Multispectral image segmentation method for abnormal discrimination driven projection fuzzy clustering

Abstract

The invention belongs to the technical field of image segmentation and machine learning. The invention provides a multispectral image segmentation method for abnormal discrimination driving projection fuzzy clustering. According to the embodiment of the invention, the multispectral image segmentation model comprises an abnormal discrimination projection fuzzy clustering item, an entropy regularization item and a principal component analysis item. The anomaly discrimination projection fuzzy clustering item allocates a corresponding weight coefficient for each pixel, so that the influence of the abnormal pixels on the image segmentation precision is reduced; the entropy regularization item is used as a supplement of the fuzzy clustering item, so that the problem that trivial solutions are very easy to appear in the fuzzy clustering solving process is solved, and the solving precision is further improved; the principal component analysis item obtains the global structure of the data, reduces the influence of the noise in the image on the algorithm precision, and solves the problem that a fuzzy clustering method is sensitive to noise points. The three are optimized together, so that the image segmentation precision of the clustering algorithm is effectively improved, and more accurate perception and interpretation of the multispectral remote sensing image are realized.

Description

Technical Field

[0001] The disclosed embodiments relate to the technical field of image segmentation and machine learning, and in particular to a multispectral image segmentation method driven by abnormal discrimination projection fuzzy clustering. Background Art

[0002] Clustering algorithm is an unsupervised learning method that aims to group samples in a given data set so that samples in the same group are highly similar, while samples in different groups are less similar. Clustering algorithm does not require pre-labeled data for model training. Instead, it discovers potential patterns and relationships by analyzing the internal structure and features of the data, saving a lot of economic and human costs. The algorithm itself has strong interpretability and is widely used in many fields, such as remote sensing image processing, social network analysis, medical diagnosis, recommendation systems, etc.

[0003] Multispectral image segmentation is a key technology in remote sensing image processing. It aims to divide a multispectral image into multiple regions so that the pixels in the same region are similar, while the features between different regions are significantly different. This is consistent with the usage scenarios of clustering algorithms to a certain extent. Therefore, clustering algorithms have been widely used in the field of multispectral image segmentation and have achieved excellent performance.

[0004] Fuzzy clustering achieves image segmentation by minimizing the objective function for multispectral remote sensing image segmentation, obtaining the membership value of each pixel sample to the cluster center. The algorithm is simple in principle, highly interpretable and easy to implement, and has excellent performance in characterizing the attribution of pixel samples. However, the fuzzy clustering algorithm has significant defects when segmenting multispectral images, that is, it is very sensitive to noise values. In response to the above problems, Wu Jiaxin et al. proposed an improved fuzzy clustering method combining subspace and KL information measurement, using local fuzzy factors, eliminating noise interference through similarity measurement and adaptive constraint parameters, and extracting local spatial information of the image. Although the influence of noise on fuzzy clustering segmentation images is alleviated to a certain extent, this method only focuses on the local area of ​​the image and does not accurately characterize the global image properties. In addition, this method gives the same weight to all samples in the data set, ignoring the influence of abnormal data on accuracy.

[0005] Therefore, it is necessary to improve one or more problems existing in the above-mentioned related technical solutions.

[0006] It should be noted that this section is intended to provide background or context for the technical solutions of the present disclosure stated in the claims. The description herein is not admitted to be prior art by virtue of being included in this section. Summary of the invention

[0007] The purpose of the embodiments of the present disclosure is to provide a multispectral image segmentation method driven by abnormal discrimination projection fuzzy clustering, thereby overcoming one or more problems caused by the limitations and defects of related technologies at least to a certain extent.

[0008] According to an embodiment of the present disclosure, a multispectral image segmentation method for abnormal discrimination driven projection fuzzy clustering is provided, the method comprising: Anomaly discrimination mechanism based on isolation forest algorithm, constructing anomaly discrimination matrix according to multispectral images; Based on the abnormal discrimination matrix, a multispectral image segmentation model is constructed; The multispectral image segmentation model is solved to obtain a segmentation result.

[0009] Furthermore, based on the abnormality discrimination mechanism of the isolation forest algorithm, the step of constructing an abnormality discrimination matrix according to the multispectral image includes: Stretching and flattening the multispectral image along the spatial dimension and reconstructing it into a second-order matrix; Constructing a distance matrix according to the second-order matrix, and using the distance matrix as input of the isolation forest algorithm to obtain an abnormality score corresponding to each pixel; The abnormality discrimination matrix is ​​constructed according to the abnormality scores.

