A method, device, equipment and medium for removing mixed noise from hyperspectral images

Abstract

The present application discloses a method, device, equipment and medium for removing mixed noise from a hyperspectral image, which relates to the field of image processing technology and is used to solve the problem that the existing technology cannot effectively remove mixed noise in a hyperspectral image. The method comprises: constructing a mathematical model of the hyperspectral source image according to the image data of the hyperspectral source image; wherein the image data comprises four parts: a target image, sparse noise, Gaussian noise and linear noise; constructing a denoising model of the hyperspectral source image according to the nuclear norm low-rank constraint regularization term of the target image, the spatial gradient sparse constraint regularization term of the linear noise, the low-rank constraint term of the linear noise and the mathematical model; optimizing the denoising model according to a plurality of preset auxiliary variables to obtain an optimized denoising model; solving the optimized denoising model to obtain a denoised hyperspectral image to effectively remove mixed noise in the hyperspectral image.

Description

Technical Field

[0001] The present application relates to the technical field of image processing, and provides a method, device, equipment and medium for removing mixed noise from a hyperspectral image. Background Art

[0002] As we all know, hyperspectral images contain rich spatial and spectral information. However, due to photon effects, random errors in light counting, calibration errors and other reasons, hyperspectral images are always corrupted by different noises, thus hindering the subsequent tasks.

[0003] However, the following limitations are still unavoidable in the existing hyperspectral image denoising methods: (1) When the intensity of the stripe noise is strong enough to break the sparse condition, or when the stripe noise has a low-rank structure due to periodic patterns across spatial or spectral dimensions, it is difficult to separate the stripe noise from the hyperspectral image; (2) Since the low-rank matrix or tensor modeling in some existing methods inevitably destroys the internal high-order structural information or the correlation between different modes, it is easy to cause loss of image details and spectral distortion; (3) Since only some simple features of hyperspectral images and several types of noise are used, while the key information of simultaneously utilizing the hyperspectral image and each type of noise is ignored, it is easy to cause insufficient noise removal and loss of valuable signals.

[0004] Therefore, how to effectively remove mixed noise in hyperspectral images has become a problem that needs to be solved urgently. Summary of the invention

[0005] The present application provides a method, device, equipment and medium for removing mixed noise from a hyperspectral image, which are used to solve the problem that the prior art cannot effectively remove mixed noise in a hyperspectral image.

[0006] On the one hand, a method for removing mixed noise from a hyperspectral image is provided, the method comprising:

[0007] According to the image data of the hyperspectral source image, a mathematical model of the hyperspectral source image is constructed; wherein the image data includes four parts: target image, sparse noise, Gaussian noise and linear noise;

[0008] Constructing a denoising model for the hyperspectral source image according to the nuclear norm low-rank constraint regularization term of the target image, the spatial dimension gradient sparse constraint regularization term of the linear noise, the low-rank constraint term of the linear noise and the mathematical model;

[0009] Optimizing the denoising model according to a plurality of preset auxiliary variables to obtain an optimized denoising model;

[0010] The optimized denoising model is solved to obtain a denoised hyperspectral image.

[0011] Optionally, the nuclear norm low-rank constraint regularization term of the target image includes the following formula:

[0012]

[0013] Among them, X represents the target image, G represents the kernel tensor, A1, A2 and A3 represent the basis of the factor matrix or subspace, and all three are orthogonal matrices , represents the basis of the i-th factor matrix or subspace, express The transpose of represents the identity matrix, 0 represents a constant, and , represents the matrix nuclear norm, Indicates A nuclear tension quantum problem.

[0014] Optionally, the spatial dimension gradient sparse constraint regularization term of the linear noise includes the following formula:

[0015]

[0016] Where S represents linear noise, represents the first regularization term of linear noise, represents the L1 norm, and Represent the gradient operators along the x-axis and y-axis respectively.

[0017] Optionally, the low-rank constraint term of the linear noise includes the following formula:

[0018]

[0019] in, represents the second regularization term of linear noise, represents a constant, and , Indicates linear noise edge The expansion matrix of the module.

[0020] Optionally, the denoising model of the hyperspectral source image includes the following formula:

[0021]

[0022] Where Y represents the hyperspectral source image, B represents sparse noise, and N represents Gaussian noise. represents the parameters that balance the regularization terms, represents a constant, and , Represents the 2-Frobenius norm of a tensor.

[0023] Optionally, the optimized denoising model includes the following formula:

[0024]

[0025] Among them, T, P and Q represent different preset auxiliary variables respectively.

