Remote sensing image fusion method and system based on local variance image mutual information

Abstract

The application belongs to the technical field of image fusion, and particularly relates to a remote sensing image fusion method and system based on local variance image mutual information. The method comprises the following steps: firstly, using a multivariate linear regression component replacement fusion method to perform fusion to obtain an initial fusion image; then, on the basis of the initial fusion image, similarity between an intensity component of a high-resolution multispectral image and an intensity component of the initial fusion image is calculated with the aim of obtaining an optimal Gaussian filter; finally, the original high-resolution panchromatic image is filtered by using the estimated optimal Gaussian filter to obtain a low-resolution panchromatic image, and image fusion is performed on the basis of the low-resolution panchromatic image. The application makes the fused image more conducive to the preservation of spectral characteristics and reduces spatial distortion in the spatial scale.

Description

Technical Field

[0001] This invention belongs to the field of image fusion technology, and specifically relates to a remote sensing image fusion method and system based on local variance image mutual information. Background Technology

[0002] In optical remote sensing systems, spatial resolution, spectral resolution, and temporal resolution are often difficult to achieve simultaneously due to factors such as optical diffraction limit, modulation transfer function, and signal-to-noise ratio. Utilizing complementary and redundant information to alleviate the inherent contradictions among these three resolutions is the intrinsic motivation behind the emergence of remote sensing image fusion technology. Among the many tasks of remote sensing image information fusion, multispectral and panchromatic image fusion is the most representative scenario. This involves simulating low-resolution panchromatic band P based on the spectral response function. L Component substitution fusion algorithm and simulation of low-resolution panchromatic band I based on multiple linear regression method L The component substitution fusion algorithm is one of the two most representative image fusion methods that improve spectral distortion by constructing substitute bands.

[0003] Assume that an ideal low-resolution panchromatic image P′ exists. L It should have two characteristics: First, the spatial information features should be consistent with the original multispectral image MS, which can ensure the spatial information of the cancelling multispectral image MS is correct; second, the spectral features should be consistent with the original panchromatic image P, which can ensure the spectral features of the cancelling panchromatic image P are correct.

[0004] In practical applications, I obtained from multispectral image MS multiple linear regression simulation LThe spatial information of the low-resolution panchromatic image obtained is unchanged compared with that of MS, so the fused image is clearer, but the spectral distortion is more serious. The reasons are as follows: First, in terms of the model itself, linear regression has certain applicable conditions: (1) whether the independent variable and the dependent variable are linearly related. (2) whether the dependent variable conforms to a normal distribution. (3) whether the values ​​of the dependent variable are independent. (4) whether the variance is homogeneous. However, due to the influence of uncertain factors such as satellite working status, atmospheric conditions and imaging errors, most actual remote sensing images are not normally distributed, and there is a certain correlation between the various bands of multispectral imaging, especially between the visible light bands. Second, the spectral response curve coverage of panchromatic images and multispectral images is inconsistent, which is more prominent between heterogeneous panchromatic and multispectral images. For example, the spectral response curves of QuickBird and IKONOS satellite panchromatic and multispectral images have inconsistent coverage ranges. When the wavelength is greater than 950μm, only the panchromatic band is covered. The spectral curves of each multispectral image are independent of each other across the entire wavelength range, with some overlap and discontinuity. If multiple linear regression is used, it means that each multispectral response curve should be distributed across the entire spectral coverage range of the panchromatic light image, which does not match the reality.

