Hyperspectral data wave band selection method and device based on spatial spectrum abnormal distribution

Abstract

The invention provides a hyperspectral data wave band selection method and device based on spatial spectrum abnormal distribution, and is applied to the technical field of data processing. The method comprises the following steps: acquiring a hyperspectral image; the hyperspectral image is decomposed into a background subspace and an abnormal subspace, a joint feature subspace of each wave band is determined according to the background subspace and the abnormal subspace, and the abnormal subspace is a subspace mixed with noise and abnormity; calculating a local abnormal factor of each pixel in the abnormal subspace to obtain an abnormal reference graph, and calculating a local abnormal factor of each pixel in the combined feature subspace of each wave band to obtain an abnormal degree graph of the corresponding wave band; and respectively calculating the similarity between the abnormal degree graph of each wave band and the abnormal reference graph to obtain the wave band scores of all the wave bands, and selecting a target wave band set according to the wave band scores of all the wave bands.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a method and device for selecting a hyperspectral data band based on spatial spectrum abnormal distribution. Background Art

[0002] Hyperspectral Image (HSI) has hundreds of spectral bands and contains rich spectral information, which can record the spectral fingerprints of different materials.

[0003] Hyperspectral Anomaly Detection (HAD) is used to find abnormal pixels in hyperspectral images that are different from surrounding pixels without prior information. In the prior art, in order to improve the accuracy of information in hyperspectral images, abnormal pixels in hyperspectral images can be detected by HAD.

[0004] However, the performance of hyperspectral anomaly detection tasks is usually severely limited by redundant spectral bands, and the detection accuracy tends to increase first and then decrease as the number of bands increases. Summary of the invention

[0005] The present invention provides a method and device for selecting a hyperspectral data band based on spatial-spectral anomaly distribution, so as to solve the problem in the prior art that the performance of a hyperspectral anomaly detection task is severely limited by redundant spectral bands.

[0006] The present invention provides a hyperspectral data band selection method based on space-spectrum anomaly distribution, comprising: acquiring a hyperspectral image; decomposing the hyperspectral image into a background subspace and an anomaly subspace, and determining a joint feature subspace of each band according to the background subspace and the anomaly subspace, wherein the anomaly subspace is a subspace mixed with noise and anomaly; obtaining an anomaly reference map by calculating a local anomaly factor of each pixel in the anomaly subspace, and obtaining an anomaly degree map of a corresponding band by calculating a local anomaly factor of each pixel in the joint feature subspace of each band; respectively calculating the similarity between the anomaly degree map of each band and the anomaly reference map to obtain band scores of all bands, and selecting a target band set according to the band scores of all bands.

[0007] According to a hyperspectral data band selection method based on spatial-spectral anomaly distribution provided by the present invention, after calculating the local anomaly factor of each pixel in the anomaly subspace, the method further includes: screening abnormal pixels according to the local anomaly factor of the pixel and the local sparsity of the abnormal target to obtain the abnormal reference image.

[0008] According to a hyperspectral data band selection method based on spatial-spectral anomaly distribution provided by the present invention, the similarity between the anomaly degree map of each band and the anomaly reference map is calculated respectively to obtain the band scores of all bands, including: using a mutual information method to measure the similarity between the anomaly degree map of each band and the anomaly reference map to obtain the band scores of all bands.

[0009] According to a hyperspectral data band selection method based on spatial-spectral anomaly distribution provided by the present invention, the target band set is selected according to the band scores of all bands, including: sorting all bands in order from high to low according to the band scores, and determining a preset number of bands with the highest band scores as the target band set.

[0010] The present invention also provides a hyperspectral data band selection device based on space-spectral anomaly distribution, comprising the following modules: an acquisition module and a processing module; the acquisition module is used to acquire a hyperspectral image; the processing module is used to decompose the hyperspectral image into a background subspace and an anomaly subspace, and determine a joint feature subspace of each band according to the background subspace and the anomaly subspace, wherein the anomaly subspace is a subspace mixed with noise and anomalies; an anomaly reference image is obtained by calculating a local anomaly factor of each pixel in the anomaly subspace, and an anomaly degree image of the corresponding band is obtained by calculating a local anomaly factor of each pixel in the joint feature subspace of each band; the similarity between the anomaly degree image of each band and the anomaly reference image is calculated respectively to obtain the band scores of all bands, and a target band set is selected according to the band scores of all bands.