[0026] Optionally, the step of solving the optimized denoising model to obtain a denoised hyperspectral image includes:

[0027] The optimized denoising model is solved by using the alternating direction multiplier method ADMM to obtain a denoised hyperspectral image.

[0028] On the one hand, a device for removing mixed noise from a hyperspectral image is provided, the device comprising:

[0029] A mathematical model building unit is used to build a mathematical model of the hyperspectral source image according to the image data of the hyperspectral source image; wherein the image data includes four parts: target image, sparse noise, Gaussian noise and linear noise;

[0030] A denoising model construction unit, used to construct a denoising model of the hyperspectral source image according to the nuclear norm low-rank constraint regularization term of the target image, the spatial dimension gradient sparse constraint regularization term of the linear noise, the low-rank constraint term of the linear noise and the mathematical model;

[0031] A denoising model optimization unit, used to optimize the denoising model according to a plurality of preset auxiliary variables to obtain an optimized denoising model;

[0032] The hyperspectral image denoising unit is used to solve the optimized denoising model to obtain a denoised hyperspectral image.

[0033] On the one hand, an electronic device is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein any one of the above methods is implemented when the processor executes the computer program.

[0034] In one aspect, a storage medium is provided, on which computer program instructions are stored, and when the computer program instructions are executed by a processor, any of the above methods is implemented.

[0035] Compared with the prior art, the beneficial effects of this application are:

[0036] In an embodiment of the present application, when removing mixed noise from a hyperspectral image, first, a mathematical model of the hyperspectral source image can be constructed based on the image data of the hyperspectral source image; wherein the image data includes four parts: a target image, sparse noise, Gaussian noise, and linear noise; then, a denoising model of the hyperspectral source image can be constructed based on the nuclear norm low-rank constraint regularization term of the target image, the spatial dimension gradient sparse constraint regularization term of the linear noise, the low-rank constraint term of the linear noise, and the mathematical model; next, the denoising model can be optimized based on a plurality of preset auxiliary variables to obtain an optimized denoising model; finally, the optimized denoising model can be solved to obtain a denoised hyperspectral image. Therefore, in the embodiment of the present application, since the denoising model of the hyperspectral source image is constructed by "the nuclear norm low-rank constraint regularization term of the target image, the spatial dimension gradient sparse constraint regularization term of the linear noise, and the low-rank constraint term of the linear noise", it is not only possible to effectively remove the mixed noise in the hyperspectral source image, but also to restore the texture information of the target image, effectively reduce the spectral distortion and improve the practicability of the denoising method. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the related technologies, the drawings required for use in the embodiments or the related technical descriptions are briefly introduced below. Obviously, the drawings described below are only the embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0038] Figure 1 A schematic diagram of an application scenario provided for an embodiment of the present application;

[0039] Figure 2 A schematic diagram of a process for removing mixed noise from a hyperspectral image provided in an embodiment of the present application;

[0040] Figure 3 A schematic diagram of image data of a hyperspectral source image provided in an embodiment of the present application;

[0041] Figure 4 A schematic diagram of the denoising result of the measured urban data set 141-143 band provided in an embodiment of the present application;

[0042] Figure 5 A schematic diagram of the denoising results of the measured AVIRIS data set 154-156 bands provided in an embodiment of the present application;

[0043] Figure 6 A schematic diagram of the denoising results of the measured GF-5 data set 327-329 band provided in an embodiment of the present application;

[0044] Figure 7 A schematic diagram of a device for removing mixed noise from a hyperspectral image provided in an embodiment of the present application.

[0045] Markings in the figure: 10 - hyperspectral image mixed noise removal device, 101 - processor, 102 - memory, 103 - I / O interface, 104 - database, 70 - hyperspectral image mixed noise removal device, 701 - mathematical model construction unit, 702 - denoising model construction unit, 703 - denoising model optimization unit, 704 - hyperspectral image denoising unit. DETAILED DESCRIPTION

[0046] In order to make the purpose, technical scheme and advantages of the present application clearer, the technical scheme in the embodiment of the present application will be clearly and completely described below in conjunction with the drawings in the embodiment of the present application. Obviously, the described embodiment is only a part of the embodiment of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of the present application. In the absence of conflict, the embodiments in the present application and the features in the embodiments can be combined with each other arbitrarily. In addition, although the logical order is shown in the flow chart, in some cases, the steps shown or described can be performed in an order different from that here.

[0047] As we all know, hyperspectral images contain rich spatial and spectral information. However, due to photon effects, random errors in light counting, calibration errors and other reasons, hyperspectral images are always corrupted by different noises, thus hindering the subsequent tasks.

