Remote sensing feature extraction method and device

Abstract

The embodiment of the present invention discloses a remote sensing feature extraction method and device, the method comprising: projecting the remote sensing data to be processed onto a preset number of principal component subspaces to obtain a preset number of principal component projection data; wherein the preset number is N+2, N is a positive integer, and the remote sensing data to be processed includes remote sensing target domain data; projecting the remote sensing data to be processed onto a preset number of correlation subspaces to obtain a preset number of correlation projection data; determining remote sensing projection data based on the principal component projection data and the correlation projection data, and performing dimensionality reduction processing on the remote sensing projection data to obtain remote sensing feature data corresponding to the remote sensing data to be processed. Through the technical solution of the embodiment of the present invention, the effect of improving the quality of remote sensing feature extraction is achieved.

Description

Technical Field

[0001] The embodiments of the present invention relate to remote sensing image processing technology, and in particular to a remote sensing feature extraction method and device. Background Art

[0002] Remote sensing image classification technology is one of the key technologies in remote sensing digital image processing. It has been widely used in many fields such as agriculture and military, and has played an important role in all walks of life.

[0003] However, in the process of remote sensing image classification, the training data and test data often come from two different remote sensing images. At this time, the test data and the training data are not identically distributed. In this case, the use of traditional machine learning has the problem of poor remote sensing feature extraction effect. Therefore, it cannot be applied to remote sensing image classification. Summary of the invention

[0004] The embodiments of the present invention provide a remote sensing feature extraction method and device to achieve the effect of improving the quality of remote sensing feature extraction.

[0005] In a first aspect, an embodiment of the present invention provides a remote sensing feature extraction method, the method comprising:

[0006] Projecting the remote sensing data to be processed onto a preset number of principal component subspaces to obtain a preset number of principal component projection data; wherein the preset number is N+2, N is a positive integer, and the remote sensing data to be processed includes remote sensing target domain data;

[0007] Projecting the remote sensing data to be processed onto a preset number of correlation subspaces to obtain a preset number of correlation projection data;

[0008] According to the principal component projection data and the correlation projection data, remote sensing projection data are determined, and dimension reduction processing is performed on the remote sensing projection data to obtain remote sensing feature data corresponding to the remote sensing data to be processed.

[0009] In a second aspect, an embodiment of the present invention further provides a remote sensing feature extraction device, the device comprising:

[0010] A principal component projection module is used to project the remote sensing data to be processed onto a preset number of principal component subspaces to obtain a preset number of principal component projection data; wherein the preset number is N+2, N is a positive integer, and the remote sensing data to be processed includes remote sensing target domain data;

[0011] A correlation projection module, used for projecting the remote sensing data to be processed onto a preset number of correlation subspaces to obtain a preset number of correlation projection data;

[0012] The feature extraction module is used to determine the remote sensing projection data according to the principal component projection data and the correlation projection data, and perform dimensionality reduction processing on the remote sensing projection data to obtain remote sensing feature data corresponding to the remote sensing data to be processed.

[0013] In a third aspect, an embodiment of the present invention further provides an electronic device, the electronic device comprising:

[0014] one or more processors;

[0015] a storage device for storing one or more programs,

[0016] When the one or more programs are executed by the one or more processors, the one or more processors implement the remote sensing feature extraction method as described in any one of the embodiments of the present invention.

[0017] In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a remote sensing feature extraction method as described in any one of the embodiments of the present invention.

[0018] The technical solution of the embodiment of the present invention is to project the remote sensing data to be processed onto a preset number of principal component subspaces to obtain a preset number of principal component projection data, project the remote sensing data to be processed onto a preset number of correlation subspaces to obtain a preset number of correlation projection data, determine the remote sensing projection data based on the principal component projection data and the correlation projection data, and perform dimensionality reduction processing on the remote sensing projection data to obtain remote sensing feature data corresponding to the remote sensing data to be processed, thereby solving the problem of poor feature extraction effect of remote sensing image data and achieving the effect of improving the quality of remote sensing feature extraction. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions of the exemplary embodiments of the present invention, the following briefly introduces the drawings required to describe the embodiments. Obviously, the drawings introduced are only drawings of a part of the embodiments to be described in the present invention, rather than all the drawings. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0020] Figure 1 A schematic diagram of a flow chart of a remote sensing feature extraction method provided in the first embodiment of the present invention;

[0021] Figure 2 A schematic diagram of a flow chart of a correlation subspace determination method provided in Embodiment 2 of the present invention;

[0022] Figure 3 A schematic diagram of a flow chart of a method for determining a principal component subspace provided in Embodiment 2 of the present invention;

[0023] Figure 4 A schematic diagram of the principle of intermediate subspace alignment provided by Embodiment 3 of the present invention;

[0024] Figure 5 A data distribution diagram of the PIC56 before feature alignment provided in the third embodiment of the present invention;

[0025] Figure 6 A data distribution diagram after PIC56 feature alignment provided in the third embodiment of the present invention;

[0026] Figure 7 A data distribution comparison diagram before and after alignment of the first type of data features in PIC56 provided in the third embodiment of the present invention;

[0027] Figure 8 A data distribution comparison diagram before and after alignment of the second type of data features in PIC56 provided in the third embodiment of the present invention;

[0028] Fig. 9 A data distribution comparison diagram of the third type of data features before and after alignment in PIC56 provided in the third embodiment of the present invention;

[0029] Fig.10 A data distribution comparison diagram of the fourth type of data features before and after alignment in PIC56 provided in the third embodiment of the present invention;

[0030] Fig.11 A data distribution comparison diagram of the fifth type of data features before and after alignment in PIC56 provided in the third embodiment of the present invention;

[0031] Fig.12 A data distribution comparison diagram of the sixth type of data features before and after alignment in PIC56 provided in the third embodiment of the present invention;

[0032] Fig.13 A data distribution comparison diagram before and after alignment of the seventh type of data features in PIC56 provided in the third embodiment of the present invention;

[0033] Fig.14 A data distribution comparison diagram before and after alignment of the eighth type of data features in PIC56 provided in the third embodiment of the present invention;

[0034] Fig.15 A data distribution comparison diagram before and after alignment of the ninth type of data features in PIC56 provided in the third embodiment of the present invention;

[0035] Fig.16 A schematic diagram of the structure of a remote sensing feature extraction device provided in Embodiment 4 of the present invention;

[0036] Fig.17This is a schematic diagram of the structure of an electronic device provided in Embodiment 5 of the present invention. DETAILED DESCRIPTION

[0037] The present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments. It is to be understood that the specific embodiments described herein are only used to explain the present invention, rather than to limit the present invention. It should also be noted that, for ease of description, only parts related to the present invention, rather than all structures, are shown in the accompanying drawings.

