Radar target high-resolution one-dimensional range profile recognition method

Abstract

The invention discloses a radar target high-resolution one-dimensional range profile recognition method, and relates to the field of radar signal processing. Comprising the following steps: S1, constructing a target high-resolution one-dimensional range profile training sample set X, a label matrix H and a label block diagonal matrix Q; s2, extracting features from the training sample set X and the tag block diagonal matrix Q, and optimizing to obtain a feature dictionary matrix D and a feature representation matrix Z; s3, constructing a linear classifier C by using the feature representation matrix Z and the label matrix H; s4, constructing a feature representation matrix Zt of the test sample set Xt according to the feature dictionary matrix D; and S5, performing target classification by using the feature representation matrix Zt of the test sample set Xt and the linear classifier C. The problem that a traditional target recognition method is poor in training effect is solved.

Description

Technical Field

[0001] This application relates to the field of radar signal processing technology, and in particular to a method for high-resolution one-dimensional range profile recognition of radar targets. Background Technology

[0002] Radar automatic target recognition (RATR) is a technique that determines the category and attributes of a target by analyzing the echoes scattered by the target. In RATR, the high-resolution range profile (HRRP) contains rich physical information such as the target's size and structure, and the echoes are easier to acquire and process compared to synthetic aperture images, thus possessing significant research importance and practical value.

[0003] Traditional target recognition methods feed manually extracted features into classification models such as machine learning for classification and recognition. In addition, representation learning and deep learning models are also used for HRRP recognition, offering the advantage of autonomous feature learning. Representation learning is a method for automatically extracting feature representations from input data, which can be used for classification and recognition tasks. Methods utilizing sparse representations and combining them with dictionary learning have also been applied in HRRP recognition. Furthermore, various deep learning models have achieved good results in HRRP recognition.

[0004] The methods described above have achieved good results in the field of target HRRP recognition, but they still have certain limitations. On the one hand, they only focus on the features of a single sample, ignoring the structural information between samples of the same type. On the other hand, in the complex environments of real-world applications, target HRRP recognition faces the problem of deteriorating training performance due to noise interference in the training sample data. Summary of the Invention

[0005] The purpose of this application is to provide a high-resolution one-dimensional range image recognition method for radar targets, overcoming the problem of poor training effect of traditional target recognition methods.

[0006] To achieve the above objectives, this application adopts the following technical solution:

[0007] On the one hand, this application provides a method for high-resolution one-dimensional range profile recognition of radar targets, including:

[0008] S1. Construct a high-resolution one-dimensional distance image training sample set X, a label matrix H, and a label block diagonal matrix Q;

[0009] S2. Extract features from the training sample set X and the label block diagonal matrix Q to obtain the feature dictionary matrix D and the feature representation matrix Z;

[0010] S3. Construct a linear classifier C using the feature representation matrix Z and the label matrix H;

[0011] S4. Construct a test sample set X based on the feature dictionary matrix D. t The characteristic representation matrix Z t ;

[0012] S5. Using the test sample set X t The characteristic representation matrix Z t The target is classified using a linear classifier C.

[0013] On the other hand, this application also provides a radar target high-resolution one-dimensional range image identification device, comprising:

[0014] The matrix construction module is used to construct the target high-resolution one-dimensional distance image training sample set X, the label matrix H, and the label block diagonal matrix Q;

[0015] The feature extraction module is used to extract features from the training sample set X and the label block diagonal matrix Q to obtain the feature dictionary matrix D and the feature representation matrix Z;

[0016] A classifier generation module is used to construct a linear classifier C using the feature representation matrix Z and the label matrix H;

[0017] The test sample extraction module is used to construct a test sample set X based on the feature dictionary matrix D. t The characteristic representation matrix Z t ;

[0018] The target classification module is used to utilize the test sample set X. t The characteristic representation matrix Z t The target is classified using a linear classifier C.