[0048] However, the following limitations are still unavoidable in the existing hyperspectral image denoising methods: (1) When the intensity of the stripe noise is strong enough to break the sparse condition, or when the stripe noise has a low-rank structure due to periodic patterns across spatial or spectral dimensions, it is difficult to separate the stripe noise from the hyperspectral image; (2) Since the low-rank matrix or tensor modeling in some existing methods inevitably destroys the internal high-order structural information or the correlation between different modes, it is easy to cause loss of image details and spectral distortion; (3) Since only some simple features of hyperspectral images and several types of noise are used, while the key information of simultaneously utilizing the hyperspectral image and each type of noise is ignored, it is easy to cause insufficient noise removal and loss of valuable signals.

[0049] Based on this, an embodiment of the present application provides a method for removing mixed noise from a hyperspectral image. In the method, first, a mathematical model of a hyperspectral source image can be constructed based on the image data of the hyperspectral source image; wherein the image data includes four parts: a target image, sparse noise, Gaussian noise, and linear noise; then, a denoising model of the hyperspectral source image can be constructed based on the nuclear norm low-rank constraint regularization term of the target image, the spatial dimension gradient sparse constraint regularization term of the linear noise, the low-rank constraint term of the linear noise, and the mathematical model; next, the denoising model can be optimized according to a plurality of preset auxiliary variables to obtain an optimized denoising model; finally, the optimized denoising model can be solved to obtain a denoised hyperspectral image. Therefore, in the embodiment of the present application, since the denoising model of the hyperspectral source image is constructed by "the nuclear norm low-rank constraint regularization term of the target image, the spatial dimension gradient sparse constraint regularization term of the linear noise, and the low-rank constraint term of the linear noise", it is not only possible to effectively remove the mixed noise in the hyperspectral source image, but also to restore the texture information of the target image, effectively reduce the spectral distortion and improve the practicability of the denoising method.

[0050] After introducing the design ideas of the embodiments of the present application, the following briefly introduces the application scenarios to which the technical solutions of the embodiments of the present application can be applied. It should be noted that the application scenarios introduced below are only used to illustrate the embodiments of the present application and are not limited. In the specific implementation process, the technical solutions provided by the embodiments of the present application can be flexibly applied according to actual needs.

[0051] like Figure 1 FIG. 1 is a schematic diagram of an application scenario provided by an embodiment of the present application. The application scenario may include a device 10 for removing mixed noise from a hyperspectral image.

[0052] The device 10 for removing mixed noise from a hyperspectral image can be used to remove mixed noise from a hyperspectral image, for example, a personal computer (PC), a server, a laptop, etc. The device 10 for removing mixed noise from a hyperspectral image can include one or more processors 101, a memory 102, an I / O interface 103, and a database 104. Specifically, the processor 101 can be a central processing unit (CPU), or a digital processing unit, etc. The memory 102 can be a volatile memory, such as a random-access memory (RAM); the memory 102 can also be a non-volatile memory, such as a read-only memory, a flash memory, a hard disk drive (HDD), or a solid-state drive (SSD); or the memory 102 can be any other medium that can be used to carry or store a desired program code in the form of an instruction or data structure and can be accessed by a computer, but is not limited thereto. The memory 102 can be a combination of the above memories. The memory 102 may store some program instructions of the method for removing mixed noise from a hyperspectral image provided in the embodiment of the present application. When these program instructions are executed by the processor 101, they can be used to implement the steps of the method for removing mixed noise from a hyperspectral image provided in the embodiment of the present application, so as to solve the problem that the prior art cannot effectively remove mixed noise in a hyperspectral image. The database 104 may be used to store data such as the hyperspectral source image, the mathematical model of the hyperspectral source image, the denoising model of the hyperspectral source image, the optimized denoising model, and the denoised hyperspectral image involved in the solution provided in the embodiment of the present application.

[0053] In the embodiment of the present application, the device 10 for removing mixed noise from a hyperspectral image can obtain image data of a hyperspectral source image through the I / O interface 103, and then the processor 101 of the device 10 for removing mixed noise from a hyperspectral image can effectively remove mixed noise in the hyperspectral image according to the program instructions of the method for removing mixed noise from a hyperspectral image provided by the embodiment of the present application in the memory 102. In addition, data such as the hyperspectral source image, the mathematical model of the hyperspectral source image, the denoising model of the hyperspectral source image, the optimized denoising model, and the denoised hyperspectral image can also be stored in the database 104.