[0011] According to a hyperspectral data band selection device based on spatial spectrum anomaly distribution provided by the present invention, the processing module is used to screen abnormal pixels according to the local abnormal factors of the pixels and the local sparsity of the abnormal targets to obtain the abnormal reference map.

[0012] According to a hyperspectral data band selection device based on spatial spectrum anomaly distribution provided by the present invention, the processing module is used to use the mutual information method to measure the similarity between the anomaly degree map of each band and the anomaly reference map to obtain the band scores of all bands.

[0013] According to a hyperspectral data band selection device based on spatial spectrum anomaly distribution provided by the present invention, the processing module is used to sort all bands in order from high to low according to the band scores, and determine a preset number of bands with the highest band scores as the target band set.

[0014] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, a hyperspectral data band selection method based on spatial spectrum anomaly distribution as described above is implemented.

[0015] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any of the above-described methods for selecting a hyperspectral data band based on spatial spectrum anomaly distribution.

[0016] The present invention also provides a computer program product, comprising a computer program, which, when executed by a processor, implements any of the above-mentioned methods for selecting a hyperspectral data band based on spatial spectrum anomaly distribution.

[0017] The hyperspectral data band selection method and device based on space-spectral anomaly distribution provided by the present invention can obtain a hyperspectral image; decompose the hyperspectral image into a background subspace and an anomaly subspace, and determine the joint feature subspace of each band according to the background subspace and the anomaly subspace, wherein the anomaly subspace is a subspace mixed with noise and anomaly; obtain an anomaly reference map by calculating the local anomaly factor of each pixel in the anomaly subspace, and obtain an anomaly degree map of the corresponding band by calculating the local anomaly factor of each pixel in the joint feature subspace of each band; respectively calculate the similarity between the anomaly degree map of each band and the anomaly reference map to obtain the band scores of all bands, and select a target band set according to the band scores of all bands. Through this scheme, on the one hand, since the target band set can be determined from the hyperspectral image, data dimensionality reduction can be achieved, thereby avoiding the trend of detection accuracy first increasing and then decreasing during hyperspectral anomaly detection; on the other hand, since band selection can be achieved based on the anomaly degree map and the anomaly reference map, the dependence on the detector in the process of band selection is avoided; on the other hand, since the hyperspectral image can be decomposed into background subspace and anomaly subspace, and the similarity between the anomaly degree map and the anomaly reference map of each band is calculated, it is possible to avoid the influence of background pixels and obtain more differentiated information in different bands, thereby improving the effectiveness of band selection. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the present invention or the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0019] Figure 1This is one of the flow charts of the hyperspectral data band selection method based on spatial spectrum abnormal distribution provided by the present invention; Figure 2 This is the second flow chart of the hyperspectral data band selection method based on spatial spectrum abnormal distribution provided by the present invention; Figure 3 It is a structural schematic diagram of a hyperspectral data band selection device based on abnormal spatial spectrum distribution provided by the present invention; Figure 4 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0020] In order to make the purpose, technical solutions and advantages of this application clearer, the technical solutions in this application will be clearly and completely described below in conjunction with the drawings in this application. Obviously, the described embodiments are part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0021] It should be noted that, in the embodiments of the present application, words such as "exemplary" or "for example" are used to indicate examples, illustrations or descriptions. Any embodiment or design described as "exemplary" or "for example" in the embodiments of the present application should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of words such as "exemplary" or "for example" is intended to present related concepts in a specific way.

[0022] It should be noted that, in this article, the terms "comprise", "include" or any other variant thereof are intended to cover non-exclusive inclusion, so that the process, method, article or device including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise one..." do not exclude the presence of other identical elements in the process, method, article or device including the element. In addition, it should be pointed out that the scope of the methods and devices in the embodiments of the present application is not limited to performing functions in the order shown or discussed, and may also include performing functions in a substantially simultaneous manner or in reverse order according to the functions involved, for example, the described method may be performed in an order different from that described, and various steps may also be added, omitted, or combined. In addition, the features described with reference to certain examples may be combined in other examples.

