SAR image target dimension reduction method based on supervised learning algorithm
Through the method based on the supervised learning algorithm, SAR image data is reduced in dimension, and a priori information is used to establish a local reconstruction weight matrix, which solves the problem of high-dimensional recognition of small samples in SAR images, and achieves a more efficient recognition effect.
Patent Information
- Application Number
- CN202411965167.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-30
- Publication Date
- 2025-04-25
AI Technical Summary
There is a problem of small samples high-dimensional recognition in SAR images, and traditional algorithms are difficult to effectively utilize prior information, resulting in poor recognition results.
The method based on the supervised learning algorithm is used to reduce the dimensionality of high-dimensional SAR image data. By using prior information to determine the adjacent data points of each sampling point, a local reconstruction weight matrix is established, and the dimensionality reduction output value is obtained.
Small sample high-dimensional recognition is realized, reducing the scale of the original data, and improving the accuracy and efficiency of SAR image target recognition.
Smart Images
Figure CN119888386A_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the technical field of image dimensionality reduction, and in particular relates to a SAR image target dimensionality reduction method based on a supervised learning algorithm. Background Art
[0002] Satellite-borne, airborne and missile-borne synthetic aperture radars have become an important monitoring method for the development and utilization of marine resources, marine scientific research and marine military activities. With the advancement of technology, the resolution of SAR imaging is getting higher and higher, and more information about the shape and structural details of the target can be obtained from SAR images. However, due to the differences in imaging principles, traditional mature optical image recognition algorithms cannot be well applied to SAR images. Research on target recognition based on SAR images has become a hot topic and a difficult point, and there are many problems that need to be solved urgently. The image data generated by the SAR system usually has a high dimensionality and a very limited number of samples. Target recognition based on SAR images is a typical small sample high-dimensional recognition and is one of the most difficult problems in modern pattern classification research.
[0003] Most popular learning algorithms are used in semi-supervised or unsupervised scenarios, focusing on revealing the intrinsic structure and characteristics of popular models. For classification problems, they do not fully utilize samples to identify the distribution of information. However, in the problem of SAR image target recognition, prior knowledge such as the attributes and typical features of the target is known, and making full use of these prior attributes will be more conducive to classification. Summary of the invention
[0004] The purpose of the present invention is to provide a SAR image target dimensionality reduction method based on a supervised learning algorithm to solve the technical problem of small sample high-dimensional image recognition.
[0005] To achieve the above object, the present invention adopts the following technical solution:
[0006] A SAR image target dimensionality reduction method based on a supervised learning algorithm comprises: for a high-dimensional data point set, using prior information to determine a preset number of data points adjacent to each sampling point; establishing a local reconstruction weight matrix based on the data points of each sampling point; and obtaining a dimensionality reduction output value based on the local reconstruction weight matrix and each data point.
[0007] A SAR image target dimensionality reduction system based on a supervised learning algorithm comprises: a sampling unit for determining a preset number of data points adjacent to each sampling point using prior information for a high-dimensional data point set; an establishment unit for establishing a local reconstruction weight matrix based on the data points of each sampling point; and an output unit for obtaining a dimensionality reduction output value based on the local reconstruction weight matrix and each data point.
[0008] A computer-readable storage medium stores a computer program, wherein the computer program is configured to execute the above-mentioned SAR image target dimensionality reduction method based on a supervised learning algorithm when running.
[0009] An electronic device comprises a memory and a processor, wherein the memory stores a computer program, and the processor is configured to execute the SAR image target dimensionality reduction method based on a supervised learning algorithm through the computer program.
[0010] In the present invention, a preset number of data points adjacent to each sampling point are determined by utilizing prior information for a high-dimensional data point set; a local reconstruction weight matrix is established based on the data points of each sampling point; a dimensionality reduction output value is obtained based on the local reconstruction weight matrix and each data point, thereby finding multiple adjacent points through the sampling point, establishing a local reconstruction weight matrix, and then obtaining a dimensionality reduction result, thereby realizing small sample high-dimensional recognition, reducing the size of the original data, solving the technical problem of small sample high-dimensional image recognition, and achieving the technical effect of small sample high-dimensional recognition. BRIEF DESCRIPTION OF THE DRAWINGS
[0011] Figure 1 It is a flowchart of a SAR image target dimensionality reduction method based on a supervised learning algorithm in an embodiment of the present invention;
[0012] Figure 2 It is a structural schematic diagram of a SAR image target dimensionality reduction system based on a supervised learning algorithm in an embodiment of the present invention. DETAILED DESCRIPTION
[0013] The present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. The advantages and features of the present invention will become more apparent from the following description and claims. It should be noted that the drawings are all in very simplified form and are not in precise proportions, and are only used to conveniently and clearly assist in explaining the purpose of the embodiments of the present invention.
