Large-scale hyperspectral image clustering method, system and equipment based on superpixel enhanced anchor diagram

Abstract

The invention discloses a large-scale hyperspectral image clustering method, system and device based on a superpixel enhanced anchor diagram, and relates to the technical field of hyperspectral image processing. A hyperspectral image has the characteristics of large data scale and complex noise, and the invention provides a large-scale hyperspectral image clustering method, system and equipment based on a super-pixel enhanced anchor diagram in order to solve the problems of high calculation complexity and insufficient precision and robustness of an existing method. According to the method, the clustering efficiency and the clustering precision of the large-scale hyperspectral image are improved by integrating the superpixel enhanced anchor diagram of the superpixel-level spatial spectrum information; and super-pixel enhanced anchor diagram decomposition is carried out under a correlation entropy criterion, so that the robustness of a hyperspectral clustering task containing complex noise is improved. A large number of experiments show that the method is obviously superior to the prior art in clustering efficiency, precision and robustness, an effective solution is provided for large-scale hyperspectral image clustering, and the method can be applied to the fields of precision agriculture, environment monitoring, military reconnaissance, mineral exploration and the like.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of hyperspectral image processing, and particularly relates to a large-scale hyperspectral image clustering method, system and device based on superpixel enhanced anchor graph. BACKGROUND

[0002] Hyperspectral image clustering plays a key role in many fields, which can mine the potential information of objects in the image and provide strong support for subsequent image analysis. It is widely used in fine agriculture, environmental monitoring, military reconnaissance, mineral exploration and other fields.

[0003] The number of hyperspectral image pixels usually reaches tens of thousands or even millions. Traditional graph clustering needs to calculate the similarity of all pixel pairs, and the computational complexity is quadratic. At the same time, the spectral decomposition of large-scale graph structure further aggravates the computational burden, so that these methods are difficult to deploy in time-sensitive applications.

[0004] Hyperspectral images have rich spectral information, but the spatial correlation between pixels is the key to improving clustering accuracy. Most existing methods only rely on spectral similarity measurement and completely ignore the inherent spatial structure of hyperspectral images. Although some recent methods introduce spatial constraints into graph learning, such as incorporating neighborhood relationships or using spatial regularization, the effect is still far from satisfactory, especially for large-scale hyperspectral images with complex structure patterns.

[0005] Hyperspectral images are inevitably disturbed by factors such as sensor defects, atmospheric influence, data transmission errors, and other factors, and contain complex noise such as Gaussian noise, impulse noise, stripe noise, and mixed distribution noise. Existing anti-noise methods include methods based on l1 norm and l 2,1 Norm methods are only effective for sparse or simple noise and are difficult to effectively deal with nonlinear, non-Gaussian or mixed distribution noise.

[0006] In summary, the existing hyperspectral image clustering methods have obvious deficiencies in improving clustering efficiency, improving clustering accuracy, and resisting noise interference, and are difficult to cope with the characteristics of large-scale hyperspectral images and containing complex noise. There is an urgent need for a new method to solve these problems. SUMMARY

[0007] The purpose of the present application is to provide a large-scale hyperspectral image clustering method, system and device based on superpixel enhanced anchor graph, to solve the problems of low clustering efficiency, insufficient clustering accuracy and poor clustering robustness in large-scale hyperspectral image clustering in the prior art.

[0008] A large-scale hyperspectral image clustering method based on superpixel enhanced anchor graph, comprising the following steps:

[0009] Superpixel segmentation: performing superpixel segmentation on the hyperspectral image, dividing the hyperspectral image into a series of superpixel regions that are mutually non-overlapping and contain highly homogenized pixels, to generate a superpixel spectral set;

[0010] Superpixel spectral-spatial enhanced set construction: for the superpixel spectral set, average spectral information of neighbors of the superpixel points is calculated, and the average spectral information of the neighbors is fused with the original features of the superpixel points to obtain a superpixel spectral-spatial enhanced set, each superpixel feature in the superpixel spectral-spatial enhanced set containing both spectral information and neighborhood spatial information;