[0019] Based on the above technical solution, this application can achieve the following technical effects:

[0020] By employing low-rank constraints, the global structural features of high-resolution one-dimensional range images are characterized, reducing the impact of noise. While retaining the advantages of low-rank models, the addition of structured constraints makes similar samples more similar in representation, while also amplifying the differences between different categories. Compared to traditional machine learning, this approach is more effective for target classification and recognition under noisy conditions, and has practical value for research on high-resolution one-dimensional range image recognition of targets. Attached Figure Description

[0021] Figure 1This is a flowchart illustrating a high-resolution one-dimensional range profile recognition method for radar targets provided in an embodiment of this application;

[0022] Figure 2 This is a schematic diagram illustrating the implementation process of a radar target high-resolution one-dimensional range profile recognition method provided in an embodiment of this application;

[0023] Figure 3 This is an example diagram of a tag block diagonal matrix provided in one embodiment of this application. Detailed Implementation

[0024] The present application will be further described in detail below with reference to the accompanying drawings and specific embodiments. The advantages and features of the present application will become clearer from the following description and claims. It should be noted that the drawings are all in a very simplified form and are not to scale, and are only used to facilitate and clarify the illustration of the embodiments of the present application.

[0025] It should be noted that, in order to clearly illustrate the content of this application, several embodiments are provided to further explain the different implementations of this application. These embodiments are enumerated rather than exhaustive. Furthermore, for the sake of brevity, content mentioned in the preceding embodiments is often omitted in the following embodiments. Therefore, content not mentioned in the following embodiments can be referred to in the preceding embodiments.

[0026] Example 1

[0027] like Figure 1 , Figure 2 The diagram shows a flowchart and implementation process of a high-resolution one-dimensional range profile recognition method for radar targets provided in this embodiment. It includes:

[0028] S1. Construct a high-resolution one-dimensional distance image training sample set X, a label matrix H, and a label block diagonal matrix Q;

[0029] It should be noted that one implementation of S1 can be:

[0030] The training sample set X of the high-resolution one-dimensional distance image of the target is composed of a combination of HRRP samples that are as rich and diverse as possible.

[0031] like Figure 3 As shown, a label block diagonal matrix Q is constructed based on the actual labels of the training samples, where Q satisfies X≈DQ, and D is the desired feature dictionary matrix, which consists of c sub-dictionary matrices, i.e., D=[D1,D2,…,D…]. c ], D i Let represent the i-th sub-dictionary of D, where c is the number of sub-dictionaries, which is also the number of classes in the training samples; Q = [q1, q2, ..., q N ], N is the sample size, q iLet q4 be the idealized label representation of the i-th sample. Taking q4 as an example, if the 4th sample corresponding to it belongs to the second class, that is, it corresponds to the sub-dictionary D2, then the element value in q4 corresponding to D2 is set to 1, and the other elements are 0.

[0032] S2. Extract features from the training sample set X and the label block diagonal matrix Q to obtain the feature dictionary matrix D and the feature representation matrix Z;

[0033] It should be noted that one implementation of S2 can be:

[0034] Features are extracted by solving the following objective function:

[0035]

[0036] Among them, ||·|| * Let ||·||1 be the dual norm of the matrix, and let ||·||1 be the l1-norm of the matrix. F Let Z be the F-norm of the matrix, and let Z be as similar to Q as possible; α, β, γ, and λ are weight parameters, and E is the noise matrix contained in the training sample data.

[0037] To solve the above objective function, the following steps are required:

[0038] S21. Introducing variables J and L, the function is rewritten using the augmented Lagrange multiplier method to facilitate solving. The rewritten objective function is:

[0039]

[0040] S22. The rewritten objective function is further rewritten as the augmented Lagrange function corresponding to the objective function:

[0041]

[0042] in,<A,B> =tr(A T ,B), tr() represents the trace of the matrix, superscript T This represents the matrix transpose operation; Y1, Y2, and Y3 are Lagrange multipliers, and μ > 0 is the penalty factor;

[0043] S23. By iteratively updating variables Z, L, E, D, and J, the augmented Lagrangian function corresponding to the objective function is solved, yielding the feature dictionary matrix D and the feature representation matrix Z. (Now using superscript...) k Indicates variables that have not been updated, superscript k+1 The following is an example of one iteration in the solution process, representing the updated variable:

[0044] (1) Fix Z, L, E, and D, and update J according to the following formula:

[0045]

[0046] in, SVD stands for Singular Value Decomposition of a matrix. The definition is as follows:

[0047]

[0048] (2) Fix J, L, E, and D, and update Z according to the following formula:

[0049]

[0050] (3) Fix Z, J, E, and D, and update L according to the following formula:

[0051]

[0052] (4) Fix Z, J, L, and E, and update D:

[0053]

[0054] By taking the partial derivative of the above equation and setting it to 0, we obtain its closed-form solution:

[0055]

[0056] Where I is the identity matrix, and the superscript - 1 This represents the matrix inversion operation.