[0010] Furthermore, the expression of the abnormal discrimination matrix is:

[0011] in, is the abnormal discrimination value corresponding to each pixel; is the anomaly score corresponding to each pixel.

[0012] Furthermore, the expression of the multispectral image segmentation model is:

[0013] in, is the abnormal discrimination value of each pixel, is the membership matrix, is the cluster center matrix, is the projection matrix, is the pixel in the original data space, is the fuzzy cluster center in the projection subspace, represents the entropy regularization term, is the principal component analysis item, is the balance coefficient between the projection clustering term and the entropy regularization term, is the membership degree of each data sample to the fuzzy cluster center in the projection space, is the identity matrix.

[0014] Furthermore, the step of solving the multispectral image segmentation model to obtain a segmentation result includes: Fixing the membership matrix, updating the projection matrix and the cluster center matrix to obtain an optimal solution of the projection matrix and an optimal solution of the cluster center matrix; The optimal solution of the projection matrix and the optimal solution of the cluster center matrix are fixed to obtain the optimal membership matrix; wherein the category to which the maximum value in the membership vector corresponding to each of the multispectral images belongs represents the category to which the multispectral image belongs.

[0015] Furthermore, if the membership matrix is ​​fixed, the multispectral image segmentation model is converted to:

[0016] in, is the fuzzy cluster center in the projection space, which is the same as in the above formula. .

[0017] Furthermore, by fixing the optimal solution of the projection matrix and the optimal solution of the cluster center matrix, the multispectral image segmentation model is converted to:

[0018] in, n is the number of data samples, c is the number of fuzzy cluster centers.

[0019] The technical solution provided by the embodiments of the present disclosure may have the following beneficial effects: In the embodiment of the present disclosure, the multispectral image segmentation method driven by the above-mentioned abnormal discrimination projection fuzzy clustering, on the one hand, designs a weight coefficient matrix based on the isolation forest method, calculates the abnormal score of each pixel, assigns corresponding weights to realize abnormal discrimination, and solves the problem that abnormal points reduce the image segmentation accuracy. At the same time, the global principal component analysis regularization term is introduced to realize global dimension reduction and feature extraction, eliminate noise interference, and improve the segmentation accuracy of fuzzy clustering. On the other hand, the multispectral image segmentation model includes an abnormal discrimination projection fuzzy clustering term, an entropy regularization term and a principal component analysis term. The abnormal discrimination projection fuzzy clustering term assigns a corresponding weight coefficient to each pixel, reducing the influence of abnormal pixels on the image segmentation accuracy; the entropy regularization term, as a supplement to the fuzzy clustering term, solves the problem that trivial solutions are very likely to appear in the fuzzy clustering solution process, and further improves the solution accuracy; the principal component analysis term obtains the global structure of the data, and reduces the influence of noise in the image on the algorithm accuracy, solving the problem that the fuzzy clustering method is sensitive to noise points. The three are optimized together, effectively improving the accuracy of the clustering algorithm to segment the image, and realizing more accurate perception and interpretation of multispectral remote sensing images. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The accompanying drawings herein are incorporated into the specification and constitute a part of the specification, illustrate embodiments consistent with the present disclosure, and together with the specification are used to explain the principles of the present disclosure. Obviously, the accompanying drawings described below are only some embodiments of the present disclosure, and for ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without creative work.

[0021] Figure 1 A step diagram showing a multispectral image segmentation method for abnormal discrimination driven projected fuzzy clustering in an exemplary embodiment of the present disclosure; Figure 2 A specific flow chart of the multispectral image segmentation method of abnormality discrimination driven projected fuzzy clustering in an exemplary embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0022] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided so that the disclosure will be more comprehensive and complete and to fully convey the concepts of the example embodiments to those skilled in the art. The described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

[0023] In addition, the accompanying drawings are only schematic illustrations of the embodiments of the present disclosure and are not necessarily drawn to scale. The same reference numerals in the figures represent the same or similar parts, and their repeated descriptions will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically independent entities.

[0024] This example embodiment provides a multispectral image segmentation method using anomaly discrimination driven projection fuzzy clustering. Figure 1 As shown in , the multispectral image segmentation method of abnormal discrimination driven projection fuzzy clustering may include: step S101 to step S103.

