A hyperspectral image band selection method, system, device and medium

Abstract

The application belongs to the technical field of spectral data classification, and discloses a hyperspectral image wave band selection method, system, device and medium; the hyperspectral image wave band selection method comprises the following steps: based on a hyperspectral image to be selected in a wave band, a multi-scale similar graph matrix is obtained; the obtained multi-scale similar graph matrix is fused to obtain a fused similar graph matrix; the obtained fused similar graph matrix is substituted into an edge graph self-representation model to obtain a multi-scale edge enhancement graph self-representation model; the obtained multi-scale edge enhancement graph self-representation model is solved to obtain a wave band sorting based on importance; and based on the wave band sorting, a hyperspectral image wave band selection result is obtained. The application can effectively improve the accuracy, robustness and generalization performance of hyperspectral image wave band selection, can reduce the case of false selection of wave bands, and can improve the wave band representation ability after selection of wave bands.

Description

Technical Field

[0001] This invention belongs to the field of spectral data classification technology, and specifically relates to a method, system, device and medium for selecting bands in hyperspectral images. Background Technology

[0002] Hyperspectral images, also known as remote sensing images with high spectral resolution, utilize hundreds of bands to provide rich spectral information, improving the ability to identify surface features, materials, and structures. They are widely used in national defense, military, and resource exploration. However, the high redundancy of hyperspectral images leads to reduced classification accuracy and increased algorithm complexity. Therefore, band selection that simplifies features, preserves original spectral information, and reduces algorithm complexity has become a hot topic in current hyperspectral image processing research.

[0003] In existing technologies, hyperspectral image band selection is typically based on Marginalized Graph Self-Representation (MGSR). Specifically, MGSR utilizes the graph structure in graph convolutional neural networks to construct similar graphs through superpixel segmentation, recording the spatial relationships between adjacent pixels to explore spatial information in different homogeneous regions. It also introduces edge corruption to enhance data and improve the model's generalization ability and robustness. However, the MGSR model does not consider the limited ability of a single similar graph structure to extract spatial-spectral information, nor does it delve into the weight matrices within the similar graph structure, and it does not account for the external differences in superpixels, resulting in limited generalization ability. These shortcomings lead to band selection errors and weak representational capabilities of band combinations after band selection. Summary of the Invention

[0004] The purpose of this invention is to provide a method, system, device, and medium for selecting bands in hyperspectral images, thereby solving one or more of the aforementioned technical problems. The technical solution provided by this invention can effectively improve the accuracy, robustness, and generalization performance of hyperspectral image band selection, reduce the occurrence of incorrect band selection, and improve the band characterization capability after selection.

[0005] To achieve the above objectives, the present invention adopts the following technical solution:

[0006] The first aspect of this invention provides a method for selecting bands in a hyperspectral image, comprising the following steps:

[0007] Based on the hyperspectral image of the band to be selected, obtain a multi-scale similarity map matrix;

[0008] The obtained multi-scale similarity map matrices are fused to obtain a fused similarity map matrix;

[0009] The obtained fused similarity map matrix is ​​substituted into the edge map self-representation model to obtain a multi-scale edge enhancement map self-representation model; the obtained multi-scale edge enhancement map self-representation model is solved to obtain a band ranking based on importance; based on the band ranking, the hyperspectral image band selection result is obtained.

[0010] A further improvement of the present invention is that, in the multi-scale similarity graph matrix, the similarity graph matrix A obtained at the y-th scale is... y middle,

[0011]

[0012]

[0013] In the formula, N k For superpixel ratio, n k H represents the number of pixels in the k-th superpixel block, where n is the total number of pixels in the hyperspectral image; k and H l These are the k-th and l-th superpixel blocks, respectively;

[0014] If pixels i and j are assigned to different superpixel blocks, the similarity weight value is 0; if pixels i and j are in the same superpixel block, the similarity weight value is obtained by extracting the spatial spectral information of the two pixels.

[0015] A further improvement of the present invention is that the fused similarity graph matrix is ​​represented as follows:

[0016]

[0017] In the formula, A S The matrix represents the fused similarity graphs; Y represents the total number of scales.