[0057] (5) Fix Z, J, L, D, and update E:

[0058]

[0059] S3. Construct a linear classifier C using the feature representation matrix Z and the label matrix H;

[0060] It should be noted that one implementation of S3 can be:

[0061] Construct a linear classifier C as follows:

[0062]

[0063] Where H = [h1, h2, ..., h N ], h i Let h be the label of the i-th sample. i If it belongs to category 2, then h i The second element has a value of 1, and the other elements have a value of 0. N is the number of samples; ||·||2 is the l2 norm of the matrix.

[0064] S4. Construct a test sample set X based on the feature dictionary matrix D. t The characteristic representation matrix Z t ;

[0065] It should be noted that one implementation of S4 can be:

[0066] The feature representation matrix Z is extracted by solving the following equation. t :

[0067]

[0068] Among them, E t It is the noise matrix of the test sample set.

[0069] S5. Using the test sample set X t The characteristic representation matrix Z t The target is classified using a linear classifier C.

[0070] It should be noted that one implementation of S5 can be:

[0071] Target classification is performed using the following formula:

[0072]

[0073] Among them, z tm The characteristic representation matrix Z t The m-th column vector is the feature representation of the m-th test sample; l is the predicted label of the m-th test sample, that is, the class of the m-th test sample is predicted to be the l-th class.

[0074] In summary, this method, through low-rank constraints, characterizes the global structural features of high-resolution one-dimensional range images, reducing the impact of noise. While retaining the advantages of low-rank models, the addition of structured constraints makes similar samples more similar in representation, while also amplifying the differences between different categories. Compared to traditional machine learning, this method is more effective for target classification and recognition under noisy conditions, and has practical value for research on high-resolution one-dimensional range image recognition of targets.

[0075] Example 2

[0076] This embodiment provides a radar target high-resolution one-dimensional range image recognition device. The device includes:

[0077] The matrix construction module is used to construct the target high-resolution one-dimensional distance image training sample set X, the label matrix H, and the label block diagonal matrix Q;

[0078] The feature extraction module is used to extract features from the training sample set X and the label block diagonal matrix Q to obtain the feature dictionary matrix D and the feature representation matrix Z;

[0079] A classifier generation module is used to construct a linear classifier C using the feature representation matrix Z and the label matrix H;

[0080] The test sample extraction module is used to construct a test sample set X based on the feature dictionary matrix D. t The characteristic representation matrix Z t ;

[0081] The target classification module is used to utilize the test sample set X. t The characteristic representation matrix Z t The target is classified using a linear classifier C.

[0082] In summary, this device characterizes the global structural features of high-resolution one-dimensional range images through low-rank constraints, reducing the impact of noise. While retaining the advantages of low-rank models, the addition of structured constraints makes similar samples have more similar representations, while also amplifying the differences between different categories. Compared with traditional machine learning, this is more conducive to target classification and recognition under noisy conditions, and has practical value for research on high-resolution one-dimensional range image recognition of targets.

[0083] Example 3

[0084] In yet another feasible embodiment, this embodiment provides a device for high-resolution one-dimensional range profile recognition of radar targets, the device specifically including:

[0085] A processor; and a memory for storing computer-executable instructions, which, when executed, cause the processor to perform the steps as described in any of the above method embodiments.

[0086] Example 4

[0087] In another feasible embodiment, this embodiment provides a storage medium for high-resolution one-dimensional range profile identification of radar targets, the storage medium specifically including:

[0088] The storage medium stores a processing program for high-resolution one-dimensional range image recognition of radar targets. When the processor executes the processing program for high-resolution one-dimensional range image recognition of radar targets, it implements the steps as described in any of the above method embodiments.

