Hyperspectral image processing method, apparatus, device, and medium

Abstract

The application provides a hyperspectral image processing method, device, equipment and medium, wherein the method comprises: in response to obtaining a hyperspectral image to be processed and a corresponding label, determining a first data set and a second data set based on the hyperspectral image to be processed and the label by using a preset data set division rule; expanding the first data set by using the second data set to obtain a third data set based on a preset spatial spectrum self-training rule; determining a measure matrix based on the hyperspectral image to be processed and the third data set by using a preset measure constraint manifold rule; and performing dimension reduction on the hyperspectral image to be processed by using the measure matrix to obtain a dimension-reduced hyperspectral image. The method of the application can obtain better hyperspectral dimension-reduced data.

Description

Technical Field

[0001] This invention relates to the field of remote sensing image processing technology, and in particular to a hyperspectral image processing method, apparatus, electronic device, non-transitory computer-readable storage medium, and computer program product. Background Technology

[0002] Hyperspectral remote sensing images provide continuous spectral bands and carry rich information about ground features, thus enabling their use in various applications, such as land use classification, environmental monitoring, mineral exploration, and medical imaging. Dimensionality reduction is a key task in hyperspectral remote sensing image processing; it involves mapping the original hyperspectral image to a lower-dimensional space.

[0003] Existing hyperspectral image dimensionality reduction methods are mainly based on manifold learning, which aims to preserve the local structural relationships between data while reducing the dimensionality of hyperspectral images. However, existing manifold learning-based dimensionality reduction methods cannot classify samples while simplifying the data, and they struggle to achieve good dimensionality reduction results in the absence of label information. Summary of the Invention

[0004] This invention provides a hyperspectral image processing method, apparatus, electronic device, non-transitory computer-readable storage medium, and computer program product to address the deficiencies in the prior art.

[0005] This invention provides a hyperspectral image processing method, comprising: In response to acquiring the hyperspectral image to be processed and its corresponding label, a first dataset and a second dataset are determined based on the hyperspectral image to be processed and the label using a preset dataset partitioning rule; wherein, the label represents the category of the target object in the hyperspectral image to be processed, the first dataset includes multiple initially labeled sample points marked based on the label, and the second dataset includes multiple unlabeled sample points. Based on a preset spatial spectral self-training rule, the first dataset is expanded using the second dataset to obtain a third dataset; wherein, the third dataset includes multiple pseudo-labeled sample points added on the basis of the first dataset; Based on the hyperspectral image to be processed and the third dataset, the measure matrix is ​​determined using a preset measure-constrained manifold rule; The dimensionality of the hyperspectral image to be processed is reduced using the metric matrix to obtain the dimensionality-reduced hyperspectral image.

[0006] The present invention also provides a hyperspectral image processing apparatus, comprising: The data preprocessing module is configured to: in response to acquiring a hyperspectral image to be processed and its corresponding label, determine a first dataset and a second dataset based on the hyperspectral image to be processed and the label using a preset dataset partitioning rule; wherein the label represents the category of the target object in the hyperspectral image to be processed, the first dataset includes multiple initially labeled sample points marked based on the label, and the second dataset includes multiple unlabeled sample points; The spatial spectrum self-training module is configured to: expand the first dataset using the second dataset based on preset spatial spectrum self-training rules to obtain a third dataset; wherein the third dataset includes multiple pseudo-labeled sample points added on the basis of the first dataset; The metric matrix determination module is configured to: determine the metric matrix based on the hyperspectral image to be processed and the third dataset, using a preset metric constraint manifold rule; The dimension reduction execution module is configured to: use the metric matrix to reduce the dimension of the hyperspectral image to be processed, and obtain the dimension-reduced hyperspectral image.

[0007] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the hyperspectral image processing method as described above.

[0008] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the hyperspectral image processing method as described above.

[0009] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the hyperspectral image processing method as described above.