[0011] Superpixel enhanced anchor graph construction: anchor points are selected for the superpixel spectral set and the superpixel spectral-spatial enhanced set to obtain a superpixel spectral anchor set and a superpixel spectral-spatial enhanced anchor set, and then, by jointly learning the relationship between the original data and the anchor points in the spectral features and the spectral-spatial features, a superpixel enhanced anchor graph is constructed;

[0012] Non-negative matrix factorization based on correlation entropy: the superpixel enhanced anchor graph constructed is subjected to non-negative matrix factorization based on a correlation entropy criterion, and a graph regularization constraint is imposed on the factorization process, and finally, the categories of the superpixel spectral points are obtained directly from the non-negative matrix obtained by the factorization;

[0013] Hyperspectral point clustering result generation: the hyperspectral points are mapped by the superpixel spectral points to obtain the clustering results of all the pixel points in the final hyperspectral image.

[0014] The large-scale hyperspectral image clustering method based on the superpixel enhanced anchor graph applies an entropy rate superpixel segmentation algorithm based on graph theory to the input large-scale hyperspectral image data to divide the superpixel regions and generate a superpixel spectral set.

[0015] The large-scale hyperspectral image clustering method based on the superpixel enhanced anchor graph calculates the average spectral information of the neighbors of each superpixel point in the superpixel spectral set, and splices the average spectral information of the neighbors with the superpixel information to obtain a superpixel spectral-spatial enhanced set.

[0016] The large-scale hyperspectral image clustering method based on the superpixel enhanced anchor graph uses a k-means method on the superpixel spectral set to obtain a superpixel spectral anchor set, maps and calculates the superpixel spectral-spatial enhanced set according to the obtained superpixel spectral anchor set to obtain a superpixel spectral-spatial enhanced anchor set corresponding to the superpixel spectral-spatial enhanced set, and then measures the structural relationship between the superpixel spectral anchor set and the superpixel spectral-spatial enhanced anchor set and between the superpixel spectral set and the superpixel spectral-spatial enhanced set by using the Euclidean distance, and constructs a superpixel enhanced anchor graph containing both spectral information and spatial information based on the two structural relationships.

[0017] The large-scale hyperspectral image clustering method based on the superpixel enhanced anchor graph performs non-negative matrix decomposition on the superpixel enhanced anchor graph under the measurement of relevant entropy to improve the calculation efficiency and robustness; meanwhile, the Laplacian matrix obtained from the superpixel enhanced anchor graph is used for graph regularization constraint to improve the clustering accuracy; the decomposition process is converted into a convex problem for solving by using a semi-quadratic technique; and finally, the classes of the superpixel points are obtained directly from the non-negative matrix obtained by the decomposition.

[0018] A large-scale hyperspectral image clustering system based on a superpixel enhanced anchor graph, comprising:

[0019] A superpixel segmentation module configured to divide the hyperspectral image into superpixel regions and generate a superpixel set;

[0020] A superpixel spatial-spectral enhancement set construction module configured to calculate the average spectral information of the neighbors of the superpixel points and fuse the original features of the superpixel points to obtain a superpixel spatial-spectral enhancement set;

[0021] A superpixel enhanced anchor graph construction module configured to jointly learn the relationship between the original data and anchor points in the spectral features and spatial-spectral features to obtain a superpixel enhanced anchor graph;

[0022] A non-negative matrix decomposition module based on relevant entropy, configured to perform non-negative matrix decomposition on the superpixel enhanced anchor graph and obtain the classes of the superpixel points directly from the non-negative matrix obtained by the decomposition.

[0023] A hyperspectral point clustering result generation module configured to map the hyperspectral points through the superpixel points to obtain the clustering results of all the pixel points in the final hyperspectral image.

[0024] A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the large-scale hyperspectral image clustering method based on the superpixel enhanced anchor graph when executing the computer program.

