Hyperspectral fusion method and device combining local non-local manifold and truncation operator

Abstract

The present invention discloses a hyperspectral fusion method and device combining local non-local manifolds and truncation operators, which divides an input high-resolution multispectral image into different sub-blocks and reconstructs them into a matrix form, wherein all sub-blocks are distributed near multiple low-dimensional manifolds; for each sub-block, a nearest neighbor search algorithm is used to find non-local similar image blocks, and a non-local similarity relationship is constructed. At the same time, a similarity relationship between each image block and four surrounding adjacent blocks is constructed, and the local and non-local similarity relationships are integrated to obtain a manifold mapping matrix L ;right L Perform eigendecomposition to obtain the eigenbase and construct the truncation operator. L The truncation operator is incorporated into the high-resolution HSI fusion model to obtain the final model. The alternating direction multiplier method ADMM algorithm is used to optimize the above model, and the decoupling operation is performed, and finally the conjugate gradient algorithm PCG is used to solve it band by band.

Description

Technical Field

[0001] The present invention relates to a fusion technology of low-resolution hyperspectral and high-resolution multispectral to obtain a high-resolution hyperspectral image, and in particular to a hyperspectral fusion method and device based on a joint local-nonlocal manifold and a truncation operator. Background Art

[0002] Hyperspectral images (HSI) are collected by an imaging spectrometer with narrow continuous wavelength intervals (such as 10nm). Compared with traditional images, hyperspectral images have higher spectral resolution and wider frequency bands, so they can provide accurate material identification capabilities. This reliable phenomenon has given HSI more application value in remote sensing, mineral exploration, military monitoring, medical diagnosis, etc.

[0003] However, since limited solar radiation is distributed in most spectral bands, obtaining rich spectral information is inevitably accompanied by a compromise in spatial resolution, which makes HSI often have a low spatial resolution and hinders its wider application. Therefore, maintaining the spectral resolution of HSI while improving its spatial resolution is a necessary task with great development potential in HSI analysis. A direct approach is to directly use HSI sensors with high spatial resolution. However, due to basic physical limitations, hardware breakthroughs are quite difficult and uneconomical, especially for hyperspectral images, where the sensors used need to meet the requirements of spectral and spatial resolution at the same time. In reality, sensors can also obtain images with higher spatial resolution but fewer spectral bands in the same scene, such as panchromatic images and multispectral images (MSI). Therefore, image fusion is an effective and economical alternative strategy to improve the spatial resolution of HSI. Common fusion methods include low-rank approximation, sparse dictionary learning, deep learning, etc. Summary of the invention

[0004] In order to solve the shortcomings of the prior art and achieve the purpose of improving the fusion effect, the present invention adopts the following technical solutions:

[0005] The hyperspectral fusion method combining local non-local manifold and truncation operator includes the following steps:

[0006] Step 1, obtain a high-resolution multispectral image MSI, divide it into different sub-blocks and reconstruct it into a matrix form, where all sub-blocks are distributed near multiple low-dimensional manifolds;

[0007] Step 2: For each sub-block, use the nearest neighbor search algorithm (such as kd-tree) to find non-local similar image blocks, build non-local similarity relations, and build similarity relations between each image block and a group of surrounding adjacent image blocks. Integrate local and non-local similarity relations to obtain the manifold mapping matrix L.

[0008] Step 3, perform feature decomposition on L to obtain the feature basis, and construct a truncation operator, and incorporate L and the truncation operator into the high-resolution hyperspectral image HSI fusion model to obtain the final fusion model;

[0009] Step 4: Use the alternating direction multiplier method to optimize the fusion model, perform decoupling operations, and solve band by band to obtain the fusion result.