[0054] Of course, the method provided in the embodiment of the present application is not limited to Figure 1 The application scenarios shown can also be used in other possible application scenarios, and the embodiments of the present application are not limited thereto. Figure 1The functions that can be realized by each device in the application scenario shown will be described in the subsequent method embodiments, and will not be described in detail here. Below, the method of the embodiment of the present application will be introduced in conjunction with the accompanying drawings.

[0055] like Figure 2 FIG. 1 is a flow chart of a method for removing mixed noise from a hyperspectral image provided by an embodiment of the present application. The method can be performed by Figure 1 The method is performed by the device 10 for removing mixed noise from a hyperspectral image. Specifically, the process of the method is described as follows.

[0056] Step 201: constructing a mathematical model of a hyperspectral source image according to image data of the hyperspectral source image.

[0057] In the embodiment of the present application, the hyperspectral source image is a hyperspectral image with mixed noise, and the image data includes four parts: target image, sparse noise (for example, salt and pepper noise and bad pixels), Gaussian noise and linear noise (for example, stripe noise and dead lines). Figure 3 , which is a schematic diagram of image data of a hyperspectral source image provided in an embodiment of the present application.

[0058] Specifically, the mathematical model of the hyperspectral source image can be expressed by the following formula (1):

[0059] (1)

[0060] in, They are all third-order tensors, Y represents the hyperspectral source image polluted by mixed noise, X represents the target image to be solved, S represents linear noise, N represents Gaussian noise, and B represents sparse noise; and Respectively represent the number of rows and columns of each spectral image, Indicates the number of spectral bands of the hyperspectral source image, Represents the field of real numbers.

[0061] Step 202: construct a denoising model for the hyperspectral source image according to the nuclear norm low-rank constraint regularization term of the target image, the spatial dimension gradient sparse constraint regularization term of the linear noise, the low-rank constraint term of the linear noise and the mathematical model.

[0062] Specifically, first, a nuclear norm low-rank constraint regularization term of the target image can be determined.

[0063] In practical applications, because the expansion operation will destroy the tensor structure, the existing matrix subspace generated by the hyperspectral image expansion matrix cannot fully represent a tensor, while the tensor subspace can be directly learned from the original tensor. Therefore, the tensor subspace can more realistically express the intrinsic structure of the tensor than the matrix subspace. That is, for an ideal, clean target image X, the following formula (2) can be used to represent it:

[0064] (2)

[0065] Among them, G represents the kernel tensor, A1, A2 and A3 represent the basis of factor matrices or subspaces, and all three are orthogonal matrices, that is, the target image X is equal to the product of the kernel tensor G and A1, A2 and A3.

[0066] Furthermore, in the embodiment of the present application, combined with the low rank of the ideal, clean target image X, the low rank constraint of the target image X can be converted into a low rank constraint of the nuclear norm to improve the denoising performance. That is, the present application designs the nuclear norm low rank constraint regularization term of the target image of the following formula (3):

[0067] (3)

[0068] Among them, X represents the target image, G represents the kernel tensor, A1, A2 and A3 represent the basis of the factor matrix or subspace, and all three are orthogonal matrices. , represents the basis of the i-th factor matrix or subspace, express The transpose of represents the identity matrix, 0 represents a constant, and , represents the matrix nuclear norm, Indicates A nuclear tension quantum problem.

[0069] Then, the spatial dimension gradient sparse constraint regularization term of the linear noise can be determined.

[0070] In traditional methods, dead lines are usually classified as sparse noise together with impulse noise, but in fact, dead lines and strip noise have the same structural characteristics, so the gradients of the two along the x-axis (horizontal direction) and y-axis (vertical direction) show the same sparsity. Based on this, in the embodiment of the present application, the first regularization term of linear noise (i.e., the sparse constraint regularization term of the spatial gradient of linear noise) is determined, which can be specifically expressed by the following formula (4):

[0071] (4)

[0072] Where S represents linear noise, represents the first regularization term of linear noise, represents the L1 norm, and Represent the gradient operators along the x-axis and y-axis respectively.

[0073] Next, the low-rank constraint term for the linear noise can be determined.

[0074] In practical applications, since linear noise also exhibits low-rank characteristics, in the embodiment of the present application, a second regularization term of linear noise (i.e., a low-rank constraint term of linear noise) is designed, which can be specifically expressed by the following formula (5):

[0075] (5)

[0076] in, represents the second regularization term of linear noise, represents a constant, and , Indicates linear noise edge The expansion matrix of the module.