[0038] Embodiment 1

[0039] Figure 1 A flowchart of a remote sensing feature extraction method provided in Example 1 of the present invention is provided. This embodiment is applicable to the case where transfer learning is performed on remote sensing images to improve classification accuracy. The method can be executed by a remote sensing feature extraction device, which can be implemented in the form of software and / or hardware. The hardware can be an electronic device. Optionally, the electronic device can be a mobile terminal, a PC, a server, etc.

[0040] like Figure 1 The method of this embodiment specifically includes the following steps:

[0041] S110 , projecting the remote sensing data to be processed onto a preset number of principal component subspaces to obtain a preset number of principal component projection data.

[0042] Among them, the preset number is N+2, and N is a positive integer. The remote sensing data to be processed includes remote sensing target domain data, and the remote sensing source data can be remote sensing data in the source domain to be classified. Source domain and target domain are related terms in transfer learning. The source domain refers to a known field in transfer learning that contains a large amount of training data, and the target domain refers to an unknown field in transfer learning that contains no training data or only a small amount of training data. The principal component subspace can be each low-dimensional space determined based on PCA (Principal Component Analysis). The principal component projection data can be the projection data of the remote sensing data to be processed on each principal component subspace.

[0043] Specifically, the remote sensing data to be processed are projected in each principal component subspace respectively, so as to obtain the projection data of the remote sensing data to be processed in each principal component subspace, that is, to obtain a preset number of principal component projection data.

[0044] S120, projecting the remote sensing data to be processed onto a preset number of correlation subspaces to obtain a preset number of correlation projection data.

[0045] The correlation subspace may be each low-dimensional space determined based on CCA (Canonical Correlation Analysis). The correlation projection data may be the projection data of the remote sensing data to be processed on each correlation subspace.

[0046] Specifically, the remote sensing data to be processed are projected in each correlation subspace respectively, so as to obtain the projection data of the remote sensing data to be processed in each correlation subspace, that is, to obtain a preset number of correlation projection data.

[0047] S130, determining remote sensing projection data according to the principal component projection data and the correlation projection data, and performing dimensionality reduction processing on the remote sensing projection data to obtain remote sensing feature data corresponding to the remote sensing data to be processed.

[0048] The remote sensing projection data may be data obtained by connecting the principal component projection data and the correlation projection data in series. The dimension reduction processing may be data obtained by processing the remote sensing projection data by a dimension reduction method, and the dimension reduction method may be a dimension reduction method such as PLS (Partial Least Squares). The remote sensing feature data may be feature data used for subsequent classification or analysis processing.

[0049] Specifically, the principal component projection data and the correlation projection data obtained by projection are spliced ​​to obtain remote sensing projection data. Since the dimension of remote sensing projection data is large, in order to facilitate subsequent classification processing, a dimensionality reduction algorithm is used to reduce the dimension of remote sensing projection data to obtain remote sensing feature data to retain the main features.

[0050] It should be noted that after obtaining the remote sensing feature data, it can be considered that the data in the target domain has been mapped to the source domain, and the classification method of the feature data in the source domain can be used for subsequent classification.

[0051] The technical solution of the embodiment of the present invention is to project the remote sensing data to be processed onto a preset number of principal component subspaces to obtain a preset number of principal component projection data, project the remote sensing data to be processed onto a preset number of correlation subspaces to obtain a preset number of correlation projection data, determine the remote sensing projection data based on the principal component projection data and the correlation projection data, and perform dimensionality reduction processing on the remote sensing projection data to obtain remote sensing feature data corresponding to the remote sensing data to be processed, thereby solving the problem of poor feature extraction effect of remote sensing image data and achieving the effect of improving the quality of remote sensing feature extraction.

[0052] Embodiment 2

[0053] Figure 2This is a flow chart of a method for determining a correlation subspace provided in Embodiment 2 of the present invention. This embodiment is based on any optional technical solution in the embodiments of the present invention. Optionally, the method for determining the correlation subspace can refer to the technical solution of this embodiment. The explanations of the terms that are the same or corresponding to the above embodiments are not repeated here.

[0054] like Figure 2 As shown, the method of this embodiment may specifically include:

[0055] S210: Obtain sample source domain data and sample target domain data.

[0056] The sample source domain data may be remote sensing data in the source domain, that is, data in a domain containing a large amount of training data. The sample target domain data may be remote sensing data in the target domain, that is, data in a domain with less training data.

[0057] Specifically, the domain of the data to be classified is taken as the target domain, and sample data in the target domain is obtained. The domain with sufficient data and data with certain similarities to the data in the target domain is taken as the source domain, and sample data in the source domain is obtained.

[0058] S220. According to a preset dimension, perform correlation analysis on the sample source domain data to obtain a correlation source domain projection matrix.

[0059] The preset dimension may be the target dimension when performing correlation analysis dimensionality reduction, that is, the dimension to be reduced. The preset dimension may be a dimension suitable for the source domain and the target domain obtained through training, or may be a pre-set dimension. The correlation source domain projection matrix may be a projection matrix of correlation analysis dimensionality reduction of sample source domain data.

[0060] Specifically, by performing correlation analysis on the sample source domain data in a preset dimension, a correlation source domain projection matrix corresponding to the sample source domain data can be obtained.

[0061] S230. According to a preset dimension, the sample target domain data is subjected to correlation analysis to obtain a correlation target domain projection matrix.

[0062] The correlation target domain projection matrix may be a projection matrix of correlation analysis dimensionality reduction of sample target domain data.

[0063] Specifically, by performing correlation analysis on the sample target domain data in a preset dimension, a correlation target domain projection matrix corresponding to the sample target domain data can be obtained.

[0064] S240. Determine N correlation intermediate spaces according to the sample source domain data, the sample target domain data, the correlation source domain projection matrix, and the correlation target domain projection matrix.

[0065] The correlation intermediate space is the intermediate space between the correlation source domain projection matrix and the correlation target domain projection matrix determined based on the Greismann manifold.

