Hyperspectral abnormal target detection method and device based on three-dimensional Markov model

Abstract

The invention discloses a hyperspectral abnormal target detection method and device based on a three-dimensional Markov model, and belongs to the field of hyperspectral data processing. The method comprises the following steps: dividing all wavebands of a hyperspectral image subjected to background Gaussian processing according to a preset standard to obtain a plurality of waveband subintervals; markov vector extraction is carried out on each wave band sub-interval, and extraction results of all the wave band sub-intervals are combined to obtain a feature vector of the hyperspectral image; and performing abnormal target detection on the feature vector by using an RX algorithm to obtain an abnormal detection result. The method can effectively solve the problems of low target detection precision and slow calculation speed caused by large redundant information between hyperspectral image bands.

Description

Technical Field

[0001] The present invention relates to the technical field of hyperspectral data processing, and particularly relates to a hyperspectral anomaly target detection method and device based on a three-dimensional Markov model. Background Art

[0002] Hyperspectral data has a very high spectral resolution and can effectively detect the differences in the spectral characteristics of ground objects. However, the improvement of spectral resolution leads to information redundancy. In the field of remote sensing image processing, the Markov random field (Gauss-Markov random field, Markov) model is often used in fields such as surface morphology classification and texture analysis. In recent years, this model has gradually been introduced into anomaly detection algorithms.

[0003] In related technologies, when the RX detection algorithm is directly used for hyperspectral image processing, the covariance matrix of samples needs to be calculated, and the dimension of the covariance matrix will increase rapidly with the increase in the number of bands, which will bring huge computational complexity and lead to a high false alarm probability in the detection results.

[0004] Based on this, there is an urgent need for a hyperspectral anomaly target detection method and device based on a three-dimensional Markov model to solve the above technical problems. Summary of the Invention

[0005] The present invention provides a hyperspectral anomaly target detection method and device based on a three-dimensional Markov model, which can solve the problem of low anomaly target detection accuracy in related technologies. The technical solutions are as follows:

[0006] On the one hand, a hyperspectral anomaly target detection method based on a three-dimensional Markov model is provided. The method includes:

[0007] Dividing all bands of the hyperspectral image after background Gaussianization processing according to a preset standard to obtain a plurality of waveband subintervals;

[0008] Extracting Markov vectors for each of the waveband subintervals, and combining the extraction results of all the waveband subintervals to obtain a feature vector of the hyperspectral image;

[0009] Using the RX algorithm to perform anomaly target detection on the feature vector to obtain an anomaly detection result.

[0010] On the other hand, a hyperspectral anomaly target detection device based on a three-dimensional Markov model is provided. The device includes:

[0011] A dividing module, configured to divide all bands of the hyperspectral image after background Gaussianization processing according to a preset standard to obtain a plurality of waveband subintervals;

[0012] An extraction module, configured to perform Markov vector extraction on each of the waveband sub-intervals, and combine the extraction results of all the waveband sub-intervals to obtain a feature vector of the hyperspectral image;

[0013] A detection module, configured to perform abnormal target detection on the feature vector by using the RX algorithm to obtain an abnormal detection result.

[0014] On the other hand, a computer device is provided, which includes a memory and a processor. The memory is used to store a computer program, and the processor is used to execute the computer program stored on the memory to implement the steps of the above-mentioned hyperspectral abnormal target detection method based on a three-dimensional Markov model.

[0015] On the other hand, a computer-readable storage medium is provided. The storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the above-mentioned hyperspectral abnormal target detection method based on a three-dimensional Markov model are implemented.

[0016] On the other hand, a computer program product is provided, which includes a computer program. When the computer program is executed by a processor, the steps of the above-mentioned hyperspectral abnormal target detection method based on a three-dimensional Markov model are implemented.