[0005] P is obtained by low-pass filtering the panchromatic image P. L Its spectral characteristics are consistent with the original panchromatic image P, which involves two questions: which low-pass filter to use, and how to measure the low-resolution panchromatic image P after low-pass filtering. L Is the spatial resolution consistent with the multispectral spatial resolution? Regarding the selection of low-pass filters, the ideal low-pass filter should be the modulation transfer function (MTF) of the remote sensor. However, most remote sensing imaging systems' MTF compensation methods are not publicly disclosed in the data products provided by satellite data providers. A Gaussian filter is used to match the MTF of the low-resolution multispectral image (MS sensor). That is, after a high-resolution multispectral HMS image is processed by an MS sensor, it becomes a low-resolution multispectral image (MS). A Gaussian filter is used to simulate this process, and the resulting Gaussian filter is the filter estimate that conforms to the MS sensor's point spread function (PSF). The estimated filter is then used to convolve the P image, and the extracted details are those missing from the low-resolution MS image. Numerous studies have experimentally demonstrated that this filter estimate has strong robustness; even with white noise added to the initial fused image, it does not affect detail extraction. Currently, most low-pass filter estimations use the correlation coefficient as a criterion, with the filtered low-resolution panchromatic P image as the reference. L and multispectral fitting intensity I L The filter parameters that maximize correlation are the optimal Gaussian filter parameters.

[0006] However, regardless of the linear regression weighted fitting low-resolution I LAlternatively, Gaussian filtering matching based on the correlation coefficient can be used to obtain a low-resolution panchromatic P. L These can only represent the linear correlation between two images. In fact, multispectral and panchromatic images exhibit both linear and nonlinear relationships. Furthermore, as mentioned above, an ideal low-resolution panchromatic image P′... L The spatial information features are consistent with the original multispectral image MS, but a low-resolution panchromatic P is obtained by Gaussian filtering based on the correlation coefficient. L In the case of spatial information features, the multispectral fitting strength I is not necessarily related to the spatial information features. L The spatial information features are consistent. Summary of the Invention

[0007] To address the problems existing in the prior art, this invention proposes a remote sensing image fusion method and system based on local variance image mutual information, which makes the fused image more conducive to the preservation of spectral features and reduces spatial distortion at the spatial scale.

[0008] To achieve the above objectives, the present invention adopts the following technical solution:

[0009] This invention provides a remote sensing image fusion method based on local variance image mutual information, comprising the following steps:

[0010] The initial fused image was obtained by using a multivariate linear regression component substitution fusion method.

[0011] Based on the initial fused image, with the goal of obtaining the optimal Gaussian filter, the similarity between the intensity components of the high-resolution multispectral image and the intensity components of the initial fused image is calculated.

[0012] The original high-resolution panchromatic image is filtered using the estimated optimal Gaussian filter to obtain a low-resolution panchromatic image, and then image fusion is performed on this basis.

[0013] Furthermore, the initial fused image is obtained by using a multiple linear regression component substitution fusion method, specifically including the following steps:

[0014] Step 1.1: Upsample the original low-resolution multispectral image to obtain the upsampled image MS. up The size remains consistent with the original high-resolution panchromatic image P;

[0015] Step 1.2, upsample the image MS up The multispectral intensity component I′ is obtained by band averaging. L ;

[0016] Step 1.3, upsample the image MS up The multispectral intensity component I was obtained using multiple linear regression. L ;

[0017] Step 1.4, based on I L Based on the correlation of each band of the multispectral image, the detail injection coefficient g of each multispectral band is calculated. i Then calculate the multispectral intensity component I. L The spatial detail image is obtained by comparing the difference between the original high-resolution panchromatic image P and the original image P.

[0018] Step 1.5, calculate the initial fused image HMS;

[0019] Step 1.6: Extract the intensity component I0 from the initial fused image HMS.

[0020] Furthermore, in step 1.3, the multispectral intensity component I L The calculation formula is as follows:

[0021]

[0022] In the formula, i represents the number of multispectral bands, MS iup This represents the upsampled image for each band, w i This indicates the weight of each band in the multispectral spectrum.

[0023] Furthermore, in step 1.4, the detail injection coefficients g for each band of the multispectral spectrum... i The calculation formula is as follows:

[0024] g i =COV(MS iup ,I L ) / VAR(I L (2)

[0025] In the formula, COV represents the covariance function, and VAR represents the variance function.