[0023] In order to facilitate a clear description of the technical solutions of the embodiments of the present application, in the embodiments of the present application, words such as "first" and "second" are used to distinguish between identical or similar items with substantially the same functions and effects. Those skilled in the art can understand that words such as "first" and "second" do not limit the quantity and order of execution.

[0024] The embodiments of the present application describe some exemplary embodiments for the purpose of explanation. It should be understood that the present application can be implemented in other ways that are not specifically shown in the drawings.

[0025] like Figure 1 As shown, the embodiment of the present application provides a hyperspectral data band selection method based on spatial spectrum abnormal distribution, and the hyperspectral data band selection method based on spatial spectrum abnormal distribution can be applied to a hyperspectral data band selection device based on spatial spectrum abnormal distribution. The hyperspectral data band selection method based on spatial spectrum abnormal distribution may include S101-S104: S101, a hyperspectral data band selection device based on spatial spectrum abnormal distribution obtains a hyperspectral image.

[0026] Alternatively, the hyperspectral image can be represented as ,in, Indicates the number of pixels, Indicates the number of bands.

[0027] S102, a hyperspectral data band selection device based on spatial spectrum abnormal distribution decomposes the hyperspectral image into a background subspace and an abnormal subspace, and determines a joint feature subspace of each band according to the background subspace and the abnormal subspace.

[0028] The abnormal subspace is a subspace mixed with noise and abnormality.

[0029] Specifically, Figure 2 As shown in the figure, the hyperspectral data band selection device based on the abnormal distribution of the spatial spectrum can use the low rank sparse decomposition model (LSDM) to transform the hyperspectral image Background subspace and the anomaly subspace mixed with noise and anomalies . The process can be expressed as: ; (1)

[0030] According to the assumption that the background of the hyperspectral image has low rank, the background subspace Can be viewed as two smaller matrices and The product of It can be expressed as: ; (2)

[0031] in, , are predefined parameters. and can be obtained from two multivariate Gaussian distributions: ; (3)

[0032] in, and Respectively and The identity matrix, parameters The prior distribution of can be expressed as: ; (4)

[0033] In order to adaptively separate the abnormal distribution and background distribution from the hyperspectral image, it is not enough to consider only the background subspace, but also the abnormal subspace. The features corresponding to the anomaly are reflected in the discriminability and correlation between the bands. Through the Hierarchical Dirichlet Process-based Gaussian Mixture Model (HDPGMM), the difference between the bands can be modeled with different Gaussian distribution forms. The difference between the bands is represented in the first level of the HDPGMM, and the correlation between adjacent bands can be represented by sharing certain Gaussian distributions in the second level of the HDPGMM.

[0034] Specifically, the abnormal subspace Line The column pixels follow the following Gaussian mixture distribution: ; (5)

[0035] in, Represents the total number of Gaussian mixture distributions, the above formula can be transformed into the following form: ; (6)

[0036] in, Is the length A vector of .

[0037] It follows a hierarchical Dirichlet distribution, namely: ; (7)

[0038] To better understand the hierarchical Dirichlet allocation ( ), here we introduce the knowledge of Dirichlet Process (DP). DP is a random process that can be represented by the concentration parameter and the basic distribution Definition, that is . is a measurable set that can be partitioned into random finite partitions If G obeys , then the following conditions must be met: ; (8)

[0039] DP has two basic properties: First, from Any sample drawn from Both are right The discrete distribution of .

[0040] Second, from the discrete distribution The sample drawn from is repeatable, i.e. ,in , .

[0041] According to these two properties, hyperspectral image pixels containing similar spectral characteristics will share the same , and pixels with different characteristics will be placed in different On, each of which can be regarded as a set of parameters consisting of the mean and variance of a Gaussian distribution. Therefore, the Gaussian Mixture Model (GMM) can be automatically clustered with the help of DP. In order to sample discrete distributions from DP , we can use the "stick breaking" construction method, which can be expressed as: ; (9)

[0042] in, and are two sets of infinitely many independent random variables. Representatives are located in Dirac function. It should be noted that can be regarded as a mixing coefficient, so the “stick breaking” construction can naturally describe the clustering properties of DP. At the same time, the parameter Adjustable The degree of discreteness. The smaller, The more discrete, the fewer the number of Gaussian distributions of GMM. Therefore, this method can adaptively adjust the number of Gaussian distributions of GMM.