[0066] Specifically, according to the sample source domain data, the sample target domain data, the correlation source domain projection matrix and the correlation target domain projection matrix, a correlation intermediate space is constructed along a path on the Greismann manifold.

[0067] Optionally, the correlation intermediate space may be determined based on the following steps:

[0068] Step 1: Determine the correlation orthogonal complete matrix of the correlation source domain projection matrix, and perform cosine-sine decomposition based on the product of the transposed correlation orthogonal complete matrix and the correlation target domain projection matrix to obtain the first orthogonal matrix, the second orthogonal matrix, the first reference matrix and the second reference matrix.

[0069] The correlation orthogonal complete matrix may be an orthogonal complete matrix corresponding to the correlation source domain projection matrix. The cosine positive decomposition may be a CS (Cosine-sine decomposition) decomposition. The first orthogonal matrix, the second orthogonal matrix, the first reference matrix, and the second reference matrix may be matrices obtained by performing cosine-sine decomposition.

[0070] Specifically, the orthogonal complete matrix of the correlation source domain projection matrix is ​​obtained, which is the correlation orthogonal complete matrix. Then, the product of the transposed correlation orthogonal complete matrix and the correlation target domain projection matrix is ​​subjected to cosine-sine decomposition to obtain various decomposition items, which are respectively recorded as the first orthogonal matrix, the second orthogonal matrix, the first reference matrix, and the second reference matrix.

[0071] Optionally, the first orthogonal matrix, the second orthogonal matrix, the first reference matrix, and the second reference matrix may be determined based on the following method:

[0072] The first orthogonal matrix, the second orthogonal matrix, the first reference matrix and the second reference matrix are determined based on the following cosine-sine decomposition formula:

[0073]

[0074] Among them, Q′ represents the transposed correlation orthogonal complete matrix, S 2 represents the correlation target domain projection matrix, represents the second orthogonal matrix, V 1 represents the first orthogonal matrix, Γ(1) represents the first reference matrix, Σ(1) represents the second reference matrix, and V′ represents the total orthogonal matrix.

[0075] Step 2: Determine a singular value decomposition matrix according to the first orthogonal matrix, the second orthogonal matrix, the first reference matrix and the second reference matrix.

[0076] The singular value decomposition matrix may be a matrix obtained by performing singular value decomposition (SVD) on a matrix.

[0077] Specifically, the diagonal elements in the first reference matrix are taken out, and the inverse sine function is calculated for these diagonal elements. The diagonal elements in the second reference matrix are taken out, and the inverse cosine function is calculated for these diagonal elements. The calculated inverse sine function and inverse cosine function are used as diagonal elements to construct a diagonal matrix. Then, a reference matrix is ​​obtained by multiplying the constructed diagonal matrix with the second orthogonal matrix and the transposed first orthogonal matrix, and the reference matrix is ​​subjected to singular value decomposition to obtain a singular value decomposition matrix.

[0078] Optionally, the singular value decomposition matrix may be determined based on the following method:

[0079] A reference diagonal matrix is ​​determined according to the arc sine values ​​of each diagonal element in the first reference matrix and the arc cosine values ​​of each diagonal element in the second reference matrix.

[0080] The arcsine value may be a value calculated based on the arcsine function, and the arccosine value may be a value calculated based on the arccosine function. The reference diagonal matrix may be a diagonal matrix with arcsine values ​​and arccosine values ​​as diagonal elements.

[0081] Specifically, the arc sine value is calculated for each diagonal element in the first reference matrix, and the arc cosine value is calculated for each diagonal element in the second reference matrix. The calculated arc sine values ​​and arc cosine values ​​are used as diagonal elements to construct a diagonal matrix, namely, a reference diagonal matrix.

[0082] A reference matrix is ​​determined based on the reference diagonal matrix, the first orthogonal matrix, and the second orthogonal matrix, and a singular value decomposition matrix is ​​obtained by performing singular value decomposition based on the reference matrix.

[0083] The reference matrix may be a matrix obtained by performing matrix multiplication on the second orthogonal matrix, the reference diagonal matrix, and the transposed first orthogonal matrix.

[0084] Specifically, the reference matrix is ​​determined based on the following formula:

[0085]

[0086] Where A represents the reference matrix, represents the second orthogonal matrix, V 1 ′ represents the first orthogonal matrix after transposition, Θ 1 Represents the base diagonal matrix

[0087] And, the singular value decomposition matrix is ​​determined based on the following formula:

[0088]

[0089] Where A represents the reference matrix, represents the second orthogonal matrix, V 1 represents the first orthogonal matrix, Θ 2 represents the singular value decomposition matrix.

[0090] Step 3: Based on the N to-be-processed values ​​within a preset interval and the diagonal elements in the singular value decomposition matrix, determine a first intermediate matrix corresponding to each to-be-processed value and a second intermediate matrix corresponding to each to-be-processed value.

[0091] The preset interval may be a pre-set interval, such as [0, 1], etc. The value to be processed may be a value randomly selected within the preset interval. The first intermediate matrix may be a matrix corresponding to the value to be processed determined based on a cosine function, and the second intermediate matrix may be a matrix corresponding to the value to be processed determined based on a sine function.

[0092] Specifically, N to-be-processed values ​​are randomly selected within a preset interval, and the cosine value of the product of each diagonal element in the singular value decomposition matrix and each to-be-processed value is calculated to obtain a first intermediate matrix corresponding to each to-be-processed value. The sine value of the product of each diagonal element in the singular value decomposition matrix and each to-be-processed value is calculated to obtain a second intermediate matrix corresponding to each to-be-processed value.

[0093] Optionally, a first intermediate matrix corresponding to each value to be processed and a second intermediate matrix corresponding to each value to be processed are determined based on the following method:

[0094] For each value to be processed, a first intermediate matrix corresponding to the value to be processed and a second intermediate matrix corresponding to the value to be processed are determined based on the following formula:

[0095] r i (t n )=cos(t n θ i )

[0096] σ i (t n )=sin(t n θ i )

[0097] Among them, t n represents the nth value to be processed, θ i represents the i-th diagonal element in the singular value decomposition matrix, r i (t n) represents the element value corresponding to the nth value to be processed and the i-th diagonal element in the first intermediate matrix, σ i (t n ) represents the element value corresponding to the nth value to be processed and the i-th diagonal element in the second intermediate matrix.

