CN118135205A - A hyperspectral image anomaly detection method - Google Patents

Abstract

The invention discloses a hyperspectral image anomaly detection method, which comprises the following steps: dividing a hyperspectral image matrix to be detected to obtain a fixed window area and a super-pixel area; converting the hyperspectral image matrix to be detected into a global two-dimensional image matrix; aiming at the global two-dimensional image matrix, the fixed window area and the super-pixel area, anchor point generation model processing and local Markov distance model processing are respectively used to obtain three abnormal detection results; and fusing the three abnormal detection results by considering logical OR operation and AND operation to obtain a final abnormal detection result. The accuracy of hyperspectral image anomaly detection is improved.

Description

A hyperspectral image anomaly detection method

Technical Field

The invention relates to a hyperspectral image anomaly detection method and belongs to the technical field of hyperspectral image processing.

Background technique

Hyperspectral images contain dozens or even hundreds of spectral bands and have high spectral resolution. They contain rich spectral feature information of ground objects, which can greatly improve the ability to accurately detect and identify ground object categories. In hyperspectral images, pixels with large spectral differences from the surrounding background are defined as anomalies.

The RX algorithm is one of the most typical hyperspectral anomaly detection methods. The RX algorithm models the image as a multivariate Gaussian distribution, estimates the mean and covariance matrix of the global image, and then uses the Mahalanobis distance between the pixel to be tested and the background as the anomaly decision criterion. However, the RX algorithm still has some shortcomings. The RX algorithm uses all pixels to calculate the background statistics (mean and covariance matrix), which often leads to contamination because abnormal pixels (which are very different from the background) are used in the calculation, so the background statistics are not completely accurate; the RX algorithm is based on the Gaussian distribution assumption, but in practical applications, many hyperspectral image data do not conform to the Gaussian distribution. In response to some shortcomings of the RX algorithm, many improvements have been made to the RX algorithm, such as the Local RX (LRX) algorithm and the Kernel RX (KRX) algorithm. LRX is the most typical window-based method. The LRX algorithm introduces the concept of local information, that is, only the data within a certain range around a pixel is considered when calculating the covariance matrix, without considering the entire image. The detection performance of LRX is usually better than that of RX because the background pixels in the local window tend to obey the Gaussian distribution. However, the LRX algorithm has poor processing capabilities for sparse data. When the hyperspectral data is sparse, the performance of the LRX algorithm may be affected.

Summary of the invention

Objective: In view of at least one of the above technical problems, the present invention provides a hyperspectral image anomaly detection method, which can improve the accuracy of hyperspectral image anomaly detection results.

Technical solution: To solve the above technical problems, the technical solution adopted by the present invention is:

In a first aspect, the present invention provides a method for detecting anomalies in a hyperspectral image, comprising:

Segment the hyperspectral image matrix to be detected to obtain the fixed window area and super-pixel area;

Converting the hyperspectral image matrix to be detected into a global two-dimensional image matrix;

For the global two-dimensional image matrix, fixed window area and superpixel area, the anchor point generation model and local Mahalanobis distance model are used to process respectively, and three types of anomaly detection results are obtained;

The anchor point generation model processing includes: calculating the Mahalanobis distance between each pixel point and the mean vector in the target area; sorting all pixels in ascending order according to the Mahalanobis distance between each pixel point and the mean vector to obtain the image matrix ; Loop to calculate the image matrix/> The Mahalanobis distance between each pixel and other pixels in the point set is obtained by filtering pixels whose Mahalanobis distance is less than the distance threshold. ; Filter out the point set whose number of pixels is greater than the number threshold, analyze the average value, and obtain the anchor point and anchor point set;

The processing using the local Mahalanobis distance model includes: calculating the Mahalanobis distance between the pixel point and the nearest anchor point to obtain the anomaly score of the pixel point, thereby obtaining an anomaly detection result;

The three anomaly detection results are merged by considering the logical “OR” and “AND” operations to obtain the final anomaly detection result.