[0043] In order to describe the abnormal correlation between the bands, two DPs can be used to extract from the first DP so that some of the same Gaussian distributions exist in different bands, characterizing the abnormal correlation between consecutive bands. This method based on the hierarchical Dirichlet process (HDP) can be expressed as follows: ; (10)

[0044] in, Indicates that it belongs to The distribution of band clusters , which follows DP. For the previous formula , can be derived through its conjugate prior: ; (11)

[0045] in, and are two predefined default parameters. To increase the flexibility of the proposed model, it can be assumed that Let's restate the model: ; (12)

[0046] To express it more concretely, we can introduce two coupled latent variables: and , combined with the "stick breaking" construction mentioned above, the model can be restated as follows: ; (13)

[0047] ; (14)

[0048] Among them, in formula (14) Compared with the formula (12) The following relationship exists: ; (15)

[0049] In order to further improve the adaptability of the model to adjusting the number of GMMs, we constrain formula (12) and Subject to the following conditions: ; (16)

[0050] It should be noted that are the hyperparameters of the prior distribution. They are set to small values ​​to make the prior distribution non-informative and thus unaffected by various hyperparameter changes. In addition, a truncation strategy is used to improve the solvability, that is, K, the number of the first and second layer GMMs are set to K and T respectively (T and K are large enough). Then, the The range can be adjusted to . It refers to the index of Gaussian components generated from the second layer GMM of band j. ( ) represents the index of the Gaussian component in the first layer GMM. Since the abnormal pixels of the abnormal target usually exist in one Gaussian distribution, and the noise (which can also be regarded as a kind of abnormal pixel) is depicted in the remaining Gaussian distribution, it is crucial to use the optimal Gaussian distribution as the abnormal subspace in each band to better represent the abnormality. That is, not only the abnormal subspace should be considered , it is also necessary to derive the corresponding .

[0051] Finally, combining the above formulas, the complete Bayesian model for characterizing abnormal distribution can be expressed as follows: ; (17)

[0052] In order to optimize the Bayesian problem, a variational Bayesian method is used to estimate the posterior distribution. Assume that the variable set Contains all variables in the full Bayesian model, and estimates the posterior distribution of the kth variable It can be expressed as follows: ; (18)

[0053] Among them, E(.) represents the mathematical expectation operator, Represents a variable set Medium All variables except . All variables can be inferred by the above method until the low-rank matrix Convergence or maximum number of iterations reached.

[0054] The corresponding inference stage can be described by the pseudo code in Table 1: Table 1

[0055] The closed-form solution of the specific inference formula is as follows: Update according to the following formula and : ; (19)

[0056] ; (20)

[0057] in,

[0058] ; (twenty one)

[0059] ; (twenty two)

[0060] ; (twenty three)

[0061] ; (twenty four)

[0062] about , , , and , according to the conjugate prior, their closed-form solutions can be expressed as follows: ; (25)

[0063] ; (26)

[0064] ; (27)

[0065] ; (28)

[0066] ; (29)

[0067] For variables , and , they can be formulated as follows: ; (30)

[0068] ; (31)

[0069] ; (32)

[0070] in, Indicates about The diagonal matrix of .

[0071] ; (33)

[0072] ; (34)

[0073] ; (35)

[0074] ; (36)

[0075] According to the above derivation, It can be obtained by recording the indices of the Gaussian distribution of different pixels. Therefore, the matrix Can be and The difference operation between them is In general, the first The abnormal subspace of the bands ( ) has a larger pixel value gap, which can be expressed as: ; (37)

[0076] in, represents the Hadamard product.

[0077] S103, the hyperspectral data band selection device based on spatial spectrum anomaly distribution obtains an anomaly reference map by calculating the local anomaly factor of each pixel in the anomaly subspace, and obtains an anomaly degree map of the corresponding band by calculating the local anomaly factor of each pixel in the joint feature subspace of each band.