[0010] Furthermore, step 1 includes the following steps:

[0011] Step 1.1: Acquire high-resolution multispectral images With any coordinate x=[N 1 ]×[N 2 ] is the origin and the size is s 1 ×s 2 ×B image blocks, where [N 1 ]={1,2,…,N 1},[N 2 ]={1,2,…,N 2},N 1 Indicates the size of the first dimension of the image space, N 2 represents the size of the second dimension of the image, b represents the number of image spectra, s 1 Indicates the first dimension size of the image block, s 2 represents the second dimension size of the image block, B represents the number of spectra of the image block, represents any image block, represents the set of all image blocks, where d = s 1 ×s 2 ×B, the formula for constructing the data manifold dimension is as follows:

[0012]

[0013] where u is the coordinate function on the manifold, It means to find the gradient of u. Indicates l 2 norm, Approximate using similar blocks in a discrete set:

[0014]

[0015] in represents the manifold where the image block is located, p and q represent different points on u, that is, different image blocks, and w is used to measure the similarity between p and q, which is expressed as:

[0016]

[0017] Where σ represents the normalization parameter and exp(·) represents the exponential function. So far, the calculation method of the manifold dimension is obtained. Then, the manifold dimension is calculated from both local and non-local aspects.

[0018] Furthermore, step 2 includes the following steps:

[0019] Step 2.1, construct a non-local manifold. For each image block, search for similar image blocks and obtain the non-local similarity matrix W by formula (2): nl ;

[0020] Step 2.2: construct a local manifold. For each image block, use formula (2) to calculate the similarity with a group of adjacent image blocks to obtain the local similarity matrix W: l ;

[0021] Step 2.3, define W as W = W l +W nl , and get the manifold mapping matrix L.

[0022] Furthermore, the manifold mapping matrix in step 2.3 is a Laplacian matrix:

[0023] L=DW (4)

[0024] Where D(x,x)=∑ y W(x,y), x represents any coordinate, and y represents a coordinate other than x.

[0025] Further, step 3 includes the following steps:

[0026] Step 3.1, according to spectral theory, decompose the L features into:

[0027] L=UΛU T (5)

[0028] where U={u 1 ,u 2 ,...,u k} represents the characteristic matrix, Λ represents the matrix containing the eigenvalues ​​{θ k} is a diagonal matrix, the eigenvector U is the manifold basis, u k represents the kth component in U, θ k represents the corresponding graph frequency, then for a manifold

[0029] Step 3.2, construct the truncation operator

[0030]

[0031] in represents the expansion matrix of the high-resolution HSI to be determined along the spectral direction, <,> represents the dot product of the vectors, t represents the t-th band, and S represents the truncation operator:

[0032]

[0033] τ represents the manually set threshold;

[0034] Step 3.3, incorporate the manifold model and truncation operator into the fusion framework, and the overall model is:

[0035]

[0036] in represents the F-norm, represents the unfolded low-resolution HSI, represents the point spread function of the spectral imager, represents the spatial downsampling matrix, Represents an expanded MSI image, represents the spectral downsampling matrix, The manifold regularization term corresponding to equation (5) is expressed as: represents the truncation operator of equation (6), where λ and β represent the corresponding equilibrium parameters;

[0037] In step 3.4, the alternating direction multiplier method (ADMM) algorithm is used to rewrite formula (8) to obtain the final fusion model:

[0038]

[0039] E and Q represent two additional split variables.

[0040] Further, step 4 includes the following steps:

[0041] Step 4.1, solve equation (9) and rewrite it as:

[0042]

[0043] Where V 1 and V 2 is the Lagrange multiplier, μ 1 and μ 2 is the penalty parameter;

[0044] Step 4.2, fix E and Q, and update Z:

[0045]

[0046] Equation (11) is the Sylvester equation, which has a unique specific solution;

[0047] Step 4.3, update E and Q:

[0048]

[0049] E and Q are two different images. For E, it can be expanded as follows:

[0050]

[0051] Further written as:

[0052]

[0053] where s z =s 1 ×s 2 , is an image block The i-th pixel in ;

[0054] For any band t∈[B], we can decompose equation (14) by using the standard variational method:

[0055]

[0056] yes The adjoint operator of get:

[0057]

[0058] in, represents the i-th pixel after pixel x in the same image block. Substituting equation (16) into equation (15), we get:

[0059]

[0060] in, express The degree matrix of is solved by the conjugate gradient algorithm PCG (17);

[0061] For the update of Q, the truncation strategy in formula (6) is used for update;

[0062] The updates of the two Lagrange multipliers are as follows:

[0063]

[0064] Step 4.4: Iteratively update until convergence. When the error between two adjacent iterations is less than the threshold, terminate the iteration and output the final fusion result.