In some embodiments, the method for obtaining a fixed window area includes: Divide into z/> Fixed window area of non-overlapping blocks of /> ; Where L, M, and B represent the length, width, and number of bands of the hyperspectral image, respectively./> ,/> Indicates the length and width of the fixed window area;

In some embodiments, a method for obtaining a region of a superpixel includes:

The hyperspectral image matrix is analyzed using principal component analysis (PCA). Dimensionality reduction is performed, where L, M, and B represent the length, width, and number of bands of the hyperspectral image, respectively;

The superpixel segmentation algorithm SLIC is used to segment the hyperspectral image matrix after dimensionality reduction into c superpixel regions. .

In some embodiments, converting the hyperspectral image matrix to be detected into a global two-dimensional image matrix includes:

The hyperspectral image matrix Convert to a global 2D image matrix , where L, M, and B represent the length, width, and number of bands of the hyperspectral image, respectively. The global two-dimensional image matrix is composed of K 1×B pixels, where K = L*M.

In some embodiments, calculating the Mahalanobis distance between each pixel point in the target area and the mean vector includes:

in, is the Mahalanobis distance between pixel point X and mean vector, μ is the mean vector, T represents the transpose of the matrix, is the covariance matrix, is the inverse matrix of the covariance matrix; M represents the width of the hyperspectral image; the expression of the mean vector is:

in, Represents the lth pixel, K=L*M.

In some embodiments, the image matrix is calculated The Mahalanobis distance between each pixel and other pixels in includes:

The calculation formula of the Mahalanobis distance between two pixels is:

in, is the Mahalanobis distance between pixel X and pixel Y; M represents the width of the hyperspectral image; is the covariance matrix, is the inverse of the covariance matrix.

In some embodiments, pixel points whose Mahalanobis distance is less than a distance threshold are selected to obtain a point set. ,include:

in, represents the g-th point set, Represents the image matrix The hth vector in , Represents the first vectors, is a non-negative number, and gamma is the distance threshold.

In some embodiments, a point set whose number of pixel points is greater than a number threshold is screened out for analysis and average value to obtain an anchor point _Sg and an anchor point set S, including:

in, Represents the number of elements in the g-th point set after filtering by the quantity threshold, represents the qth element in the gth point set after filtering by the quantity threshold, Indicates 1 to The sum of all the elements of

The set of all anchor points is called an anchor point set; the expression of the anchor point set S is:

in, Represents the average value of the p-th point set after screening, that is, the p-th anchor point.

In some embodiments, calculating the Mahalanobis distance between a pixel point and a nearest anchor point to obtain an abnormality score of the pixel point includes:

Among them, K is the total number of pixels, is the lth pixel The Mahalanobis distance to the nearest anchor point, is the lth pixel The Mahalanobis distance with the g-th anchor point S _g ; M represents the width of the hyperspectral image.

In some embodiments, the three anomaly detection results are merged by considering logical “OR” and “AND” operations to obtain a final anomaly detection result, including:

Among them, Final represents the final anomaly detection result, OR represents logical "or", AND represents logical "and", R ₁ , R ₂ , and R ₃ represent three types of anomaly detection results respectively. Represents the weight parameter.

In a second aspect, the present invention provides a hyperspectral image anomaly detection device, including a processor and a storage medium;

The storage medium is used to store instructions;

The processor is configured to operate according to the instructions to execute the method according to the first aspect.

In a third aspect, the present invention provides a device, comprising:

Memory;

processor;

as well as

Computer program;

The computer program is stored in the memory and is configured to be executed by the processor to implement the method described in the first aspect above.

In a fourth aspect, the present invention provides a storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the method described in the first aspect.