[0010] As described above, the hyperspectral image processing method provided in this embodiment of the invention addresses the problems of existing methods by combining spatial spectral self-training and metric constraints on manifolds, thereby obtaining better hyperspectral dimensionality reduction data. Attached Figure Description

[0011] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0012] Figure 1This is one of the flowcharts of the hyperspectral image processing method provided in the embodiments of the present invention; Figure 2 This is a second schematic flowchart of the hyperspectral image processing method provided in this embodiment of the invention; Figure 3 This is the third flowchart of the hyperspectral image processing method provided in this embodiment of the invention; Figure 4 This is the fourth flowchart of the hyperspectral image processing method provided in this embodiment of the invention; Figure 5 This is a schematic diagram of the structure of the hyperspectral image processing device provided in an embodiment of the present invention; Figure 6 This is a schematic diagram of the structure of the electronic device provided in an embodiment of the present invention. Detailed Implementation

[0013] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0014] To address the problems of the prior art, the inventors of this invention, after research, propose a hyperspectral image processing scheme as described in the following embodiments.

[0015] Example: The implementation details of the hyperspectral image processing scheme of the present invention will be described below with reference to the accompanying drawings.

[0016] Figure 1 This is a schematic flowchart of a hyperspectral image processing method provided in an exemplary embodiment of the present invention. This embodiment can be applied to electronic devices (e.g., servers or cloud computing platforms), such as... Figure 1 As shown, the hyperspectral image processing method includes the following steps: S110. In response to obtaining the hyperspectral image to be processed and the corresponding label, a first dataset and a second dataset are determined based on the hyperspectral image to be processed and the label using a preset dataset partitioning rule.

[0017] The label represents the category of the target object in the hyperspectral image to be processed. The first dataset includes multiple initially labeled sample points based on the label, and the second dataset includes multiple unlabeled sample points.

[0018] Reference Figure 2As an optional example, initially labeled sample points may include category 1 samples and category 2 samples. Unlabeled sample points may include category 1 samples to be identified.

[0019] The present invention does not limit the method of acquisition. For example, an electronic device (e.g., a server or cloud computing platform) acting as the execution subject can communicate with a server storing the "hyperspectral image to be processed and the corresponding tag" to obtain the "hyperspectral image to be processed and the corresponding tag".

[0020] S120. Based on the preset spatial spectrum self-training rules, the first dataset is expanded using the second dataset to obtain the third dataset.

[0021] The third dataset includes multiple pseudo-labeled sample points added on top of the first dataset.

[0022] S130. Based on the hyperspectral image to be processed and the third dataset, determine the metric matrix using a preset metric constraint manifold rule.

[0023] S140. Using the metric matrix, the hyperspectral image to be processed is reduced in dimension to obtain a reduced hyperspectral image.

[0024] As described above, the hyperspectral image processing method provided in this embodiment of the invention addresses the problems of existing methods by combining spatial spectral self-training and metric constraints on manifolds, thereby obtaining better hyperspectral dimensionality reduction data.

[0025] exist Figure 1 Based on the embodiments, as an optional implementation method, refer to Figure 3 Step S110 may include the following steps: S1110. Based on the tensor representation of the hyperspectral image, the hyperspectral image is divided into multiple pixels as sample points.

[0026] As an optional example, a hyperspectral remote sensing image can be represented as The labels corresponding to hyperspectral remote sensing images can be represented as: Where H, W, and B represent the length, width, and spectral dimension of the hyperspectral image, respectively.

[0027] Reference Figure 2 A hyperspectral image is a data cube, which can be represented by a tensor. The vector of each sample point in the tensor is the pixel radiation value corresponding to each band.

[0028] Reference Figure 2 The small squares shown in the figure are the pixels, or sample points.

[0029] S1120. In response to the pixel having the corresponding label, the pixel is assigned to the first dataset and denoted as the initial labeled sample point.

[0030] S1130. In response to the fact that the pixel does not have a corresponding label, the pixel is assigned to the second dataset and recorded as the unlabeled sample point.