[0098] Step 4: For each value to be processed, determine the correlation intermediate subspace corresponding to the value to be processed based on the correlation orthogonal complete matrix, the first orthogonal matrix, the second orthogonal matrix, and the first intermediate matrix and the second intermediate matrix corresponding to the value to be processed.

[0099] Specifically, a matrix to be processed corresponding to each value to be processed can be constructed according to the first orthogonal matrix, the second orthogonal matrix, the first intermediate matrix and the second intermediate matrix. The correlation orthogonal complete matrix is ​​multiplied with each matrix to be processed respectively to obtain a correlation intermediate subspace corresponding to each value to be processed, that is, to obtain a correlation intermediate subspace of N.

[0100] Optionally, the correlation intermediate subspace corresponding to the value to be processed can be determined in the following manner:

[0101] The correlation intermediate subspace corresponding to the value to be processed is determined based on the following formula:

[0102]

[0103] Among them, t n represents the nth value to be processed, ψ(t n ) represents the correlation intermediate subspace corresponding to the nth value to be processed, Q represents the correlation orthogonal complete matrix, represents the second orthogonal matrix, V 1 represents the first orthogonal matrix, Γ 2 (t n ) represents the first intermediate matrix corresponding to the nth value to be processed, Σ 2 (t n ) represents the second intermediate matrix corresponding to the nth value to be processed.

[0104] S250, determining a correlation subspace according to each correlation intermediate space, a correlation source domain projection matrix, and a correlation target domain projection matrix.

[0105] Specifically, N correlation intermediate spaces, the correlation source domain projection matrix, and the correlation target domain projection matrix are determined as N+2 correlation subspaces.

[0106] Accordingly, Figure 3 A schematic flow chart of a method for determining a principal component subspace provided in Embodiment 2 of the present invention.

[0107] S310: Obtain sample source domain data and sample target domain data.

[0108] S320. Perform principal component analysis on the sample source domain data according to the preset dimension to obtain a principal component source domain projection matrix.

[0109] S330. Perform principal component analysis on the sample target domain data according to the preset dimension to obtain a principal component target domain projection matrix.

[0110] S340, determining N principal component intermediate spaces according to the sample source domain data, the sample target domain data, the principal component source domain projection matrix, and the principal component target domain projection matrix.

[0111] S350, determining the principal component subspace according to the intermediate space of each principal component, the principal component source domain projection matrix, and the principal component target domain projection matrix.

[0112] It should be noted that S310 - S350 are similar to S210 - S250 and will not be described in detail in this embodiment.

[0113] It should also be noted that determining the principal component subspace and the correlation subspace is to establish a path on the Greismann manifold between the source domain and the target domain. After determining the principal component subspace and the correlation subspace, the weights, preset dimensions and preset numbers of the principal component subspace and the correlation subspace can be obtained by training.

[0114] The technical solution of the embodiment of the present invention obtains sample source domain data and sample target domain data, performs correlation analysis on the sample source domain data according to a preset dimension, obtains a correlation source domain projection matrix, performs correlation analysis on the sample target domain data according to a preset dimension, obtains a correlation target domain projection matrix, determines N correlation intermediate spaces according to the sample source domain data, the sample target domain data, the correlation source domain projection matrix, and the correlation target domain projection matrix, and then determines a correlation subspace according to each correlation intermediate space, the correlation source domain projection matrix, and the correlation target domain projection matrix; obtains sample source domain data and sample target domain data according to a preset dimension, performs correlation analysis on the sample target domain data, obtains a correlation target domain projection matrix, and determines N correlation intermediate spaces according to each correlation intermediate space, the correlation source domain projection matrix, and the correlation target domain projection matrix; Dimension, perform principal component analysis on the sample source domain data to obtain the principal component source domain projection matrix, perform principal component analysis on the sample target domain data according to the preset dimension, obtain the principal component target domain projection matrix, determine N principal component intermediate spaces according to the sample source domain data, the sample target domain data, the principal component source domain projection matrix and the principal component target domain projection matrix, and then determine the principal component subspace according to each principal component intermediate space, the principal component source domain projection matrix and the principal component target domain projection matrix, which solves the problem of single determination method and inaccurate determination of the intermediate feature space and realizes the accurate determination of the correlation subspace and the principal component subspace, so as to improve the effect of feature extraction quality.

[0115] Embodiment 3

[0116] As an optional implementation scheme of the above embodiments, the third embodiment of the present invention provides a flow chart of a method for determining a principal component subspace and a correlation subspace, wherein the explanations of the terms that are the same or corresponding to the above embodiments are not repeated here.

[0117] The method of this embodiment specifically includes the following steps:

[0118] 1. Find the source domain data X s (sample source domain data) and target domain data X t The CCA dimension reduction projection matrix S of (sample target domain data) 1 (correlation source domain projection matrix) and S 2 (correlation target domain projection matrix), and find S 1 The orthogonal complete matrix Q of .

[0119] 2. Find Q's 2 CS decomposition form

[0120]

[0121] Among them, Q′ represents the transposed correlation orthogonal complete matrix, S 2 represents the correlation target domain projection matrix, represents the second orthogonal matrix, V 1 represents the first orthogonal matrix, Γ(1) represents the first reference matrix, Σ(1) represents the second reference matrix, and V′ represents the total orthogonal matrix.

[0122] 3. Take the diagonal elements of Γ and Σ, calculate the inverse sine function of the elements in Γ, and calculate the inverse cosine function of the elements in Σ to obtain the set Matrix Θ 1 (The base diagonal matrix) is is a diagonal matrix with elements.

[0123] 4. By formula Get the matrix A (reference matrix), according to the formula Perform SVD decomposition on A to get Θ 2 (Singular Value Decomposition Matrix).

[0124] 5. Randomly select N values ​​t (to be processed) in [0,1] and calculate the diagonal matrix Γ 2 (t n ),Σ 2 (t n ), r i (t n )=cos(t n θ i ), σi (t n )=sin(t n θ i ), r i (t n ) and σ i (t n ) are Γ 2 (t n ) and Σ 2 (t n ), θ i is Θ 2 diagonal elements of .

[0125] 6. Calculation ψ(t n ) means S 1 and S 2 The intermediate subspace between (correlation intermediate subspace).

