The first Doppler channel l If a distance-Doppler cell is the distance-Doppler cell to be detected, then the normalized vectors of the landform types of its two adjacent distance-Doppler cells are respectively... (9) (10) The weighted normalized vector is then... (11) The weighting coefficients represent the influence of the Doppler channel on the sample selection decision during the sample selection process; The criteria for selecting the weight coefficients corresponding to adjacent distance-Doppler cells are given, assuming the first... k The first Doppler channel l The weighted scattering coefficient of each range-Doppler unit is σ k，l Its calculation formula is (12) in i k，l，t Indicates the first t The percentage of each landform type in this distance-Doppler cell, σ k，l，t Indicates the first t The empirical backscattering coefficients for various landform types are calculated; similarly, the weighted scattering coefficients of two adjacent distance-Doppler cells are calculated as σ. k-1，l and σ k+1，l Then the formula for calculating the optimal weight coefficient is: (13) Combining equations (11), we get: (14) The normalized vector after further considering the influence of different landform types is: (15) in Represents the Hardmard product; (16) a i Indicates the first i The normalized scattering intensity of this landform type relative to all eight landform types was calculated; the proportion of landform types with larger scattering coefficients was increased through weighting. No. The training unit and the first l The Euclidean distance between each detection distance-Doppler unit is (17) The Euclidean distances of each training unit are obtained from the above formula. These units are then sorted from smallest to largest, and the units with the smallest distances are selected. L One sample was used as the training sample, among which L It must meet the RMB standard.

5. The method for selecting training samples based on prior knowledge for post-Doppler STAP according to claim 4, characterized in that, Step 4 specifically includes: Let the echo power of the distance-Doppler element to be detected be... P j Then all the distance units to be detected contain J The power values ​​at each Doppler frequency point are averaged, i.e. (18) Similarly, we obtain the first... l Average power of training units ; By measuring the echo power of the distance unit to be detected, the first l Each training unit is normalized to obtain a power difference correction factor due to distance and antenna pattern modulation. (19) After power difference correction, the first k The first Doppler channel to be detected corresponds to the first l The training samples are (20) in X k，l This represents the original training samples.

6. The method for selecting training samples based on prior knowledge for post-Doppler STAP according to claim 5, characterized in that, Step 5 specifically includes: Based on this estimate, the first... k The clutter noise covariance matrix of each Doppler channel to be detected is: (21) At this time, its corresponding space-time filtering weight vector is (22) in S The space-time steering vector for the target signal.

7. A prior knowledge-based training sample selection device for post-Doppler STAP, characterized in that: The method includes at least one processor and a memory, which are connected via a data bus. The memory stores instructions that can be executed by the at least one processor. After being executed by the processor, the instructions are used to complete the prior knowledge-based training sample selection method for post-Doppler STAP as described in any one of claims 1-6.

Images