[0026] Furthermore, the calculation formula for the initial fused image HMS in step 1.5 is as follows:

[0027] HMS i =MS iup +g i (PI L (3)

[0028] In the formula, HMS is the fused high-resolution multispectral image.

[0029] Furthermore, based on the initial fused image, with the goal of obtaining the optimal Gaussian filter, the similarity between the intensity components of the high-resolution multispectral image and the intensity components of the initial fused image is calculated, specifically including the following steps:

[0030] Step 2.1, calculate the multispectral intensity component I′ LLocal variance image I′ δL This allows us to obtain the edge information of the image.

[0031] Step 2.2: Process the intensity component I0 using a Gaussian low-pass filter to obtain...

[0032] Step 2.3, Calculation Local variance image

[0033] Step 2.4, Calculation and I′ δL MI i This characterizes the similarity between the two in terms of spatial features;

[0034] Step 2.5: Repeat steps 2.2, 2.3, and 2.4, recording the mutual information MI. i The number of iterations m at which the maximum value is obtained yields the optimal Gaussian filter estimate H. Gm .

[0035] Furthermore, in step 2.2, the intensity component I0 is processed using a Gaussian low-pass filter to obtain... Represented as:

[0036]

[0037] In the formula, H G This represents a Gaussian low-pass filter.

[0038] Furthermore, the original high-resolution panchromatic image is filtered using the estimated optimal Gaussian filter to obtain a low-resolution panchromatic image. Image fusion is then performed on this low-resolution panchromatic image, specifically including the following steps:

[0039] Step 3.1, filter H Gm Applying this to the original high-resolution panchromatic image P yields a low-resolution panchromatic image P. GSF ;

[0040] Step 3.2, calculate the low-resolution panchromatic image P. GSF and MS of upsampled images in each band iup The injection coefficient g′ between i The optimized detail injection coefficients are obtained.

[0041] Step 3.3, calculate the low-resolution panchromatic image P. GSF The difference image D between the original high-resolution panchromatic image P and the original high-resolution panchromatic image P;

[0042] Step 3.4: Calculate the final fused image.

[0043] Furthermore, the calculation formula for the final fused image in step 3.4 is as follows:

[0044] P GSF =H Gm (P) (5)

[0045] g′ i =COV(MS iup ,P GSF ) / VAR(P GSF (6)

[0046] D = PP GSF (7)

[0047] HMS i =MS iup +g′ i D (8).

[0048] This invention provides a remote sensing image fusion system based on local variance image mutual information, comprising:

[0049] The initial fused image calculation module is used to perform fusion by replacing the components of a multiple linear regression method to obtain the initial fused image.

[0050] The optimal Gaussian filter estimation module is used to calculate the similarity between the intensity components of the high-resolution multispectral image and the intensity components of the initial fused image, with the goal of obtaining the optimal Gaussian filter based on the initial fused image.

[0051] The final image fusion calculation module is used to filter the original high-resolution panchromatic image using the estimated optimal Gaussian filter to obtain a low-resolution panchromatic image, and then perform image fusion on this basis.

[0052] Compared with the prior art, the present invention has the following advantages:

[0053] 1. Local variance images can describe the structural information of an image. Compared with intensity component images, this gradient domain-based local variance image is more suitable for measuring the similarity of two images in spatial features. The resulting low-resolution panchromatic image and the original low-resolution multispectral image have more consistent spatial information, so that the spatial information of the fused image is consistent with the original low-resolution multispectral image, reducing the spatial distortion caused by differences in spatial scale.

[0054] 2. The mutual information-based evaluation method replaces the correlation coefficient as the evaluation criterion for the similarity between two images. It considers both the linear relationship between multispectral and panchromatic images and the nonlinear relationship between them. Compared with multiple linear regression, it improves the correlation coefficient between the low-resolution panchromatic image and the original high-resolution panchromatic image.