[0017] The technical solution provided by the present invention can at least bring the following beneficial effects: By performing background Gaussianization processing on the original hyperspectral image; then dividing all wavebands into multiple sub-intervals, and then performing Markov vector extraction for each sub-interval, and forming new hyperspectral data by combining the Markov vectors of all sub-spaces; finally, using the RX algorithm to perform abnormal target detection on the new hyperspectral data to obtain an abnormal detection result. This method can effectively solve the problems of low target detection accuracy and slow calculation speed caused by large redundant information between wavebands in hyperspectral images. Description of the Drawings

[0018] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0019] Figure 1 It is a flowchart of a hyperspectral abnormal target detection method based on a three-dimensional Markov model provided by an embodiment of the present invention;

[0020] Figure 2 It is a schematic diagram of a sub-unit of a three-dimensional Markov model provided by an embodiment of the present invention;

[0021] Figure 3 It is a schematic diagram of a first-order Markov model provided by an embodiment of the present invention;

[0022] Figure 4 It is the expansion process of a vector provided by an embodiment of the present invention;

[0023] Figure 5 It is the experimental area of the AVIRIS hyperspectral image provided by an embodiment of the present invention;

[0024] Figure 6 It is the result graph of a hyperspectral anomaly target detection algorithm based on a three-dimensional Markov model provided by an embodiment of the present invention;

[0025] Figure 7 It is the result graph of directly using the RX anomaly detection algorithm provided by an embodiment of the present invention;

[0026] Figure 8 It is the structural diagram of a hyperspectral anomaly target detection device based on a three-dimensional Markov model provided by an embodiment of the present invention;

[0027] Figure 9 It is the hardware architecture diagram of a computer device provided by an embodiment of the present invention. Detailed implementation manners

[0028] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0029] As mentioned above, directly using the RX algorithm to process hyperspectral images will result in a high false alarm probability. When directly used for hyperspectral image processing, it is necessary to calculate the covariance matrix of samples, and the dimension of the covariance matrix will increase rapidly with the increase in the number of bands, thus bringing a huge amount of calculation.

[0030] Based on this, the concept of the present invention is to use a Markov model to describe the spectral characteristics of hyperspectral images to obtain new hyperspectral data, and then perform RX anomaly detection. Compared with the method of directly performing RX anomaly detection, it can improve the accuracy of anomaly target detection.

[0031] The following describes the specific implementation manners of the above concept.

[0032] Please refer to Figure 1, a hyperspectral anomaly target detection method based on a three-dimensional Markov model provided by an embodiment of the present invention, the method comprising:

[0033] Step 100, dividing all bands of a hyperspectral image subjected to background Gaussianization processing according to a preset standard to obtain a plurality of waveband subintervals;

[0034] Step 102, extracting Markov vectors for each of the waveband subintervals, and combining the extraction results of all the waveband subintervals to obtain a feature vector of the hyperspectral image;

[0035] Step 104, using the RX algorithm to perform anomaly target detection on the feature vector to obtain an anomaly detection result.

[0036] In the embodiment of the present invention, the original hyperspectral image is subjected to background Gaussianization processing; then all bands are divided into a plurality of subintervals, and then Markov vector extraction is performed for each subinterval, and the Markov vectors of all subspaces are combined into new hyperspectral data; finally, the RX algorithm is used to perform anomaly target detection on the new hyperspectral data to obtain an anomaly detection result. This method can effectively solve the problems of low target detection accuracy and slow calculation speed caused by large redundant information between bands of hyperspectral images.

[0037] The following describes Figure 1 the execution manners of the respective steps shown.

[0038] First, for step 100, all bands of a hyperspectral image subjected to background Gaussianization processing are divided according to a preset standard to obtain a plurality of waveband subintervals.

[0039] Hyperspectral data is data with a three-dimensional structure. In terms of the spatial structure, the pixel values of ground objects have obvious Markov properties; extended to the spectral dimension, when there is no strong mutation in the spectral information, it also has obvious Markov properties. In fact, due to the influence of atmospheric effects and spectral aliasing effects, the spectral response of remote sensors does not yet have an obvious spectral line structure. Therefore, it is feasible to use the Markov model to describe the spectral characteristics of hyperspectral images. Figure 2 is a schematic diagram of a subunit in a three-dimensional Markov model.

[0040] Among them, N k is determined by the number of bands of the hyperspectral data, and N i and N j are determined by the order of the Markov model. The embodiment of the present invention adopts a first-order Markov model. Figure 3 is a schematic diagram of a first-order three-dimensional Markov model.