[0065] Further, the specific conditions for Step 4.4 are as follows:

[0066] ||Z m+1 -Z m || F / ||Z m || F <1e-3 (19)

[0067] Where Z m+1 represents the current iteration output, Z m represents the previous iteration output, m represents the number of iterations. When the convergence formula (19) is satisfied, terminate the iteration and output the final fusion result Z m+1 . The obtained high-resolution HSI can be used for subsequent applications such as mineral exploration and medical detection.

[0068] Further, a group of adjacent image patches in Step 2 are four adjacent image patches.

[0069] Further, the conjugate gradient algorithm PCG is used in Step 4 for band-by-band solution.

[0070] The hyperspectral fusion device combining local non-local manifolds and truncation operators includes one or more processors for implementing the hyperspectral fusion method combining local non-local manifolds and truncation operators.

[0071] The advantages and beneficial effects of the present invention are as follows:

[0072] The hyperspectral fusion method and device based on combining local non-local manifolds and truncation operators of the present invention conduct in-depth research on the fusion task, use the manifold method to learn the spatial structure in MSI, and supplement it with a truncation operator to enhance the robustness under noise interference, thereby improving the fusion effect. Description of the Drawings

[0073] Figure 1 is the flowchart of the hyperspectral remote sensing image fusion method in the present invention.

[0074] Figure 2a is the original Figure 1 .

[0075] Figure 2b is the original Figure 1 corresponding low-resolution image in the embodiment of the present invention.

[0076] Figure 2c is the original Figure 1The corresponding super-resolution effect of band#15 on toys hyperspectral data at 8 times downsampling rate based on the joint local-nonlocal manifold and truncation operator.

[0077] Figure 3a It is original picture 2 in the embodiment of the present invention.

[0078] Figure 3b It is a low-resolution image corresponding to the original image 2 in the embodiment of the present invention.

[0079] Figure 3c This is a super-resolution effect diagram of band #30 on the Pavia University hyperspectral data at a 4-fold downsampling rate corresponding to the original image 2 in an embodiment of the present invention based on the joint local non-local manifold and truncation operator.

[0080] Figure 4 It is a structural diagram of a hyperspectral fusion device combining local non-local manifolds and truncation operators in the present invention. DETAILED DESCRIPTION

[0081] The specific implementation of the present invention is described in detail below in conjunction with the accompanying drawings. It should be understood that the specific implementation described here is only used to illustrate and explain the present invention, and is not used to limit the present invention.

[0082] Hyperspectral fusion method based on joint local non-local manifold and truncation operator, hyperspectral image restoration based on manifold weighted non-local curvature regularization, such as Figure 1 The specific steps are as follows:

[0083] Step 1: Given a high-resolution multispectral image With any coordinate x=[N 1 ]×[N 2 ] is the origin and the size is s 1 ×s 2 ×B image blocks, where [N 1 ]={1,2,…,N 1},[N 2 ]={1,2,…,N 2},N 1 Indicates the size of the first dimension of the image space, N 2 represents the size of the second dimension of the image, b represents the number of image spectra, s 1 Indicates the first dimension size of the image block, s 2 represents the second dimension size of the image block, B represents the number of spectra of the image block, For any image block, define is the set of all image patches and is defined as Where d = s 1 ×s 2×B, the data manifold dimension can be expressed by the following formula:

[0084]

[0085] where u is the coordinate function on the manifold, It means to find the gradient of u. Indicates l 2 norm, In discrete sets, similar blocks can be used for approximation:

[0086]

[0087] in represents the manifold where the image block is located, p and q represent different points on u, that is, different image blocks, and w is used to measure the similarity between p and q, which can be expressed as:

[0088]

[0089] Where σ represents the normalization parameter and exp(·) represents the exponential function. So far, the calculation method of the manifold dimension is obtained. Then, the manifold dimension is calculated from both local and non-local aspects.