[0025] A computer readable storage medium storing a computer program, wherein the computer program is executable by a processor to implement the steps of the large-scale hyperspectral image clustering method based on the superpixel enhanced anchor graph.

[0026] Compared with the prior art, the present application has the following advantages:

[0027] 1) The present application constructs a superpixel enhanced anchor graph and performs non-negative matrix decomposition on the constructed superpixel enhanced anchor graph, which has linear complexity and significantly improves the efficiency of hyperspectral image clustering.

[0028] 2) The present application significantly improves the accuracy of hyperspectral image clustering by constructing a superpixel spectral enhancement set and jointly learning the relationship between the original data and anchor points in the spectral feature and spatial-spectral feature in the process of constructing a superpixel enhanced anchor graph.

[0029] 3) The present application significantly improves the robustness of hyperspectral image clustering in the real world by introducing a correlation entropy measure into the anchor graph non-negative matrix factorization process, which reduces the influence of a large number of nonlinear and non-Gaussian noise in the real world.

[0030] A large number of experiments show that the method of the present application is significantly better than the prior art in clustering efficiency, accuracy and robustness, and provides an effective solution for large-scale hyperspectral image clustering, which can be used in the fields of precision agriculture, environmental monitoring, military reconnaissance, mineral exploration, etc. BRIEF DESCRIPTION OF DRAWINGS

[0031] Figure 1 is a flowchart of the large-scale hyperspectral image clustering method based on superpixel enhanced anchor graph according to the present application;

[0032] Figure 2 is a feature splicing graph of the original spectral information and the neighborhood average spectral information;

[0033] Figure 3a and Figure 3b are the convergence curves of the method of the present application on the Indian Pines and Pavia Centre data sets, respectively.

[0034] Figure 4a is the final segmentation result graph of the S3AGC method on the Indian Pines data set, Figure 4b is the final segmentation result graph of the BGPC method on the Indian Pines data set, Figure 4c is the final segmentation result graph of the SDST method on the Indian Pines data set, Figure 4d is the final segmentation result graph of the method of the present application on the Indian Pines data set. DETAILED DESCRIPTION

[0035] The embodiments of the present application will be described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals represent the same or similar elements or elements having the same or similar functions throughout. The embodiments described below by reference to the accompanying drawings are exemplary and are only used to explain the present application, and cannot be understood as a limitation of the present application.

[0036] In the first embodiment of the present application, as shown in Figure 1 is the large-scale hyperspectral image clustering method based on superpixel enhanced anchor graph according to the present application, which includes the following processes:

[0037] Step S01: For the input hyperspectral image X, we have:

[0038]

[0039] where X is the set of all pixels in the hyperspectral image, x i represents the i-th hyperspectral point, d is the number of spectral bands, n is the number of hyperspectral pixels, and i is the index. Apply the entropy rate superpixel segmentation algorithm based on graph theory to the input X to generate the superpixel set where m represents the number of superpixels, s i represents the i-th superpixel spectral point.

[0040] Step S02: Superpixel spectral enhancement set construction, for the superpixel s i , the average spectral information of its direct neighbors, i.e., the feature mean, is denoted as f avg (s i ), which satisfies:

[0041]

[0042] where |H1(s i )| represents the number of direct neighbors of s i . The feature mean of the direct neighbors of the superpixel is concatenated and fused with the original feature of s i , and the resulting new is denoted as:

[0043] is the superpixel spectral feature point fused with local spatial information, and thus the superpixel spectral enhancement set fused with spatial information Each point in the superpixel spectral enhancement set contains both the original spectral information and the neighborhood average spectral information.