[0018] A further improvement of the present invention is that the multi-scale edge enhancement graph self-representation model is expressed as follows:

[0019]

[0020] In the formula, M represents the total number of times the data is destroyed and reconstructed, M→∞; X is the input hyperspectral data; To normalize and fuse similarity graph matrices; For the m-th destruction and reconstruction data; W is the self-representation matrix to be obtained; λ is the regularization coefficient; ||W|| 2,1 Let W be the 2,1 norm of matrix W. This indicates that the F2 norm of the expression is to be calculated.

[0021] in, In the formula, It is a fusion similarity graph matrix with self-looping, IN It is an N×N identity matrix; It is a degree matrix, with diagonal elements of 1. This represents the i-th row and j-th column of the degree matrix. Let represent the i-th row and j-th column of a fusion similarity graph matrix with self-looping characteristics.

[0022] A second aspect of the present invention provides a hyperspectral image band selection system, comprising:

[0023] The multi-scale similarity map matrix acquisition module is used to acquire a multi-scale similarity map matrix based on the hyperspectral image to be selected for band selection.

[0024] The fusion module is used to fuse the obtained multi-scale similarity map matrices to obtain a fused similarity map matrix;

[0025] The sorting module is used to substitute the obtained fused similarity map matrix into the edge map self-representation model to obtain a multi-scale edge enhancement map self-representation model; solve the obtained multi-scale edge enhancement map self-representation model to obtain a band sort based on importance; and obtain the hyperspectral image band selection result based on the band sort.

[0026] A further improvement of the present invention is that, in the multi-scale similarity graph matrix, the similarity graph matrix A obtained at the y-th scale is... y middle,

[0027]

[0028]

[0029] In the formula, N k For superpixel ratio, n k H represents the number of pixels in the k-th superpixel block, where n is the total number of pixels in the hyperspectral image; k and H l These are the k-th and l-th superpixel blocks, respectively;

[0030] If pixels i and j are assigned to different superpixel blocks, the similarity weight value is 0; if pixels i and j are in the same superpixel block, the similarity weight value is obtained by extracting the spatial spectral information of the two pixels.

[0031] A further improvement of the present invention is that the fused similarity graph matrix is ​​represented as follows:

[0032]

[0033] In the formula, A S The matrix represents the fused similarity graphs; Y represents the total number of scales.

[0034] A further improvement of the present invention is that the multi-scale edge enhancement graph self-representation model is expressed as follows:

[0035]

[0036] In the formula, M represents the total number of times the data is destroyed and reconstructed, M→∞; X is the input hyperspectral data; To normalize and fuse similarity graph matrices; For the m-th destruction and reconstruction data; W is the self-representation matrix to be obtained; λ is the regularization coefficient; ||W|| 2,1 Let W be the 2,1 norm of matrix W. This indicates that the F2 norm of the expression is to be calculated.

[0037] in, In the formula, It is a fusion similarity graph matrix with self-looping, I N It is an N×N identity matrix; It is a degree matrix, with diagonal elements of 1. This represents the i-th row and j-th column of the degree matrix. Let represent the i-th row and j-th column of a fusion similarity graph matrix with self-looping characteristics.

[0038] A third aspect of the present invention provides an electronic device comprising:

[0039] At least one processor; and,

[0040] A memory communicatively connected to the at least one processor; wherein,

[0041] The memory stores instructions executable by the at least one processor, which, when executed by the at least one processor, enables the at least one processor to perform the hyperspectral image band selection method as described in any of the first aspects of the invention.

[0042] The fourth aspect of the present invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the hyperspectral image band selection method described in any one of the first aspects of the present invention.

[0043] Compared with the prior art, the present invention has the following beneficial effects:

[0044] The technical solution provided by this invention can effectively improve the accuracy, robustness and generalization performance of hyperspectral image band selection, reduce the occurrence of incorrect band selection, and improve the band characterization ability after band selection.

[0045] To address the issue of insufficient extraction of spatial-spectral information by the MGSR model, this invention specifically designs a multi-scale similarity graph structure, which can fully extract spatial-spectral information. Furthermore, in image processing, the representation of tiny and extremely large objects significantly impacts model performance. Therefore, this invention samples images at different granularities, allowing the observation of different features at different scales, thereby achieving a more powerful feature representation capability.

