Hyperspectral image classification method and device based on subgraph adaptive neural network

Abstract

The application discloses a hyperspectral image classification method and equipment based on a subgraph adaptive neural network, and the method comprises the following steps: S1, constructing a pair of probability graph structures based on an SSAPGCN method; S2, reading a feature index and generating a feature matrix, performing PCA dimension reduction on input data, and performing data preprocessing; S3, obtaining a global graph structure S based on the dimension reduction preprocessing, calling a Metis graph segmentation algorithm to perform graph segmentation, and dividing the global probability graph structure into a plurality of subgraphs; S4, calculating an intra-class distance intra (C) as an adaptive feedback threshold value, determining an optimal graph convolution layer number k according to spectral differences of the subgraphs, and avoiding over-smoothing or under-smoothing; and S5, training and testing the data by using the optimal graph convolution layer number k, and completing hyperspectral image classification. The application is mainly used for processing large graph classification, can reduce the calculation cost, fully excavate spectral and spatial features, and improve the accuracy and efficiency of classification.

Description

Technical Field

[0001] This invention relates to the field of hyperspectral image classification technology, and more specifically to a hyperspectral image classification method and device based on a subgraph adaptive neural network. Background Technology

[0002] Hyperspectral image classification is an important research direction in the field of remote sensing. Traditional hyperspectral image classification methods are mainly based on spectral features, such as maximum likelihood classification and support vector machines. These methods are effective when dealing with simple scenes and small amounts of data. However, with the continuous increase in the amount of hyperspectral image data and the increasing complexity of scenes, traditional methods have gradually revealed some limitations. On the one hand, traditional methods often ignore the rich spatial information in hyperspectral images and rely solely on spectral features for classification, which can easily lead to inaccurate classification results, especially when dealing with ground features with similar spectral features but different spatial distributions. On the other hand, traditional methods have low computational efficiency when processing large-scale hyperspectral image data, making it difficult to meet the needs of practical applications.

[0003] In recent years, deep learning technology has achieved remarkable results in the field of image classification, and some methods based on Convolutional Neural Networks (CNNs) have been applied to hyperspectral image classification. These methods can automatically extract features from hyperspectral images, improving classification accuracy to some extent. However, CNNs are mainly based on local convolution operations, which do not make sufficient use of the global structural information of the image, and they have certain limitations when dealing with the complex topological structures of hyperspectral images.

[0004] To overcome the aforementioned problems, researchers began exploring hyperspectral image classification methods based on graph neural networks (GNNs). GNNs can effectively process graph-structured data, representing pixels in a hyperspectral image as nodes in a graph, and the relationships between pixels as edges, thus fully utilizing spectral and spatial information for classification. However, existing GNN methods still face challenges such as high computational complexity and long training times when processing large-scale hyperspectral images.

[0005] Therefore, researchers began to focus on Subgraph Neural Networks (SGNNs). However, although SGNN-based methods can preserve structural information between nodes and reduce the loss of structural information to some extent, the intra-class spectral variability and inter-class spectral similarity in hyperspectral images (HSI) lead to topological imbalances within the subgraph. In message passing mechanisms, using a fixed K-order neighbor aggregation strategy for subgraphs with imbalanced graph structures makes it difficult to effectively smooth the feature information within the subgraph, potentially leading to over-smoothing or under-smoothing phenomena. Summary of the Invention

[0006] The present invention provides a hyperspectral image classification method and device based on subgraph adaptive neural networks that effectively reduces computational costs, fully exploits the spectral and spatial features of images, and significantly improves the accuracy and efficiency of classification, thereby solving at least one of the above-mentioned technical problems.

[0007] To solve the above-mentioned technical problems, the present invention adopts the following technical solution:

[0008] The hyperspectral image classification method based on subgraph adaptive neural networks includes the following steps:

[0009] S1. Based on the SSAPGCN method, a pairwise probability graph structure is constructed to effectively integrate the spatial and spectral information of hyperspectral images;

[0010] S2. Read the feature index of each spectral band in the original spectral image and generate a feature matrix containing spectral information. Perform PCA dimensionality reduction on the input feature matrix data, perform data preprocessing, and remove redundant information.