[0044] Step S03: Superpixel enhancement anchor point graph construction, obtain the superpixel spectral anchor point set where l is the number of anchor points. Map the superpixel spectral anchor point set to to calculate the superpixel spectral enhancement anchor point set fused with spectral information where u i represents the i-th superpixel spectral anchor point, represents the i-th superpixel spectral enhancement anchor point. The superpixel enhancement anchor point graph is constructed through an efficient neighbor assignment strategy that can fully utilize spatial information and reduce the need for manual setting of superparameters, as follows:

[0045]

[0046] where, is the constructed superpixel enhanced anchor graph, b ij denotes the element in the i-th row and j-th column of matrix B, while denotes the Euclidean distance between superpixel point s i and superpixel anchor u j . is the superpixel spectral enhancement fused with spatial information. The above objective function contains three terms: learning B using spectral features; learning B combining local spatial and spectral features, where a is the trade-off coefficient between the first two terms; is the regularization term, where m is the regularization parameter. By optimizing the above objective function, the sparse superpixel enhanced anchor graph B is obtained.

[0047] Step S04: Correlation entropy based non-negative matrix factorization: In this step, a superpixel level anchor graph non-negative matrix factorization method based on a correlation entropy measurement criterion is designed. The correlation entropy measurement criterion ensures the robustness of clustering, and the class of samples obtained directly from the non-negative matrix obtained by decomposition greatly improves the efficiency of clustering. In addition, by imposing a light graph regularization constraint in the process of implementing non-negative matrix factorization, the clustering accuracy is ensured without affecting the clustering efficiency. The objective function of the correlation entropy based non-negative matrix factorization is as follows:

[0048]

[0049] s.t.Q≥0,H≥0.

[0050] where and are the factor matrices of the non-negative matrix, c represents the number of classes of samples, G{} represents a Gaussian function, m represents the balance parameter of the graph regularization constraint, H T represents the transpose matrix of matrix H, L AG is the Laplacian matrix, which is defined as where W AG = B T B is a similarity matrix constructed by the anchor graph, D AG is the corresponding degree matrix of W AG .

[0051] Since the above problem is a non-convex problem, it is difficult to solve directly. The present application converts it into a convex optimization problem through semi-quadratic technology, so the above objective function can be transformed into the following form:

[0052]

[0053] s.t. Q ³ 0, H ³ 0.

[0054] where V is a diagonal auxiliary matrix derived from semi-quadratic technique, the elements on its diagonal are where s is the Gaussian kernel bandwidth. Finally, by optimizing the objective function in the above equation, the class of the superpixel spectral point can be directly obtained from Q.

[0055] Step S05: hyperspectral point clustering result generation: mapping the hyperspectral points through the superpixel spectral points to obtain the clustering result of all pixel points in the final hyperspectral image.

[0056] By designing a spatial perception type superpixel enhanced anchor point graph construction method that fuses superpixel neighborhood information, local spatial information is integrated when constructing the anchor point graph, not only the spectral characteristics are considered, but also the spatial neighborhood correlation is used to effectively obtain high-quality spatial information, and the overall clustering accuracy of the method is improved. Experiments on multiple mainstream data sets show that, compared with existing clustering methods, the present application achieves the best or second-best results in overall accuracy (OA), Kappa coefficient and normalized mutual information (NMI).

[0057] The non-negative matrix factorization algorithm based on correlation entropy is introduced, and the semi-quadratic technique is combined to convert the original problem into a convex optimization problem. The optimization algorithm for matrix decomposition can be completed with only a few iterations, the clustering result can be directly obtained from the factor matrix in the non-negative matrix factorization, and the calculation redundancy caused by the use of singular value decomposition and post-processing steps is avoided. Experimental results show that, when processing large-scale hyperspectral images, the present application requires less time than similar algorithms, and even in some cases only a fraction of the time required by other algorithms, greatly improving the clustering efficiency.

[0058] The correlation entropy is applied to the non-negative matrix factorization process, which significantly enhances the ability of the present application to resist noise and data anomalies, and improves the robustness of the clustering result, which shows higher noise resistance performance on different sizes of data sets.