[0041] For step 102, Markov vector extraction is performed on each of the sub-bands, and the extraction results of all the sub-bands are combined to obtain the feature vector of the hyperspectral image.

[0042] In the embodiment of the present invention, the feature vector is determined through the following steps: according to the number of processed image bands and the order of the Markov model, a three-dimensional Markov field model for characterizing the relationship between the central pixel point of a sub-band and its surrounding pixel points is established; the three-dimensional Markov field model is expanded to obtain a relationship vector for characterizing the relationship between the central pixel points of all sub-bands and their surrounding pixel points; texture features composed of Markov model parameters are calculated according to the relationship vector; and the texture features are combined to obtain a feature vector for characterizing the texture features of the hyperspectral image.

[0043] Specifically, a point in the three-dimensional Markov field is set as where 1 ≤ i ≤ N i , 1 ≤ j ≤ N j , 1 ≤ k ≤ N k m represents the index value of the Markov window, that is, the index value in the entire scene. Then the three-dimensional Markov field model can be represented by the following formula:

[0044]

[0045] where, β h , β v and β s represent the weight factors in the horizontal, vertical, and spectral directions in sequence; is Gaussian noise with a mean of 0.

[0046] The above formula is an expression of a pixel point, which describes the relationship between the value of the central point and its surrounding points. In order to describe the relationship between the values of all points and their surrounding points, the data needs to be expanded into a vector form, that is, the relationship vector:

[0047] W m = [w m (1) T , w m (2) T ... w m (N k ) T T

[0048] In the formula, the superscript T represents transpose.

[0049] The entire expansion process is as Figure 4 shown and can be expressed as:

[0050] AW m = ε m

[0051] where ε m is also Gaussian noise.

[0052]

[0053] That is:

[0054]

[0055] where denotes the Kronecker matrix product, is the identity matrix, is a Toeplitz matrix, which has values of 1 only at positions on both sides of the diagonal and values of 0 at other positions. The subscript indicates the size of the matrix. If N i = N j = N k = 2, then B, C, and D in the above equation are:

[0056]

[0057] β h , β v and β s The values of these parameters can be obtained using maximum likelihood estimation:

[0058]

[0059] where χ h , χ ν , χ s are all maximum likelihood estimates:

[0060]

[0061]

[0062] α = (N k (N j - 1) / N j (N k - 1)), ξ = 0.5 - δ

[0063] Typically, δ is a very small value, with a typical value of 0.01, used to ensure that the estimated values of β h , β v and β s satisfy their constraint conditions.

[0064] The data structure of hyperspectral images can be well described by using a three-dimensional Markov model to achieve the division of the surface morphology of hyperspectral images. The texture feature vector composed of the three-dimensional Markov model parameters obtained by solving the above formula fully characterizes the texture characteristics of the hyperspectral image. The texture feature is as follows:

[0065] β = (β h , β ν , β s ) T

[0066] After extracting and determining the texture feature vector of a sub-interval, the extraction results of all waveband sub-intervals are combined to obtain the total feature vector of the hyperspectral image.

[0067] For step 104, the RX algorithm is used to detect abnormal targets for the said feature vector to obtain the abnormal detection result.

[0068] The feature vector determined through the above steps can be used for surface morphology classification; or the RX algorithm can be used for abnormal target detection. The specific detection processes are all well-known to those skilled in the art and will not be elaborated here.

[0069] The following uses an embodiment to prove the feasibility of the above method.

[0070] The data selected in the embodiment of the present invention is the 224-band AVIRIS hyperspectral image obtained at the Naval Air Station in San Diego, California, USA. The wavelength range is 0.37 - 2.51um, the spatial resolution is 3.5m, and the size is 400 * 400 pixels. After removing the water vapor absorption bands and low signal-to-noise ratio bands, 189 bands are retained. Due to the large amount of data, in order to improve the experimental efficiency, a region with a size of 100 * 100 pixels is intercepted, and this region contains a large number of aircraft targets to be extracted, such as Figure 5 shown.