[0055] 3. The optimal Gaussian filtering estimation method is applicable to the generation of low-resolution panchromatic images from various satellite remote sensing images. It has a certain degree of generalization and has improved the quality evaluation indicators of multiple image fusion. Attached Figure Description

[0056] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0057] Figure 1 This is one of the flowcharts of the remote sensing image fusion method based on local variance image mutual information according to an embodiment of the present invention;

[0058] Figure 2 This is the second flowchart of the remote sensing image fusion method based on local variance image mutual information according to an embodiment of the present invention;

[0059] Figure 3 This is a structural block diagram of a remote sensing image fusion system based on local variance image mutual information according to an embodiment of the present invention. Detailed Implementation

[0060] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0061] like Figure 1 and Figure 2 As shown, the remote sensing image fusion method based on local variance image mutual information in this embodiment includes the following steps:

[0062] Step S1: Use the multiple linear regression component replacement fusion method to perform fusion and obtain the initial fused image.

[0063] Step S101: Upsample the original low-resolution multispectral image to obtain the upsampled image MS. up And it maintains the same size as the original high-resolution panchromatic image P.

[0064] Step S102, upsample the image MS up The multispectral intensity component I′ is obtained by band averaging. L .

[0065] Step S103, upsample the image MS up The multispectral intensity component I was obtained using multiple linear regression. L Its spatial information characteristics remain consistent with the original low-resolution multispectral image MS, ensuring the cancellation of the spatial information of the original low-resolution multispectral image MS; simultaneously, the minimum variance constraint is used to reduce its spectral difference with the original high-resolution panchromatic image P. Multispectral intensity component I L The calculation formula is as follows:

[0066]

[0067] In the formula, i represents the number of multispectral bands, MS iup This represents the upsampled image for each band, w i This indicates the weight of each band in the multispectral spectrum.

[0068] Where w i The constraints are:

[0069] minG(w)=∑(∑ i (w i *MS iup -P)) 2 +5∑ i (max(0,-w i ) 2 (2)

[0070] Due to the coefficient w i Since it is non-negative, w can be increased using the Lagrange multiplier λ. i The non-negativity constraint can be obtained by gradient descent iterations to obtain w. i value.

[0071] Step S104, according to I L Based on the correlation of each band of the multispectral image, the detail injection coefficient g of each multispectral band is calculated. i Then calculate the multispectral intensity component I. L The spatial detail image is obtained by comparing the difference between the original high-resolution panchromatic image P and the original image P.

[0072] Considering the upsampled image MS up Global content characteristics and local contextual information of each band and the original high-resolution panchromatic image P; different multispectral bands do not share the same detail map, g i This represents the detail injection coefficient corresponding to the i-th band, used to adjust the proportion of injected spatial information to reduce the degree of spatial distortion. Detail injection coefficients g for each band of the multispectral array. i The calculation formula is as follows:

[0073] g i =COV(MSiup ,I L ) / VAR(I L (3)

[0074] In the formula, COV represents the covariance function, and VAR represents the variance function.

[0075] Step S105, calculate the initial fused image HMS, using the following formula:

[0076] HMS i =MS iup +g i (PI L (4)

[0077] In the formula, HMS represents the fused high-resolution multispectral image, and P and I... L The greater the correlation, the better the fusion quality.

[0078] Furthermore, the initial fused image HMS can also be expressed using the following formula:

[0079] HMS i =MS iup +g i (PP L (5)

[0080] Therefore, spatial detail images can also be obtained by comparing the original high-resolution panchromatic image P with its low-pass filtered low-frequency image P. L The difference is obtained. As can be seen from formula (5), the algorithm processes spectral and spatial information separately, which largely preserves the spectral diversity of the original multispectral image. However, because the original low-resolution multispectral image is upsampled (i.e., magnified), the upsampled image and the original low-resolution multispectral image will have local spatial differences, which can easily cause spatial distortion phenomena such as blurring and artifacts in the fused image. Therefore, it is clear how to obtain the low-resolution panchromatic image P L It is one of the key variables that determines the quality of fused images.