[0014] It should be noted that, in order to clearly explain the content of the present invention, the present invention specifically cites multiple embodiments to further illustrate different implementations of the present invention, wherein the multiple embodiments are enumerated rather than exhaustive. In addition, for the sake of brevity of explanation, the contents mentioned in the previous embodiments are often omitted in the subsequent embodiments. Therefore, the contents not mentioned in the subsequent embodiments can refer to the previous embodiments accordingly.
[0015] Example 1
[0016] A SAR image target dimensionality reduction method based on supervised learning algorithm, such as Figure 1 As shown, the method includes:
[0017] S102, for the high-dimensional data point set, using prior information to determine a preset number of data points adjacent to each sampling point;
[0018] S104, establishing a local reconstruction weight matrix based on the data points of each sampling point;
[0019] S106, obtaining a dimension reduction output value based on the local reconstruction weight matrix and each data point.
[0020] Specifically, first, the prior information of the sampling points is used to find multiple adjacent sampling points; second, a local reconstruction weight matrix is established; third, the dimensionality reduction result is obtained based on the local reconstruction weight matrix and the neighborhood of the original data.
[0021] As an optional implementation, for a high-dimensional data point set, a priori information is used to determine a preset number of data points adjacent to each sampling point, including: in the high-dimensional data point set, calculating the distance between the sampling points; and searching for a preset number of data points adjacent to each sampling point according to the distance between the sampling points.
[0022] As an optional implementation, in a high-dimensional data point set, calculating the distance between sampling points includes: calculating the Euclidean distance between the sampling points; and calculating the distance between the sampling points based on the Euclidean distance and the classification result.
[0023] As an optional implementation, a local reconstruction weight matrix is established based on the data points of each sampling point, including: reconstructing the sampling point using the data points of each sampling point to obtain the local reconstruction weight matrix corresponding to the data point; and optimizing the reconstruction error between the reconstructed sampling point and the sampling point to determine the local reconstruction weight matrix.
[0024] As an optional implementation, the local reconstruction weight matrix is determined by optimizing the reconstruction error between the reconstructed sampling points, including: establishing a minimization function to represent the reconstruction error between the reconstructed sampling points; and solving the minimum value of the minimization function to determine the local reconstruction weight matrix.
[0025] As an optional implementation, a dimension reduction output value is obtained based on a local reconstruction weight matrix and each data point, including: using an objective function to calculate a dimension reduction output value of a target dimension based on the local reconstruction weight matrix and each data point.
[0026] As an optional implementation, the objective function is used to calculate the reduced dimension output value of the target dimension based on the local reconstruction weight matrix and each data point, including: constructing an objective function including the local reconstruction of all matrices, data points and reduced dimension output values according to the minimization function.
[0027] As an alternative implementation, by using an objective function, based on the local reconstruction weight matrix and each data point, a dimensionality reduction output value of the target dimension is calculated, including: keeping the local reconstruction weight matrix unchanged, and calculating the dimensionality reduction output value of the target dimension with the minimum value of the objective function as the constraint condition.
[0028] As an alternative implementation, with the minimum value of the objective function as the constraint condition, a dimensionality reduction output value of the target dimension is calculated, including: adding a feature vector as a constraint condition and calculating the dimensionality reduction output value of the target dimension.
[0029] Specifically, the implementation of the SAR image target dimensionality reduction method based on the supervised learning algorithm is not limited to including the following steps:
[0030] Given X = [x1,..., x n , x i ∈R D of the high-dimensional data point set, sample from the D-dimensional manifold and find the low-dimensional coordinates of Y = [y1,..., y n .
[0031] Step 1: Find the k points adjacent to each sampling point.
[0032] Given the observed data set X = [x1,..., x i ,..., x n , x i ∈R D , calculate the distances between the sampling points through Equation (1), search for the k (<n) adjacent data points of each sampling point x i , and find the data set (xi1, xi2,..., x ik ), where x ij ∈X.
[0033] D' = D + αmax(D)A
[0034]
[0035] where D' is the calculation result of the distance, max(D) is the maximum Euclidean distance between classes, A is the classification information, A = 0 when two points belong to the same class, otherwise, A = 1; α is the distance parameter for controlling the distance, reflecting the degree of supervision.
[0036] Step 2: Calculate the reconstruction weight matrix W.
