Multi-source image fusion method and system based on subject and detail features

Abstract

The present invention discloses a multi-source image fusion method and system based on subject and detail features, and relates to the field of image processing technology. The method acquires an optical image and a SAR image to generate an optical image intensity component and a simulated SAR image; then, the optical image intensity component and the simulated SAR image are subjected to subject and detail feature decomposition, and the subject feature component and the detail feature component are fused to obtain a fused subject feature fusion result and a fused detail feature fusion result; finally, the subject feature fusion result and the detail feature fusion result are added to obtain a preliminary feature decomposition fusion image, and the feature decomposition fusion image is further quickly fused with the original optical image to obtain a final fused image. The present invention adopts different fusion strategies for the subject and detail feature components of the image, thereby reducing the fusion differences caused by different feature components of the image and improving the overall effect of image fusion. The final fused image has more vivid color information, achieving a more detailed and effective fusion processing effect.

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and in particular to a multi-source image fusion method and fusion system based on subject and detail features, which are mainly used for effectively fusing optical images and SAR images. Background Art

[0002] Optical images and SAR images have different imaging mechanisms, geometric characteristics, and radiation characteristics. When fusing optical images and SAR images, the main features and detail features are considered separately, and fusion strategies suitable for each are adopted. While retaining the unique information of each optical and SAR image, the introduction of unnecessary redundant information can be reduced, which is conducive to the effective integration of the information of the two, thereby more effectively serving subsequent related applications.

[0003] The model-based fusion method regards the fusion of optical images and SAR images as an image generation or image reconstruction problem. By establishing a mathematical model that describes the mapping relationship from the source image to the fusion result, or establishing a constraint relationship between the fusion result and the source image, the final fusion result is obtained by optimization. In order to obtain a better fusion effect, a probabilistic model and prior constraints can be introduced into the model, and the solution of the model will also be more complicated. The present invention proposes a modular fusion strategy for different feature components of the image, so that image fusion can better select the best strategy according to the characteristics of different images, and solve the problem of poor quality of fusion results due to large differences in multi-source image features. Summary of the Invention

[0004] The purpose of the present invention is to overcome the deficiencies of the prior art and provide a multi-source image fusion method and system based on subject and detail features for performing more detailed and effective fusion processing on multi-source images.

[0005] On the one hand, the present invention provides a multi-source image fusion method based on subject and detail features. The method acquires an optical image and a SAR image to generate an optical image intensity component and a simulated SAR image; then, the optical image intensity component and the simulated SAR image are decomposed into subject and detail features respectively, and the subject feature components and the detail feature components are fused respectively to obtain a fused subject feature fusion result and a fused detail feature fusion result; finally, the subject feature fusion result and the detail feature fusion result are added together to obtain a preliminary feature decomposition fusion image, and the feature decomposition fusion image is further quickly fused with the original optical image to obtain a final fused image.

[0006] On the other hand, the present invention further provides a multi-source image fusion system based on subject and detail features, wherein the image fusion system is implemented based on the steps of any of the aforementioned multi-source image fusion method embodiments, and the system includes:

[0007] A preprocessing module, used for acquiring optical images and SAR images, and preprocessing the optical images and SAR images;

[0008] Spectral correction module, used to generate optical image intensity components and generate simulated SAR images;

[0009] A decomposition module is used to decompose the grayscale image into main features and detail features, and obtain the main feature component and detail feature component of each image respectively;

[0010] The main feature fusion module is used to fuse the main feature components to obtain the main feature fusion result;

[0011] A detail feature fusion module is used to fuse detail feature components to obtain a detail feature fusion result;

[0012] The final fusion module is used to add the main feature fusion results and the detail feature fusion results to obtain a preliminary feature decomposition fusion image, and further quickly fuse the feature decomposition fusion image with the original optical image to obtain the final fusion image.

[0013] In summary, due to the adoption of the above-mentioned multi-source image fusion technology solution based on subject and detail features, the beneficial effects of the present invention are:

[0014] The multi-source image fusion method and system constructed by the present invention can perform a rapid IHS transformation on an optical image to obtain its intensity component image. The optical intensity component image and simulated SAR image are then decomposed into main and detail features. Different fusion rules are then designed for different feature components. The main feature components are fused based on a variational model, preserving the pixel intensity distribution of the optical image and structural information such as the main edges and contours of the SAR image. The detail feature components are fused based on feature similarity, preserving image information with richer detail texture features in the optical and SAR images. Finally, the resulting decomposed feature fused image is rapidly fused with the original optical image, resulting in a fused image with more vivid color information, achieving a more detailed and effective fusion processing effect for optical and SAR images. The present invention adopts different fusion strategies for the main and detail feature components of the image, reducing fusion differences caused by different image feature components and improving the overall image fusion effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 This is a flow chart of an optical and SAR remote sensing image fusion method disclosed in the present invention.

[0016] Figure 2 This is a comparison diagram of fused images according to an embodiment of the method of the present invention.

[0017] Figure 3 This is the specific operation of the present invention to decompose the main body and detail features of the remote sensing image.

[0018] Figure 4 A comparison chart of image fusion results provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0019] In order to enable those skilled in the art to better understand the technical solution of the present invention, the technical solution of the present invention is clearly and completely described below in conjunction with the accompanying drawings of the present invention. Based on the embodiments in this application, other similar embodiments obtained by ordinary technicians in this field without making any creative work should fall within the scope of protection of this application.