[0081] Step S106: Extract the intensity component I0 (mean value of each band of the initial fused image HMS) from the initial fused image HMS.

[0082] Step S2: Based on the initial fused image, with the goal of obtaining the optimal Gaussian filter, calculate the similarity between the intensity components of the high-resolution multispectral image and the intensity components of the initial fused image.

[0083] Step S201, calculate the multispectral intensity component I′ L Local variance image I′ δL The edge information of the image is obtained and used as a reference image for Gaussian filter estimation.

[0084] Step S202: Process the intensity component I0 using a Gaussian low-pass filter to obtain... Spatial fuzziness is achieved, and the calculation formula is as follows:

[0085]

[0086] In the formula, H G This represents a Gaussian low-pass filter.

[0087] Step S203, Calculate Local variance image

[0088] Suppose we have a grayscale image x(k,l) of size M*N, and a window size of [(2*n+1), (2*n+1)] (n is usually taken as 1), then we can calculate the local average value m of the image. x (i, j) represents the low-frequency components of the image:

[0089]

[0090] The local variance image is:

[0091]

[0092] Calculate according to formula (8) Local variance image Simultaneously, calculate the multispectral intensity component I′ L Local variance image I′ δL .

[0093] Step S204, calculate and I′ δL MI i This characterizes the similarity between the two in terms of spatial features.

[0094] Mutual information is a useful measure of information in information theory; it refers to the correlation between two sets of events. The mutual information of two variables X and Y is defined as follows:

[0095] MI(X,Y)=H(X)+H(Y)-H(X,Y) (9)

[0096] H(X)=-∑p(x)logp(x) (10)

[0097] H(X,Y)=-∑p(x,y)logp(x,y) (11)

[0098] Where H(X,Y) is the joint entropy of the two variables, and H(X) and H(Y) are the entropies of each variable; the mutual information of two images reflects the degree of mutual inclusion of information between them through their entropy and joint entropy. When the similarity between two images is higher or the overlap is larger, their correlation is greater, the joint entropy is smaller, that is, the mutual information is larger. Therefore, mutual information is widely used as a standard for image registration, especially medical image registration, feature selection and feature transformation in machine learning, to characterize the correlation and redundancy of variables, and a loss function in decision tree learning, etc. Unlike ordinary similarity measurement methods, mutual information can capture the nonlinear statistical correlation between variables. Calculated according to formulas (9), (10) and (11) and I′ δL MI i .

[0099] Step S205: Repeat steps S202, S203, and S204, recording the mutual information MI. i The number of iterations m at which the maximum value is obtained yields the optimal Gaussian filter estimate H. Gm .

[0100] When MI i The number of iterations m at which the maximum value is obtained is the optimal Gaussian filter estimate H. Gm .

[0101]

[0102] Step S3: Filter the original high-resolution panchromatic image using the estimated optimal Gaussian filter to obtain a low-resolution panchromatic image, and then perform image fusion on this basis.

[0103] Step S301, filter H Gm Applying the process to the original high-resolution panchromatic image P yields a low-resolution panchromatic image p. GSF The formula is: P GSF =H Gm (P) (13). In this way, its spatial information features are consistent with the original low-resolution multispectral image MS, which can ensure the cancellation of the spatial information of MS; at the same time, its spectral features are consistent with the original high-resolution panchromatic image P, which can ensure the cancellation of the spectral features of P.

[0104] Step S302: Calculate the low-resolution panchromatic image P GSF and MS of upsampled images in each band iup The injection coefficient g′ between i The optimized detail injection coefficients are obtained; P is used. GSF Replace I in formula (3) L Calculate g′ i The formula is: g′ i=COV(MS iup ,P GSF ) / VAR(P GSF (14).