[0126] 7. Find the source domain data X s (sample source domain data) and target domain data X t The PCA dimension reduction projection matrix S of (sample target domain data) 3 (principal component source domain projection matrix) and S 4 (principal component target domain projection matrix), and find S 3 The orthogonal complete matrix Q 2 .

[0127] 8. Ask for Q 2 'S 4 CS decomposition form

[0128]

[0129] Among them, Q 2 ′ represents the transposed orthogonal complete matrix of the principal components, S 4 represents the principal component target domain projection matrix, represents the fourth orthogonal matrix, V 3 represents the third orthogonal matrix, Γ 3 (1) represents the third reference matrix, Σ 3 (1) represents the fourth reference matrix, and V″ represents the total orthogonal matrix.

[0130] 9. Take out Γ 3 and Σ 3 The diagonal elements of Γ 3 Find the inverse sine function of the elements in Σ 3 The elements in the set {ω i}, matrix Θ 3 is a diagonal matrix with ω as its element.

[0131] 10. By formula Get the matrix A 2 , according to the formula Perform SVD decomposition on A to get Θ 4 .

[0132] 11. Randomly select N values ​​t (to be processed) in [0,1] and calculate the diagonal matrix Γ 4 (t n ),Σ 4 (t n ), p i (t n )=cos(t n τ i ), q i (t n )=sin(t n τ i ), p i (t n ) and q i (t n ) are Γ 4 (t n ) and Σ 4 (t n ), τ i is Θ 4 diagonal elements of .

[0133] 12. Calculation φ(t n ) means S 3 and S 4 The intermediate subspace between (the intermediate subspace of principal components).

[0134] 13. By S 1 and S 2 and N intermediate subspaces ψ(t n ), S 3 and S 4 And N intermediate subspaces φ(t n ), the intermediate subspace of the two series is obtained, the remote sensing data to be processed are projected into the intermediate subspace of the two series respectively, and the two results are connected in series. The dimension of the obtained data is relatively large. In order to make the data easier to classify, the PLS algorithm is used to further reduce the dimension of the series data and retain the most important features.

[0135] It should be noted that Figure 4 Schematic diagram of the principle of intermediate subspace alignment, G N,dis a Greismann manifold, S1 and S2 are the feature spaces obtained by CCA in the source domain and target domain respectively, S1.3 and S1.6 are the intermediate subspaces obtained by sampling along the path from S1 to S2 on the Greismann manifold, S3 and S4 are the feature spaces obtained by PCA in the source domain and target domain respectively, S3.3 and S3.6 are the intermediate subspaces obtained by sampling along the path from S3 to S4 on the Greismann manifold. Figure 4 It can be seen that after combining PCA and CCA, the feature space of the source domain is the feature space of the target domain along two paths. Finally, the source domain and the target domain are simultaneously projected onto the intermediate subspace obtained by the two paths to achieve alignment of the source domain and the target domain.

[0136] For example, three remote sensing images of a certain region are used as experimental data for verification. These three images are remote sensing images taken in May, June and July of a certain region. The experiment selects data with real labels of 9 categories from the three images.

[0137] In the experiment, the data of May is recorded as PIC5, the data of June is recorded as PIC6, and the data of July is recorded as PIC7. The source domain data used in the experiment is labeled, and the target domain data is not labeled. For the sake of convenience, if PIC5 is used as the source domain data and PIC6 is used as the target domain data, this group of data is recorded as PIC56, and so on, PIC5, PIC6, and PIC7 constitute 6 groups of experimental data. Table 1 shows the classification accuracy comparison of the three algorithms.

[0138] Table 1

[0139] data KNN Classification Intermediate Subspace Alignment Algorithm The method of this embodiment PIC56 71.21％ 86.02％ 87.38％ PIC65 55.00％ 80.13％ 82.15％ PIC57 73.85％ 84.96％ 85.78％ PIC75 57.78％ 79.81％ 79.99％ PIC67 89.04％ 90.22％ 91.39％ PIC76 81.87％ 90.76％ 90.69％

[0140] In order to reflect the role of algorithm feature alignment, the experiment compares the experimental data before and after feature alignment (such as Figure 5 and Figure 6 As shown), the experimental data is PIC56, the experiment is carried out under the optimal parameters, and the band 1 and band 2 of the data are selected to make a scatter plot to represent the data distribution. Figure 5 It can be seen that before the PIC56 data alignment, the data distribution is not the same. There are many data in the source domain that are not distributed the same as the target domain data. Figure 6 It can be seen that after alignment, the distribution of source domain and target domain data becomes very similar, and the data overlap is higher. Figure 5 and Figure 6 Only the distribution of the overall data is given. Similar distribution of the overall data does not mean that the distribution of each type of data will also become similar. Therefore, the experiment further gives a comparison chart of each type of data of PIC56 before and after alignment, such as Figure 7-Figure 15As shown. From the comparison before and after the alignment of various features of PIC56, it can be seen that before the alignment, the distribution of various data is very far apart, with basically no overlapping parts. After the feature alignment, the distribution of the data becomes similar and overlaps, especially for categories 3, 5, 6, and 8. It can be seen that after the feature alignment, the distribution of the data overlaps very well. It can be seen that the feature alignment effect of the method of this embodiment is significant.

[0141] The technical solution of the embodiment of the present invention obtains the feature space of the data by retaining the maximum amount of information of the data through PCA, determines the correlation between the source domain and the target domain through CCA and then obtains the feature space of the two domains, and the connection between the two domains established through the feature space of PCA and CCA is closer, which solves the problem of poor feature extraction effect of remote sensing image data and achieves the effect of improving the quality of remote sensing feature extraction.

[0142] Embodiment 4

[0143] Fig.16 This is a schematic diagram of the structure of a remote sensing feature extraction device provided in Embodiment 4 of the present invention. The device includes: a principal component projection module 310, a correlation projection module 320 and a feature extraction module 330.

[0144] Among them, the principal component projection module 310 is used to project the remote sensing data to be processed onto a preset number of principal component subspaces to obtain a preset number of principal component projection data; wherein the preset number is N+2, N is a positive integer, and the remote sensing data to be processed includes remote sensing target domain data; the correlation projection module 320 is used to project the remote sensing data to be processed onto a preset number of correlation subspaces to obtain a preset number of correlation projection data; the feature extraction module 330 is used to determine the remote sensing projection data based on the principal component projection data and the correlation projection data, and perform dimensionality reduction processing on the remote sensing projection data to obtain remote sensing feature data corresponding to the remote sensing data to be processed.