[0046] To address the limited generalization ability of the MGSR model, this invention specifically discloses a design for multi-scale similar graph structure fusion and superpixel segmentation. By fusing similar graphs at multiple scales, the feature representation ability of the model is enhanced, resulting in strong robustness and improved generalization ability. In addition, this invention designs a similar graph structure for superpixel segmentation, which is more robust in representing features compared to the similar graph structure of MGSR. Attached Figure Description

[0047] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art are briefly introduced below; obviously, the drawings described below are some embodiments of the present invention, and those skilled in the art can obtain other drawings based on these drawings without creative effort.

[0048] Figure 1 This is a flowchart illustrating a hyperspectral image band selection method provided in an embodiment of the present invention;

[0049] Figure 2 This is a schematic block diagram of the hyperspectral image band method based on edge graph self-representation in an embodiment of the present invention;

[0050] Figure 3 This is a schematic diagram comparing the classification methods of various approaches using the IP dataset and the K-means classifier in an embodiment of the present invention; wherein, Figure 3 (a) is a reference figure. Figure 3 (b) represents Allbands. Figure 3 In the middle (c), EFDPC is represented. Figure 3 In the middle (d), EGCSR is represented. Figure 3 In the middle (e), MSGR is represented. Figure 3 (f) stands for Ours;

[0051] Figure 4 This is a schematic diagram comparing the classification methods of various methods on the KSC dataset and under the SVM classifier in an embodiment of the present invention; wherein, Figure 4 (a) is a reference figure. Figure 4 (b) represents Allbands. Figure 4 In the middle (c), EFDPC is represented. Figure 4 In the middle (d), EGCSR is represented. Figure 4 In the middle (e), MSGR is represented. Figure 4 (f) stands for Ours;

[0052] Figure 5 This is a schematic diagram comparing the classification methods for the Botswana dataset and the few-shot problem in this embodiment of the invention; wherein, Figure 5 (a) is a reference figure. Figure 5 (b) represents Allbands. Figure 5 In the middle (c), EFDPC is represented. Figure 5 In the middle (d), EGCSR is represented. Figure 5 In the middle (e), MSGR is represented. Figure 5 (f) stands for Ours;

[0053] Figure 6 This is a schematic diagram of a hyperspectral image band selection system provided in an embodiment of the present invention. Detailed Implementation

[0054] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0055] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0056] The present invention will now be described in further detail with reference to the accompanying drawings:

[0057] Please see Figure 1 and Figure 2 The present invention provides a hyperspectral image band selection method, specifically a hyperspectral image band selection method based on edge map self-representation, comprising the following steps:

[0058] Step 1: Based on the hyperspectral image of the band to be selected, obtain a multi-scale similarity map matrix;

[0059] Step 2: Fuse the multi-scale similarity map matrix obtained in Step 1 to obtain a fused similarity map matrix;

[0060] Step 3: Substitute the fused similarity graph matrix obtained in Step 2 into the edge graph self-representation model to obtain a multi-scale edge enhancement graph self-representation model; solve the obtained multi-scale edge enhancement graph self-representation model to obtain a band ranking based on importance; based on the band ranking, obtain the hyperspectral image band selection result.

[0061] The technical solution provided by the embodiments of this invention can effectively improve the accuracy, robustness, and generalization performance of hyperspectral image band selection, enhance the band representation ability after band selection, and reduce the occurrence of incorrect band selection. Compared with the MGSR model, the model proposed in the embodiments of this invention has better classification performance, stronger generalization ability and robustness, and outperforms the MGSR model in small sample problems.

[0062] In this embodiment of the invention, further explanation is provided regarding the construction of the multi-scale similarity graph matrix in step 1 above:

[0063] Different superpixel blocks contain different spatial information. The larger the proportion of a superpixel block in space, the greater the possibility of spatial misclassification. Therefore, the similarity graph matrix A obtained at the y-th scale... y In this invention, the present invention creatively proposes the concept of superpixel ratio (i.e., the ratio of the number of pixels in a superpixel block to the total number of pixels in the image); the superpixel ratio is used to represent the external spatial difference of the superpixel block, and the variance is used to represent the internal spectral difference within the superpixel block. In order to eliminate the adverse effects caused by singular sample data, the similarity weight range is set to [0,1].