[0078] Through the above-mentioned feature subspace method, the background distribution and the anomaly distribution can be adaptively separated, and the background and anomaly subspaces of each band can be obtained. With the assistance of these two subspaces, we can evaluate the background suppression and anomaly detection capabilities of different bands. However, since more background pixels are unfavorable for the HAD task, this application proposes the MILOF method, which explores the correlation of local sparse priors between bands while reducing the influence of background pixels.

[0079] like Figure 2 As shown, since the anomaly subspace takes into account the anomalies present in each band, the overall combination involving the overall band anomalies can be regarded as a better anomaly reference basis compared to the individual anomaly subspaces. In order to explore the anomaly reference map from the overall combination of anomaly subspaces, it is necessary to observe the definition of the HAD task in advance. The HAD task usually refers to highlighting pixels with significant spectral differences compared to the surrounding pixels. Local Outlier Factor (LOF), as a common anomaly detection method, has been successfully applied to color images, but is rare in hyperspectral images. It gives each pixel an anomaly score by comparing the difference between the pixel and the surrounding pixels, which is consistent with the definition of the HAD task. Therefore, LOF can give higher scores to abnormal pixels, thereby highlighting abnormal pixels.

[0080] Here are some conceptual definitions: represents the combination of the overall abnormal subspace. The k-th distance of pixel O can be defined as , represents the distance of the kth pixel closest to pixel O. Let pixel and The reachable distance between ,in represents the Euclidean distance between pixels P and O. Subsequently, the local reachability density of pixel P can be expressed as Indicates that is the set of k nearest neighbors of pixel P. According to the above definition, the LOF of pixel P can be expressed as follows: ; (38)

[0081] By calculation The LOF of each pixel in can be used to obtain the abnormal reference map and is defined as ,in .

[0082] Optionally, the hyperspectral data band selection device based on the spatial spectrum abnormal distribution can screen abnormal pixels according to the local abnormal factors of the pixels and the local sparsity of the abnormal targets to obtain the abnormal reference map.

[0083] Specifically, due to the large number of background pixels, the task is mainly affected by background pixels. In order to avoid this unreasonable situation, it is necessary to filter out abnormal pixels based on the anomaly score and the local sparsity of the abnormal target. Since pixels with higher scores tend to be more like abnormal pixels, Perform abnormal filtering and obtain , expressed as follows: ; (39)

[0084] in, Indicates greater than Center front % value. It should be noted that is a preset constant. Although most of the interference from background pixels has been eliminated by the above method, some interference still exists because the local sparsity of abnormal targets is not considered.

[0085] Connected component labeling can obtain abnormal regions of different sizes after anomaly filtering, which helps to measure the sparsity of abnormal regions. Therefore, we use the Two-Pass method to further eliminate the influence of interference pixels, which is a common method for obtaining connected regions. Specifically, is considered as the initial matrix According to the following formula, each non-zero point in the initial matrix is ​​traversed.

[0086] ; (40)

[0087] According to the above conditions, traverse the label matrix from left to right For each non-zero element in , and get the updated Note that the same connected region may have multiple labels. In order to make the same connected region have the same label, a second traversal is required. Specifically, the second traversal is to set the label of each pixel in each connected region to the minimum label in the current connected region. We define the obtained connected region set as It should be noted that Represents the first In order to reflect the sparse nature of abnormal targets, the pixels with fewer pixels can be retained. , and discard the pixels in the residual set, which is specifically expressed as follows: ; (41)

[0088] in, express The number of pixels in Represents a hyperparameter, which is only saved in certain The pixels in appear in the final abnormal reference image and can be defined as Therefore, it can be used as a criterion to measure the difference in local sparsity of abnormal targets in each band.

[0089] In order to obtain more differentiation in different bands, each band uses a combination of its own background and abnormal subspace. The combination of the i-th band can be expressed as , then, the LOF method is The anomaly degree map is obtained by repeating the application in . Subsequently, in order to delay the influence of the main background pixels and retain the different sparsity of the anomaly targets in each band, the anomaly degree map of each band is anomaly filtered, which is different from the process of generating anomaly reference maps using connected component markers. The anomaly degree map of the i-th band after anomaly filtering can be expressed as .