[0031] exist Figure 1 Based on the embodiments, as an optional implementation method, refer to Figure 2 Step S120 may include the following steps: Step 1, in the second dataset (by...) Figure 2 In the blank squares representing the composition of unlabeled samples shown, determine the spatial nearest neighbor (i.e., ...) corresponding to each of the initially labeled sample points in the first dataset. Figure 2 (See "Locating Spatial Nearest Neighbors").

[0032] Step 2: According to the preset judgment rules, determine the spatial spectral nearest neighbor points corresponding to each initial labeled sample point from the spatial nearest neighbor points corresponding to the initial labeled sample point.

[0033] As an optional example, step 2 can be implemented as follows: Step 21: For each of the initial labeled sample points, calculate the Euclidean distance between the initial labeled sample point and each of its corresponding spatial nearest neighbors.

[0034] Step 22: In response to any spatial nearest neighbor point having a Euclidean distance less than a preset distance threshold, the spatial nearest neighbor point is determined to be the spatial spectral nearest neighbor point of the corresponding initial labeled sample point.

[0035] Alternatively, steps 21 and 22 above can be performed using the following calculation formula;

[0036]

[0037] in, d This represents the function for calculating Euclidean distance; Indicates the first dataset, the first... i 100 labeled sample points; express The j-th spatial nearest neighbor; b express Adaptive threshold (refer to) Figure 2 It can also be used t express (adaptive threshold) Indicates the first dataset, the first... ik10 labeled sample points (and their correlation) (Labeled samples of the same category).

[0038] Step 3: Assign a label to the spatial spectral nearest neighbor of each initial labeled sample point, so that the spatial spectral nearest neighbor forms the pseudo-labeled sample point.

[0039] Step 4: Merge the spatial spectral nearest neighbors that have been labeled into the first dataset; Step 5: Iterate through steps 1 to 4 until the preset iteration termination condition is met, and obtain the third dataset from the expanded first dataset.

[0040] Reference Figure 2 In each iteration, spatial spectral nearest neighbors can be learned, and pseudo-label learning feedback can be performed.

[0041] The preset iteration termination condition can be that no more pseudo-labels are generated.

[0042] exist Figure 1 Based on the embodiments, as an optional implementation method, refer to Figure 2 Step S130 can be achieved in the following way: Optionally, based on the hyperspectral image to be processed and the third dataset, the metric matrix A is solved using the following calculation formulas (1) to (5); (1) (2) (3) (4) (5) in, This indicates the first hyperspectral image to be processed. i One sample point; This indicates the first hyperspectral image to be processed. j One sample point; This refers to the hyperspectral image to be processed; Indicates the first i Euclidean distance between each sample point and its corresponding spatial nearest neighbor; Indicates the first j Euclidean distance between each sample point and its corresponding spatial nearest neighbor; tr represents the Gaussian kernel function; tr represents the sum of the elements on the diagonal of the matrix. represents the hyperparameters of the measure constraint part; L represents the Laplacian matrix; The hyperparameter represents the manifold regularization term; n represents the number of labeled sample points in the third dataset; Represents a loss term; Represents the regularization term for the complexity of the metric matrix; This represents the i-th labeled sample point in the third dataset; This represents the j-th labeled sample point in the third dataset; Indicates correspondence Tags; Indicates correspondence The tag.

[0043] It should be noted that formula (3) is the specific calculation of the Gaussian kernel function; refer to Figure 2 Equation (2) is a regularization term defined based on the Gaussian kernel function and Mahalanobis distance metric, used to achieve low-dimensional manifold embedding; refer to Figure 2 Equation (1) is the objective function that imposes a "regularized distance metric constraint" on the manifold, used to separate intra-class and inter-class samples. The metric matrix A is obtained by minimizing the objective function, i.e. Figure 2 The metric matrix A is shown below.

[0044] Among them, reference Figure 2 Purple arrows represent inter-class distances, black arrows represent intra-class distances, green squares and green circles represent pseudo-labeled samples, and blank squares and blank circles represent unlabeled samples.

[0045] exist Figure 1 Based on the embodiments, as an optional implementation method, refer to Figure 4 The hyperspectral image processing method further includes: S150. Using a preset classifier, classify the dimensionality-reduced hyperspectral image to obtain the classification result.