[0064] Specifically, the final constructed enhanced region similarity map matrix takes the following form:

[0065]

[0066]

[0067] In the formula, N k For superpixel ratio, n k H represents the number of pixels in the k-th superpixel block, where n is the total number of pixels in the hyperspectral image; k and H l These are the k-th and l-th superpixel blocks, respectively;

[0068] If pixels i and j are assigned to different superpixel blocks H k H l In the middle, their similarity weight A y If (i,j) is 0, and these two pixels are within the same superpixel block, then the similarity weight value A is obtained by extracting the spatial spectral information of these two pixels. y (i,j);

[0069] Within the same superpixel block, the greater the spectral variance between pixels, the lower the similarity. When superpixels have the same variance, smaller superpixel blocks are less likely to be misclassified, and their spatial information is more accurate, so pixels within smaller superpixel blocks are more similar.

[0070] In this embodiment of the invention, the graph structure at multiple scales reflects the spatial relationships and spectral information of the image at different scales, which can make more comprehensive use of the information contained in the image and improve robustness and generalization ability. By constructing a multi-scale graph structure, the feature representation ability of the selected bands for classification tasks can be enhanced. By designing similar graph structures, the situation of bands being incorrectly selected can be effectively reduced, and more robust feature representation ability can be obtained.

[0071] In this embodiment of the invention, further explanation is provided regarding the acquisition of the fused similarity graph matrix in step 2:

[0072] For a selected hyperspectral image, Y different scales of superpixel numbers are chosen to measure spatial relationships. Superpixel images obtained based on different superpixel numbers correspond to different scales. A y Let A represent the similarity graph matrix obtained at the y-th scale, and let A be the new fused similarity graph matrix. S :

[0073]

[0074] In order to ensure A during information transmission S The original distribution is obtained by normalizing the matrix. Represented as,

[0075]

[0076] in, It is a fusion similarity graph matrix with self-looping, I N It is an N×N identity matrix. It is a degree matrix, with diagonal elements of 1. in, This represents the i-th row and j-th column of the degree matrix. Let represent the i-th row and j-th column of a fusion similarity graph matrix with self-looping characteristics.

[0077] In this embodiment of the invention, further explanation is provided regarding the acquisition of the multi-scale edge enhancement map self-representation model in step 3:

[0078] The resulting multi-scale edge enhancement map self-representation model is represented as follows:

[0079]

[0080] In the formula, M represents the total number of times the data is destroyed and reconstructed, M→∞, and X is the input hyperspectral data. This is the normalized fusion similarity graph matrix proposed in this invention. Let W be the self-representation matrix to be obtained for the m-th destruction and reconstruction, λ be the regularization coefficient, and ||W|| 2,1 Let W be the 2,1 norm of matrix W. This indicates that the F2 norm of the expression is to be calculated.

[0081] In this embodiment of the invention, the solution process is summarized as follows: The model solution is transformed into a trace equation; the trace equation is differentiated by matrix, and the derivative is set to zero to solve the equation; since there are corrupted terms in the equation, to avoid explicit use of these terms, according to the weak law of large numbers, assuming M→∞, explicit corrupted terms are eliminated to obtain the desired output self-representation matrix W; the bands are ranked by importance based on the self-representation matrix W to obtain the selected bands. For example, the bands with the highest importance ranking can be used as the final selection result.

[0082] For example, to facilitate the solution, the present invention defines the auxiliary matrix as follows: M represents the total number of times the data has been destroyed and reconstructed, M→∞. This is the normalized fusion similarity graph matrix proposed in this invention. This represents the data from the m-th destruction and reconstruction.

[0083] The multi-scale edge enhancement graph self-representation model is rewritten as follows:

[0084]

[0085] In the formula, C is a B×B diagonal matrix, with diagonal elements... C ii W represents the i-th row and i-th column of a diagonal matrix. i This represents a 1×N matrix formed by the i-th row of the weight matrix W.

[0086] According to the principle of matrix derivatives, the derivative of J(W) with respect to W can be expressed as:

[0087]

[0088] make We can obtain:

[0089]

[0090] because It contains explicit impairment terms. To avoid directly using explicit impairment terms, according to the weak large number theorem, when M→∞, It can be rewritten to its expectation. Therefore, we get W = (T + λC) -1 U;

[0091] in,

[0092]

[0093]

[0094] make for and For the undamaged term, 1-p is the feature survival probability, resulting in the following formula:

[0095]

[0096] U ij =Q ij (1-p);

[0097] For T, when the features are different, the survival probability of the feature is (1-p). 2 Conversely, the survival probability of a feature is (1-p). For U, since there is only one feature, the survival probability of the feature is (1-p).

