Underwater hyperspectral clustering method based on bipartite graph

Abstract

The invention discloses an underwater hyperspectral clustering method based on a bipartite graph, and relates to the technical field of underwater image processing. The method comprises the following steps: acquiring spectral images of a to-be-processed image in different spectral intervals, taking the spectral image of each spectral interval as a view, and constructing hyperspectral data based on the acquired views; constructing a hyperspectral clustering model based on subspace clustering learning, and processing the hyperspectral image based on the hyperspectral clustering model; the hyperspectral clustering model comprises an adaptive dynamic anchor point selection module, an anchor point centroid learning module and a bipartite graph decomposition module; and iterating the hyperspectral clustering model based on an alternative update variable strategy until an iteration stop condition is met, and outputting a clustering result based on the clustering indication index matrix. According to the method, a model based on subspace clustering learning is constructed, uniform self-adaptive anchor points are dynamically learned in all pixel points, and the optimal cluster with accurate pixel division can be directly obtained from bipartite graph decomposition.

Description

Technical Field

[0001] The present invention relates to the technical field of underwater image processing. Specifically, it relates to an underwater hyperspectral clustering method based on a bipartite graph. Background Art

[0002] An underwater hyperspectral image is an image obtained through underwater hyperspectral imaging technology, which can simultaneously provide high spatial resolution and high spectral resolution information of underwater targets and is one of the important means of underwater optical detection. Through the analysis of underwater spectral images, this technology can obtain the spatial shape and spectral characteristics of targets, providing an important basis for the analysis of the composition and structure of substances. The underwater hyperspectral image classification technology is a method for classifying based on these images and is widely used in fields such as underwater target recognition and environmental monitoring.

[0003] In the field of underwater hyperspectral image classification, various methods and technologies have been applied. Among them, conditional diffusion models are used to identify the structural features of images, and by denoising and generating different sample data, prior variable spectral groups are obtained to simulate the changes in the spectral curves of underwater targets. With the development of deep learning technology, deep learning-based methods have gradually emerged. For example, an asymmetric residual network with a SE module can effectively improve the classification performance of underwater hyperspectral images. The progress of these technologies provides strong support for the efficient classification of underwater hyperspectral images.

[0004] However, in practical applications, obtaining the label information of large-scale hyperspectral data is often costly and time-consuming. Therefore, hyperspectral unsupervised clustering methods have emerged. However, most of the existing unsupervised clustering methods are only based on the spectral features of hyperspectral images and ignore their spatial information, resulting in some pixels being unable to be correctly classified. In addition, the complexity of the underwater environment makes there be a large number of background noise pixels in hyperspectral images, and at the same time, the number of pixel points is numerous, which further increases the difficulty of classification. Therefore, how to effectively solve the problems of background noise pixels and a large number of pixel points has become an urgent challenge to be solved. Summary of the Invention

[0005] In view of the deficiencies of the prior art, the present invention provides an underwater hyperspectral clustering method based on a bipartite graph. The present invention constructs a model based on subspace clustering learning, dynamically learns a unified adaptive anchor point among all pixel points, and can directly obtain the optimal clustering with precise pixel partitioning from the bipartite graph decomposition.

[0006] The technical means adopted by the present invention are as follows:

[0007] An underwater hyperspectral clustering method based on a bipartite graph, comprising the following steps:

[0008] S1. Obtain the image to be processed, and obtain the spectral images of the image to be processed in different spectral intervals. Take the spectral image of each spectral interval as a view, and construct hyperspectral data based on the obtained views;

[0009] S2. Construct a hyperspectral clustering model based on subspace clustering learning, and process the hyperspectral image based on the hyperspectral clustering model; the hyperspectral clustering model includes: an adaptive dynamic anchor selection module, an anchor centroid learning module, and a bipartite graph decomposition module.

[0010] The adaptive dynamic anchor selection module performs adaptive dynamic anchor selection on the hyperspectral data to obtain an anchor information matrix.

[0011] The anchor centroid learning module performs centroid learning on the anchors based on the anchor information matrix to obtain a potential centroid matrix of the anchors.

[0012] The bipartite graph decomposition module constructs a clustering indicator matrix, and performs bipartite graph decomposition on the anchors based on the potential centroid matrix of the anchors and the clustering indicator matrix to obtain a consensus bipartite graph.

[0013] S3. Iterate the hyperspectral clustering model based on the alternating update variable strategy until the iteration stop condition is met, and then output the clustering result based on the clustering indicator matrix.