[0077] Finally, the denoising model of the hyperspectral source image can be constructed based on the determined nuclear norm low-rank constraint regularization term of the target image, the spatial dimension gradient sparse constraint regularization term of the linear noise, the gradient sparse constraint term of the linear noise, the mathematical model of the hyperspectral source image, the L1 norm of the sparse noise and the L2 norm constraint of the Gaussian noise. Specifically, the denoising model of the hyperspectral source image can be expressed by the following formula (6):

[0078] (6)

[0079] Where Y represents the hyperspectral source image, B represents sparse noise, and N represents Gaussian noise. represents the parameters that balance the regularization terms, represents a constant, and , Represents the 2-Frobenius norm of the tensor. Based on the denoising model of the hyperspectral source image, it can be seen that the input of the denoising model is the hyperspectral source image, that is, the polluted hyperspectral image, and the output is the target image, that is, the clean image with the mixed noise removed.

[0080] Step 203: Optimize the denoising model according to the preset multiple auxiliary variables to obtain an optimized denoising model.

[0081] In the embodiment of the present application, in order to eliminate the correlation between the data, auxiliary variables T, Q and P are introduced to optimize the denoising model, so as to obtain the optimized denoising model. Specifically, the optimized denoising model can be expressed by the following formula (7):

[0082] (7)

[0083] Among them, T, P and Q represent different preset auxiliary variables respectively.

[0084] Furthermore, the augmented Lagrangian form of formula (7) can be obtained, as shown in the following formula (8):

[0085] (8)

[0086] in, is the Lagrange multiplier, and Both represent penalty parameters. According to formula (8), as long as , then the target image X can be obtained according to formula (2).

[0087] Step 204: Solve the optimized denoising model to obtain a denoised hyperspectral image.

[0088] In an embodiment of the present application, the alternating direction multiplier method ADMM can be used to solve the optimized denoising model (ie, the optimized nuclear norm low-rank constrained dictionary learning denoising and debanding method model) to obtain a denoised hyperspectral image.

[0089] Specifically, step 2041: other variables (other variables except the core tensor G in the optimized denoising model) may be fixed, and the core tensor G may be updated.

[0090] The k+1th iteration value of the core tensor G can be expressed using the following formula (9):

[0091] (9)

[0092] in, .

[0093] The formula (9) can be further divided into solving There are 3 sub-problems to solve. The sub-problem can be expressed by the following formula (10):

[0094] (10)

[0095] in, .

[0096] Furthermore, formula (10) can be solved as follows:

[0097] (11)

[0098] in, , , represents the singular value decomposition of the matrix X, . is a singular value contraction operator, and .

[0099] Step 2042: other variables (other variables except A1, A2 and A3 in the optimized denoising model) may be fixed, and A1, A2 and A3 may be updated.

[0100] To this end, we can extract the terms containing A1, A2 and A3, as shown in the following formula (12):

[0101] (12)

[0102] Furthermore, in order to express formula (12) concisely, we can let , and then, the optimization problem (12) can be simplified to the following formula (13):

[0103] (13)

[0104] Next, we can fix A2 and A3 and update the objective function of A1 to the following formula (14):

[0105] (14)

[0106] And, since for any tensor Z and satisfying The orthogonal matrix ,have Therefore, there is . Therefore, the above formula (14) is equivalent to the following formula (15):

[0107] (15)

[0108] in, , .

[0109] The closed-form solution of formula (15) is as follows:

[0110] (16)

[0111] in, for The singular value decomposition of .

[0112] Similarly, A2 and A3 can be updated using the following formula (17):

[0113] (17)

[0114] in, , , express The singular value decomposition of . , , express The singular value decomposition of .

[0115] Then, according to the above formulas (9)-(17), we can update .

[0116] Step 2043: other variables (other variables except the linear noise S in the optimized denoising model) may be fixed, and the linear noise S may be updated.

[0117] To this end, the term containing linear noise S can be extracted, as shown in the following formula (18):

[0118] (18)

[0119] Furthermore, in order to express formula (18) more concisely, we can let

[0120] , then, the k+1th iteration value of the linear noise S can be expressed by the following formula (19):

[0121] (19)

[0122] The formula (19) can be further divided into solving There are 3 sub-problems to solve. The sub-problem can be expressed by the following formula (20):

[0123] (20)

[0124] in, .

[0125] Furthermore, formula (20) can be solved as follows:

[0126] (twenty one)

[0127] in, , , represents the singular value decomposition of the matrix X, . is a singular value contraction operator, and .

[0128] Step 2044: other variables (other variables except the kernel auxiliary variable T in the optimized denoising model) may be fixed, and the kernel auxiliary variable T may be updated.