[0071] After processing using the above method and then using the RX algorithm for abnormal target detection, the abnormal detection result is obtained, as shown in Figure 6 shown. At the same time, directly using the RX algorithm for abnormal target detection on the original hyperspectral image, the obtained abnormal detection result is as shown in Figure 7 shown.

[0072] Comparing Figure 6 and Figure 7 , it can be seen that the effect of abnormal target detection by the method proposed by the present invention has been greatly improved.

[0073] Please refer to Figure 8 , the embodiment of the present invention provides a hyperspectral abnormal target detection device based on a three-dimensional Markov model. The device includes:

[0074] A partitioning module 800, configured to partition all bands of a hyperspectral image that has undergone background Gaussianization processing according to a preset criterion, to obtain a plurality of sub-band intervals;

[0075] An extraction module 802, configured to perform Markov vector extraction on each of the sub-band intervals, and combine the extraction results of all the sub-band intervals to obtain a feature vector of the hyperspectral image;

[0076] A detection module 804, configured to perform anomaly target detection on the feature vector by using the RX algorithm to obtain an anomaly detection result.

[0077] In an embodiment of the present invention, when the extraction module 802 performs Markov vector extraction on each of the sub-band intervals, and combines the extraction results of all the sub-band intervals to obtain a feature vector of the hyperspectral image, it is specifically configured to perform the following operations: establish a three-dimensional Markov field model for characterizing the relationship between the central pixel point of a sub-band interval and its surrounding pixel points according to the number of processed image bands and the order of the Markov model; expand the three-dimensional Markov field model to obtain a relationship vector for characterizing the relationship between the central pixel points of all the sub-band intervals and their surrounding pixel points; calculate texture features composed of Markov model parameters according to the relationship vector; and combine the texture features to obtain a feature vector for characterizing the texture features of the hyperspectral image.

[0078] In an embodiment of the present invention, the three-dimensional Markov field model is established by the following formula:

[0079]

[0080] wherein, is a pixel point of the Markov field; m is a Markov window index value; β h , β v and β s respectively represent weight factors in the horizontal, vertical, and spectral directions; is Gaussian noise with a mean of 0.

[0081] In an embodiment of the present invention, the relationship vector is established by the following formula:

[0082] W m =[w m (1) T , w m (2) T ... w m (N k ) T T

[0083] wherein, the superscript T represents transpose.​

[0084] In the embodiments of the present invention, the texture feature β composed of Markov model parameters is calculated by the following formula:

[0085]

[0086] β = (β h , β ν , β s ) T

[0087] where χ h , χ ν , χ s are all maximum likelihood estimates; ξ = 0.5 - δ, and δ typically has a value of 0.01.

[0088] In the embodiments of the present invention, the maximum likelihood estimate is calculated by the following formula:

[0089]

[0090] where N k is the number of bands of the hyperspectral data, and N i and N j are the orders of the Markov model.

[0091] It should be noted that: the hyperspectral anomaly target detection device based on the three-dimensional Markov model provided in the above embodiments is only illustrated by the division of the above functional modules. In practical applications, the above functions can be allocated to different functional modules according to needs, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. In addition, the hyperspectral anomaly target detection device based on the three-dimensional Markov model provided in the above embodiments and the embodiments of the hyperspectral anomaly target detection method based on the three-dimensional Markov model belong to the same concept. The specific implementation process is detailed in the method embodiments and will not be elaborated here.

[0092] An embodiment of the present application also provides a computer device. Please refer to Figure 9 , which includes a processor and a memory. The memory stores at least one instruction, at least one program, a code set, or an instruction set. The at least one instruction, at least one program, a code set, or an instruction set is loaded and executed by the processor to implement the hyperspectral anomaly target detection method based on the three-dimensional Markov model provided in the above method embodiments.

[0093] An embodiment of the present application further provides a computer-readable storage medium, on which at least one instruction, at least one program segment, a code set or an instruction set is stored. The at least one instruction, at least one program segment, the code set or the instruction set is loaded and executed by a processor to implement the hyperspectral anomaly target detection method based on a three-dimensional Markov model provided in the above method embodiments.