[0014] Further, the adaptive dynamic anchor selection module is set as:

[0015]

[0016] s.t. A (v)T A (v) = I

[0017] where, X (v) represents the hyperspectral data, v represents the number of spectral intervals existing in the hyperspectral data, A (v) represents the anchor information matrix, and Z represents the consensus bipartite graph.

[0018] Further, the anchor centroid learning module is set as:

[0019]

[0020] s.t. U (v)T U (v) = I

[0021] where, A (v) represents the anchor information matrix, v represents the number of spectral intervals existing in the hyperspectral data, U (v) represents the basis matrix, and R represents the potential centroid matrix of the anchors.

[0022] Further, the bipartite graph decomposition module is configured to:

[0023]

[0024] s.t. P ∈ {0, 1}

[0025] where Z represents the consensus bipartite graph, P represents the clustering indicator matrix, and R represents the potential centroid matrix of the anchor points.

[0026] Further, the hyperspectral clustering model based on subspace clustering learning is configured to:

[0027]

[0028] s.t. A (v)T A (v) = I, U (v)T U (v) = I

[0029] where X (v) represents the hyperspectral data, v represents the number of spectral intervals existing in the hyperspectral data, A (v) represents the anchor point information matrix, Z represents the consensus bipartite graph, U (v) represents the basis matrix, P represents the clustering indicator matrix, R represents the potential centroid matrix of the anchor points, λ1 represents the first penalty parameter, and λ2 represents the second penalty parameter.

[0030] Further, the hyperspectral clustering model is iterated based on an alternating update variable strategy, including:

[0031] Updating and solving the consensus bipartite graph, where the anchor point information matrix, the basis matrix, the clustering indicator matrix, and the potential centroid matrix of the anchor points are fixed;

[0032] Updating and solving the anchor point information matrix, where the consensus bipartite graph, the basis matrix, the clustering indicator matrix, and the potential centroid matrix of the anchor points are fixed

[0033] Updating and solving the basis matrix, where the consensus bipartite graph, the anchor point information matrix, the clustering indicator matrix, and the potential centroid matrix of the anchor points are fixed;

[0034] Updating and solving the clustering indicator matrix, where the consensus bipartite graph, the anchor point information matrix, the basis matrix, and the potential centroid matrix of the anchor points are fixed;

[0035] Updating and solving the potential centroid matrix of the anchor points, where the consensus bipartite graph, the anchor point information matrix, the basis matrix, and the clustering indicator matrix are fixed.

[0036] Compared with the prior art, the present invention has the following advantages:

[0037] 1. The present invention proposes a new unsupervised clustering method to fully exploit the consistent information hidden in different bands, significantly improving the clustering accuracy.

[0038] 2. The present invention designs an adaptive dynamic anchor selection module, regards different spectral intervals as the feature information of ground objects and inputs them into the subspace clustering framework, fully utilizes the spatial information and the ground object consistency information, reduces the spatial complexity and the time complexity, and reduces the background noise interference.

[0039] 3. The present invention proposes a bipartite graph decomposition module to avoid the generation of suboptimal solutions in the bipartite graph clustering process, directly obtains the optimal clustering of the precise pixel division from the bipartite graph, fully utilizes the underlying information shared by samples in different bands, ensures the high representativeness of the anchors, and achieves the effect that the centroid of the anchor coincides with the centroid of the pixel cluster.

[0040] 4. The present invention adopts an optimization strategy to repair other variables in the sub-problem while alternately updating each variable, further optimizing the clustering process and realizing the improvement of the overall performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0042] Figure 1 It is a flowchart of an underwater hyperspectral clustering method based on a bipartite graph in an embodiment of the present invention.

[0043] Figure 2 It is a hyperspectral clustering model architecture based on subspace clustering learning in an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0044] In order to enable those skilled in the art to better understand the solution of the present invention, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0045] It should be noted that the terms "first", "second", etc. in the description, claims and above-mentioned drawings of the present invention are used to distinguish similar objects, and do not necessarily need to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged under appropriate circumstances, so that the embodiments of the present invention described here can be implemented in an order other than those illustrated or described here. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device that includes a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0046] As Figure 1 shown, the present invention provides a hyperspectral clustering method based on a bipartite graph, including the following steps:

[0047] S1. Obtain the image to be processed, and obtain the spectral images of the image to be processed in different spectral intervals. Take the spectral image of each spectral interval as a view, and construct hyperspectral data based on the obtained views where X υ represents the data of the υ-th view, n represents the number of pixel points in the hyperspectral image, and d v represents the number of bands in the v-th spectral interval. In this application, the bands are uniformly divided, and the entire spectral range is uniformly divided into several bands, and each band covers the same spectral width. During the division process, principles such as consistent band width, no overlap between bands, and clear band boundaries are followed to facilitate subsequent data processing and analysis.