[0011] S3. Based on the dimensionality reduction preprocessing, the global graph structure S is obtained. The Metis graph segmentation algorithm is called to perform graph segmentation, dividing the global probabilistic graph structure obtained by S1 into multiple local subgraphs in order to mine local features.

[0012] S4. Calculate the intra-class distance intra(C) as the adaptive feedback threshold, and determine the optimal number of convolutional layers k based on the spectral differences of the subgraphs to avoid over-smoothing or under-smoothing.

[0013] S5. Use the optimal number of graph convolutional layers k to train and test the global probabilistic graph structure data to complete the hyperspectral image classification.

[0014] Furthermore, S1 further includes:

[0015] S11. Input the spectral feature matrix and spatial coordinate matrix of the hyperspectral image:

[0016] X∈R N×d

[0017] L∈R N×2

[0018] In this formula, X is the spectral feature matrix, L is the spatial coordinate matrix, N is the number of pixels, and d is the spectral dimension.

[0019] S12. To establish graph topological relationships that balance spectral similarity and spatial proximity, a probabilistic graph structure A∈R is constructed by minimizing the following objective function. N×N :

[0020]

[0021] In this formula:

[0022] a ij Let x be the similarity between pixels i and j, expressed as spectral difference ||x. i -x j ‖and spatial distance‖l i -l j ‖To be decided jointly;

[0023] P∈R N×N A guidance matrix based on expert prior knowledge;

[0024] δ is the balance parameter between spectral and spatial information.

[0025] Furthermore, in S2, PCA dimensionality reduction is performed to remove redundant information:

[0026] X PCA =X·V top

[0027] In this formula, X PCA V is the eigenma matrix after dimensionality reduction. top The eigenvector matrix is ​​dx = 60.

[0028] Retaining the first 60 principal components yields the dimensionality-reduced feature matrix X. PCA ∈R n×60 .

[0029] Furthermore, in S3, the multiple sub-graphs are as follows:

[0030] A = [A1, A2, ..., A m ],X=[X1,X2,...,X m ]

[0031] In this formula, each subgraph A i ∈R n×n Preserve the local topology, including the corresponding node features Xi ∈R n ×60 This enables parallel data processing.

[0032] Furthermore, in S4, considering the heterogeneity of the subgraphs, for each subgraph i, the following dynamic mechanism is used to iteratively determine the optimal number of graph convolutional layers k:

[0033] S41. Initialize the feature matrix

[0034] S42. The update rule for the node features from the original features of layer 0 to the node features of layer k+1 is as follows:

[0035]

[0036] In this formula:

[0037] Let i be the feature vector of node i in the l-th layer;

[0038] a ij Let A be the adjacency matrix. i The connection weights between nodes i and j in the middle node;

[0039] n is the total number of nodes in the current subgraph;

[0040] S43. Introduce a degree matrix residual mechanism, embedding the degree information of nodes as residuals into the feature representation, as shown in the following formula:

[0041]

[0042] In this formula:

[0043] De d ∈R n×n This is the degree matrix of the subgraph, with the diagonal elements representing the degree of the nodes.

[0044] β is a balance coefficient used to control the weights of the original features and degree information;

[0045] S44. Design an adaptive adjustment feedback threshold ξ to obtain the optimal number of graph convolutional layers k based on the spectral differences within different subgraphs.

[0046] S45. To determine the optimal number of convolutional layers k in the graph, ξ = intra(C) is used as the early stopping condition for adaptive subgraph iteration. The intra-class distance intra(C) is calculated as follows:

[0047]

[0048] In this formula, C is the set of categories, |C| is the total number of categories, and c i For class i;

[0049] The number of convolutional layers required to connect sparse or dense subgraphs differs. If the number of convolutional layers is too small, the feature information within the subgraph will not be fully aggregated, resulting in significant differences in the features of samples of the same class, which cannot effectively capture intra-class consistency, a phenomenon known as undersmoothing. If the number of convolutional layers is too large, the feature information within the subgraph will be over-aggregated, causing the features of samples of different classes to become similar and lose discriminative power, a phenomenon known as oversmoothing.