[0059] As shown in Figure 2 , the present application fuses the average value of the features of the direct neighbors of the superpixels with the original features of s i , to obtain a superpixel spectral enhancement set that fuses local spatial information. The perception of local spatial information by the pixel enhancement anchor point graph is effectively improved.

[0060] As shown in Figure 3a and Figure 3b , the Indian Pines and Pavia Centre data sets both converge rapidly after iteration and optimization by the present application, proving that the present application has good rapid convergence characteristics for different large-scale hyperspectral image data sets.

[0061] The embodiments of the present invention can effectively improve the segmentation results of hyperspectral images while maintaining good robustness. Compared with other clustering methods for hyperspectral images in related technologies, the image segmentation effectiveness of the embodiments of the present invention is higher, such as... Figures 4a-4d As shown, Figure 4a , Figure 4b , Figure 4c , Figure 4d The final segmentation results of the three comparison methods and the method of this invention on the Indian Pines dataset are shown sequentially. The segmentation map obtained by the method of this invention is smoother than the other methods. It exhibits higher precision and naturalness in image smoothness and region transition processing, with softer boundaries, stronger internal consistency, and a more harmonious and unified overall appearance. Specifically, the algorithm proposed in this invention outperforms S3AGC, BGPC, and SDST, indicating that the superpixel segmentation and neighbor feature fusion method has a more accurate representation ability for homogeneous HRSI regions.

[0062] Therefore, this invention is a high-precision, robust, and efficient large-scale hyperspectral image clustering method based on superpixel-enhanced anchor maps. It improves the clustering efficiency of large-scale hyperspectral images by integrating superpixel-level spatial-spectral information into superpixel-enhanced anchor maps. Under the correlation entropy criterion, superpixel-enhanced anchor map decomposition is performed, which improves the robustness to hyperspectral clustering tasks containing complex noise.

[0063] The following are embodiments of the apparatus of the present invention, which can be used to execute embodiments of the method of the present invention. For details not omitted in the apparatus embodiments, please refer to the embodiments of the method of the present invention.

[0064] In the second embodiment of the present invention, a large-scale hyperspectral image clustering system is provided. This large-scale hyperspectral image clustering system can be used to implement the above-mentioned large-scale hyperspectral image clustering method. Specifically, the large-scale hyperspectral image clustering system includes a superpixel segmentation module, a superpixel spatial spectrum enhancement set construction module, a superpixel enhancement anchor point map construction module, a non-negative matrix factorization module based on correlation entropy, and a hyperspectral point clustering result generation module.

[0065] The superpixel segmentation module is used to divide the hyperspectral image into superpixel regions and generate a superpixel set.

[0066] The superpixel spatial spectrum enhancement set construction module is used to calculate the average spectral information of the superpixel's neighbors and fuse it with the original features of the superpixel to obtain the superpixel spatial spectrum enhancement set.

[0067] The superpixel-enhanced anchor graph construction module is used to jointly learn the relationship between the original data and anchor points in spectral features and spatial spectral features to construct an anchor graph containing spatial information, thus obtaining the superpixel-enhanced anchor graph.

[0068] The non-negative matrix factorization module based on correlation entropy performs non-negative matrix factorization on the constructed superpixel enhancement anchor point map based on the correlation entropy criterion, and applies graph regularization constraints to the factorization process. Finally, the category of superpixel points is directly obtained from the non-negative matrix obtained by factorization.

[0069] The hyperspectral point clustering result generation module is used to map hyperspectral points through superpixels to obtain the clustering result of all pixels in the final hyperspectral image.