[0048] S2. Construct a hyperspectral clustering model based on subspace clustering learning, and process the hyperspectral image based on the hyperspectral clustering model; the hyperspectral clustering model includes: an adaptive dynamic anchor point selection module, an anchor point centroid learning module, and a bipartite graph decomposition module.

[0049] Further, the adaptive dynamic anchor point selection module performs adaptive dynamic anchor point selection on the hyperspectral data to obtain an anchor point information matrix. Specifically, the Frobenius norm of the hyperspectral data is constrained and iteratively updated, aiming to achieve adaptive dynamic anchor point selection. In this application, the anchor point selection is expressed as follows:

[0050]

[0051] s.t.A (v)T A (v) =I

[0052] where, X (v)denotes the hyperspectral data, v denotes the number of spectral intervals existing in the hyperspectral data, A (v) denotes the anchor information matrix, which stores the anchor information of each view. The purpose of imposing an orthogonality constraint on it is to ensure the diversity of the anchors. Z denotes the consensus bipartite graph, which stores the weights of each anchor and its neighboring pixels.

[0053] Dynamically learn the anchors in different spectral intervals and store the anchor information in the matrix where m represents the predefined number of anchors, and the number of anchors is much smaller than the number of pixel points.

[0054] Furthermore, the anchor centroid learning module performs centroid learning on the anchors based on the anchor information matrix to obtain the latent centroid matrix of the anchors. Specifically, during the process of dynamically learning the anchors, in order to avoid the interference of background noise on the anchor positions, centroid learning is applied to the anchors, which is expressed as follows:

[0055]

[0056] s.t. U (v)T U (v) = I

[0057] where denotes the basis matrix. The purpose of imposing an orthogonality constraint on the matrix U (v) is to avoid infinite solutions, denotes the latent centroid matrix of the anchors, where c represents the category of the anchors. The latent centroid matrix R naturally classifies a large number of anchors, and the anchors of each category are associated with pixel clusters to ensure that the generated anchors are of high quality.

[0058] Furthermore, the bipartite graph decomposition module constructs a clustering indicator matrix and performs bipartite graph decomposition on the anchors based on the latent centroid matrix of the anchors and the clustering indicator matrix to obtain the consensus bipartite graph. Specifically, in order to avoid the generation of suboptimal solutions during the bipartite graph clustering process and directly obtain the optimal clustering with accurate pixel partitioning from the bipartite graph, make full use of the underlying information shared by samples in different bands to ensure that the anchors have high representativeness, and perform bipartite graph decomposition on the anchors, which is expressed as follows:

[0059]

[0060] s.t. P ∈ {0, 1}

[0061] where is the centroid matrix of the pixel clusters and is also the clustering indicator matrix. The meaning of its matrix elements can be expressed as follows:

[0062]

[0063] In summary, the hyperspectral clustering model constructed in this application based on subspace clustering learning is expressed as:

[0064]

[0065] s.t. A (v)T A (v) = I, U (v)T U (v) = I

[0066] where λ1 and λ2 are two penalty parameters, whose purpose is to balance the weights of the two modules in the overall framework, facilitating the optimization of the performance of the overall algorithm.

[0067] S3. Iterate the hyperspectral clustering model based on the alternating update variable strategy until the iteration stop condition is met, and then output the clustering result based on the clustering indicator matrix. Since it is difficult to jointly update all variables in the overall model, in order to achieve the optimal solution, an optimization strategy is adopted to fix other variables in the sub-problem while alternately updating each variable. The steps to update the variables are as follows:

[0068] S301. Update variable Z: When fixing other variables A (v) , U (v) , P, R, the optimization problem of Z can be transformed into:

[0069]

[0070] Take the derivative of the matrix Z in the above formula and set the derivative equal to 0, and we can get:

[0071]

[0072] S302. Update variable A (v) : When fixing other variables Z, U (v) , P, R, the optimization problem of A (v) can be transformed into:

[0073]

[0074] s.t. A (v)T A (v) = I

[0075] Take the derivative of the matrix A (v) in the above formula and set its derivative to 0, and we can get:

[0076] ZZ T A (v) + λ1A (v) - λ1U (v) P - X (v) ZT = 0

[0077] A (v) = (λ1U (v) P + X (v) Z T )(ZZ T + λ1I) -1

[0078] S303. Update variable U (v) : Fix other variables, the sub - problem of matrix U (v) can be transformed into:

[0079]

[0080] s.t. U (v)T U (v) = I

[0081] This can be equivalent to:

[0082]

[0083] s.t. U (v)T U (v) = I

[0084] The above equation is an orthogonal problem, which can be calculated by computing the singular value decomposition (SVD) of A (v) R T = UΣV T So U (v) = UV T .