[0098] Please see Figures 3 to 5 The following is an exemplary experiment using real hyperspectral data from an embodiment of the present invention:

[0099] Based on the method described in the embodiments of the present invention, three sets of publicly available real hyperspectral image datasets are used to test and illustrate the hyperspectral image band selection method based on edge graph self-representation provided in the embodiments of the present invention, as well as to analyze and evaluate its application effects.

[0100] To evaluate the proposed model, embodiments of the present invention conduct experiments on three hyperspectral datasets, namely the Indian Pines (IP) dataset, the KSC, and the Botswana dataset. The overall accuracy (OA), average accuracy (AA), and Kappa coefficient are selected as quantitative evaluation indicators. To demonstrate the performance of the self-representation model, embodiments of the present invention are compared with the Enhanced Fast Density-Peak-Based Clustering (EFDPC). To illustrate the improvement of embodiments of the present invention, embodiments of the present invention are compared with the original full-band Allbands, the non-Euclidean domain self-representation model EGCSR, and the benchmark model MGSR method.

[0101] To qualitatively measure the proposed model, embodiments of the present invention use two classifiers, Kmeans and SVM, to verify the effectiveness of the algorithm. Hyperspectral data has the characteristics of being easy to capture and difficult to label. Most hyperspectral images are unlabeled images, which greatly increases the classification difficulty. To further demonstrate the generalization ability and robustness of the model, embodiments of the present invention conduct experimental comparisons for the hyperspectral small sample problem.

[0102] Table 1 shows the dataset comparison; among them, the KSC dataset has a total of 176 bands, and 21 low signal-to-noise ratio bands numbered 1, 132, 133, 134, 139, 142, 156, 157, 160, 161, 164, 167-176 are removed, and finally 155 effective bands remain.

[0103] Table 1. Dataset Comparison

[0104]

[0105] The model proposed by embodiments of the present invention has a total of Y + 2 hyperparameters. Considering the calculation time and complexity, for the scale number Y, Y = 3 is selected, namely a, b, c (a < b < c), and the value range is set to [3, 5, 10, 15, 25, 35, 45, 47, 50, 65, 90, 100]. The value range of the regularization term coefficient λ is [10 -3 , 10 -2 , 10 -1 , 10 0 , 10 1 , 10 2 , 10 3 , 10 4 , 10 5The damage probability p ranges from [0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9]. The optimal hyperparameters are found through iterative searching. The dataset undergoes uniform preprocessing by normalizing the maximum and minimum values ​​of each band, and the top 25 sorted bands are selected for evaluating classification performance.

[0106] For the K-means classifier, to avoid errors, the embodiments of this invention employ an averaging operation of ten times. The hyperparameters of the model on the three datasets are [25, 35, 50, 10000, 0.5], [25, 35, 50, 100, 0.1], and [25, 35, 100, 100, 0.8], respectively. The experimental results are shown in Table 2. On the IP dataset, the Kappa coefficient is 1.1% lower than the full band and 0.7% higher than the benchmark algorithm MGSR. OA is 1.8% higher than the full band and 0.1% higher than the benchmark algorithm MGSR. AA is 0.6% higher than the benchmark algorithm. On the KSC dataset, OA is 2.3% higher than the full band and 1.2% higher than MGSR. AA and Kappa coefficients are 3.8% and 4.3% higher than MGSR, respectively. On the Botswana dataset, OA is slightly lower than the full band but higher than the benchmark model MGSR. AA and Kappa coefficients are the best among the compared models.

[0107] Table 2. Classification results of different algorithms on the K-means classifier

[0108]

[0109] For the SVM classifier, 20% of the samples in each class were used as the training set, and all samples were used as the test set for testing. The trainer was libsvm, with default parameter settings. The hyperparameters for the three datasets were [5, 15, 50, 0.01, 0.1], [25, 35, 50, 100, 0.4], and [25, 47, 50, 10, 0.1]. The experimental results are shown in Table 3. In the IP dataset, the model proposed in this embodiment improved the coefficients of OA and Kappa by approximately 0.3% compared to MGSR, while reducing AA by 0.4%, showing little difference in performance. In the KSC dataset, the model provided in this embodiment improved the coefficients of OA, AA, and Kappa by approximately 9%, 11%, and 11% respectively compared to MGSR, all of which were better than the MGSR model. In the Botswana dataset, the coefficients of OA, AA, and Kappa were all improved by approximately 3% compared to the baseline model MGSR, which was better than the MGSR model.