[0105] Step S303: Calculate the low-resolution panchromatic image P GSF The difference image D between the original high-resolution panchromatic image P and the original high-resolution panchromatic image P is expressed by the formula: D = PP GSF (15).

[0106] Step S304: Calculate the final fused image.

[0107] The final fused image is obtained by calculating using formula (5): HMS i =MS iup +g′ i (PH Gm (P)) (16).

[0108] Therefore, the final fused image obtained in this embodiment has the same spatial information as the original low-resolution multispectral image and the same spectral characteristics as the original high-resolution panchromatic image, thus avoiding spatial distortion in the fused image and achieving good fusion quality.

[0109] Corresponding to the aforementioned remote sensing image fusion method based on local variance image mutual information, such as Figure 3 As shown, this embodiment also proposes a remote sensing image fusion system based on local variance image mutual information, including:

[0110] The initial fused image calculation module is used to perform fusion by replacing the components of a multiple linear regression method to obtain the initial fused image.

[0111] The optimal Gaussian filter estimation module is used to calculate the similarity between the intensity components of the high-resolution multispectral image and the intensity components of the initial fused image, with the goal of obtaining the optimal Gaussian filter based on the initial fused image.

[0112] The final image fusion calculation module is used to filter the original high-resolution panchromatic image using the estimated optimal Gaussian filter to obtain a low-resolution panchromatic image, and then perform image fusion on this basis.

[0113] It should be noted that, in this document, the terms “comprising,” “including,” or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0114] Finally, it should be noted that the above description is merely a preferred embodiment of the present invention and is only used to illustrate the technical solution of the present invention, and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention are included within the scope of protection of the present invention.

Claims

1. A remote sensing image fusion method based on local variance image mutual information, characterized in that, Includes the following steps: Step 1: Use the multiple linear regression component substitution fusion method to perform fusion and obtain the initial fused image. This includes the following steps: Step 1.1: Upsample the original low-resolution multispectral image to obtain the upsampled image MS. up The size remains consistent with the original high-resolution panchromatic image P; Step 1.2, upsample the image MS up The multispectral intensity component I′ is obtained by band averaging. L ; Step 1.3, upsample the image MS up The multispectral intensity component I was obtained using multiple linear regression. L ; Step 1.4, based on I L Based on the correlation of each band of the multispectral image, the detail injection coefficient g of each multispectral band is calculated. i Then calculate the multispectral intensity component I. L The spatial detail image is obtained by comparing the difference between the original high-resolution panchromatic image P and the original image P. Step 1.5, calculate the initial fused image HMS; Step 1.6: Extract the intensity component I0 from the initial fused image HMS; Step 2: Based on the initial fused image, with the goal of obtaining the optimal Gaussian filter, calculate the similarity between the intensity components of the high-resolution multispectral image and the intensity components of the initial fused image. This specifically includes the following steps: Step 2.1, calculate the multispectral intensity component I L The local variance image I′ δL This allows us to obtain the edge information of the image. Step 2.2: Process the intensity component I0 using a Gaussian low-pass filter to obtain... Step 2.3, Calculation Local variance image Step 2.4, Calculation and I′ δL MI i This characterizes the similarity between the two in terms of spatial features; Step 2.5: Repeat steps 2.2, 2.3, and 2.4, recording the mutual information MI. i The number of iterations m at which the maximum value is obtained yields the optimal Gaussian filter estimate H. Gm ; Step 3: Filter the original high-resolution panchromatic image using the estimated optimal Gaussian filter to obtain a low-resolution panchromatic image, and then perform image fusion on this basis.

2. The remote sensing image fusion method based on local variance image mutual information according to claim 1, characterized in that, Step 1.3 Multispectral intensity component I L The calculation formula is as follows: In the formula, i represents the number of multispectral bands, MS iup This represents the upsampled image for each band, w i This indicates the weight of each band in the multispectral spectrum.