[0090] Step 2: First, construct a non-local manifold. For each image block, search for similar image blocks. The number of searches is 10. And the non-local similarity matrix W is obtained by formula (2): nl Then, a local manifold is constructed. For each image block, the similarity with its four adjacent image blocks is calculated using formula (2) to obtain the local similarity matrix W l , define W as W = W l +W nl The Laplace matrix L can be defined as:

[0091] L=DW (4)

[0092] Where D(x,x)=∑ y W(x,y), x represents any coordinate, y represents a coordinate other than x;

[0093] Step 3: According to spectral theory, the L feature can be decomposed into:

[0094] L=UΛU T (5)

[0095] where U={u 1 ,u 2 ,...,u k} represents the characteristic matrix, Λ represents the matrix containing the eigenvalues ​​{θ k} is a diagonal matrix, the eigenvector U can be regarded as a manifold basis, u krepresents the kth component in U, θ k is the corresponding graph frequency, then for a manifold Truncation operator It can be defined as:

[0096]

[0097] in represents the expansion matrix of the high-resolution HSI to be obtained along the spectral direction, <,> represents the dot product of the vectors, t represents the tth band, is the truncation operator:

[0098]

[0099] τ is a manually set threshold. Incorporating the above manifold model and truncation operator into the fusion framework, the overall model can be written as:

[0100]

[0101] in represents the F-norm, represents the unfolded low-resolution HSI, represents the point spread function of the spectral imager, represents the spatial downsampling matrix, Represents an expanded MSI image, represents the spectral downsampling matrix, The manifold regularization term corresponding to equation (5) is expressed as: It represents the truncation operator of equation (6), where λ and β represent the corresponding equilibrium parameters.

[0102] Using the ADMM (Alternating Direction Multiplier Method) algorithm for formula (8), adding two splitting variables E and Q, formula (8) can be rewritten as:

[0103]

[0104] Step 4, solve equation (9), which can be rewritten as:

[0105]

[0106] Where V 1 and V 2 is the Lagrange multiplier, μ 1 and μ 2 is the penalty parameter.

[0107] Then fix E and Q and update Z:

[0108]

[0109] Equation (11) is the Sylvester equation, which has a unique specific solution.

[0110] Then update E and Q:

[0111]

[0112] E and Q can be regarded as two different images. For E, it can be expanded as:

[0113]

[0114] It can be further written as:

[0115]

[0116] where s z =s 1 ×s 2 , is an image block The i-th pixel in .

[0117] For any band t∈[B], using the standard variational method, equation (14) can be decoupled as:

[0118]

[0119] yes The adjoint operator of Easy to get:

[0120]

[0121] in, represents the i-th pixel after pixel x in the same image block. Substituting equation (16) into equation (15) yields:

[0122]

[0123] in, express The degree matrix of . Solve equation (17) using the PCG algorithm.

[0124] For the update of Q, the truncation strategy mentioned in formula (6) is used for update.

[0125] The updates of the two Lagrange multipliers are as follows:

[0126]

[0127] Iterate until convergence. When the error between two consecutive iterations is very small, the iteration is terminated. The specific conditions are as follows:

[0128] ||Z m+1 -Z m || F / ||Z m || F <1e-3 (19)

[0129] Among them, Z m+1 Represents the current iteration output, Z m represents the output of the last iteration, and m represents the number of iterations. When the convergence equation (19) is satisfied, the iteration is terminated and the final fusion result Z is output. m+1 The output results Z of the two times are substituted into formula 19. When the error is less than 0.001, the iteration is terminated. The final fusion result is Z m+1 The resulting high-resolution HSI can be used in subsequent applications such as mineral exploration and medical testing.

[0130] like Figure 2a-2c As shown in the figure, the super-resolution effect of band#15 on toys hyperspectral data at 8 times downsampling rate based on the joint local non-local manifold and truncation operator. Figure 2a For the original Figure 1 , Figure 2b For low-resolution images, Figure 2c In order to demonstrate the super-resolution effect of the method of the present invention on the hyperspectral image, a part of the image is enlarged.

[0131] like Figure 3a-3c As shown in the figure, the super-resolution effect of band#30 on the Pavia University hyperspectral data at a 4-fold downsampling rate based on the joint local non-local manifold and truncation operator. Figure 3a This is the original picture 2. Figure 3b For low-resolution images, Figure 3c In order to demonstrate the super-resolution effect of the method of the present invention on the hyperspectral image, a part of the image is enlarged.

[0132] Corresponding to the above-mentioned embodiment of the hyperspectral fusion method combining local non-local manifold and truncation operator, the present invention also provides an embodiment of a hyperspectral fusion device combining local non-local manifold and truncation operator.