[0145] Optionally, the correlation subspace is obtained based on the following correlation subspace determination module, which is used to obtain sample source domain data and sample target domain data; perform correlation analysis on the sample source domain data according to a preset dimension to obtain a correlation source domain projection matrix; perform correlation analysis on the sample target domain data according to the preset dimension to obtain a correlation target domain projection matrix; determine N correlation intermediate spaces based on the sample source domain data, the sample target domain data, the correlation source domain projection matrix and the correlation target domain projection matrix; determine the correlation subspace based on each of the correlation intermediate spaces, the correlation source domain projection matrix and the correlation target domain projection matrix.

[0146] Optionally, the correlation subspace determination module is also used to determine the correlation orthogonal complete matrix of the correlation source domain projection matrix, perform cosine-sine decomposition based on the product of the transposed correlation orthogonal complete matrix and the correlation target domain projection matrix to obtain a first orthogonal matrix, a second orthogonal matrix, a first reference matrix and a second reference matrix; determine a singular value decomposition matrix based on the first orthogonal matrix, the second orthogonal matrix, the first reference matrix and the second reference matrix; determine a first intermediate matrix corresponding to each value to be processed and a second intermediate matrix corresponding to each value to be processed based on N values ​​to be processed within a preset interval and each diagonal element in the singular value decomposition matrix; for each value to be processed, determine the correlation intermediate subspace corresponding to the value to be processed based on the correlation orthogonal complete matrix, the first orthogonal matrix, the second orthogonal matrix, and the first intermediate matrix and the second intermediate matrix corresponding to the value to be processed.

[0147] Optionally, the correlation subspace determination module is also used to determine a reference diagonal matrix based on the arcsine values ​​of each diagonal element in the first reference matrix and the arccosine values ​​of each diagonal element in the second reference matrix; determine a reference matrix based on the reference diagonal matrix, the first orthogonal matrix and the second orthogonal matrix, and perform singular value decomposition based on the reference matrix to obtain a singular value decomposition matrix.

[0148] Optionally, the correlation subspace determination module is further used to determine a reference matrix based on the following formula:

[0149]

[0150] Wherein, A represents the reference matrix, Denotes the second orthogonal matrix, V 1 ' represents the first orthogonal matrix after transposition, Θ 1 represents the reference diagonal matrix;

[0151] Accordingly, the singular value decomposition matrix is ​​determined based on the following formula:

[0152]

[0153] Wherein, A represents the reference matrix, Denotes the second orthogonal matrix, V 1 represents the first orthogonal matrix, Θ 2 represents the singular value decomposition matrix.

[0154] Optionally, the correlation subspace determination module is further used to determine the first orthogonal matrix, the second orthogonal matrix, the first reference matrix and the second reference matrix based on the following cosine-sine decomposition formula:

[0155]

[0156] Where Q′ represents the transposed orthogonal complete matrix of the correlation, S 2 represents the correlation target domain projection matrix, Denotes the second orthogonal matrix, V 1 represents the first orthogonal matrix, Γ(1) represents the first reference matrix, Σ(1) represents the second reference matrix, and V′ represents the total orthogonal matrix.

[0157] Optionally, the correlation subspace determination module is further used to determine, for each value to be processed, a first intermediate matrix corresponding to the value to be processed and a second intermediate matrix corresponding to the value to be processed based on the following formula:

[0158] r i (t n )=cos(t n θ i )

[0159] σ i (t n )=sin(t n θ i )

[0160] Among them, t n represents the nth value to be processed, θ i represents the i-th diagonal element in the singular value decomposition matrix, r i (t n ) represents the element value corresponding to the nth value to be processed and the i-th diagonal element in the first intermediate matrix, σ i (t n ) represents the element value in the second intermediate matrix corresponding to the nth value to be processed and the i-th diagonal element.

[0161] Optionally, the correlation subspace determination module is further configured to determine a correlation intermediate subspace corresponding to the value to be processed based on the following formula:

[0162]

[0163] Among them, t n represents the nth value to be processed, ψ(t n ) represents the correlation intermediate subspace corresponding to the nth value to be processed, Q represents the correlation orthogonal complete matrix, Denotes the second orthogonal matrix, V 1 represents the first orthogonal matrix, Γ 2 (t n) represents the first intermediate matrix corresponding to the nth value to be processed, Σ 2 (t n ) represents the second intermediate matrix corresponding to the nth value to be processed.

[0164] Optionally, the correlation subspace is obtained based on the following principal component subspace determination module, which is used to obtain sample source domain data and sample target domain data; perform principal component analysis on the sample source domain data according to a preset dimension to obtain a principal component source domain projection matrix; perform principal component analysis on the sample target domain data according to the preset dimension to obtain a principal component target domain projection matrix; determine N principal component intermediate spaces according to the sample source domain data, the sample target domain data, the principal component source domain projection matrix and the principal component target domain projection matrix; determine the principal component subspace according to each of the principal component intermediate spaces, the principal component source domain projection matrix and the principal component target domain projection matrix.

[0165] The technical solution of the embodiment of the present invention is to project the remote sensing data to be processed onto a preset number of principal component subspaces to obtain a preset number of principal component projection data, project the remote sensing data to be processed onto a preset number of correlation subspaces to obtain a preset number of correlation projection data, determine the remote sensing projection data based on the principal component projection data and the correlation projection data, and perform dimensionality reduction processing on the remote sensing projection data to obtain remote sensing feature data corresponding to the remote sensing data to be processed, thereby solving the problem of poor feature extraction effect of remote sensing image data and achieving the effect of improving the quality of remote sensing feature extraction.

[0166] The remote sensing feature extraction device provided in the embodiment of the present invention can execute the remote sensing feature extraction method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0167] It is worth noting that the various units and modules included in the above-mentioned device are only divided according to functional logic, but are not limited to the above-mentioned division, as long as the corresponding functions can be achieved; in addition, the specific names of the functional units are only for the convenience of distinguishing each other, and are not used to limit the protection scope of the embodiments of the present invention.

[0168] Embodiment 5

[0169] Fig.17 This is a schematic diagram of the structure of an electronic device provided in Embodiment 5 of the present invention. Fig.17 A block diagram of an exemplary electronic device 40 suitable for implementing exemplary embodiments of the present invention is shown. Fig.17 The electronic device 40 shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present invention.