[0050] The intra-class distance intra(C) is used as a threshold to determine whether the optimal number of convolutional layers has been reached. During the convolution process, the intra-class distance intra(C) first decreases and then increases as the number of convolutional layers increases. The optimal number of convolutional layers, i.e., the optimal number of convolutional layers k, is found based on the trend of change.

[0051] S46. In each training round, calculate the intra-class distance integer(C) for the current round and compare it with the intra-class distance integer(C) for round t. t The intra-class distance intra(C) in round t+1. t+1 If intra(C) t <intra(C) t+1 If ..., then continue iterating; if intra(C) ... t >intra(C) t+1 If the node features embedded by the residual information have reached a reasonable similarity, the iteration stops and the current iteration layer number is recorded.

[0052] S47. The first appearance of intra(C) t >intra(C) t+1 When the intra-class distance is considered to be at its minimum, the iteration stops to prevent oversmoothing.

[0053] Furthermore, the testing process in S5 further includes:

[0054] S51, finalize the features of each subgraph. Concatenate into a global feature matrix Generating classification probabilities using fully connected layers and the Softmax function:

[0055]

[0056] In this formula, W cls To provide evidence for the classification weights, the dimension is 60x|C|;

[0057] S52. Set the maximum number of iterations to 100, and average the obtained overall accuracy, average accuracy and KAP coefficient results to verify the classification accuracy.

[0058] A computer device includes a memory and a processor, the memory storing a computer program that, when executed by the processor, causes the processor to perform the steps of the hyperspectral image classification method based on a subgraph adaptive neural network described above.

[0059] The beneficial effects of this invention are reflected in:

[0060] 1. Reduce computational costs: By partitioning graphs, large-scale graph processing is transformed into processing multiple small-scale subgraphs, which reduces computational complexity and storage requirements, and improves the efficiency of processing large graphs.

[0061] 2. Fully exploit features: By integrating spatial and spectral information into the graph structure using the SSAPGCN method and combining it with adaptive subgraph convolution, the spectral and spatial features of hyperspectral images can be fully exploited, improving the accuracy of feature representation.

[0062] 3. Avoid smoothing problems: By adaptively determining the optimal number of convolutional layers for each subgraph, the over-smoothing or under-smoothing phenomenon caused by the fixed K-order neighbor information aggregation strategy is avoided, thus improving the effective smoothness of feature information.

[0063] 4. Improved classification performance: Introducing the adaptive subgraph convolutional neural network model into hyperspectral image classification significantly improves the classification effect and accuracy of hyperspectral images, providing more reliable support for the practical application of hyperspectral images. Attached Figure Description

[0064] The accompanying drawings, which are provided to further illustrate this application and form part of this application, illustrate exemplary embodiments of this application and are used to explain this application, but do not constitute an undue limitation of this application.

[0065] Figure 1 This is an overall flowchart of the classification method according to an embodiment of the present invention.

[0066] Figure 2 This is a schematic diagram of the overall process of the classification method according to an embodiment of the present invention.

[0067] Figure 3 This is a simplified flowchart of the adaptive algorithm according to an embodiment of the present invention.

[0068] Figure 4 This is a structural block diagram of a computer device according to an embodiment of the present invention. Detailed Implementation

[0069] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Unless otherwise specified, the embodiments and features in the embodiments of this application can be combined with each other. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0070] It should be noted that the meaning of "and / or" throughout the text includes three parallel solutions. Taking "A and / or B" as an example, it includes solution A, solution B, or a solution that simultaneously satisfies A and B. Furthermore, "multiple" refers to two or more. Additionally, the technical solutions of the various embodiments can be combined with each other, but this must be based on the ability of those skilled in the art to implement them. When the combination of technical solutions is contradictory or impossible to implement, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed by this invention.