[0025] Step S101: constructing an abnormality discrimination matrix based on the multispectral image based on the abnormality discrimination mechanism of the isolation forest algorithm; Step S102: constructing a multispectral image segmentation model based on the abnormality discrimination matrix; Step S103: solving the multispectral image segmentation model to obtain a segmentation result.

[0026] Through the above-mentioned multispectral image segmentation method driven by abnormal discrimination projection fuzzy clustering, on the one hand, a weight coefficient matrix is ​​designed based on the isolation forest method, the abnormal score of each pixel is calculated, and the corresponding weight is assigned to realize abnormal discrimination, so as to solve the problem that abnormal points reduce the image segmentation accuracy. At the same time, the global principal component analysis regularization term is introduced to realize global dimension reduction and feature extraction, eliminate noise interference, and improve the segmentation accuracy of fuzzy clustering. On the other hand, the multispectral image segmentation model includes abnormal discrimination projection fuzzy clustering term, entropy regularization term and principal component analysis term respectively. The abnormal discrimination projection fuzzy clustering term assigns a corresponding weight coefficient to each pixel, reducing the influence of abnormal pixels on the image segmentation accuracy; the entropy regularization term, as a supplement to the fuzzy clustering term, solves the problem that trivial solutions are very likely to appear in the fuzzy clustering solution process, and further improves the solution accuracy; the principal component analysis term obtains the global structure of the data, and reduces the influence of noise in the image on the algorithm accuracy, solving the problem that the fuzzy clustering method is sensitive to noise points. The three are optimized together, which effectively improves the accuracy of image segmentation by the clustering algorithm and realizes more accurate perception and interpretation of multispectral remote sensing images.

[0027] Next, we will refer to Figure 1 to Figure 2 Each step of the above-mentioned multispectral image segmentation method of abnormality discrimination driven projected fuzzy clustering in this example implementation is described in more detail.

[0028] In step S101, the traditional isolation forest algorithm uses the original data as input, randomly selects features and divides the left and right subtrees, and calculates the path length from the root node of each sample point when it is segmented to the end based on the idea that "abnormal points" are easier to isolate than "normal points", thereby determining the abnormal score of each sample. In view of the problem that the insufficient feature utilization rate when the original data is input leads to insufficient discrimination of the abnormal score, the present invention proposes to use the distance matrix between data as input to determine the abnormal score of each original data sample.

[0029] For multispectral images ,in Represent the height, width and number of bands of the image respectively. Stretch and flatten it along the spatial dimension and reconstruct it into a second-order tensor ,in Indicates the number of pixels in each band. .in Indicates that each pixel has Further, construct the data The distance matrix ,in Represents pixel and pixel The Euclidean distance between As the input of the isolation forest algorithm, each pixel is obtained The corresponding anomaly score The higher the anomaly score, the higher the probability that the sample is an outlier, and it should be given a lower weight in the cluster segmentation model. Therefore, the anomaly discrimination matrix is ​​constructed. : (1) in, , is the abnormal discrimination value corresponding to each pixel.

[0030] In step S102 and step S103, based on the obtained abnormal discrimination matrix, the following model is established: (2) in, represents the abnormal discrimination value of each pixel, represents the projection matrix, represents the pixel in the original data space, represents the fuzzy cluster center in the projected subspace, represents the entropy regularization term, which is used to suppress the influence of trivial solutions on clustering accuracy. As the principal component analysis item, it can realize the extraction of global structural information while reducing the interference and influence of noise on image segmentation.

[0031] By solving model (2), we obtain the final membership matrix , each pixel is assigned a membership value corresponding to a different cluster center, where the category with the highest membership value is the category to which the pixel belongs, and the segmentation of the multispectral image is achieved accordingly.

[0032] In a specific embodiment, Figure 2 As shown, it is a specific flow chart of the multispectral image segmentation method driven by anomaly discrimination projected fuzzy clustering.