[0170] like Fig.17 As shown, the electronic device 40 is in the form of a general computing device. The components of the electronic device 40 may include but are not limited to: one or more processors or processing units 401, a system memory 402, and a bus 403 connecting different system components (including the system memory 402 and the processing unit 401).

[0171] Bus 403 represents one or more of several types of bus structures, including a memory bus or memory controller, a peripheral bus, an accelerated graphics port, a processor or a local bus using any of a variety of bus structures. For example, these architectures include but are not limited to Industry Standard Architecture (ISA) bus, Micro Channel Architecture (MAC) bus, Enhanced ISA bus, Video Electronics Standards Association (VESA) local bus and Peripheral Component Interconnect (PCI) bus.

[0172] The electronic device 40 typically includes a variety of computer system readable media. These media can be any available media that can be accessed by the electronic device 40, including volatile and non-volatile media, removable and non-removable media.

[0173] System memory 402 may include computer system readable media in the form of volatile memory, such as random access memory (RAM) 404 and / or cache 405. Electronic device 40 may further include other removable / non-removable, volatile / non-volatile computer system storage media. By way of example only, storage system 406 may be used to read and write non-removable, non-volatile magnetic media ( Fig.17 not shown, usually called a "hard drive"). Although Fig.17 Not shown in the figure, a disk drive for reading and writing to a removable non-volatile disk (e.g., a "floppy disk"), and an optical disk drive for reading and writing to a removable non-volatile optical disk (e.g., a CD-ROM, a DVD-ROM, or other optical media) may be provided. In these cases, each drive may be connected to the bus 403 via one or more data medium interfaces. The system memory 402 may include at least one program product having a set (e.g., at least one) of program modules that are configured to perform the functions of the embodiments of the present invention.

[0174] A program / utility 408 having a set (at least one) of program modules 407 may be stored, for example, in system memory 402, such program modules 407 including, but not limited to, an operating system, one or more application programs, other program modules, and program data, each of which or some combination may include an implementation of a network environment. Program modules 407 generally perform the functions and / or methods of the embodiments described herein.

[0175] The electronic device 40 may also communicate with one or more external devices 409 (e.g., keyboards, pointing devices, displays 410, etc.), one or more devices that enable a user to interact with the electronic device 40, and / or any device that enables the electronic device 40 to communicate with one or more other computing devices (e.g., a network card, a modem, etc.). Such communication may be performed through an I / O interface (input / output interface) 411. Furthermore, the electronic device 40 may also communicate with one or more networks (e.g., a local area network (LAN), a wide area network (WAN), and / or a public network, such as the Internet) through a network adapter 412. As shown, the network adapter 412 communicates with other modules of the electronic device 40 through the bus 403. It should be understood that although Fig.17 Not shown, other hardware and / or software modules may be used in conjunction with the electronic device 40, including but not limited to microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0176] The processing unit 401 executes various functional applications and data processing by running the programs stored in the system memory 402, such as implementing the remote sensing feature extraction method provided in the embodiment of the present invention.

[0177] Embodiment 6

[0178] Embodiment 6 of the present invention further provides a storage medium containing computer executable instructions, wherein the computer executable instructions are used to perform a remote sensing feature extraction method when executed by a computer processor, the method comprising:

[0179] Projecting the remote sensing data to be processed onto a preset number of principal component subspaces to obtain a preset number of principal component projection data; wherein the preset number is N+2, N is a positive integer, and the remote sensing data to be processed includes remote sensing target domain data;

[0180] Projecting the remote sensing data to be processed onto a preset number of correlation subspaces to obtain a preset number of correlation projection data;

[0181] According to the principal component projection data and the correlation projection data, remote sensing projection data are determined, and dimension reduction processing is performed on the remote sensing projection data to obtain remote sensing feature data corresponding to the remote sensing data to be processed.

[0182] The computer storage medium of the embodiment of the present invention can adopt any combination of one or more computer-readable media. The computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples (non-exhaustive list) of computer-readable storage media include: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this document, a computer-readable storage medium can be any tangible medium containing or storing a program, which can be used by an instruction execution system, device or device or used in combination with it.

[0183] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, which carry computer-readable program code. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Computer-readable signal media may also be any computer-readable medium other than a computer-readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0184] Program code embodied on a computer readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.

[0185] Computer program code for performing the operation of embodiments of the present invention may be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages, such as Java, Smalltalk, C++, and conventional procedural programming languages, such as "C" or similar programming languages. The program code may be executed entirely on a user's computer, partially on a user's computer, as an independent software package, partially on a user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0186] Note that the above are only preferred embodiments of the present invention and the technical principles used. Those skilled in the art will understand that the present invention is not limited to the specific embodiments described herein, and that various obvious changes, readjustments and substitutions can be made by those skilled in the art without departing from the scope of protection of the present invention. Therefore, although the present invention has been described in more detail through the above embodiments, the present invention is not limited to the above embodiments, and may include more other equivalent embodiments without departing from the concept of the present invention, and the scope of the present invention is determined by the scope of the appended claims.

Claims

1. A remote sensing feature extraction method, It is characterized in that include: Projecting the remote sensing data to be processed onto a preset number of principal component subspaces to obtain a preset number of principal component projection data; wherein the preset number is N+2, N is a positive integer, and the remote sensing data to be processed includes remote sensing target domain data; Projecting the remote sensing data to be processed onto a preset number of correlation subspaces to obtain a preset number of correlation projection data; According to the principal component projection data and the correlation projection data, remote sensing projection data is determined, and dimension reduction processing is performed on the remote sensing projection data to obtain remote sensing feature data corresponding to the remote sensing data to be processed, wherein the remote sensing projection data is data obtained by serially splicing the principal component projection data and the correlation projection data.

2. The method according to claim 1, It is characterized in that The correlation subspace is obtained based on the following method: Obtain sample source domain data and sample target domain data; According to the preset dimension, the sample source domain data is subjected to correlation analysis to obtain a correlation source domain projection matrix; According to the preset dimension, the sample target domain data is subjected to correlation analysis to obtain a correlation target domain projection matrix; Determine N correlation intermediate spaces according to the sample source domain data, the sample target domain data, the correlation source domain projection matrix, and the correlation target domain projection matrix; A correlation subspace is determined according to each of the correlation intermediate spaces, the correlation source domain projection matrix, and the correlation target domain projection matrix.