[0071] Hyperspectral images contain rich spectral information, providing crucial data support for numerous fields such as land cover identification, environmental monitoring, and agricultural yield estimation. Accurate classification of hyperspectral images can effectively extract the characteristics and distribution information of different land cover features, showing broad application prospects in resource exploration, urban planning, and ecological protection. This method utilizes an adaptive subgraph neural network to process hyperspectral images, aiming to improve the accuracy and efficiency of classification. Specifically, the method process is as follows:

[0072] See Figure 1 This invention provides a hyperspectral image classification method based on a subgraph adaptive neural network, comprising the following steps:

[0073] S1. Based on the SSAPGCN method, a pairwise probability graph structure is constructed to effectively integrate the spatial and spectral information of hyperspectral images, providing a foundation for subsequent analysis.

[0074] S2. Read the feature index of each spectral band in the original spectral image and generate a feature matrix containing spectral information. Perform PCA dimensionality reduction on the input feature matrix data to preprocess the data, remove redundant information, reduce computational complexity, and improve processing efficiency.

[0075] S3. Based on the dimensionality reduction preprocessing, the global graph structure S is obtained. The Metis graph segmentation algorithm is called to perform graph segmentation, dividing the global probabilistic graph structure obtained by S1 into multiple local subgraphs in order to mine local features.

[0076] S4. Calculate the intra-class distance intra(C) as the adaptive feedback threshold, and determine the optimal number of convolutional layers k based on the spectral differences of the subgraphs to avoid over-smoothing or under-smoothing.

[0077] S5. Use the optimal number of graph convolutional layers k to train and test the global probabilistic graph structure data to complete hyperspectral image classification and improve classification accuracy.

[0078] This invention provides a hyperspectral image classification method based on an adaptive subgraph neural network. On the one hand, by decomposing a large-scale graph into multiple subgraphs, the processing complexity is reduced, allowing the model to focus more on local features. On the other hand, by adopting an adaptive method to determine the optimal number of convolutional layers, the problem of feature smoothing under different subgraph structures is cleverly solved, avoiding the over-smoothing or under-smoothing problems common in traditional methods. Thus, the intrinsic information of hyperspectral images can be extracted more accurately, achieving more accurate image classification.

[0079] The following is a description of some of the technical terms used in this application:

[0080] The SSAPGCN method is used to construct graph structures in hyperspectral image (HSI) classification. It adaptively learns local graph structures and output features, and optimizes graph construction through a feedback mechanism.

[0081] PCA dimensionality reduction: Projects the original high-dimensional data (e.g., 220 bands) into a low-dimensional space (e.g., 60 dimensions) through linear transformation, retaining the principal components with the largest variance.

[0082] Metis graph partitioning algorithm: Through multi-level coarsening, partitioning and thinning steps, it divides the original graph into high-density subgraphs and reduces the edge connections between subgraphs.

[0083] See Figure 2 In this embodiment, S1 further includes:

[0084] S11. Input the spectral feature matrix and spatial coordinate matrix of the hyperspectral image:

[0085] X∈R N×d

[0086] L∈R N×2

[0087] In this formula, X is the spectral feature matrix, L is the spatial coordinate matrix, N is the number of pixels, and d is the spectral dimension.

[0088] S12. To establish graph topological relationships that balance spectral similarity and spatial proximity, a probabilistic graph structure A∈R is constructed by minimizing the following objective function. N×N :

[0089]

[0090] In this formula:

[0091] a ij Let x be the similarity between pixels i and j, expressed as spectral difference ||x. i -x j ‖and spatial distance‖l i -l j ‖To be decided jointly;

[0092] P∈R N×N A guidance matrix based on expert prior knowledge;

[0093] δ is the balance parameter between spectral and spatial information.

[0094] By minimizing the above objective function, a graph structure that simultaneously considers spectral and spatial information can be constructed.

[0095] In this step, the SSAPGCN method is used to fuse spatial and spectral information through pairwise probability graphs, which can effectively enhance the representation ability of graph structures.

[0096] See Figure 2 In this embodiment, in step S2, PCA dimensionality reduction is performed on the input image data to remove redundant information, reduce high-dimensional noise, and improve computational efficiency.