[0090] S104, the hyperspectral data band selection device based on the spatial spectrum anomaly distribution respectively calculates the similarity between the anomaly degree map of each band and the anomaly reference map, obtains the band scores of all bands, and selects the target band set according to the band scores of all bands.

[0091] Optionally, the hyperspectral data band selection device based on spatial-spectral anomaly distribution can use a mutual information method to measure the similarity between the anomaly degree map of each band and the anomaly reference map to obtain the band scores of all bands; then all bands are sorted in order from high to low according to the band scores, and a preset number of bands with the highest band scores are determined as the target band set.

[0092] Specifically, Figure 2 As shown in the figure, since mutual information (MI) can measure the correlation between variables, in order to consider the relationship between the local sparsity of abnormal targets in each band and assign priorities to different bands, the hyperspectral data band selection device based on spatial spectrum abnormal distribution can adopt the mutual information method to achieve this goal. That is, mutual information is used to measure the correlation between variables. and The difference between them is the degree of closeness between the anomaly of the ith band and the anomaly reference map that is closer to the truth. The corresponding formula is as follows: ; (42)

[0093] in, Indicates the number of different grayscale values ​​in the image. Since the anomaly reference image is a grayscale image, it is set to 255. Representing images The mid-gray value is equal to probability. Indicated in the image and images When the pixels in the image appear at the same position, The gray value is The image The gray value is probability.

[0094] Through the above process, the band score of each band can be obtained. After that, the hyperspectral data band selection device based on the abnormal distribution of the spatial spectrum can sort all the bands in the order of the band scores from high to low, and obtain ,Finally, the hyperspectral data band selection device based on ,abnormal spatial spectrum distribution can determine a preset number of bands with the highest band scores as the ,target band set.

[0095] In the embodiments of the present application, on the one hand, since the target band set can be determined from the hyperspectral image, data dimensionality reduction can be achieved, thereby avoiding the trend of detection accuracy first increasing and then decreasing during hyperspectral anomaly detection; on the other hand, since band selection can be achieved based on the anomaly degree map and the anomaly reference map, dependence on the detector in the process of band selection is avoided; on the other hand, since the hyperspectral image can be decomposed into a background subspace and an anomaly subspace, and the similarity between the anomaly degree map and the anomaly reference map of each band is calculated, it is possible not only to avoid the influence of background pixels, but also to obtain more differentiated information in different bands, thereby improving the effectiveness of band selection.

[0096] The above mainly introduces the solution provided by the embodiment of the present application from the perspective of the method. In order to realize the above functions, it includes hardware structures and / or software modules corresponding to the execution of each function. Those skilled in the art should easily realize that, in combination with the units and algorithm steps of each example described in the embodiment disclosed herein, the embodiment of the present application can be implemented in the form of hardware or a combination of hardware and computer software. Whether a function is executed in the form of hardware or computer software driving hardware depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present application.

[0097] It should be noted that the device in the embodiment of the present application includes a virtual device and a physical device. The virtual device can be a hyperspectral data band selection device based on spatial spectrum anomaly distribution, and the physical device can include an electronic device, a computer storage medium, and a computer program product.

[0098] The method for selecting a hyperspectral data band based on spatial-spectral anomaly distribution provided in the embodiment of the present application may be performed by a hyperspectral data band selection device based on spatial-spectral anomaly distribution, or a control module for selecting a hyperspectral data band based on spatial-spectral anomaly distribution in the hyperspectral data band selection device based on spatial-spectral anomaly distribution. In the embodiment of the present application, the method for selecting a hyperspectral data band based on spatial-spectral anomaly distribution is performed by a hyperspectral data band selection device based on spatial-spectral anomaly distribution as an example to illustrate the hyperspectral data band selection device based on spatial-spectral anomaly distribution provided in the embodiment of the present application.

[0099] It should be noted that the embodiment of the present application can divide the functional modules of the hyperspectral data band selection device based on the abnormal distribution of the spatial spectrum according to the above method example. For example, each functional module can be divided according to each function, or two or more functions can be integrated into one processing module. The above integrated module can be implemented in the form of hardware or in the form of software functional modules. Optionally, the division of modules in the embodiment of the present application is schematic and is only a logical functional division. There may be other division methods in actual implementation.