Beneficial effects: The hyperspectral image anomaly detection method provided by the present invention segments the original image, and divides the image matrix into two local areas: a fixed window area of non-overlapping blocks and an area of superpixels; an anchor point set is obtained through an anchor point generation algorithm model, and the spectral vectors of all pixels in the area are averaged first, and the Mahalanobis distance between each pixel point and the mean vector in the area is analyzed to obtain the Mahalanobis distance between each pixel point and the mean vector, and the pixels are sorted in ascending order according to the Mahalanobis distance between each pixel point and the mean vector to obtain an image matrix; according to the image matrix, the Mahalanobis distance between each pixel point in the area and other pixels is cyclically analyzed, and the Mahalanobis distance is analyzed whether the Mahalanobis distance Less than the distance threshold, obtain the point set, analyze whether the number of pixels in the point set is greater than the number threshold, if greater than the number threshold, analyze the average value of the point set to obtain the anchor point set; analyze the Mahalanobis distance between each pixel and the nearest anchor point as the corresponding abnormal score through the local Mahalanobis local model to determine the abnormality detection result; take the pixels in the global area, the pixels in the fixed window area pixel block and the pixels in the super pixel block of the super pixel area as input, and use the anchor point generation model and the local Mahalanobis distance model to obtain three abnormality detection results; consider the logical "or" and logical "and" operations of the three abnormality detection results, and then fuse them to obtain the final abnormality detection result. Therefore, this method improves the accuracy of abnormality detection in hyperspectral images through the steps of image segmentation, anchor point generation, local Mahalanobis distance analysis and logical fusion.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG1 is a schematic diagram of a method for detecting anomalies in a hyperspectral image according to an embodiment of the present invention.

FIG. 2 is an RGB pseudo-color image of an Urban city hyperspectral image according to an embodiment of the present invention.

FIG3 is a map showing the real location of objects in an Urban city hyperspectral image of a comparative example of the present invention.

FIG. 4 is a structural diagram of abnormal target detection of an Urban city hyperspectral image according to a comparative example of the present invention (Method 1).

FIG. 5 is a diagram showing abnormal target detection results of an Urban city hyperspectral image according to a comparative example of the present invention (Method 2).

FIG6 is a diagram showing the abnormal target detection structure of the Urban city hyperspectral image of the comparative example of the present invention (Method 3)

FIG. 7 is a structural diagram of abnormal target detection of an Urban city hyperspectral image according to a comparative example of the present invention (Method 4).

FIG8 is a structural diagram of abnormal target detection of an Urban city hyperspectral image according to a comparative example of the present invention (Method 5.

FIG9 is a structural diagram of abnormal target detection of an Urban city hyperspectral image according to a comparative example of the present invention (Method 6).

FIG. 10 is a diagram showing abnormal target detection results of an Urban city hyperspectral image according to an embodiment of the present invention.

Detailed ways

The present invention will be further described below in conjunction with the accompanying drawings and embodiments. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and cannot be used to limit the protection scope of the present invention.

In the description of the present invention, "several" means more than one, "many" means more than two, "greater than", "less than", "exceed", etc. are understood to exclude the number itself, and "above", "below", "within", etc. are understood to include the number itself. If there is a description of "first" or "second", it is only used for the purpose of distinguishing technical features, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features or implicitly indicating the order of the indicated technical features.

In the description of the present invention, the description with reference to the terms "one embodiment", "some embodiments", "illustrative embodiments", "examples", "specific examples", or "some examples" 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 invention. In this specification, the schematic representation of the above terms does 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.

Example 1

In a first aspect, as shown in FIG1 , this embodiment provides a method for detecting anomalies in a hyperspectral image, including:

S1. Segment the hyperspectral image matrix to be detected to obtain the fixed window area and the super-pixel area;

In some embodiments, the method for obtaining a fixed window area includes: Divide into z Fixed window area of non-overlapping blocks of ; Where L, M, and B represent the length, width, and number of bands of the hyperspectral image, respectively. , Indicates the length and width of the fixed window area;

In some embodiments, a method for obtaining a region of a superpixel includes:

The hyperspectral image matrix is analyzed using principal component analysis (PCA). Dimensionality reduction is performed, where L, M, and B represent the length, width, and number of bands of the hyperspectral image, respectively;

The superpixel segmentation algorithm SLIC is used to segment the hyperspectral image matrix after dimensionality reduction into c superpixel regions. .

S2, converting the hyperspectral image matrix to be detected into a global two-dimensional image matrix;

In order to facilitate parallel computing, the three-dimensional hyperspectral image matrix needs to be Convert to a global 2D image matrix.