[0133] See also Figure 4 The hyperspectral fusion device combining local non-local manifolds and truncation operators provided in an embodiment of the present invention includes one or more processors for implementing the hyperspectral fusion method combining local non-local manifolds and truncation operators in the above embodiment.

[0134] The embodiment of the hyperspectral fusion device combining local non-local manifold and truncation operator of the present invention can be applied to any device with data processing capability, and the device with data processing capability can be a device or apparatus such as a computer. The device embodiment can be implemented by software, or by hardware or a combination of software and hardware. Taking software implementation as an example, as a device in a logical sense, the processor of any device with data processing capability reads the corresponding computer program instructions in the non-volatile memory into the internal memory and runs them. From the hardware level, if Figure 4 As shown, it is a hardware structure diagram of any device with data processing capability where the hyperspectral fusion device combining local non-local manifold and truncation operator of the present invention is located, except Figure 4 In addition to the processor, memory, network interface, and non-volatile memory shown, any device with data processing capabilities in which the apparatus in the embodiments is located may also include other hardware, generally based on the actual functions of the device with data processing capabilities, which will not be described in detail.

[0135] The implementation process of the functions and effects of each unit in the above-mentioned device is specifically described in the implementation process of the corresponding steps in the above-mentioned method, and will not be repeated here.

[0136] For the device embodiment, since it basically corresponds to the method embodiment, the relevant parts can refer to the partial description of the method embodiment. The device embodiment described above is only schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of the present invention. Ordinary technicians in this field can understand and implement it without paying creative work.

[0137] An embodiment of the present invention further provides a computer-readable storage medium having a program stored thereon. When the program is executed by a processor, the hyperspectral fusion method combining local and non-local manifolds and truncation operators in the above embodiment is implemented.

[0138] The computer-readable storage medium may be an internal storage unit of any device with data processing capability described in any of the aforementioned embodiments, such as a hard disk or a memory. The computer-readable storage medium may also be an external storage device of any device with data processing capability, such as a plug-in hard disk, a smart media card (SMC), an SD card, a flash card, etc. equipped on the device. Furthermore, the computer-readable storage medium may also include both an internal storage unit and an external storage device of any device with data processing capability. The computer-readable storage medium is used to store the computer program and other programs and data required by any device with data processing capability, and may also be used to temporarily store data that has been output or is to be output.

[0139] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some or all of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. Hyperspectral fusion method combining local non-local manifold and truncation operator, Features The steps include: Step 1, obtain a high-resolution multispectral image MSI, divide it into different sub-blocks and reconstruct it into a matrix form, and all sub-blocks are distributed in multiple low-dimensional manifolds; Step 2: For each sub-block, use the nearest neighbor search algorithm to find non-local similar image blocks, build non-local similarity relationships, and build similarity relationships between each image block and a group of surrounding adjacent image blocks. Integrate local and non-local similarity relationships to obtain the manifold mapping matrix L. Step 3, perform feature decomposition on L, obtain feature bases, and construct a truncation operator, and incorporate L and the truncation operator into the high-resolution hyperspectral image HSI fusion model to obtain the final fusion model; specifically, the steps include: Step 3.1, according to spectral theory, decompose the L features into: L = UΛU T (5) where U={u 1 ,u 2 ,...,u k } represents the characteristic matrix, Λ represents the matrix containing the eigenvalues ​​{θ k } is a diagonal matrix, the eigenvector U is the manifold basis, u k represents the kth component in U, θ k represents the corresponding graph frequency, then for a manifold Step 3.2, construct the truncation operator in represents the expansion matrix of the high-resolution HSI to be obtained along the spectral direction, <,> represents the dot product of the vectors, t represents the tth band, Represents the truncation operator: τ represents the manually set threshold; Step 3.3, incorporate the manifold model and truncation operator into the fusion framework, and the overall model is: where denotes the F-norm, denotes the unfolded low-resolution HSI, denotes the point spread function of the spectral imager, denotes the spatial downsampling matrix, denotes the unfolded MSI image, denotes the spectral downsampling matrix, denotes the manifold regularization term corresponding to Equation (5), denotes the truncation operator of Equation (6), where λ and β denote the corresponding balance parameters; Step 3.4, rewrite formula (8) to get the final fusion model: E and Q represent two added split variables; Step 4: Use the alternating direction multiplier method to optimize the fusion model, perform decoupling operations, and solve band by band to obtain the fusion result.