[0100] like Figure 3 As shown, the embodiment of the present application provides a hyperspectral data band selection device 300 based on spatial spectrum anomaly distribution. The hyperspectral data band selection device 300 based on spatial spectrum anomaly distribution includes: an acquisition module 301 and a processing module 302. The acquisition module 301 can be used to acquire a hyperspectral image; the processing module 302 is used to decompose the hyperspectral image into a background subspace and an anomaly subspace, and determine the joint feature subspace of each band according to the background subspace and the anomaly subspace, and the anomaly subspace is a subspace mixed with noise and anomaly; by calculating the local anomaly factor of each pixel in the anomaly subspace, an anomaly reference map is obtained, and by calculating the local anomaly factor of each pixel in the joint feature subspace of each band, an anomaly degree map of the corresponding band is obtained; the similarity between the anomaly degree map of each band and the anomaly reference map is calculated respectively, to obtain the band scores of all bands, and a target band set is selected according to the band scores of all bands.

[0101] Optionally, the processing module 302 is used to filter abnormal pixels according to the local abnormal factors of the pixels and the local sparsity of the abnormal targets to obtain the abnormal reference image.

[0102] Optionally, the processing module 302 is used to measure the similarity between the abnormality degree map of each band and the abnormal reference map by using a mutual information method to obtain band scores of all bands.

[0103] Optionally, the processing module 302 is configured to sort all the bands in descending order of the band scores, and determine a preset number of bands with the highest band scores as the target band set.

[0104] In the embodiments of the present application, on the one hand, since the target band set can be determined from the hyperspectral image, data dimensionality reduction can be achieved, thereby avoiding the trend of detection accuracy first increasing and then decreasing during hyperspectral anomaly detection; on the other hand, since band selection can be achieved based on the anomaly degree map and the anomaly reference map, dependence on the detector in the process of band selection is avoided; on the other hand, since the hyperspectral image can be decomposed into a background subspace and an anomaly subspace, and the similarity between the anomaly degree map and the anomaly reference map of each band is calculated, it is possible not only to avoid the influence of background pixels, but also to obtain more differentiated information in different bands, thereby improving the effectiveness of band selection.

[0105] Figure 4 An example of a physical structure diagram of an electronic device is shown in FIG. Figure 4 As shown, the electronic device may include: a processor (processor) 410 , a communication interface (Communications Interface) 420 , a memory (memory) 430 and a communication bus 440 , wherein the processor 410 , the communication interface 420 , and the memory 430 communicate with each other through the communication bus 440 . The processor 410 can call the logic instructions in the memory 430 to execute a hyperspectral data band selection method based on spatial-spectral anomaly distribution, the method comprising: acquiring a hyperspectral image; decomposing the hyperspectral image into a background subspace and an anomaly subspace, and determining a joint feature subspace of each band based on the background subspace and the anomaly subspace, wherein the anomaly subspace is a subspace mixed with noise and anomalies; obtaining an anomaly reference map by calculating a local anomaly factor of each pixel in the anomaly subspace, and obtaining an anomaly degree map of the corresponding band by calculating a local anomaly factor of each pixel in the joint feature subspace of each band; respectively calculating the similarity between the anomaly degree map of each band and the anomaly reference map to obtain band scores of all bands, and selecting a target band set based on the band scores of all bands.

[0106] In addition, the logic instructions in the above-mentioned memory 430 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0107] On the other hand, the present invention also provides a computer program product, which includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the hyperspectral data band selection method based on spatial-spectral anomaly distribution provided by the above methods, the method including: acquiring a hyperspectral image; decomposing the hyperspectral image into a background subspace and an anomaly subspace, and determining a joint feature subspace of each band based on the background subspace and the anomaly subspace, wherein the anomaly subspace is a subspace mixed with noise and anomalies; obtaining an anomaly reference map by calculating the local anomaly factor of each pixel in the anomaly subspace, and obtaining an anomaly degree map of the corresponding band by calculating the local anomaly factor of each pixel in the joint feature subspace of each band; respectively calculating the similarity between the anomaly degree map of each band and the anomaly reference map to obtain the band scores of all bands, and selecting a target band set based on the band scores of all bands.