[0110] Table 3. Classification results of different algorithms on the SVM classifier

[0111]

[0112] For the problem of small sample size in hyperspectral imaging, in this embodiment of the invention, five samples are selected as the training set for each class, and all samples are used as the test set for testing. The hyperparameters of the three datasets are [40, 50, 65, 0.01, 0.1], [25, 35, 65, 100, 0.5], and [25, 35, 50, 100, 0.1]. The experimental results are shown in Table 4. In the IP dataset, the coefficients of OA, AA, and Kappa of the new model are slightly lower than those of the full band, but 1.2%, 4.1%, and 1.8% higher than those of MGSR, respectively. In the KSC dataset, the coefficients of OA, AA, and Kappa are slightly lower than those of the full band, but 10.6%, 9.7%, and 12.2% higher than those of MGSR, and 1.2%, 1.4%, and 1.2% higher than those of EGCSR. This indicates that in small-sample classification tasks, the robustness and classification performance of the new model are superior to the comparison models. In the Botswana dataset, the coefficients of OA, AA, and Kappa are improved by approximately 3.5%, 3%, and 3.6% compared to the full band and EGCSR, and 10.2%, 10.6%, and 10.9% higher than the baseline model MGSR, respectively.

[0113] Table 4. Classification results of different algorithms on few-sample classification

[0114]

[0115] Please see Figure 3 A comparison was made between various algorithms for the IP dataset and the Kmeans classifier; the model proposed in this invention has a significantly better classification effect than the baseline model in the lower right corner square part of the image, and its overall classification performance is better than the full band.

[0116] Please see Figure 4 A comparison was made between various algorithms for the KSC dataset and SVM classifiers; the overall accuracy of the model proposed in this invention is slightly lower than that of the full-band model, but better than other algorithms. Figure 4 As can be seen, the algorithm proposed in this invention performs well in classifying the lower part of the image, but it fails to classify the right side, which limits the classification effect of the model.

[0117] Please see Figure 5 For the Botswana dataset and the few-sample problem, various algorithms were compared. Compared to other algorithms, the algorithm proposed in this embodiment of the invention has the best classification effect. Compared to the EFDPC model, the model proposed in this invention can classify the upper right corner and lower region very well. This invention, along with the EGCSR and MGSR models, misclassified the middle region of the image, but for the right side of the image, the method proposed in this invention is superior to other algorithms.

[0118] In summary, the experimental results show that the new model proposed in this invention has better generalization ability and performance than other algorithms.

[0119] 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 disclosed in the apparatus embodiments, please refer to the embodiments of the method of the present invention.

[0120] Please see Figure 6 In another embodiment of the present invention, a hyperspectral image band selection system is provided, comprising:

[0121] The multi-scale similarity map matrix acquisition module is used to acquire a multi-scale similarity map matrix based on the hyperspectral image to be selected for band selection.

[0122] The fusion module is used to fuse the obtained multi-scale similarity map matrices to obtain a fused similarity map matrix;

[0123] The sorting module is used to substitute the obtained fused similarity map matrix into the edge map self-representation model to obtain a multi-scale edge enhancement map self-representation model; solve the obtained multi-scale edge enhancement map self-representation model to obtain a band sort based on importance; and obtain the hyperspectral image band selection result based on the band sort.

[0124] In another embodiment of the present invention, a computer 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 in the computer storage medium to achieve a corresponding method flow or corresponding function. The processor described in this embodiment of the present invention can be used for the operation of a hyperspectral image band selection method.

[0125] In another embodiment of the present invention, a storage medium is provided, specifically a computer-readable storage medium (Memory), which is a memory device in a computer device used to store programs and data. It is understood that the computer-readable storage medium here can include both the built-in storage medium in the computer device and extended storage media supported by the computer device. The computer-readable storage medium provides storage space that stores the terminal's operating system. Furthermore, the storage space also stores one or more instructions suitable for loading and execution by a processor. These instructions can be one or more computer programs (including program code). It should be noted that the computer-readable storage medium here can be high-speed RAM or non-volatile memory, such as at least one disk storage device. The processor can load and execute one or more instructions stored in the computer-readable storage medium to implement the corresponding steps of the hyperspectral image band selection method in the above embodiments.