[0097] X PCA =X·V top

[0098] In this formula, X PCA V is the eigenma matrix after dimensionality reduction. top The eigenvector matrix is ​​dx = 60.

[0099] Retaining the first 60 principal components yields the dimensionality-reduced feature matrix X. PCA ∈R n×60 This low-dimensional feature not only preserves the main spectral information, but also provides a suitable dimension for subsequent graph neural network processing.

[0100] The feature matrix X after dimensionality reduction mentioned above PCA Used for subsequent graph segmentation and convolution operations.

[0101] See Figure 2 In this embodiment, the multiple sub-graphs divided in step S3 are as follows:

[0102] A = [A1, A2, ..., A m ],X=[X1,X2,...,X m ]

[0103] In this formula, each subgraph A i ∈R n×n Preserve the local topology, including the corresponding node features X i ∈R n ×60 This enables parallel data processing.

[0104] In this step, the Metis graph partitioning algorithm is used to partition the graph after obtaining the overall graph structure.

[0105] See Figures 2-3 In this embodiment, since common SGNN-based methods typically use a fixed number of convolutional layers for subgraph convolution, and due to the imbalance in the subgraph structure, adaptive subgraph convolution is more reasonable for different subgraphs. Therefore, in S4, considering the heterogeneity of the subgraphs, for each subgraph i, the following dynamic mechanism is used to iteratively determine the optimal number of graph convolutional layers k:

[0106] S41. Initialize the feature matrix

[0107] S42. The update rule for the node features from the original features of layer 0 to the node features of layer k+1 is as follows:

[0108]

[0109] In this formula:

[0110] Let i be the feature vector of node i in the l-th layer;

[0111] a ij Let A be the adjacency matrix. i The connection weights between nodes i and j in the middle node;

[0112] n is the total number of nodes in the current subgraph;

[0113] Each node's feature representation not only includes its own feature information, but also incorporates information from its neighboring nodes;

[0114] S43. To further enhance the performance of convolutional layers and alleviate the oversmoothing problem during deep training, a degree matrix residual mechanism is introduced. The degree information of nodes is embedded as a residual in the feature representation. This mechanism can better represent the adjacency connections and topological structure of nodes. The formula is as follows:

[0115]

[0116] In this formula:

[0117] De d ∈R n×nThis is the degree matrix of the subgraph, with the diagonal elements representing the degree of the nodes.

[0118] β is a balance coefficient used to control the weights of the original features and degree information;

[0119] By dynamically balancing the original features and topological features using a learnable parameter β, the feature assimilation problem caused by deep propagation can be alleviated.

[0120] S44. Considering that determining the optimal number of graph convolutional layers k will significantly affect the quality of neighborhood information aggregation, an adaptive adjustment feedback threshold ξ is designed to obtain the optimal number of graph convolutional layers k based on the spectral differences within different subgraphs.

[0121] S45. To determine the optimal number of convolutional layers k in the graph, ξ = intra(C) is used as the early stopping condition for adaptive subgraph iteration. The intra-class distance intra(C) is calculated as follows:

[0122]

[0123] In this formula, C is the set of categories, |C| is the total number of categories, and c i For class i;

[0124] The number of convolutional layers required to connect sparse or dense subgraphs differs. If the number of convolutional layers is too small, the feature information within the subgraph will not be fully aggregated, resulting in significant differences in the features of samples of the same class, which cannot effectively capture intra-class consistency, a phenomenon known as undersmoothing. If the number of convolutional layers is too large, the feature information within the subgraph will be over-aggregated, causing the features of samples of different classes to become similar and lose discriminative power, a phenomenon known as oversmoothing.

[0125] The intra-class distance intra(C) is used as a threshold to determine whether the optimal number of convolutional layers has been reached. During the convolution process, the intra-class distance intra(C) first decreases and then increases as the number of convolutional layers increases. The optimal number of convolutional layers, i.e., the optimal number of convolutional layers k, is found based on the trend of change.