[0070] In a third embodiment of the present invention, a terminal device is provided, comprising a processor and a memory. The memory stores a computer program, which includes program instructions. The processor executes the program instructions stored in the computer storage medium. The processor may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing and control core of the terminal, suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions to achieve a corresponding method flow or corresponding function. The processor described in this embodiment can be used for large-scale hyperspectral image clustering methods, including: S01, superpixel segmentation, dividing the hyperspectral image into a series of non-overlapping superpixel regions containing highly homogeneous pixels to obtain a superpixel spectral set. S02, Superpixel Spatial-Spectral Enhancement Set Construction: For the superpixel spectral set, calculate the average spectral information of the neighbors of each superpixel spectral point, and fuse it with the original features of the superpixel points to obtain the superpixel spatial-spectral enhancement set. S03, Superpixel Enhancement Anchor Map Construction: Select anchor points for both the superpixel spectral set and the superpixel spatial-spectral enhancement set to obtain the superpixel spectral anchor set and the superpixel spatial-spectral enhancement anchor set, respectively. Then, construct the superpixel enhancement anchor map by jointly learning the relationship between the original data in the spectral features and spatial-spectral features and the anchor points. S04, Non-negative Matrix Factorization Based on Correlation Entropy: Perform non-negative matrix factorization on the constructed superpixel enhancement anchor map based on the correlation entropy criterion, and finally directly obtain the superpixel point categories from the non-negative matrix obtained by factorization. S05, Hyperspectral Point Clustering Result Generation: Map hyperspectral points to superpixel points to obtain the final clustering results of all pixels in the hyperspectral image.

[0071] In a fourth embodiment of the present invention, a computer-readable storage medium is also provided. This computer-readable storage medium is a memory device in a terminal device used to store programs and data. It is understood that the computer-readable storage medium here may include both the built-in storage medium in the terminal device and extended storage media supported by the terminal device. The computer-readable storage medium provides storage space that stores the terminal's operating system. Furthermore, this storage space also stores one or more instructions suitable for loading and execution by a processor. These instructions may be one or more computer programs (including program code). It should be noted that the computer-readable storage medium here may be high-speed RAM or non-volatile memory, such as at least one disk storage device.

[0072] One or more instructions stored in a computer-readable storage medium can be loaded and executed by a processor to implement the corresponding steps of the incomplete multi-view image segmentation method in the above embodiments.

[0073] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0074] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure One One or more processes and / or boxes Figure One A device that provides the functions specified in one or more boxes.

[0075] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure OneOne or more processes and / or boxes Figure One The function specified in one or more boxes.

[0076] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure One One or more processes and / or boxes Figure One The steps of the function specified in one or more boxes.

[0077] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0078] In the above embodiments of this application, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0079] The above description is only a preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

[0080] It should be understood that the above description is for illustrative purposes and not for limitation. Many embodiments and applications beyond the provided examples will be apparent to those skilled in the art upon reading the above description. Therefore, the scope of this patent should not be determined by reference to the above description, but rather by reference to the foregoing claims and the full scope of their equivalents. For purposes of completeness, all articles and references, including patent applications and publications, are incorporated herein by reference. The omission of any aspect of the subject matter disclosed herein in the foregoing claims is not intended as a waiver of that subject matter, nor should it be construed as an indication that the applicant has not considered that subject matter as part of the disclosed inventive subject matter.