[0126] 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.

[0127] 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 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0128] 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 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0129] 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 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0130] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for selecting bands in a hyperspectral image, characterized in that, Includes the following steps: Based on the hyperspectral image of the band to be selected, obtain a multi-scale similarity map matrix; The obtained multi-scale similarity map matrices are fused to obtain a fused similarity map matrix; The obtained fused similarity map matrix is ​​substituted into the edge map self-representation model to obtain a multi-scale edge enhancement map self-representation model; the obtained multi-scale edge enhancement map self-representation model is solved to obtain an importance-based band ranking; based on the band ranking, the hyperspectral image band selection result is obtained. Among them, in the multi-scale similarity graph matrix, the first... Similarity graph matrix obtained at each scale middle, ； ； In the formula, For superpixel ratio, For the first The number of pixels in a superpixel block This represents the total number of pixels in the hyperspectral image. and The first The and the first One superpixel block; Among them, pixels , When assigned to different superpixel blocks, the similarity weight value is... If pixels , Within the same superpixel block, spatial spectral information is extracted from two pixels to obtain similarity weight values.

2. The hyperspectral image band selection method according to claim 1, characterized in that, The fused similarity graph matrix is ​​represented as follows: ； In the formula, To fuse similarity graph matrices; Y This represents the total number of scales.

3. The hyperspectral image band selection method according to claim 2, characterized in that, The multi-scale edge enhancement map self-representation model is expressed as follows: ； In the formula, This indicates the total number of times the data has been destroyed and rebuilt. ; For input hyperspectral data; To normalize and fuse similarity graph matrices; For the first Data on secondary destruction and reconstruction; The desired self-representation matrix; The coefficient of the regularization term; For the matrix The 2,1 norm, This means to find the expression. Norm; in, In the formula, It is a fused similarity graph matrix with self-looping properties. yes The identity matrix; It is a degree matrix, with diagonal elements of 1. , The degree matrix represents the first degree. Line number List, The first digit represents the fused similarity graph matrix with self-looping characteristics. Line number List.

4. A hyperspectral image band selection system, characterized in that, include: The multi-scale similarity map matrix acquisition module is used to acquire a multi-scale similarity map matrix based on the hyperspectral image to be selected for band selection. The fusion module is used to fuse the obtained multi-scale similarity map matrices to obtain a fused similarity map matrix; The sorting module is used to substitute the obtained fused similarity map matrix into the edge map self-representation model to obtain a multi-scale edge enhancement map self-representation model; solve the obtained multi-scale edge enhancement map self-representation model to obtain an importance-based band sort; and obtain the hyperspectral image band selection result based on the band sort. Among them, in the multi-scale similarity graph matrix, the first... Similarity graph matrix obtained at each scale middle, ； ； In the formula, For superpixel ratio, For the first The number of pixels in a superpixel block This represents the total number of pixels in the hyperspectral image. and The first The and the first One superpixel block; Among them, pixels , When assigned to different superpixel blocks, the similarity weight value is... If pixels , Within the same superpixel block, spatial spectral information is extracted from two pixels to obtain similarity weight values.

5. A hyperspectral image band selection system according to claim 4, characterized in that, The fused similarity graph matrix is ​​represented as follows: ； In the formula, To fuse similarity graph matrices; Y This represents the total number of scales.

6. The hyperspectral image band selection system according to claim 5, characterized in that, The multi-scale edge enhancement map self-representation model is expressed as follows: ； In the formula, This indicates the total number of times the data has been destroyed and rebuilt. ; For input hyperspectral data; To normalize and fuse similarity graph matrices; For the first Data on secondary destruction and reconstruction; The desired self-representation matrix; The coefficient of the regularization term; For the matrix The 2,1 norm, This means to find the expression. Norm; in, In the formula, It is a fused similarity graph matrix with self-looping properties. yes The identity matrix; It is a degree matrix, with diagonal elements of 1. , The degree matrix represents the first degree. Line number List, The first digit represents the fused similarity graph matrix with self-looping characteristics. Line number List.

7. An electronic device, characterized in that, include: At least one processor; as well as, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the hyperspectral image band selection method as described in any one of claims 1 to 3.

8. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the hyperspectral image band selection method according to any one of claims 1 to 3.

Images