[0046] As an optional example, the preset classifier includes k -NN classifier.

[0047] As described above, the hyperspectral image processing method provided in this embodiment of the invention firstly, to fully utilize unlabeled samples, an adaptive threshold can be calculated based on the spectral neighborhood information of labeled samples. This threshold is then used to filter the spatial nearest neighbors of labeled samples, obtaining spatial spectral nearest neighbors. Pseudo-labels are then assigned to these spatial spectral nearest neighbors and added to the labeled sample set for multiple iterations. Secondly, to fully explore the potential local structural information of hyperspectral images and improve the discriminative power of data in low-dimensional space, a specific manifold space can be constructed based on Mahalanobis distance and Gaussian kernel function, representing the local structure of the data as a manifold. Finally, to make the distinction between intra-class and inter-class samples more obvious, regularized metric learning can be used to impose metric constraints on the manifold, obtaining good generalization ability and preventing overfitting.

[0048] In summary, the hyperspectral image processing method provided by the embodiments of the present invention addresses the problems of the prior art by combining spatial spectral self-training and metric constraints on manifolds, thereby obtaining better hyperspectral dimensionality reduction data and classification results.

[0049] The hyperspectral image processing apparatus provided by the present invention will be described below. The hyperspectral image processing apparatus described below can be referred to in correspondence with the hyperspectral image processing method described above.

[0050] Figure 5 This is a schematic diagram of the structure of a hyperspectral image processing apparatus provided in an exemplary embodiment of the present invention. Figure 5 As shown, the hyperspectral image processing apparatus includes: The data preprocessing module 510 is configured to: in response to acquiring the hyperspectral image to be processed and its corresponding label, determine a first dataset and a second dataset based on the hyperspectral image to be processed and the label using a preset dataset partitioning rule; wherein the label represents the category of the target object in the hyperspectral image to be processed, the first dataset includes multiple initially labeled sample points labeled based on the label, and the second dataset includes multiple unlabeled sample points; the spatial spectrum self-training module 520 is configured to: expand the first dataset using the second dataset based on a preset spatial spectrum self-training rule to obtain a third dataset; wherein the third dataset includes multiple pseudo-labeled sample points added on the basis of the first dataset; the metric matrix determination module 530 is configured to: determine the metric matrix based on the hyperspectral image to be processed and the third dataset using a preset metric constraint manifold rule; the dimensionality reduction execution module 540 is configured to: reduce the dimensionality of the hyperspectral image to be processed using the metric matrix to obtain a dimensionality-reduced hyperspectral image.

[0051] Figure 6 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 6 As shown, the electronic device may include: a processor 610, a communication interface 620, a memory 630, and a communication bus 640, wherein the processor 610, the communication interface 620, and the memory 630 communicate with each other through the communication bus 640. The processor 610 can call logical instructions in the memory 630 to execute a hyperspectral image processing method, the method comprising: in response to acquiring a hyperspectral image to be processed and its corresponding label, determining a first dataset and a second dataset based on the hyperspectral image to be processed and the label using a preset dataset partitioning rule; wherein the label represents the category of a target object in the hyperspectral image to be processed, the first dataset includes multiple initially labeled sample points marked based on the label, and the second dataset includes multiple unlabeled sample points; expanding the first dataset using the second dataset based on a preset spatial spectral self-training rule to obtain a third dataset; wherein the third dataset includes multiple pseudo-labeled sample points added on top of the first dataset; determining a metric matrix based on the hyperspectral image to be processed and the third dataset using a preset metric constraint manifold rule; and using the metric matrix to reduce the dimensionality of the hyperspectral image to be processed to obtain a dimensionality-reduced hyperspectral image.