[0033] Anomaly discrimination mechanism based on isolation forest algorithm: For multispectral images ,in Represent the height, width and number of bands of the image respectively. Stretch and flatten it along the spatial dimension and reconstruct it into a second-order tensor ,in Indicates the number of pixels in each band. .in Indicates that each pixel has Further, construct the data The distance matrix ,in Represents pixel and pixel The Euclidean distance between As the input of the isolation forest algorithm, each pixel is obtained The corresponding anomaly score The higher the anomaly score, the higher the probability that the sample is an outlier, and it should be given a lower weight in the cluster segmentation model. Therefore, the anomaly discrimination matrix is ​​constructed. : (1) in, , is the abnormal discrimination value corresponding to each pixel.

[0034] Multispectral image segmentation model driven by anomaly discrimination projected fuzzy clustering: According to the abnormal discrimination model, after obtaining the abnormal discrimination value corresponding to each pixel, the following model is established: (2) Among them, the first term is the projection fuzzy clustering term weighted by the abnormal discriminant value, and the projection matrix The original data Projecting to a subspace to obtain more discriminative features, represents the cluster center in the subspace. The second term is the entropy regularization term, which is used to avoid trivial solutions. The third term is the principal component analysis term, which enables the model to effectively avoid noise interference and extract the global structural characteristics of the data.

[0035] The variables that need to be updated and optimized in the above model include the projection matrix , membership matrix and the cluster center matrix .

[0036] 1. Fixed membership matrix , update the projection matrix and the cluster center matrix : Model (2) can be transformed into: (3) make , model (3) can be transformed into: (4) Since model (4) does not include variables Make any constraints, so you can Taking the derivative and setting the derivative function equal to 0 gives: (5) Therefore, we can get: (6) remember: (7) Right now ,matrix represents the cluster center of the original data. Based on this, the first term of model (2) can be expressed as: (8) in: (9) remember: (10) Then the model (7) is transformed into: (11) That is, the matrix The optimal solution is the matrix Before the corresponding The eigenvectors with the smallest eigenvalues.

[0037] 2. Fixed projection matrix and the cluster center matrix , update the membership matrix : Model (2) is transformed into: (12) make , then formula (12) can be written as: (13) Considering that formula (13) for each data sample are independent of each other, so they can be simplified to: (14) The augmented Lagrangian function is established as follows: (15) For variables and variables Taking the derivative and setting the derivative function equal to 0 gives: (16) According to formula (16), we have: (17) Also because ,therefore: (18) so: (19) We can get: (20) At this point, after satisfying the convergence accuracy or reaching the maximum number of iterations, the final membership matrix is ​​obtained , where the category to which the largest value in the membership vector corresponding to each data sample belongs represents the category to which the sample belongs.

[0038] Through the above-mentioned multispectral image segmentation method driven by abnormal discrimination projection fuzzy clustering, on the one hand, a weight coefficient matrix is ​​designed based on the isolation forest method, the abnormal score of each pixel is calculated, and the corresponding weight is assigned to realize abnormal discrimination, so as to solve the problem that abnormal points reduce the image segmentation accuracy. At the same time, the global principal component analysis regularization term is introduced to realize global dimension reduction and feature extraction, eliminate noise interference, and improve the segmentation accuracy of fuzzy clustering. On the other hand, the multispectral image segmentation model includes abnormal discrimination projection fuzzy clustering term, entropy regularization term and principal component analysis term respectively. The abnormal discrimination projection fuzzy clustering term assigns a corresponding weight coefficient to each pixel, reducing the influence of abnormal pixels on the image segmentation accuracy; the entropy regularization term, as a supplement to the fuzzy clustering term, solves the problem that trivial solutions are very likely to appear in the fuzzy clustering solution process, and further improves the solution accuracy; the principal component analysis term obtains the global structure of the data, and reduces the influence of noise in the image on the algorithm accuracy, solving the problem that the fuzzy clustering method is sensitive to noise points. The three are optimized together, which effectively improves the accuracy of image segmentation by the clustering algorithm and realizes more accurate perception and interpretation of multispectral remote sensing images.

[0039] It should be understood that the terms "center", "longitudinal", "lateral", "length", "width", "thickness", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", "clockwise", "counterclockwise", etc. in the above description indicate orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the embodiments of the present disclosure and simplifying the description, and do not indicate or imply that the referred device or element must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be understood as a limitation on the embodiments of the present disclosure.

[0040] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the features. In the description of the embodiments of the present disclosure, the meaning of "multiple" is two or more, unless otherwise clearly and specifically defined.