3. The method according to claim 2, It is characterized in that The determining of N correlation intermediate spaces according to the sample source domain data, the sample target domain data, the correlation source domain projection matrix and the correlation target domain projection matrix comprises: Determine a correlation orthogonal complete matrix of the correlation source domain projection matrix, perform cosine-sine decomposition according to the product of the transposed correlation orthogonal complete matrix and the correlation target domain projection matrix, and obtain a first orthogonal matrix, a second orthogonal matrix, a first reference matrix, and a second reference matrix; Determine a singular value decomposition matrix according to the first orthogonal matrix, the second orthogonal matrix, the first reference matrix, and the second reference matrix; Based on the N to-be-processed values ​​within a preset interval and the diagonal elements in the singular value decomposition matrix, determine a first intermediate matrix corresponding to each to-be-processed value and a second intermediate matrix corresponding to each to-be-processed value; For each value to be processed, a correlation intermediate subspace corresponding to the value to be processed is determined based on the correlation orthogonal complete matrix, the first orthogonal matrix, the second orthogonal matrix, and the first intermediate matrix and the second intermediate matrix corresponding to the value to be processed.

4. The method according to claim 3, It is characterized in that The step of determining a singular value decomposition matrix according to the first orthogonal matrix, the second orthogonal matrix, the first reference matrix, and the second reference matrix comprises: Determine a reference diagonal matrix according to the arc sine values ​​of each diagonal element in the first reference matrix and the arc cosine values ​​of each diagonal element in the second reference matrix; A reference matrix is ​​determined based on the reference diagonal matrix, the first orthogonal matrix, and the second orthogonal matrix, and a singular value decomposition matrix is ​​obtained by performing singular value decomposition based on the reference matrix.

5. The method according to claim 4, It is characterized in that The step of determining a reference matrix based on the reference diagonal matrix, the first orthogonal matrix, and performing singular value decomposition based on the reference matrix to obtain a singular value decomposition matrix includes: The reference matrix is ​​determined based on the following formula: Wherein, A represents the reference matrix, Denotes the second orthogonal matrix, V 1 ′ represents the first orthogonal matrix after transposition, Θ 1 represents the reference diagonal matrix; Accordingly, the singular value decomposition matrix is ​​determined based on the following formula: Wherein, A represents the reference matrix, Denotes the second orthogonal matrix, V 1 represents the first orthogonal matrix, Θ 2 represents the singular value decomposition matrix.

6. The method according to claim 3, It is characterized in that The method performs cosine-sine decomposition based on the product of the transposed correlation orthogonal complete matrix and the correlation target domain projection matrix to obtain a first orthogonal matrix, a second orthogonal matrix, a first reference matrix and a second reference matrix, including: The first orthogonal matrix, the second orthogonal matrix, the first reference matrix and the second reference matrix are determined based on the following cosine-sine decomposition formula: Among them, Q ′ represents the correlation orthogonal complete matrix after transposition, S 2 represents the correlation target domain projection matrix, Denotes the second orthogonal matrix, V 1 represents the first orthogonal matrix, Γ(1) represents the first reference matrix, Σ(1) represents the second reference matrix, V ′ represents the total orthogonal matrix.

7. The method according to claim 3, It is characterized in that The determining, based on the N to-be-processed values ​​within the preset interval and the diagonal elements in the singular value decomposition matrix, a first intermediate matrix corresponding to each to-be-processed value and a second intermediate matrix corresponding to each to-be-processed value comprises: For each value to be processed, a first intermediate matrix corresponding to the value to be processed and a second intermediate matrix corresponding to the value to be processed are determined based on the following formula: r i (t n )=cos(t n θ i ) s i (t n )=sin(t n i i ) Among them, t n represents the nth value to be processed, θ i represents the i-th diagonal element in the singular value decomposition matrix, r i (t n ) represents the element value corresponding to the nth value to be processed and the i-th diagonal element in the first intermediate matrix, σ i (t n ) represents the element value in the second intermediate matrix corresponding to the nth value to be processed and the i-th diagonal element.

8. The method according to claim 3, It is characterized in that The step of determining a correlation intermediate subspace corresponding to the value to be processed according to the correlation orthogonal complete matrix, the first orthogonal matrix, the second orthogonal matrix, and the first intermediate matrix and the second intermediate matrix corresponding to the value to be processed comprises: The correlation intermediate subspace corresponding to the value to be processed is determined based on the following formula: Among them, t n represents the nth value to be processed, ψ(t n ) represents the correlation intermediate subspace corresponding to the nth value to be processed, Q represents the correlation orthogonal complete matrix, Denotes the second orthogonal matrix, V 1 represents the first orthogonal matrix, Γ 2 (t n ) represents the first intermediate matrix corresponding to the nth value to be processed, Σ 2 (t n ) represents the second intermediate matrix corresponding to the nth value to be processed.

9. The method according to claim 1, It is characterized in that The principal component subspace is obtained based on the following method: Obtain sample source domain data and sample target domain data; According to the preset dimension, the sample source domain data is subjected to principal component analysis to obtain a principal component source domain projection matrix; According to the preset dimension, the sample target domain data is subjected to principal component analysis to obtain a principal component target domain projection matrix; Determine N principal component intermediate spaces according to the sample source domain data, the sample target domain data, the principal component source domain projection matrix, and the principal component target domain projection matrix; A principal component subspace is determined according to each of the principal component intermediate spaces, the principal component source domain projection matrix, and the principal component target domain projection matrix.

10. A remote sensing feature extraction device, It is characterized in that include: A principal component projection module is used to project the remote sensing data to be processed onto a preset number of principal component subspaces to obtain a preset number of principal component projection data; wherein the preset number is N+2, N is a positive integer, and the remote sensing data to be processed includes remote sensing target domain data; A correlation projection module, used for projecting the remote sensing data to be processed onto a preset number of correlation subspaces to obtain a preset number of correlation projection data; A feature extraction module is used to determine remote sensing projection data based on the principal component projection data and the correlation projection data, and perform dimensionality reduction processing on the remote sensing projection data to obtain remote sensing feature data corresponding to the remote sensing data to be processed, wherein the remote sensing projection data is the data obtained by concatenating the principal component projection data and the correlation projection data.

Images