[0089] The above description is merely an embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of this application should be included within the scope of the claims of this application.

Claims

1. A method for high-resolution one-dimensional range profile recognition of radar targets, characterized in that, include: S1. Construct a high-resolution one-dimensional distance image training sample set X, a label matrix H, and a label block diagonal matrix Q; S2. Extract features from the training sample set X and the label block diagonal matrix Q, and optimize to obtain the feature dictionary matrix D and the feature representation matrix Z; S3. Construct a linear classifier C using the feature representation matrix Z and the label matrix H; S4. Construct a test sample set X based on the feature dictionary matrix D. t The characteristic representation matrix Z t ; S5. Using the test sample set X t The characteristic representation matrix Z t The target is classified using a linear classifier C.

2. The method according to claim 1, characterized in that, S2 includes: The feature dictionary matrix D and the feature representation matrix Z are obtained by solving the following equation: Among them, ||·|| * Let ||·||1 be the dual norm of the matrix, and let ||·||1 be the l1-norm of the matrix. F Let Z be the F-norm of the matrix, and let Z be as similar to Q as possible; α, β, γ, and λ are weight parameters, and E is the noise matrix contained in the training sample data.

3. The method according to claim 2, characterized in that, S3 includes: Construct a linear classifier C as follows: Where H = [h1, h2, ..., h N ], h i Let be the label of the i-th sample, N be the number of samples, and ||·||2 be the l2-norm of the matrix.

4. The method according to claim 3, characterized in that, S4 includes: The feature representation matrix Z is extracted by solving the following equation. t : Among them, E t It is the noise matrix of the test sample set.

5. The method according to claim 4, characterized in that, S5 includes: Target classification is performed using the following formula: Among them, z tm The characteristic representation matrix Z t The m-th column vector is the feature representation of the m-th test sample; l is the predicted label of the m-th test sample.

6. The method according to claim 2, characterized in that, The feature extraction by solving the following formula includes: S21. Introducing variables J and L, and rewriting the function using the augmented Lagrange multiplier method, the rewritten objective function is: S22. The rewritten objective function is further rewritten as the augmented Lagrange function corresponding to the objective function: in,<A,B> =tr(A T ,B), tr() represents the trace of the matrix, superscript T This represents the matrix transpose operation; Y1, Y2, and Y3 are Lagrange multipliers, and μ > 0 is the penalty factor; S23. By iteratively updating variables Z, L, E, D, and J, the augmented Lagrangian function corresponding to the objective function is solved to obtain the feature dictionary matrix D and the feature representation matrix Z.

7. The method according to claim 1, characterized in that, S1 includes: Construct a label block diagonal matrix Q based on the actual labels of the training samples, where Q = [q1, q2, ..., q N ], q i Let N be the idealized label representation of the i-th sample, and N be the number of samples.

8. A radar target high-resolution one-dimensional range image recognition device, characterized in that, include: The matrix construction module is used to construct the target high-resolution one-dimensional distance image training sample set X, the label matrix H, and the label block diagonal matrix Q; The feature extraction module is used to extract features from the training sample set X and the label block diagonal matrix Q to obtain the feature dictionary matrix D and the feature representation matrix Z; A classifier generation module is used to construct a linear classifier C using the feature representation matrix Z and the label matrix H; The test sample extraction module is used to construct a test sample set X based on the feature dictionary matrix D. t The characteristic representation matrix Z t ; The target classification module is used to utilize the test sample set X. t The characteristic representation matrix Z t The target is classified using a linear classifier C.

9. An electronic device, characterized in that, include: processor; And a memory for storing computer-executable instructions, which, when executed, cause the processor to perform the steps of the method as claimed in any one of claims 1 to 7.

10. A storage medium, characterized in that, include: The storage medium stores a processing program for high-resolution one-dimensional range image recognition of radar targets. When the processing program for high-resolution one-dimensional range image recognition of radar targets is executed by the processor, it implements the method steps as described in any one of claims 1-7.

Images