[0129] The k+1th iteration value of the kernel auxiliary variable T can be expressed by the following formula (22):

[0130] (twenty two)

[0131] Furthermore, the optimization problem can be viewed as solving the linear system of formula (23):

[0132] (twenty three)

[0133] in, express The adjoint operator of , I represents the identity matrix.

[0134] Therefore, the kernel auxiliary variable T can be obtained as follows:

[0135] (twenty four)

[0136] in, and They represent fast three-dimensional Fourier transform and its inverse transform respectively, and H represents complex conjugate.

[0137] Step 2045: other variables (other variables except the sparse noise B in the optimized denoising model) may be fixed, and the sparse noise B may be updated.

[0138] The k+1th iteration value of the sparse noise B can be expressed by the following formula (25):

[0139] (25)

[0140] Furthermore, the sparse noise B can be obtained by the soft threshold shrinkage operator as follows (26):

[0141] (26)

[0142] in, is a soft threshold operator, and .

[0143] Step 2046: other variables (other variables except the auxiliary variable P in the optimized denoising model) may be fixed, and the auxiliary variable P may be updated.

[0144] The k+1th iteration value of the auxiliary variable P can be expressed by the following formula (27):

[0145] (27)

[0146] Furthermore, the auxiliary variable P of the following formula (28) can be obtained by the soft threshold shrinkage operator:

[0147] (28)

[0148] in, is a soft threshold operator, and .

[0149] Step 2047: Other variables (other variables except the auxiliary variable Q in the optimized denoising model) may be fixed, and the auxiliary variable Q may be updated.

[0150] The k+1th iteration value of the auxiliary variable Q can be expressed by the following formula (29):

[0151] (29)

[0152] Furthermore, the auxiliary variable Q of the following formula (30) can be obtained by the soft threshold shrinkage operator:

[0153] (30)

[0154] in, is a soft threshold operator, and .

[0155] Step 2048: other variables (other variables except Gaussian noise N in the optimized denoising model) may be fixed, and Gaussian noise N may be updated.

[0156] The k+1th iteration value of Gaussian noise N can be expressed by the following formula (31):

[0157] (31)

[0158] Then, the Gaussian noise N can be obtained as follows:

[0159] (32)

[0160] Step 2049: Other variables can be fixed (except Lagrange multipliers in the optimized denoising model ), update the Lagrange multiplier .

[0161] Furthermore, based on the augmented Lagrange multiplier method, the Lagrange multiplier The k+1th iteration value of can be expressed by the following formula (33):

[0162] (33)

[0163] Step 20410: Determine whether the iteration termination condition is met. If ( is a given threshold), the iteration is terminated and the noise-free target image X is output; otherwise, the iterative update continues.

[0164] In the examples of the present application, in order to further describe the technical effects of the present application, the examples of the present application are verified by actual measured data as follows:

[0165] like Figure 4 FIG. 1 is a schematic diagram of the denoising result of the measured urban data set 141-143 bands provided in an embodiment of the present application, wherein the Figure 4 (a) in the figure represents the original 141-band image of the urban dataset hyperspectral image acquired by the Hydice sensor. Figure 4 (b) in the figure shows the original 142-band image of the urban dataset hyperspectral image acquired by the Hydice sensor. Figure 4 (c) in the figure represents the original 143-band image of the urban dataset hyperspectral image acquired by the Hydice sensor. Figure 4 (d) in the figure represents the denoising result obtained by denoising the original 141-band image using the method of this application. Figure 4 (e) in the figure represents the denoising result obtained by denoising the original 142-band image using the method of this application. Figure 4 (f) in the figure represents the denoising result obtained by denoising the original 143-band image using the method of the present application.

[0166] like Figure 5 FIG. 1 is a schematic diagram of the denoising result of the measured AVIRIS data set 154-156 band provided in an embodiment of the present application, wherein the Figure 5 (a) shows the original 154-band image of the dataset acquired by the Airborne Visible Infrared Imaging Spectrometer (AVIRIS). Figure 5 (b) shows the original 155-band image of the dataset acquired by the Airborne Visible Infrared Imaging Spectrometer (AVIRIS). Figure 5 (c) shows the original 156-band image of the dataset acquired by the Airborne Visible Infrared Imaging Spectrometer (AVIRIS). Figure 5(d) in the figure represents the denoising result obtained by denoising the original 154-band image using the method of this application. Figure 5 (e) in the figure represents the denoising result obtained by denoising the original 155-band image using the method of this application. Figure 5 (f) in the figure represents the denoising result obtained by denoising the original 156-band image using the method of the present application.