[0108] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which is implemented when the computer program is executed by a processor to execute the hyperspectral data band selection method based on spatial-spectral anomaly distribution provided by the above-mentioned methods, the method comprising: acquiring a hyperspectral image; decomposing the hyperspectral image into a background subspace and an anomaly subspace, and determining a joint feature subspace of each band based on the background subspace and the anomaly subspace, the anomaly subspace being a subspace mixed with noise and anomalies; obtaining an anomaly reference map by calculating the local anomaly factor of each pixel in the anomaly subspace, and obtaining an anomaly degree map of the corresponding band by calculating the local anomaly factor of each pixel in the joint feature subspace of each band; respectively calculating the similarity between the anomaly degree map of each band and the anomaly reference map to obtain the band scores of all bands, and selecting a target band set based on the band scores of all bands.

[0109] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0110] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0111] 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 aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A hyperspectral data band selection method based on spatial spectrum abnormal distribution, characterized in that: include: Acquire hyperspectral images; Decomposing the hyperspectral image into a background subspace and an abnormal subspace, and determining a joint feature subspace of each band according to the background subspace and the abnormal subspace, wherein the abnormal subspace is a subspace mixed with noise and abnormality; By calculating the local anomaly factor of each pixel in the anomaly subspace, an anomaly reference map is obtained, and by calculating the local anomaly factor of each pixel in the joint feature subspace of each band, an anomaly degree map of the corresponding band is obtained; The similarity between the abnormality degree map of each band and the abnormal reference map is calculated respectively to obtain the band scores of all bands, and the target band set is selected according to the band scores of all bands.

2. The hyperspectral data band selection method based on spatial spectrum abnormal distribution according to claim 1 is characterized in that: After calculating the local anomaly factor of each pixel in the anomaly subspace, the method further includes: The abnormal pixels are screened according to the local abnormal factors of the pixels and the local sparsity of the abnormal targets to obtain the abnormal reference image.

3. The hyperspectral data band selection method based on spatial spectrum abnormal distribution according to claim 1 is characterized in that: The process of respectively calculating the similarity between the abnormality degree map of each band and the abnormal reference map to obtain the band scores of all bands includes: The mutual information method is used to measure the similarity between the abnormal degree map of each band and the abnormal reference map, and the band scores of all bands are obtained.

4. The method for selecting hyperspectral data bands based on spatial spectrum abnormal distribution according to claim 3, characterized in that: The step of selecting a target band set according to the band scores of all bands includes: All bands are sorted in descending order according to the band scores, and a preset number of bands with the highest band scores are determined as the target band set.

5. A hyperspectral data band selection device based on spatial spectrum abnormal distribution, characterized in that: include: Acquisition module and processing module; The acquisition module is used to acquire a hyperspectral image; The processing module is used to decompose the hyperspectral image into a background subspace and an abnormal subspace, and determine a joint feature subspace of each band according to the background subspace and the abnormal subspace, wherein the abnormal subspace is a subspace mixed with noise and abnormality; obtain an abnormal reference map by calculating a local abnormal factor of each pixel in the abnormal subspace, and obtain an abnormal degree map of the corresponding band by calculating a local abnormal factor of each pixel in the joint feature subspace of each band; respectively calculate the similarity between the abnormal degree map of each band and the abnormal reference map to obtain the band scores of all bands, and select a target band set according to the band scores of all bands.

6. The hyperspectral data band selection device based on spatial spectrum abnormal distribution according to claim 5, characterized in that: The processing module is used to screen abnormal pixels according to the local abnormal factors of the pixels and the local sparsity of the abnormal targets to obtain the abnormal reference image.

7. The hyperspectral data band selection device based on spatial spectrum abnormal distribution according to claim 5, characterized in that: The processing module is used to measure the similarity between the abnormality degree map of each band and the abnormal reference map by using the mutual information method to obtain the band scores of all bands.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the method for selecting a hyperspectral data band based on spatial spectrum anomaly distribution as described in any one of claims 1 to 4 is implemented.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for selecting a band of hyperspectral data based on spatial spectrum abnormal distribution as described in any one of claims 1 to 4 is implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the method for selecting a band of hyperspectral data based on spatial spectrum abnormal distribution as described in any one of claims 1 to 4 is implemented.

Images