[0085] S304. Update variable P: Fix other variables, take the derivative of the sub - problem of P and set the derivative to 0, we can get:

[0086]

[0087] P = (RR T ) -1 RZ

[0088] S305. Update variable R: Fix other variables, the sub - problem about R can be expressed as:

[0089]

[0090] The update of the above variables will continuously iterate and solve. When is satisfied, the update will stop, where t is the number of iterations and ε is set to 0.0001.

[0091] S306. Cluster the finally obtained P to get the clustering result.

[0092] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. An underwater hyperspectral clustering method based on a bipartite graph, characterized in that, It includes the following steps: S1. Obtain the image to be processed, and obtain the spectral images of the image to be processed in different spectral intervals. Take the spectral image of each spectral interval as a view, and construct hyperspectral data based on the obtained views; S2. Construct a hyperspectral clustering model based on subspace clustering learning, and process the hyperspectral image based on the hyperspectral clustering model; The hyperspectral clustering model includes: an adaptive dynamic anchor selection module, an anchor centroid learning module, and a bipartite graph decomposition module, The adaptive dynamic anchor selection module performs adaptive dynamic anchor selection on the hyperspectral data to obtain an anchor information matrix, The anchor centroid learning module performs centroid learning on the anchors based on the anchor information matrix to obtain a potential centroid matrix of the anchors, The bipartite graph decomposition module constructs a clustering indicator matrix, and performs bipartite graph decomposition on the anchors based on the potential centroid matrix of the anchors and the clustering indicator matrix to obtain a consensus bipartite graph; S3. Iterate the hyperspectral clustering model based on the alternating update variable strategy until the iteration stop condition is met, and then output the clustering result based on the clustering indicator matrix.

2. The underwater hyperspectral clustering method based on a bipartite graph according to claim 1, wherein The adaptive dynamic anchor selection module is set as: s.t.A (v)T A (v) = I Among them, X (v) represents hyperspectral data, v represents the number of spectral intervals existing in the hyperspectral data, and A (v) represents the anchor point information matrix, and Z represents the consensus bipartite graph.

3. A method for underwater hyperspectral clustering based on a bipartite graph according to claim 1, characterized in that, The anchor centroid learning module is set as: s.t.U (v)T U (v) =I Among them, A (v) represents the anchor point information matrix, v represents the number of spectral intervals existing in the hyperspectral data, and U (v) represents the basis matrix, and R represents the potential centroid matrix of the anchor points.

4. The underwater hyperspectral clustering method based on a bipartite graph according to claim 1, wherein The bipartite graph decomposition module is set as: s.t.P∈{0,1} where Z represents the consensus bipartite graph, P represents the clustering indicator matrix, and R represents the potential centroid matrix of the anchors.

5. The underwater hyperspectral clustering method based on bipartite graph according to claim 1, characterized in that, The hyperspectral clustering model based on subspace clustering learning is set as: s.t.A (v)T A (v) = I, U (v)T U (v) = I Among them, X (v) represents hyperspectral data, v represents the number of spectral intervals existing in the hyperspectral data, A (v) represents the anchor point information matrix, Z represents the consensus bipartite graph, U (v) represents the basis matrix, P represents the clustering indicator matrix, R represents the potential centroid matrix of the anchor points, λ1 represents the first penalty parameter, and λ2 represents the second penalty parameter.

6. The underwater hyperspectral clustering method based on a bipartite graph according to claim 5, wherein Iterating the hyperspectral clustering model based on the alternating update variable strategy includes: Updating and solving the consensus bipartite graph, where the anchor information matrix, the basis matrix, the clustering indicator matrix, and the potential centroid matrix of the anchors are fixed; Updating and solving the anchor information matrix, where the consensus bipartite graph, the basis matrix, the clustering indicator matrix, and the potential centroid matrix of the anchors are fixed; Updating and solving the basis matrix, where the consensus bipartite graph, the anchor information matrix, the clustering indicator matrix, and the potential centroid matrix of the anchors are fixed; Updating and solving the clustering indicator matrix, where the consensus bipartite graph, the anchor information matrix, the basis matrix, and the potential centroid matrix of the anchors are fixed; Updating and solving the potential centroid matrix of the anchors, where the consensus bipartite graph, the anchor information matrix, the basis matrix, and the clustering indicator matrix are fixed.

Images