In some embodiments, converting the hyperspectral image matrix to be detected into a global two-dimensional image matrix includes:

The hyperspectral image matrix Convert to a global 2D image matrix , where L, M, and B represent the length, width, and number of bands of the hyperspectral image, respectively. The global two-dimensional image matrix is composed of K 1×B pixels, where K = L*M.

S3, for the global two-dimensional image matrix, fixed window area and superpixel area, the anchor point generation model and the local Mahalanobis distance model are used to process respectively, and three types of anomaly detection results are obtained;

In this step, the pixel points in the global two-dimensional image matrix and the pixel blocks in the fixed window area are respectively The pixel points and superpixel regions in the superpixel block The pixel points in are taken as input, and three anomaly detection results are obtained by using the anchor point generation model and the local Mahalanobis distance model;

in, is the i-th window area in the fixed window area, is the jth pixel block in the superpixel region.

S31, wherein the anchor point generation model processing includes: calculating the Mahalanobis distance between each pixel point and the mean vector in the target area; sorting all the pixels in ascending order according to the Mahalanobis distance between each pixel point and the mean vector, and obtaining an image matrix ; Loop to calculate the image matrix The Mahalanobis distance between each pixel and other pixels in the point set is obtained by filtering pixels whose Mahalanobis distance is less than the distance threshold. ; Filter out the point set whose number of pixels is greater than the number threshold, analyze the average value, and obtain the anchor point and anchor point set;

Furthermore, in some embodiments, calculating the Mahalanobis distance between each pixel point in the target area and the mean vector includes:

in, is the Mahalanobis distance between pixel point X and mean vector, μ is the mean vector, T represents the transpose of the matrix, is the covariance matrix, is the inverse matrix of the covariance matrix; M represents the width of the hyperspectral image; the expression of the mean vector is:

in, Represents the lth pixel, K=L*M.

In some embodiments, the image matrix is calculated The Mahalanobis distance between each pixel and other pixels in includes:

The calculation formula of the Mahalanobis distance between two pixels is:

in, is the Mahalanobis distance between pixel X and pixel Y; M represents the width of the hyperspectral image; is the covariance matrix, is the inverse of the covariance matrix.

Furthermore, the calculation method of the covariance matrix includes:

(1) First, analyze the average value of all bands of the sample;

(2) Analyze the difference vectors between all bands of two pixels and their mean values respectively;

(3) Calculate the covariance of all bands of pixel point X and pixel point Y;

(4) Construct the covariance matrix.

The average value of the band is calculated as:

The covariance calculation formula is:

,

The formula of the covariance matrix is:

,

in, represents the i-th dimension of pixel X, represents the j-th dimension of pixel Y, express and The covariance between represents the expected value, represents the average value of all bands of X, Represents the average value of all bands of Y.

In some embodiments, pixel points whose Mahalanobis distance is less than a distance threshold are selected to obtain a point set. ,include:

in, represents the g-th point set, Represents the image matrix The hth vector in , Represents the first vectors, is a non-negative number, and gamma is the distance threshold.

In some embodiments, a point set whose number of pixel points is greater than a number threshold is screened out for analysis and average value to obtain an anchor point _Sg and an anchor point set S, including:

in, Represents the number of elements in the g-th point set after filtering by the quantity threshold, represents the qth element in the gth point set after filtering by the quantity threshold, Indicates 1 to The sum of all the elements of

The set of all anchor points is called an anchor point set; the expression of the anchor point set S is:

in, Represents the average value of the p-th point set after screening, that is, the p-th anchor point.

S32, wherein the processing using the local Mahalanobis distance model includes: calculating the Mahalanobis distance between the pixel point and the nearest anchor point to obtain an anomaly score of the pixel point, thereby obtaining an anomaly detection result;

In some embodiments, calculating the Mahalanobis distance between a pixel point and a nearest anchor point to obtain an abnormality score of the pixel point includes:

Among them, K is the total number of pixels, is the lth pixel The Mahalanobis distance to the nearest anchor point, is the lth pixel The Mahalanobis distance with the g-th anchor point S _g ; M represents the width of the hyperspectral image.

S4. The three anomaly detection results are merged by considering the logical “OR” and “AND” operations to obtain the final anomaly detection result.