2. The hyperspectral fusion method combining local non-local manifold and truncation operator according to claim 1, Features The step 1 comprises the following steps: Step 1.1: Acquire high-resolution multispectral images With any coordinate x=[N 1 ]×[N 2 ] is the origin and the size is s 1 ×s 2 ×B image blocks, where [N 1 ]={1,2,…,N 1 },[N 2 ]={1,2,…,N 2 },N 1 Indicates the size of the first dimension of the image space, N 2 represents the size of the second dimension of the image, b represents the number of image spectra, s 1 Indicates the first dimension size of the image block, s 2 represents the second dimension size of the image block, B represents the number of spectra of the image block, represents any image block, represents the set of all image blocks, where d = s 1 ×s 2 ×B, the formula for constructing the data manifold dimension is as follows: where u is the coordinate function on the manifold, It means to find the gradient of u. Indicates l 2 norm, Approximate using similar blocks in a discrete set: in represents the manifold where the image block is located, p and q represent different points on u, that is, different image blocks, and w is used to measure the similarity between p and q, which is expressed as: Where σ represents the normalization parameter and exp(·) represents the exponential function. So far, the calculation method of the manifold dimension is obtained. Then, the manifold dimension is calculated from both local and non-local aspects.

3. The hyperspectral fusion method combining local non-local manifold and truncation operator according to claim 2, Features The step 2 comprises the following steps: Step 2.1, construct a non-local manifold. For each image block, search for similar image blocks and obtain the non-local similarity matrix W by formula (2): nl ; Step 2.2: construct a local manifold. For each image block, use formula (2) to calculate the similarity with a group of adjacent image blocks to obtain the local similarity matrix W: l ; Step 2.3, define W as W = W l +W nl , and get the manifold mapping matrix L.

4. The hyperspectral fusion method combining local non-local manifold and truncation operator according to claim 3, Features The manifold mapping matrix in step 2.3 is the Laplace matrix: L=DW (4) Where D(x,x)=∑ y W(x,y), x represents any coordinate, and y represents a coordinate other than x.

5. The hyperspectral fusion method combining local non-local manifold and truncation operator according to claim 1, Features The step 4 comprises the following steps: Step 4.1, solve equation (9) and rewrite it as: where V 1 and V 2 are Lagrange multipliers, μ 1 and μ 2 are penalty parameters; Step 4.2, fix E and Q, and update Z: Equation (11) is the Sylvester equation, which has a unique specific solution; Step 4.3, update E and Q: E and Q are two different images. For E, it can be expanded as follows: Further written as: where s z =s 1 ×s 2 , is an image block The i-th pixel in ; For any band t∈[B], we can decompose equation (14) by using the standard variational method: yes The adjoint operator of get: Among them, represents the i-th pixel after pixel x in the same image block. Substituting Equation (16) into Equation (15), we get: in, express The degree matrix of is solved by the conjugate gradient algorithm PCG (17); For the update of Q, the truncation strategy in formula (6) is used for update; The updates of the two Lagrange multipliers are as follows: Step 4.4, iterate and update until convergence. When the error between two adjacent iterations is less than the threshold, terminate the iteration and output the final fusion result.

6. The hyperspectral fusion method combining local non-local manifold and truncation operator according to claim 5, Features The specific conditions of step 4.4 are as follows: ||With m+1 -Z m || F / ||Z m || F <1e-3 (19) Among them, Z m+1 Represents the current iteration output, Z m represents the output of the last iteration, m represents the number of iterations, and when the convergence formula (19) is satisfied, the iteration is terminated and the final fusion result Z is output m+1 .

7. The hyperspectral fusion method combining local non-local manifold and truncation operator according to claim 1, Features The group of adjacent image blocks in step 2 is four adjacent image blocks.

8. The hyperspectral fusion method combining local non-local manifold and truncation operator according to claim 1, Features In step 4, the conjugate gradient algorithm PCG is used to solve the problem band by band.

9. Hyperspectral fusion device combining local non-local manifold and truncation operator, It is characterized in that The method comprises one or more processors for implementing the hyperspectral fusion method of combining local and non-local manifolds and truncation operators as described in any one of claims 1 to 8.

Images