[0126] Intra(C) is used to evaluate feature consistency within homogeneous regions. Sparsely connected subgraphs may require more convolutional layers, while tightly connected subgraphs may only require fewer convolutional layers to achieve effective feature smoothing.

[0127] S46. In each training round, calculate the intra-class distance integer(C) for the current round and compare it with the intra-class distance integer(C) for round t. t The intra-class distance intra(C) in round t+1. t+1 If intra(C) t <intra(C) t+1If ..., then continue iterating; if intra(C) ... t >intra(C) t+1 If the node features embedded by the residual information have reached a reasonable similarity, the iteration stops and the current iteration layer number is recorded.

[0128] S47. The first appearance of intra(C) t >intra(C) t+1 When the intra-class distance is considered to be at its minimum, the iteration stops to prevent oversmoothing.

[0129] In this step, the intra-class distance intra(C) is used as the adaptive adjustment feedback threshold as a clustering performance metric. By comparing the size of intra(C) calculated with different convolutional layers, the optimal number of graph convolutional layers k can be obtained before adaptive convolution.

[0130] See Figure 2 In this embodiment, the testing process in S5 further includes:

[0131] S51, finalize the features of each subgraph. Concatenate into a global feature matrix Generating classification probabilities using fully connected layers and the Softmax function:

[0132]

[0133] In this formula, W cls To provide evidence for the classification weights, the dimension is 60x|C|;

[0134] S52. Set the maximum number of iterations to 100, and average the obtained overall accuracy, average accuracy and KAP coefficient results to verify the classification accuracy.

[0135] To verify the effectiveness of this method, experiments were conducted on three public datasets: Indian pines, PaviaU, and Houston. The detection results are shown in Tables 1-3 below:

[0136] Table 1 OA(STD)(%),AA(STD)(%),KC(STD)(%)OF DIFFERENTCOMPARED METHODSAND ASGCN METHOD FOR INDIAN DATASET

[0137]

[0138]

[0139] Table 2 OA(STD)(%),AA(STD)(%),KC(STD)(%)OF DIFFERENT COMPARED METHODSAND ASGCN METHOD FOR HOUSTON DATASET

[0140]

[0141]

[0142] Table 3 OA(STD)(%),AA(STD)(%),KC(STD)(%)OF DIFFERENT COMPARED METHODSAND ASGCN METHOD FOR PAVIAU DATASET

[0143] Class DPRN DSGSF morph Former GraphGST ASGCN 1 90.95(8.12) 86.98(7.33) 91.64(6.04) 91.54(6.23) 84.48(4.63) 2 97.81(2.18) 85.35(6.15) 89.91(5.96) 91.64(6.73) 85.67(6.80) 3 97.33(2.21) 99.33(1.30) 86.88(11.55) 83.99(24.79) 98.96(2.86) 4 90.89(2.08) 88.04(2.58) 90.27(5.31) 90.61(2.94) 85.28(4.0) 5 92.57(4.61) 87.55(3.65) 97.76(5.62) 97.34(7.96) 98.29(0.70) 6 50.18(35.83) 97.76(2.29) 60.78(38.45) 69.57(32.00) 88.70(5.99) 7 70.13(6.81) 72.18(7.27) 67.74(12.20) 68.91(9.80) 84.66(6.75) 8 68.92(9.26) 70.99(13.01) 63.73(7.36) 64.74(7.97) 59.01(5.93) 9 66.55(10.94) 57.87(14.95) 62.48(12.83) 75.24(10.01) 63.44(8.89) 10 82.49(11.12) 56.36(10.57) 65.34(21.69) 73.68(12.51) 91.89(4.72) 11 74.81(9.71) 57.36(17.70) 76.30(12.23) 76.32(12.56) 86.25(6.53) 12 76.55(7.04) 54.93(15.38) 61.80(25.30) 73.26(17.76) 77.07(6.43) 13 60.10(9.86) 67.60(19.08) 61.25(23.48) 61.80(25.30) 74.19(7.73) 14 91.11(5.68) 94.03(4.74) 98.40(1.77) 91.90(18.09) 99.38(1.95) 15 97.79(3.01) 94.47(6.24) 97.99(3.87) 100.00(0.00) 97.94(2.61) OA 81.09(2.24) 74.49(6.58) 78.71(1.84) 80.85(2.34) 83.54(1.17) AA 80.26(2.11) 78.05(5.60) 78.84(3.00) 80.67(2.43) 85.01(1.14) KC 79.93(2.55) 72.39(7.13) 76.98(1.99) 79.29(2.52) 82.21(1.26)