In some embodiments, the three anomaly detection results are merged by considering logical “OR” and “AND” operations to obtain a final anomaly detection result, including:

Among them, Final represents the final anomaly detection result, OR represents logical "or", AND represents logical "and", R ₁ , R ₂ , and R ₃ represent three types of anomaly detection results respectively. Represents the weight parameter.

Specific application example: The airport hyperspectral image obtained by the airborne visible light/infrared imaging spectrometer (AVIRIS) is used, as shown in Figure 2, the Urban city hyperspectral image pseudo-color image, referred to as the Urban city hyperspectral image, the spatial resolution of the Urban city hyperspectral image is 17.2 meters/pixel, and it contains 204 bands for abnormal target detection. The experimental area data size of the Urban city image is 100×100. Figure 3 is the real location of the ground object in the Urban city hyperspectral image (the white light spot in Figure 3 is the abnormal target). The existing RX, LRX, CRD, LSMAD, FrFE, RGAE and the hyperspectral image anomaly detection method of the present application are used to detect abnormal targets on the Beach hyperspectral image of the embodiment, and the abnormal target detection accuracy is shown in Table 1:

Table 1 Comparison of hyperspectral abnormal target detection accuracy

Abnormal target detection method RX LRX CRD LSMAD FWf RGAE this invention Abnormal target detection accuracy/% 99.07 99.29 98.97 98.97 98.92 98.22 99.80

As can be seen from Table 1, the abnormal target detection accuracy of the hyperspectral image anomaly detection method of the present application is as high as 99.80%, which is significantly better than several other existing methods and 0.51% higher than the high LRX method with the second highest accuracy AUC score.

And from the clarity of the abnormal targets and the degree of separation from the background in Figures 4, 5, 6, 7, 8, 9 and 10, it can be seen that the abnormal target detection effect of the hyperspectral image anomaly detection method of the present application is significantly better than that of the other methods.

The above proves that the present invention can effectively improve the accuracy of abnormal target detection, and is feasible for abnormal target detection in hyperspectral images.

Example 2

In a second aspect, based on embodiment 1, this embodiment provides a hyperspectral image anomaly detection device, including a processor and a storage medium;

The storage medium is used to store instructions;

The processor is configured to operate according to the instructions to execute the method according to embodiment 1.

Example 3

In a third aspect, based on embodiment 1, this embodiment provides a device, including:

Memory;

processor;

as well as

Computer program;

The computer program is stored in the memory and is configured to be executed by the processor to implement the method described in Example 1.

Example 4

In a fourth aspect, based on Embodiment 1, this embodiment provides a storage medium on which a computer program is stored, and when the computer program is executed by a processor, the method described in Embodiment 1 is implemented.

Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

The present application is described with reference to the flowcharts and/or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and/or box in the flowchart and/or block diagram, as well as the combination of the processes and/or boxes in the flowchart and/or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate a device for implementing the functions specified in one process or multiple processes in the flowchart and/or one box or multiple boxes in the block diagram.

These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce a manufactured product including an instruction device that implements the functions specified in one or more processes in the flowchart and/or one or more boxes in the block diagram.

These computer program instructions may also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, whereby the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one or more processes in the flowchart and/or one or more boxes in the block diagram.

The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims (10)

1. A hyperspectral image anomaly detection method, the method comprising:

Dividing a hyperspectral image matrix to be detected to obtain a fixed window area and a super-pixel area;

Converting the hyperspectral image matrix to be detected into a global two-dimensional image matrix;

Aiming at the global two-dimensional image matrix, the fixed window area and the super-pixel area, anchor point generation model processing and local Markov distance model processing are respectively used to obtain three abnormal detection results;

wherein generating the model process using the anchor points comprises: calculating the mahalanobis distance between each pixel point in the target area and the mean value vector; ascending order sorting is carried out on all the pixel points according to the mahalanobis distance between each pixel point and the mean value vector, and an image matrix is obtained ; Circularly computing the image matrix/>The Markov distance between each pixel point and other pixel points in the array is selected, and the pixel points with the Markov distance smaller than a distance threshold value are screened to obtain a point set/>; Screening out point sets with the number of pixel points being greater than a number threshold value, and analyzing an average value to obtain anchor points and anchor point sets;

Wherein the processing using the local mahalanobis distance model includes: calculating the mahalanobis distance between the pixel point and the nearest anchor point to obtain the abnormal score of the pixel point, thereby obtaining an abnormal detection result;

And fusing the three abnormal detection results by considering logical OR operation and AND operation to obtain a final abnormal detection result.