[0037] Assume x iIt can be reconstructed by its k-nn neighboring points, and equation (2) is constructed to reflect the reconstruction error. The weight matrix W is a sparse matrix that covers the inherent characteristics of the source data. The local reconstruction weight matrix W can be calculated by the neighboring points of each sampling point to minimize equation (2).
[0038]
[0039] Formula (3) represents the constraint optimization condition for solving each point:
[0040]
[0041] Step 3: Calculate the output point by fixing the weight matrix W
[0042] Keep the weight matrix W unchanged and construct the minimization objective function. The output value of the sampling point is obtained by locally reconstructing the weight matrix and the k-nn neighboring points:
[0043]
[0044] The dimension of the output is smaller than the dimension of the input, given the constraint Y T Y=I, where I is an n-order identity matrix.
[0045] Formula (5) represents the constraint optimization condition for solving each point:
[0046]
[0047] Where, M = (IW) T (IW).
[0048] This problem can be transformed into an eigenvalue problem without being limited to the use of the lag multiplier method. The eigenvector corresponding to the smallest eigenvalue can be used as a constraint, and the eigenvector with the smallest eigenvalue corresponding to the mean of Y is discarded to enforce
[0049] So far, a SAR image target dimensionality reduction method based on supervised learning algorithm has been realized.
[0050] In the embodiments of the present application, a SAR image target dimensionality reduction method based on a supervised learning algorithm provides a method for high-dimensional recognition of small samples. It is mainly used in missile-borne synthetic aperture radar imaging systems. It is simple to implement, has strong real-time performance, is easy to implement in engineering, is easy to promote, and has strong portability.
[0051] Example 2
[0052] A SAR image target dimensionality reduction system based on supervised learning algorithm, such as Figure 2 As shown, the system includes:
[0053] The sampling unit 202 is used to determine a preset number of data points adjacent to each sampling point using prior information for the high-dimensional data point set;
[0054] An establishing unit 204, configured to establish a local reconstruction weight matrix based on the data points of each sampling point;
[0055] The output unit 206 is used to obtain a dimension reduction output value based on the local reconstruction weight matrix and each data point.
[0056] Optionally, the sampling unit 202 uses prior information to determine a preset number of data points adjacent to each sampling point in the high-dimensional data point set, including: calculating the distance between the sampling points in the high-dimensional data point set; and searching for a preset number of data points adjacent to each sampling point according to the distance between the sampling points.
[0057] Optionally, the sampling unit 202 calculates the distance between the sampling points in the high-dimensional data point set, including: calculating the Euclidean distance between the sampling points; and calculating the distance between the sampling points based on the Euclidean distance and the classification result.
[0058] Optionally, the establishment unit 204 establishes a local reconstruction weight matrix based on the data points of each sampling point, including: reconstructing the sampling point using the data points of each sampling point to obtain a local reconstruction weight matrix corresponding to the data point; and optimizing and determining the local reconstruction weight matrix by reconstructing the sampling point and the reconstruction error between the sampling points.
[0059] Optionally, the establishing unit 204 optimizes and determines the local reconstruction weight matrix by reconstructing the reconstruction error between the sampling points, including: establishing a minimization function to represent the reconstruction error between the sampling points; and solving the minimum value of the minimization function to determine the local reconstruction weight matrix.
[0060] Optionally, the output unit 206 obtains a dimension reduction output value based on the local reconstruction weight matrix and each data point, including: using an objective function to calculate a dimension reduction output value of a target dimension based on the local reconstruction weight matrix and each data point.
[0061] Optionally, the output unit 206 calculates the reduced dimension output value of the target dimension based on the local reconstruction weight matrix and each data point using the objective function, including: constructing an objective function including the local reconstruction of all matrices, data points and reduced dimension output values according to the minimization function.
[0062] Optionally, the output unit 206 uses the objective function to calculate the reduced dimension output value of the target dimension based on the local reconstruction weight matrix and each data point, including: keeping the local reconstruction weight matrix unchanged, using the minimum value of the objective function as a constraint condition, and calculating the reduced dimension output value of the target dimension.
[0063] Optionally, the output unit 206 calculates the reduced-dimensional output value of the target dimension using the minimum value of the objective function as a constraint condition, including: adding a feature vector as a constraint condition to calculate the reduced-dimensional output value of the target dimension.