3. The remote sensing image fusion method based on local variance image mutual information according to claim 2, characterized in that, Step 1.4: Detail injection coefficients g for each band of the multispectral spectrum i The calculation formula is as follows: g i =COV(MS iup ,I L ) / VAR(I L ) (2) In the formula, COV represents the covariance function, and VAR represents the variance function.

4. The remote sensing image fusion method based on local variance image mutual information according to claim 3, characterized in that, The calculation formula for the initial fused image HMS in step 1.5 is as follows: HMS i =MS iup +g i (P-I L ) (3) In the formula, HMS i This represents the fused high-resolution multispectral image of the i-th band.

5. The remote sensing image fusion method based on local variance image mutual information according to claim 1, characterized in that, In step 2.2, the intensity component I0 is processed using a Gaussian low-pass filter to obtain... Represented as: In the formula, H G This represents a Gaussian low-pass filter.

6. The remote sensing image fusion method based on local variance image mutual information according to claim 1, characterized in that, The original high-resolution panchromatic image is filtered using the estimated optimal Gaussian filter to obtain a low-resolution panchromatic image. Image fusion is then performed on this low-resolution panchromatic image, specifically including the following steps: Step 3.1, filter H Gm Applying this to the original high-resolution panchromatic image P yields a low-resolution panchromatic image P. GSF ; Step 3.2, calculate the low-resolution panchromatic image P. GSF and MS of upsampled images in each band iup Injection coefficient g between i ′, thus obtaining the optimized detail injection coefficients; Step 3.3, calculate the low-resolution panchromatic image P. GSF The difference image D between the original high-resolution panchromatic image P and the original high-resolution panchromatic image P; Step 3.4: Calculate the final fused image.

7. The remote sensing image fusion method based on local variance image mutual information according to claim 6, characterized in that, The formula for calculating the final fused image in step 3.4 is as follows: P GSF =H Gm (P) (5) g′ i =COV(MS iup ,P GSF ) / VAR(P GSF ) (6) D=P-P GSF (7) HMS i =MS iup +g′ i D (8)。 8. A remote sensing image fusion system based on local variance image mutual information, characterized in that, include: The initial fused image calculation module is used to perform fusion using a multivariate linear regression component substitution fusion method to obtain the initial fused image. Specifically, it includes: Step 1.1: Upsample the original low-resolution multispectral image to obtain the upsampled image MS. up The size remains consistent with the original high-resolution panchromatic image P; Step 1.2, upsample the image MS up The multispectral intensity component I′ is obtained by band averaging. L ; Step 1.3, upsample the image MS up The multispectral intensity component I was obtained using multiple linear regression. L ; Step 1.4, based on I L Based on the correlation of each band of the multispectral image, the detail injection coefficient g of each multispectral band is calculated. i Then calculate the multispectral intensity component I. L The spatial detail image is obtained by comparing the difference between the original high-resolution panchromatic image P and the original image P. Step 1.5, calculate the initial fused image HMS; Step 1.6: Extract the intensity component I0 from the initial fused image HMS; The optimal Gaussian filter estimation module, based on the initial fused image, calculates the similarity between the intensity components of the high-resolution multispectral image and the intensity components of the initial fused image, with the goal of obtaining the optimal Gaussian filter. Specifically, it includes: Step 2.1, calculate the multispectral intensity component I L The local variance image I′ δL This allows us to obtain the edge information of the image. Step 2.2: Process the intensity component I0 using a Gaussian low-pass filter to obtain... Step 2.3, Calculation Local variance image Step 2.4, Calculation and I′ δL MI i This characterizes the similarity between the two in terms of spatial features; Step 2.5: Repeat steps 2.2, 2.3, and 2.4, recording the mutual information MI. i The number of iterations m at which the maximum value is obtained yields the optimal Gaussian filter estimate H. Gm ; The final image fusion calculation module is used to filter the original high-resolution panchromatic image using the estimated optimal Gaussian filter to obtain a low-resolution panchromatic image, and then perform image fusion on this basis.

Images