Claims

1. A method for hyperspectral image clustering based on superpixel enhanced anchor graph, characterized in that, The method comprises the following steps: Superpixel segmentation: performing superpixel segmentation on the hyperspectral image to divide the hyperspectral image into a series of superpixel regions that do not overlap with each other and contain highly homogenized pixels, and generating a superpixel spectrum set; Superpixel hyperspectral enhancement set construction: calculating average spectral information of neighbors of superpixel points for the superpixel spectrum set, and obtaining a superpixel hyperspectral enhancement set by fusing the average spectral information with original features of the superpixel points, wherein each superpixel feature in the superpixel hyperspectral enhancement set contains spectral information and neighborhood spatial information; Superpixel enhanced anchor point graph construction: selecting anchor points for the superpixel spectrum set and the superpixel hyperspectral enhancement set to obtain a superpixel spectrum anchor point set and a superpixel hyperspectral enhancement anchor point set, and then constructing a superpixel enhanced anchor point graph by jointly learning relationships between original data and anchor points in spectral features and hyperspectral features; Non-negative matrix factorization based on correlation entropy: performing non-negative matrix factorization on the constructed superpixel enhanced anchor point graph based on a correlation entropy criterion, and simultaneously imposing graph regularization constraints on the factorization process, and finally obtaining classes of superpixel spectrum points directly from non-negative matrices obtained through the factorization; Hyperspectral point clustering result generation: mapping hyperspectral points through superpixel spectrum points to obtain clustering results of all pixel points in the final hyperspectral image. 2.The method of claim 1, wherein, An entropy rate superpixel segmentation algorithm based on graph theory is applied to input large-scale hyperspectral image data to divide superpixel regions and generate a superpixel spectrum set. 3.The method of claim 1, wherein, For each superpixel point in the superpixel spectrum set, average spectral information of neighbors of the superpixel point is calculated, and the average spectral information of the neighbors is spliced with superpixel information to obtain a superpixel hyperspectral enhancement set. 4.The method of claim 1, wherein, The k-means method is used on the superpixel spectrum set to obtain a superpixel spectrum anchor point set, and the superpixel hyperspectral enhancement set is mapped and calculated according to the obtained superpixel spectrum anchor point set to obtain a superpixel hyperspectral enhancement anchor point set corresponding to the superpixel hyperspectral enhancement set. Then, Euclidean distances are used to measure structural relationships between the superpixel spectrum anchor point set and the superpixel hyperspectral enhancement anchor point set and between the superpixel spectrum set and the superpixel hyperspectral enhancement set, and a superpixel enhanced anchor point graph containing spectral information and spatial information is constructed based on the two structural relationships.

5. The method of claim 1, wherein, Non-negative matrix factorization is performed on the superpixel enhanced anchor point graph under the measurement of correlation entropy to improve calculation efficiency and robustness, and a Laplacian matrix obtained from the superpixel enhanced anchor point graph is used for graph regularization constraints to improve clustering accuracy. A semi-quadratic technique is used to convert the factorization process into a convex problem for solving, and finally, classes of superpixel points are obtained directly from non-negative matrices obtained through the factorization. 6.A large hyperspectral image clustering system based on superpixel enhanced anchor graph, characterized in that, The method comprises the following steps: A superpixel segmentation module is configured to divide a hyperspectral image into superpixel regions and generate a superpixel set; A superpixel hyperspectral enhancement set construction module is configured to calculate average spectral information of neighbors of superpixel points and fuse the average spectral information with original features of the superpixel points to obtain a superpixel hyperspectral enhancement set; A superpixel enhanced anchor point graph construction module is configured to jointly learn relationships between original data and anchor points in spectral features and hyperspectral features to obtain a superpixel enhanced anchor point graph; and A non-negative matrix factorization based on correlation entropy module is configured to perform non-negative matrix factorization on the constructed superpixel enhanced anchor point graph based on a correlation entropy criterion, and simultaneously impose graph regularization constraints on the factorization process to obtain classes of superpixel spectrum points directly from non-negative matrices obtained through the factorization. A non-negative matrix factorization module based on correlation entropy is configured to perform non-negative matrix factorization on the superpixel-enhanced anchor graph and obtain the categories of the superpixel points directly from the non-negative matrix obtained by the factorization. A hyperspectral point clustering result generation module is configured to map the hyperspectral points by using the superpixel points to obtain the clustering results of all the pixel points in the final hyperspectral image.

7. A computer device, comprising: The computer readable storage medium stores a computer program, and the computer program is executed by the processor to implement the steps of the large-scale hyperspectral image clustering method based on the superpixel-enhanced anchor graph according to any one of claims 1 to 6.

8. A computer-readable storage medium, characterized in that, The computer readable storage medium stores a computer program, and the computer program is executed by the processor to implement the steps of the large-scale hyperspectral image clustering method based on the superpixel-enhanced anchor graph according to any one of claims 1 to 6.

Images