[0041] In the embodiments of the present disclosure, unless otherwise clearly specified and limited, the terms "installed", "connected", "connected", "fixed" and the like should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium, it can be the internal connection of two elements or the interaction relationship between two elements. For ordinary technicians in this field, the specific meanings of the above terms in the present disclosure can be understood according to specific circumstances.

[0042] In the embodiments of the present disclosure, unless otherwise clearly specified and limited, a first feature being "above" or "below" a second feature may include the first and second features being in direct contact, or may include the first and second features not being in direct contact but being in contact through another feature between them. Moreover, a first feature being "above", "above" and "above" a second feature includes the first feature being directly above and obliquely above the second feature, or simply indicates that the first feature is higher in level than the second feature. A first feature being "below", "below" and "below" a second feature includes the first feature being directly below and obliquely below the second feature, or simply indicates that the first feature is lower in level than the second feature.

[0043] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example" or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present disclosure. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification.

[0044] Those skilled in the art will readily appreciate other embodiments of the present disclosure after considering the specification and practicing the invention disclosed herein. This application is intended to cover any modification, use or adaptation of the present disclosure, which follows the general principles of the present disclosure and includes common knowledge or customary techniques in the art that are not disclosed in the present disclosure. The specification and examples are intended to be exemplary only, and the true scope and spirit of the present disclosure are indicated by the appended claims.

Claims

1. A multispectral image segmentation method driven by abnormal discrimination projected fuzzy clustering, characterized in that: The method includes: Anomaly discrimination mechanism based on isolation forest algorithm, constructing anomaly discrimination matrix according to multispectral images; Based on the abnormal discrimination matrix, a multispectral image segmentation model is constructed; The multispectral image segmentation model is solved to obtain a segmentation result.

2. According to the multispectral image segmentation method of abnormal discrimination driven projection fuzzy clustering in claim 1, it is characterized in that: The abnormality discrimination mechanism based on the isolation forest algorithm and the steps of constructing the abnormality discrimination matrix according to the multispectral image include: Stretching and flattening the multispectral image along the spatial dimension and reconstructing it into a second-order matrix; Constructing a distance matrix according to the second-order matrix, and using the distance matrix as input of the isolation forest algorithm to obtain an abnormality score corresponding to each pixel; The abnormality discrimination matrix is ​​constructed according to the abnormality scores.

3. According to the multispectral image segmentation method of abnormal discrimination driven projection fuzzy clustering according to claim 2, it is characterized in that: The expression of the abnormal discrimination matrix is: in, is the abnormal discrimination value corresponding to each pixel; is the anomaly score corresponding to each pixel.

4. According to the multispectral image segmentation method of abnormal discrimination driven projection fuzzy clustering according to claim 3, it is characterized in that: The expression of the multispectral image segmentation model is: in, is the abnormal discrimination value of each pixel, is the membership matrix, is the cluster center matrix, is the projection matrix, is the pixel in the original data space, is the fuzzy cluster center in the projection subspace, represents the entropy regularization term, is the principal component analysis item, is the balance coefficient between the projection clustering term and the entropy regularization term, is the membership degree of each data sample to the fuzzy cluster center in the projection space, is the identity matrix.

5. According to the multispectral image segmentation method of abnormal discrimination driven projection fuzzy clustering according to claim 4, it is characterized in that: The step of solving the multispectral image segmentation model to obtain a segmentation result includes: Fixing the membership matrix, updating the projection matrix and the cluster center matrix to obtain an optimal solution of the projection matrix and an optimal solution of the cluster center matrix; The optimal solution of the projection matrix and the optimal solution of the cluster center matrix are fixed to obtain the optimal membership matrix; wherein the category to which the maximum value in the membership vector corresponding to each of the multispectral images belongs represents the category to which the multispectral image belongs.

6. The multispectral image segmentation method based on abnormal discrimination driven projection fuzzy clustering according to claim 5 is characterized in that: Fixing the membership matrix, the multispectral image segmentation model is converted to: in, is the fuzzy cluster center in the projection space, which is the same as in the above formula. .

7. The multispectral image segmentation method based on abnormal discrimination driven projection fuzzy clustering according to claim 6 is characterized in that: Fixing the optimal solution of the projection matrix and the optimal solution of the cluster center matrix, the multispectral image segmentation model is converted to: in, n is the number of data samples, c is the number of fuzzy cluster centers.

Images