[0144] As shown in Tables 1-3 above, the detection results obtained on the three public datasets Indian pines, PaviaU and Houston can achieve overall accuracy of 88.76%, 93.17% and 83.54%, respectively.

[0145] This invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to perform the steps of the hyperspectral image classification method based on subgraph adaptive neural networks described above.

[0146] See Figure 4 The present invention also provides a computer device, including a memory and a processor. The memory stores a computer program, and when the computer program is executed by the processor, the processor performs the steps of the hyperspectral image classification method based on subgraph adaptive neural networks as described above.

[0147] This invention also provides a computer program product containing instructions that, when run on a computer, cause the computer to perform the steps of the hyperspectral image classification method based on subgraph adaptive neural networks described above.

[0148] It is understood that the systems, devices, and storage media provided in the embodiments of the present invention correspond to the methods provided in the embodiments of the present invention, and the explanations, examples, and beneficial effects of the relevant content can be referred to the corresponding parts of the above-described hyperspectral image classification method based on subgraph adaptive neural networks.

[0149] It should be noted that those skilled in the art will understand that all or part of the steps implemented in the embodiments of the present invention can be implemented entirely or partially by software, hardware, firmware, or any combination thereof. When implemented in hardware, it can be implemented entirely or partially by purchasing standard parts or modifications. When implemented in software, it can be implemented entirely or partially in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available media may be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media (e.g., solid state disks (SSDs)).

[0150] In summary, existing hyperspectral image classification methods suffer from high computational complexity, insufficient utilization of global and local structural information, and difficulties in effectively smoothing feature information within subgraphs due to fixed K-order neighbor aggregation strategies when processing large-scale image data, potentially leading to over-smoothing or under-smoothing phenomena. Therefore, this invention provides a hyperspectral image classification method based on an adaptive subgraph neural network. When processing large-scale image data, this method can effectively reduce computational costs, fully exploit the spectral and spatial features of images, and thus significantly improve the accuracy and efficiency of classification.

[0151] It should be understood that the examples and embodiments described herein are for illustrative purposes only and are not intended to limit the invention. Those skilled in the art can make various modifications or changes based on them. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the invention should be included within the protection scope of the invention.