[0052] Furthermore, the logical instructions in the aforementioned memory 630 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0053] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the hyperspectral image processing method provided by the above methods. The method includes: in response to acquiring a hyperspectral image to be processed and a corresponding label, determining a first dataset and a second dataset based on the hyperspectral image to be processed and the label using a preset dataset partitioning rule; wherein the label represents the category of a target object in the hyperspectral image to be processed, the first dataset includes multiple initially labeled sample points labeled based on the label, and the second dataset includes multiple unlabeled sample points; expanding the first dataset using the second dataset based on a preset spatial spectral self-training rule to obtain a third dataset; wherein the third dataset includes multiple pseudo-labeled sample points added on top of the first dataset; determining a metric matrix based on the hyperspectral image to be processed and the third dataset using a preset metric constraint manifold rule; and using the metric matrix to reduce the dimensionality of the hyperspectral image to be processed to obtain a dimensionality-reduced hyperspectral image.

[0054] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the hyperspectral image processing method provided by the above methods. The method includes: in response to acquiring a hyperspectral image to be processed and corresponding labels, determining a first dataset and a second dataset based on the hyperspectral image to be processed and the labels using a preset dataset partitioning rule; wherein the labels characterize the category of target objects in the hyperspectral image to be processed, the first dataset includes multiple initially labeled sample points marked based on the labels, and the second dataset includes multiple unlabeled sample points; expanding the first dataset using the second dataset based on a preset spatial spectral self-training rule to obtain a third dataset; wherein the third dataset includes multiple pseudo-labeled sample points added to the first dataset; determining a metric matrix based on the hyperspectral image to be processed and the third dataset using a preset metric constraint manifold rule; and using the metric matrix to reduce the dimensionality of the hyperspectral image to be processed to obtain a dimensionality-reduced hyperspectral image.

[0055] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0056] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0057] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A hyperspectral image processing method, characterized in that, include: In response to acquiring the hyperspectral image to be processed and its corresponding label, a first dataset and a second dataset are determined based on the hyperspectral image to be processed and the label using a preset dataset partitioning rule; wherein, the label represents the category of the target object in the hyperspectral image to be processed, the first dataset includes multiple initially labeled sample points marked based on the label, and the second dataset includes multiple unlabeled sample points. Based on a preset spatial spectral self-training rule, the first dataset is expanded using the second dataset to obtain a third dataset; wherein, the third dataset includes multiple pseudo-labeled sample points added on the basis of the first dataset; Based on the hyperspectral image to be processed and the third dataset, the measure matrix is ​​determined using a preset measure-constrained manifold rule; Using the metric matrix, the hyperspectral image to be processed is reduced in dimension to obtain a dimension-reduced hyperspectral image; Specifically, based on the hyperspectral image to be processed and the third dataset, the metric matrix is ​​determined using a preset metric constraint manifold rule, including: Based on the hyperspectral image to be processed and the third dataset, the metric matrix A is solved using the following calculation formulas (1) to (5); （1） （2） （3） （4） （5） in, This represents the i-th sample point in the hyperspectral image to be processed; This represents the j-th sample point in the hyperspectral image to be processed; This refers to the hyperspectral image to be processed; This represents the Euclidean distance between the i-th sample point and its corresponding spatial nearest neighbor. This represents the Euclidean distance between the j-th sample point and its corresponding spatial nearest neighbor. tr represents the Gaussian kernel function; tr represents the sum of the elements on the diagonal of the matrix. represents the hyperparameters of the measure constraint part; L represents the Laplacian matrix; The hyperparameter represents the manifold regularization term; n represents the number of labeled sample points in the third dataset; Represents a loss term; Represents the regularization term for the complexity of the metric matrix; This represents the i-th labeled sample point in the third dataset; This represents the j-th labeled sample point in the third dataset; Indicates correspondence Tags; Indicates correspondence The tag.

2. The hyperspectral image processing method according to claim 1, characterized in that, In response to acquiring the hyperspectral image to be processed and its corresponding label, the process utilizes a preset dataset partitioning rule to determine a first dataset and a second dataset based on the hyperspectral image to be processed and the label, including: Based on the tensor representation of the hyperspectral image, the hyperspectral image is divided into multiple pixels as sample points; In response to the pixel having the corresponding label, the pixel is assigned to the first dataset and recorded as the initial labeled sample point; If a pixel does not have a corresponding label, then the pixel is assigned to the second dataset and denoted as the unlabeled sample point.