2. The method of claim 1, wherein the method of acquiring the fixed window area comprises: matrix hyperspectral imageDivided into z/>Fixed window area/>, of non-overlapping blocks of; Wherein L, M, B represents the length, width and band number of the hyperspectral image, respectively,/>，/>Representing the length and width of the fixed window area;

and/or, the method for acquiring the region of the super pixel comprises the following steps:

Matrix of hyperspectral images using principal component analysis PCA Performing dimension reduction, wherein L, M, B respectively represents the length, width and wave band quantity of the hyperspectral image;

dividing the hyperspectral image matrix after dimension reduction into areas of c superpixels by utilizing a superpixel division algorithm SLIC 。

3. The method according to claim 1, wherein converting the hyperspectral image matrix to be detected into a global two-dimensional image matrix comprises:

Matrix hyperspectral image Conversion to a Global two-dimensional image matrix/>Wherein L, M, B respectively represent the length, width and band number of the hyperspectral image, and the global two-dimensional image matrix is composed of K1×B pixel points, wherein K=L×M.

4. The method of claim 1, wherein calculating the mahalanobis distance of each pixel point in the target area from the mean vector comprises:

；

Wherein, The Marshall distance between the pixel point X and the mean vector, mu is the mean vector, T represents the transposition of the matrix,/>Is covariance matrix,/>Is the inverse of the covariance matrix; m represents the width of the hyperspectral image; the expression of the mean vector is:

；

Wherein, Represents the i-th pixel, k=l×m.

5. The method of claim 1, wherein the image matrix is calculatedThe mahalanobis distance between each pixel point and other pixel points comprises:

the calculation formula of the mahalanobis distance between two pixel points is as follows:

；

Wherein, Is the mahalanobis distance between the pixel point X and the pixel point Y; m represents the width of the hyperspectral image; /(I)Is covariance matrix,/>Is the inverse of the covariance matrix.

6. The method of claim 1, wherein the set of points is obtained by filtering pixels whose mahalanobis distance is less than a distance thresholdComprising:

；

Wherein, Represents the g-th set of points,/>Representing the image matrix/>The h-th vector of >/>Representing the/>, in the image matrixVectors,/>Being non-negative, gamma is the distance threshold.

7. The method of claim 1, wherein screening out a set of points with a number of pixels greater than a number threshold for analysis averages to obtain an anchor point S _g and an anchor point set S comprises:

；

Wherein, Representing the number of elements of the g-th point set after the screening of the quantity threshold value,/>Represents the q-th element in the g-th point set after passing the quantity threshold screening,/>Represent 1 to/>Is the sum of all elements of (a);

The set of all anchors is called an anchor set; the expression of the anchor point set S is:

；

Wherein, Represents the average value of the p-th point set after screening, namely the p-th anchor point.

8. The method of claim 1, wherein calculating the mahalanobis distance of a pixel point from the nearest anchor point to obtain an outlier score for the pixel point comprises:

；

wherein K is the total number of pixel points, For the first pixel/>Mahalanobis distance from the nearest anchor point,/>For the first pixel/>The mahalanobis distance from the g-th anchor point S _g; m represents the width of the hyperspectral image.

9. The method of claim 1, wherein fusing the three anomaly detection results with respect to a logical or and operation to obtain a final anomaly detection result comprises:

；

；

；

Wherein Final represents the Final abnormality detection result, OR represents the logical OR, AND represents the logical AND, R ₁、R₂、R₃ represents the three abnormality detection results, respectively, Representing the weight parameter.

10. A storage medium having stored thereon a computer program which, when executed by a processor, implements the method of any of claims 1 to 9.