Claims

1. A hyperspectral image classification method based on subgraph adaptive neural networks, characterized in that, Includes the following steps: S1. Based on the SSAPGCN method, a pairwise probability graph structure is constructed to effectively integrate the spatial and spectral information of hyperspectral images; S2. Read the feature index of each spectral band in the original spectral image and generate a feature matrix containing spectral information. Perform PCA dimensionality reduction on the input feature matrix data, perform data preprocessing, and remove redundant information. S3. Based on the dimensionality reduction preprocessing, the global graph structure S is obtained. The Metis graph segmentation algorithm is called to perform graph segmentation, dividing the global probabilistic graph structure obtained by S1 into multiple local subgraphs in order to mine local features. S4. Calculate the intra-class distance intra(C) as the adaptive feedback threshold, and determine the optimal number of convolutional layers k based on the spectral differences of the subgraphs to avoid over-smoothing or under-smoothing. S5. Use the optimal number of graph convolutional layers k to train and test the global probabilistic graph structure data to complete the hyperspectral image classification. S1 further includes: S11. Input the spectral feature matrix and spatial coordinate matrix of the hyperspectral image: In this formula, X is the spectral feature matrix, L is the spatial coordinate matrix, N is the number of pixels, and d is the spectral dimension. S12. To establish graph topological relationships that balance spectral similarity and spatial proximity, a probabilistic graph structure is constructed by minimizing the following objective function. : In this formula: a ij The similarity between pixels i and j is determined by spectral differences. and spatial distance Joint decision; A guidance matrix based on expert prior knowledge; It is a balance parameter between spectral and spatial information; In S4, considering the heterogeneity of the subgraphs, for each subgraph i, the following dynamic mechanism is used to iteratively determine the optimal number of graph convolutional layers k: S41. Initialize the feature matrix S42. The update rule for the node features from the original features of layer 0 to the node features of layer k+1 is as follows: In this formula: Let i be the feature vector of node i in the l-th layer; a ij Let A be the adjacency matrix. i The connection weights between nodes i and j in the middle node; n is the total number of nodes in the current subgraph; S43. Introduce a degree matrix residual mechanism, embedding the degree information of nodes as residuals into the feature representation, as shown in the following formula: In this formula: This is the degree matrix of the subgraph, with the diagonal elements representing the degree of the nodes. β is a balance coefficient used to control the weights of the original features and degree information; S44. Design an adaptive adjustment feedback threshold ξ to obtain the optimal number of graph convolutional layers k based on the spectral differences within different subgraphs. S45. To determine the optimal number of convolutional layers k in the graph, ξ = intra(C) is used as the early stopping condition for adaptive subgraph iteration. The intra-class distance intra(C) is calculated as follows: In this formula, C is the set of categories, |C| is the total number of categories, and c i For class i; The number of convolutional layers required to connect sparse or dense subgraphs differs. If the number of convolutional layers is too small, the feature information within the subgraph will not be fully aggregated, resulting in significant differences in the features of samples of the same class, which cannot effectively capture intra-class consistency, a phenomenon known as undersmoothing. If the number of convolutional layers is too large, the feature information within the subgraph will be over-aggregated, causing the features of samples of different classes to become similar and lose discriminative power, a phenomenon known as oversmoothing. The intra-class distance intra(C) is used as a threshold to determine whether the optimal number of convolutional layers has been reached. During the convolution process, the intra-class distance intra(C) first decreases and then increases as the number of convolutional layers increases. The optimal number of convolutional layers, i.e., the optimal number of convolutional layers k, is found based on the trend of change. S46. In each training round, calculate the intra-class distance integer(C) for the current round and compare it with the intra-class distance integer(C) for round t. t The intra-class distance intra(C) in round t+1. t+1 If intra(C) t <intra(C) t+1 If so, continue iterating; if intra(C) t >intra(C) t+1 If the node features embedded by the residual information have reached a reasonable similarity, the iteration stops and the current iteration layer number is recorded. S47. The first appearance of intra(C) t >intra(C) t+1 When the intra-class distance is considered to be at its minimum, the iteration stops to prevent oversmoothing.

2. The hyperspectral image classification method based on subgraph adaptive neural network as described in claim 1, characterized in that, In step S2, PCA dimensionality reduction is performed to remove redundant information: In this formula, X PCA V is the eigenma matrix after dimensionality reduction. top The eigenvector matrix is ​​dx = 60. Retaining the first 60 principal components yields the dimensionality-reduced feature matrix. .

3. The hyperspectral image classification method based on subgraph adaptive neural network as described in claim 1, characterized in that, In S3, the multiple subgraphs are as follows: In this formula, each subgraph Preserve the local topology, including the characteristics of the corresponding nodes. This enables parallel data processing.

4. The hyperspectral image classification method and device based on subgraph adaptive neural network as described in claim 1, characterized in that, The testing process in S5 further includes: S51, finalize the features of each subgraph. Concatenate into a global feature matrix Classification probabilities are generated using fully connected layers and the Softmax function: In this formula, W cls To provide evidence for the classification weights, the dimension is 60x|C|; S52. Set the maximum number of iterations to 100, and average the obtained overall accuracy, average accuracy and KAP coefficient results to verify the classification accuracy.

5. A computer device, characterized in that, The system includes a memory and a processor, the memory storing a computer program that, when executed by the processor, causes the processor to perform the steps of the hyperspectral image classification method based on a subgraph adaptive neural network as described in any one of claims 1-4.

Images