[0064] In an embodiment of the present application, a preset number of data points adjacent to each sampling point are determined by using prior information for a high-dimensional data point set; a local reconstruction weight matrix is established based on the data points of each sampling point; a dimensionality reduction output value is obtained based on the local reconstruction weight matrix and each data point, thereby finding multiple adjacent points through the sampling point, establishing a local reconstruction weight matrix, and then obtaining a dimensionality reduction result, thereby realizing small sample high-dimensional recognition, reducing the size of the original data, solving the technical problem of small sample high-dimensional image recognition, and achieving the technical effect of small sample high-dimensional recognition.
[0065] Example 3
[0066] Another aspect of the embodiments of the present invention further provides an electronic device for implementing the above-mentioned SAR image target dimensionality reduction method based on the supervised learning algorithm, and the electronic device is not limited to being a terminal device or a server in the system. The electronic device is not limited to including a memory and a processor, the memory stores a computer program, and the processor is configured to execute the steps in any of the above-mentioned method embodiments through the computer program.
[0067] Example 4
[0068] Another aspect of the embodiments of the present invention provides a computer program product or a computer program, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the method provided in various optional embodiments of the SAR image target dimensionality reduction method based on the supervised learning algorithm. The computer program is configured to execute the steps of any of the above method embodiments when it is run.
Claims
1. A SAR image target dimensionality reduction method based on supervised learning algorithm, characterized in that: include: For a high-dimensional data point set, a priori information is used to determine a preset number of data points adjacent to each sampling point; Establish a local reconstruction weight matrix based on the data points of each sampling point; The dimension-reduced output value is obtained based on the local reconstruction weight matrix and each data point.
2. The SAR image target dimensionality reduction method based on supervised learning algorithm as claimed in claim 1, characterized in that: For a high-dimensional data point set, a priori information is used to determine a preset number of data points adjacent to each sampling point, including: In a high-dimensional data point set, calculate the distance between sampling points; A preset number of data points adjacent to each sampling point are obtained by searching according to the distance between the sampling points.
3. The SAR image target dimensionality reduction method based on supervised learning algorithm as claimed in claim 2, characterized in that: In a high-dimensional data point set, calculate the distance between sampling points, including: Calculate the Euclidean distance between sample points; The distance between sampling points is calculated based on the Euclidean distance and the classification results.
4. The SAR image target dimensionality reduction method based on supervised learning algorithm as claimed in claim 1, characterized in that: A local reconstruction weight matrix is established based on the data points of each sampling point, including: Reconstruct the sampling point using the data point of each sampling point to obtain the local reconstruction weight matrix corresponding to the data point; The local reconstruction weight matrix is determined by optimizing the reconstruction error between the reconstruction sampling points.
5. The SAR image target dimensionality reduction method based on supervised learning algorithm as claimed in claim 4, characterized in that: The local reconstruction weight matrix is determined by optimizing the reconstruction error between the reconstruction sampling points, including: Establish a minimization function to represent the reconstruction error between the reconstructed sampling points; The minimum value of the minimization function is solved to determine the local reconstruction weight matrix.
6. The SAR image target dimensionality reduction method based on supervised learning algorithm as claimed in claim 5, characterized in that: Based on the local reconstruction weight matrix and each data point, the dimension reduction output value is obtained, including: Using the objective function, the dimension reduction output value of the target dimension is calculated based on the local reconstruction weight matrix and each data point.
7. The SAR image target dimensionality reduction method based on supervised learning algorithm as claimed in claim 6, characterized in that: Using the objective function, based on the local reconstruction weight matrix and each data point, the dimension reduction output value of the target dimension is calculated, including: According to the minimization function, an objective function including local reconstruction of all matrices, data points and dimension reduction output values is constructed.
8. The SAR image target dimension reduction method based on supervised learning algorithm as claimed in claim 6, characterized in that: Using the objective function, based on the local reconstruction weight matrix and each data point, the dimension reduction output value of the target dimension is calculated, including: Keeping the local reconstruction weight matrix unchanged, taking the minimum value of the objective function as a constraint, the dimension reduction output value of the target dimension is calculated.
9. The SAR image target dimension reduction method based on supervised learning algorithm as claimed in claim 4, characterized in that: Taking the minimum value of the objective function as a constraint, the dimension reduction output value of the target dimension is calculated, including: Add the feature vector as a constraint and calculate the reduced dimension output value of the target dimension.
10. A SAR image target dimensionality reduction system based on supervised learning algorithm, characterized in that: include: A sampling unit, for determining a preset number of data points adjacent to each sampling point using prior information for a high-dimensional data point set; An establishing unit, for establishing a local reconstruction weight matrix based on the data points of each sampling point; The output unit is used to obtain a dimension-reduced output value based on the local reconstruction weight matrix and each data point.