[0094] An embodiment of the present application further provides a computer program product, which includes a computer program. The processor of the computer device reads the computer program from the computer-readable storage medium, and the processor executes the computer program, so that the computer device executes the hyperspectral anomaly target detection method based on a three-dimensional Markov model described in any one of the above embodiments.

[0095] For the convenience of description, when describing the above system or device, various modules or units are described separately according to their functions. Of course, when implementing the present application, the functions of each unit can be realized in one or more software and / or hardware.

[0096] From the description of the above embodiments, those skilled in the art can clearly understand that the present application can be implemented by means of software plus a necessary general hardware platform. Based on such an understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. The computer software product can be stored in a storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in various embodiments or some parts of the embodiments of the present application.

[0097] Finally, it should also be noted that in this article, relational terms such as first, second, third, and fourth are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or also includes elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "including a..." does not exclude the presence of additional identical elements in the process, method, article or device including the element.

[0098] The above are only the preferred embodiments of the present application. It should be noted that for those of ordinary skill in the art, without departing from the principle of the present application, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of the present application.

Claims

1. A method for detecting abnormal hyperspectral targets based on a three-dimensional Markov model, characterized in that: The method comprises: Divide all bands of the hyperspectral image after background Gaussianization according to a preset standard to obtain multiple band sub-intervals; Performing Markov vector extraction on each of the band sub-intervals, and combining the extraction results of all the band sub-intervals to obtain a feature vector of the hyperspectral image; The RX algorithm is used to perform abnormal target detection on the feature vector to obtain an abnormal detection result.

2. The method according to claim 1, characterized in that The Markov vector extraction is performed on each of the band sub-intervals, and the extraction results of all the band sub-intervals are combined to obtain the feature vector of the hyperspectral image, including: According to the number of processed image bands and the order of the Markov model, a three-dimensional Markov field model is established to characterize the relationship between the central pixel point of a band sub-interval and its surrounding pixels; Expanding the three-dimensional Markov field model to obtain a relationship vector for representing a central pixel point and surrounding pixels of all band subintervals; Calculate the texture feature composed of Markov model parameters according to the relationship vector; The texture features are combined to obtain a feature vector for characterizing the texture features of the hyperspectral image.

3. The method according to claim 2, characterized in that The three-dimensional Markov field model is established by the following formula: In the formula, is the pixel point of the Markov field; m is the index value of the Markov window; β h , β v and β s Represent the weight factors in the horizontal, vertical and spectral directions respectively; is Gaussian noise with a mean of 0.

4. The method according to claim 2, characterized in that The relationship vector is established by the following formula: W m =[w m (1) T ,w m (2) T ...w m (N k ) T ] T In the formula, the superscript T represents transposition.

5. The method according to claim 3, characterized in that The texture feature β composed of Markov model parameters is calculated by the following formula: β＝(β h ,β ν ,β s ) T Among them, χ h , χ ν , χ s All are maximum likelihood estimates; ξ=0.5-δ, δ is typically 0.

01.

6. The method according to claim 5, characterized in that The maximum likelihood estimate is calculated by the following formula: Among them, N k is the number of bands of hyperspectral data, N i and N j is the Markov model order.

7. A hyperspectral abnormal target detection device based on a three-dimensional Markov model, characterized in that: The device comprises: A division module is used to divide all bands of the hyperspectral image after background Gaussianization processing according to a preset standard to obtain multiple band sub-intervals; An extraction module, used for performing Markov vector extraction on each of the band sub-intervals, and combining the extraction results of all the band sub-intervals to obtain a feature vector of the hyperspectral image; The detection module is used to perform abnormal target detection on the feature vector using the RX algorithm to obtain an abnormal detection result.

8. A computer device, characterized in that: The computer device includes a memory and a processor, the memory is used to store a computer program, and the processor is used to execute the computer program stored in the memory to implement the steps of any one of the methods described in claims 1-6.

9. A computer-readable storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the method described in any one of claims 1 to 6 are implemented.

10. A computer program product, characterized in that The method comprises a computer program, wherein when the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images