[0167] like Figure 6 FIG. 1 is a schematic diagram of the denoising result of the measured GF-5 data set 327-329 band provided in an embodiment of the present application, wherein the Figure 6 (a) in the figure represents the original 327-band image of the remote sensing image acquired by the Gaofen-5 satellite. Figure 6 (b) in the figure represents the original 328-band image of the remote sensing image acquired by the Gaofen-5 satellite. Figure 6 (c) in the figure represents the original 329-band image of the remote sensing image acquired by the Gaofen-5 satellite. Figure 6 (d) in the figure represents the denoising result obtained by denoising the original 327-band image using the method of this application. Figure 6 (e) in the figure represents the denoising result obtained by denoising the original 328-band image using the method of this application. Figure 6 (f) in the figure represents the denoising result obtained by denoising the original 329-band image using the method of the present application.

[0168] Obviously, according to Figure 4-6 It can be seen that the method of the present application shows good denoising effect in removing mixed noise, especially when containing linear noise. While completely denoising, it can well retain the detail information of the original image, providing data support for the subsequent application of hyperspectral images.

[0169] In summary, since the denoising model of hyperspectral source images is constructed through "the nuclear norm low-rank constraint regularization term of the target image, the spatial dimension gradient sparse constraint regularization term of the linear noise, and the low-rank constraint term of the linear noise", it can not only effectively remove the mixed noise in the hyperspectral source image, but also restore the texture information of the target image, effectively reduce the spectral distortion and improve the practicability of the denoising method.

[0170] Based on the same inventive concept, the present application embodiment provides a device 70 for removing mixed noise from a hyperspectral image, such as Figure 7 As shown, the device 70 for removing mixed noise from a hyperspectral image comprises:

[0171] The mathematical model building unit 701 is used to build a mathematical model of the hyperspectral source image according to the image data of the hyperspectral source image; wherein the image data includes four parts: target image, sparse noise, Gaussian noise and linear noise;

[0172] A denoising model construction unit 702 is used to construct a denoising model for a hyperspectral source image according to a nuclear norm low-rank constraint regularization term of a target image, a spatial dimension gradient sparse constraint regularization term of linear noise, a low-rank constraint term of linear noise and a mathematical model;

[0173] A denoising model optimization unit 703 is used to optimize the denoising model according to a plurality of preset auxiliary variables to obtain an optimized denoising model;

[0174] The hyperspectral image denoising unit 704 is used to solve the optimized denoising model to obtain a denoised hyperspectral image.

[0175] Optionally, the hyperspectral image denoising unit 704 is further used for:

[0176] The alternating direction multiplier method (ADMM) is used to solve the optimized denoising model and obtain the denoised hyperspectral image.

[0177] The hyperspectral image mixed noise removal device 70 can be used to perform Figure 2 The method implemented in the embodiment shown in FIG. 1 is a method implemented in the embodiment shown in FIG. 1 . Therefore, the functions that can be realized by each functional module of the hyperspectral image mixed noise removal device 70 can be referred to. Figure 2 The description of the illustrated embodiment is not repeated in detail.

[0178] In some possible implementations, various aspects of the method provided in the present application may also be implemented in the form of a program part, which includes a program code. When the program part is run on a computer device, the program code is used to enable the computer device to execute the steps of the method according to various exemplary embodiments of the present application described above in this specification. For example, the computer device may execute the following steps: Figure 2 The method performed in the illustrated embodiment.

[0179] Those skilled in the art can understand that all or part of the steps of the above method embodiment can be completed by hardware related to program instructions, and the aforementioned program can be stored in a computer-readable storage medium. When the program is executed, the steps of the above method embodiment are executed; and the aforementioned storage medium includes: various media that can store program codes, such as mobile storage devices, read-only memory (ROM), random access memory (RAM), magnetic disks or optical disks. Alternatively, if the above integrated unit of the present invention is implemented in the form of a software function module and sold or used as an independent part, it can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiment of the present invention can be essentially or in other words, the part that contributes to the prior art can be embodied in the form of a software part, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the methods described in various embodiments of the present invention. And the aforementioned storage medium includes: various media that can store program codes, such as mobile storage devices, ROM, RAM, magnetic disks or optical disks.

[0180] Although the preferred embodiments of the present application have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present application.

[0181] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.