3. The hyperspectral image processing method according to claim 1, characterized in that, The third dataset, obtained by expanding the first dataset using the second dataset based on a preset spatial spectral self-training rule, includes: Step 1: In the second dataset, determine the spatial nearest neighbor point corresponding to each of the initial labeled sample points in the first dataset; Step 2: According to the preset judgment rules, determine the spatial spectral nearest neighbor points corresponding to each initial labeled sample point from the spatial nearest neighbor points corresponding to the initial labeled sample point; Step 3: Assign a label to the spatial spectral nearest neighbor of each initial labeled sample point, so that the spatial spectral nearest neighbor forms the pseudo-labeled sample point; Step 4: Merge the spatial spectral nearest neighbors that have been labeled into the first dataset; Step 5: Iterate through steps 1 to 4 until the preset iteration termination condition is met, and obtain the third dataset from the expanded first dataset.

4. The hyperspectral image processing method according to claim 3, characterized in that, Step 2 includes: For each of the initial labeled sample points, calculate the Euclidean distance between the initial labeled sample point and each of its corresponding spatial nearest neighbors; If the Euclidean distance to any spatial nearest neighbor is less than a preset distance threshold, then that spatial nearest neighbor is determined to be the spatial spectral nearest neighbor of the corresponding initial labeled sample point.

5. The hyperspectral image processing method according to any one of claims 1-4, characterized in that, The hyperspectral image processing method further includes: The dimensionality-reduced hyperspectral image is classified using a preset classifier to obtain the classification result.

6. The hyperspectral image processing method according to claim 5, characterized in that, The preset classifier includes k -NN classifier.

7. A hyperspectral image processing device, characterized in that, include: The data preprocessing module is configured to: in response to acquiring a hyperspectral image to be processed and its corresponding label, determine a first dataset and a second dataset based on the hyperspectral image to be processed and the label using a preset dataset partitioning rule; wherein the label represents the category of the target object in the hyperspectral image to be processed, the first dataset includes multiple initially labeled sample points marked based on the label, and the second dataset includes multiple unlabeled sample points; The spatial spectrum self-training module is configured to: expand the first dataset using the second dataset based on preset spatial spectrum self-training rules to obtain a third dataset; wherein the third dataset includes multiple pseudo-labeled sample points added on the basis of the first dataset; The metric matrix determination module is configured to: determine the metric matrix based on the hyperspectral image to be processed and the third dataset, using a preset metric constraint manifold rule; The dimension reduction execution module is configured to: use the metric matrix to reduce the dimension of the hyperspectral image to be processed, and obtain the dimension-reduced hyperspectral image; Specifically, based on the hyperspectral image to be processed and the third dataset, the metric matrix is ​​determined using a preset metric constraint manifold rule, including: Based on the hyperspectral image to be processed and the third dataset, the metric matrix A is solved using the following calculation formulas (1) to (5); （1） （2） （3） （4） （5） in, This represents the i-th sample point in the hyperspectral image to be processed; This represents the j-th sample point in the hyperspectral image to be processed; This refers to the hyperspectral image to be processed; This represents the Euclidean distance between the i-th sample point and its corresponding spatial nearest neighbor. This represents the Euclidean distance between the j-th sample point and its corresponding spatial nearest neighbor. tr represents the Gaussian kernel function; tr represents the sum of the elements on the diagonal of the matrix. represents the hyperparameters of the measure constraint part; L represents the Laplacian matrix; The hyperparameter represents the manifold regularization term; n represents the number of labeled sample points in the third dataset; Represents a loss term; Represents the regularization term for the complexity of the metric matrix; This represents the i-th labeled sample point in the third dataset; This represents the j-th labeled sample point in the third dataset; Indicates correspondence Tags; Indicates correspondence The tag.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the hyperspectral image processing method as described in any one of claims 1 to 6.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the hyperspectral image processing method as described in any one of claims 1 to 6.

Images