Claims

1. A method for removing mixed noise from a hyperspectral image, characterized in that: The method comprises: According to the image data of the hyperspectral source image, a mathematical model of the hyperspectral source image is constructed; wherein the image data includes four parts: target image, sparse noise, Gaussian noise and linear noise; the mathematical model of the hyperspectral source image is expressed by the following formula: They are all third-order tensors, Y represents the hyperspectral source image polluted by mixed noise, X represents the target image to be solved, S represents linear noise, N represents Gaussian noise, and B represents sparse noise; and Respectively represent the number of rows and columns of each spectral image, Indicates the number of spectral bands of the hyperspectral source image, represents the field of real numbers; According to the nuclear norm low-rank constraint regularization term of the target image, the spatial dimension gradient sparse constraint regularization term of the linear noise, the low-rank constraint term of the linear noise and the mathematical model, a denoising model of the hyperspectral source image is constructed; wherein the nuclear norm low-rank constraint regularization term of the target image includes the following formula: The spatial gradient sparse constraint regularization term of the linear noise includes the following formula: The low-rank constraint term of the linear noise includes the following formula: The denoising model of the hyperspectral source image includes the following formula: G represents the kernel tensor, A1, A2 and A3 represent the basis of factor matrices or subspaces, and all three are orthogonal matrices. , represents the basis of the i-th factor matrix or subspace, express The transpose of represents the identity matrix, 0 represents a constant, and , represents the matrix nuclear norm, Indicates A nuclear tension quantum problem; represents the first regularization term of linear noise, represents the L1 norm, and Represent the gradient operators along the x-axis and y-axis respectively; represents the second regularization term of linear noise, represents a constant, and , Indicates linear noise edge The expansion matrix of the module; represents the parameters that balance the regularization terms, Represents the 2-Frobenius norm of a tensor; According to the preset multiple auxiliary variables, the denoising model is optimized to obtain an optimized denoising model; wherein the optimized denoising model includes the following formula: T, P and Q represent different preset auxiliary variables; The optimized denoising model is solved to obtain a denoised hyperspectral image.

2. The method according to claim 1, characterized in that The step of solving the optimized denoising model to obtain a denoised hyperspectral image comprises: The optimized denoising model is solved by using the alternating direction multiplier method ADMM to obtain a denoised hyperspectral image.

3. A device for removing mixed noise from a hyperspectral image, characterized in that: The device comprises: The mathematical model building unit is used to build a mathematical model of the hyperspectral source image according to the image data of the hyperspectral source image; wherein the image data includes four parts: target image, sparse noise, Gaussian noise and linear noise; the mathematical model of the hyperspectral source image is expressed by the following formula: They are all third-order tensors, Y represents the hyperspectral source image polluted by mixed noise, X represents the target image to be solved, S represents linear noise, N represents Gaussian noise, and B represents sparse noise; and Respectively represent the number of rows and columns of each spectral image, Indicates the number of spectral bands of the hyperspectral source image, represents the field of real numbers; The denoising model construction unit is used to construct a denoising model of the hyperspectral source image according to the nuclear norm low rank constraint regularization term of the target image, the spatial dimension gradient sparse constraint regularization term of the linear noise, the low rank constraint term of the linear noise and the mathematical model; wherein the nuclear norm low rank constraint regularization term of the target image includes the following formula: The spatial gradient sparse constraint regularization term of the linear noise includes the following formula: The low-rank constraint term of the linear noise includes the following formula: The denoising model of the hyperspectral source image includes the following formula: G represents the kernel tensor, A1, A2 and A3 represent the basis of factor matrices or subspaces, and all three are orthogonal matrices. , represents the basis of the i-th factor matrix or subspace, express The transpose of represents the identity matrix, 0 represents a constant, and , represents the matrix nuclear norm, Indicates A nuclear tension quantum problem; represents the first regularization term of linear noise, represents the L1 norm, and Represent the gradient operators along the x-axis and y-axis respectively; represents the second regularization term of linear noise, represents a constant, and , Indicates linear noise edge The expansion matrix of the module; represents the parameters that balance the regularization terms, Represents the 2-Frobenius norm of a tensor; The denoising model optimization unit is used to optimize the denoising model according to a plurality of preset auxiliary variables to obtain an optimized denoising model; wherein the optimized denoising model includes the following formula: T, P and Q represent different preset auxiliary variables; The hyperspectral image denoising unit is used to solve the optimized denoising model to obtain a denoised hyperspectral image.

4. An electronic device, characterized in that: The device comprises: A memory for storing program instructions; A processor is used to call the program instructions stored in the memory, and execute any method according to claims 1-2 according to the obtained program instructions.

5. A storage medium, characterized in that: The storage medium stores computer-executable instructions, and the computer-executable instructions are used to enable a computer to execute any one of the methods in claims 1-2.

Images