[0020] like Figure 1 As shown in FIG, a multi-source image fusion method based on subject-detail features is used to effectively fuse optical images and SAR images, including the following steps:

[0021] Step S01: Acquire an optical image and a SAR image, and preprocess the optical image and the SAR image.

[0022] In a preferred embodiment, the acquired optical image and SAR image are a multi-source image pair acquired by different sensors on the same scene, and the multi-source image pair is used as an input source image for image fusion processing.

[0023] The acquired optical image and SAR image are preprocessed, mainly by resampling and cropping the image pairs, so that the scale of the preprocessed optical image and the SAR image remain consistent, which facilitates the subsequent image fusion operation.

[0024] The pre-processing of step S01 is further illustrated by an example. Figure 2 As shown in Figure 1, where (a) is an optical image and (b) is a SAR image. The optical image and the SAR image are registered, then resampled to the same resolution and finally cropped into image blocks of fixed size. Opt And the corresponding SAR image I SAR As the input image, the method in the embodiment of the present invention is used to achieve fusion to obtain the final fused image I in Figure (c) fuse The multi-source image fusion method provided in the embodiment of the present application can make the final fused image I fuse It includes not only the pixel intensity distribution of optical images and structural information such as the main edges and contours of SAR images, but also image information with richer detailed texture features in optical images and SAR images.

[0025] Step S02: Calculate the multi-band mean features of each pixel in the optical image to generate the optical image intensity component; calculate the mean and standard deviation of the optical image intensity component and the SAR image respectively, and adjust the mean and standard deviation of the SAR image based on the mean and standard deviation of the optical image intensity component to generate a simulated SAR image. This makes the global statistical variables of the SAR image and the optical image similar, so as to maintain the characteristics of the optical image in the subsequent fusion.

[0026] Step S02 is further described in conjunction with the embodiment. The specific calculation of calculating the multi-band mean features of each pixel of the optical image to generate the optical image intensity component is shown in formula (1). The optical image intensity component I is the mean grayscale value of each pixel in the R, G, and B bands:

[0027]

[0028] Wherein, I is the intensity component of the optical image, R is the red channel of the optical image, G is the green channel of the optical image, and B is the blue channel of the optical image.

[0029] According to formula (2), the mean and standard deviation of the SAR image are adjusted to reduce the spectral distortion of the fusion result caused by the inconsistency of the spectral response functions of the intensity components of the SAR image and the optical image:

[0030]

[0031] where f sar is the original SAR image, is the adjusted simulated SAR image, μ sar , σ sar is the mean and standard deviation of the original SAR image, μ I , σ I are the mean and standard deviation of the optical image intensity component.

[0032] Step S03: the intensity component I of the optical image and the simulated SAR image The main and detail feature decomposition operations are performed separately to obtain the main feature and detail feature components of the corresponding image, which is convenient for designing different fusion strategies according to the respective characteristics of the two components to obtain the desired fusion effect.

[0033] In this embodiment, the main features refer to the contours and large edge structures of objects in the image, and the detail features refer to the textures and fine structures of objects in the image.

[0034] Step S03 is further described. This embodiment proposes a method for rapidly decomposing the subject / detail features of an image to effectively segment and enhance the different feature information of the image. This embodiment uses the Cartoon-Texture decomposition method; other decomposition methods may also be used in other embodiments, and the present invention is not limited thereto.

[0035] As shown in the example of formula (3), f is the intensity component of the optical image I and the simulated SAR image Grayscale image, u is the main feature component image, v is the detail feature component image:

[0036] f=u+v(3).

[0037] like Figure 3 The following is a detailed flow chart showing the decomposition of the main and detail features of a SAR image in this embodiment. For any pixel in the grayscale image, the change in local total variation (LTV) before and after passing through the nonlinear low-pass filter is calculated. Smaller changes are attributed to the main feature component, while larger changes are attributed to the detail feature component.

[0038] Step S301: define the local total variation LTV of the grayscale image f at pixel x σ (f)(x) and its relative attenuation rate λ(x) are shown in formulas (4) and (5):

[0039]

[0040] Among them L σ is a nonlinear low-pass filter, is the gradient operator.

[0041] Step S302: The relative attenuation rate λ(x) calculated in step S301 is applied to the original image f and the low-pass filtered image L. σ ×f is weighted averaged to obtain a set of fast low-pass filter and high-pass filter pairs, as shown in formulas (6) and (7), to obtain the main feature component u(x) and detail feature component v(x) corresponding to the grayscale image:

[0042] u(x)=w(λ(x))L σ ×f+(1-w(λ(x)))f(6)

[0043] v(x)=f(x)-u(x)(7)

[0044] Where w(λ(x)) is the weighted average function, which is defined as shown in formula (8):

[0045]

[0046] In step S04, the main feature components of the optical image and the main feature components of the SAR image are first fused based on the variational method, so that the main feature fusion result after the fusion of the main feature components has a similar pixel intensity distribution as the optical image, and also has similar main edge, large outline and other structural information as the SAR image.

[0047] Step S401: construct a variational fusion model of the main feature components of the optical image and the SAR image.

[0048] In order to make the ideal fusion result have less spectral distortion, the fusion result of the main feature component should have a similar pixel intensity distribution to the main feature component of the optical image. p The norm (p ≥ 1) is constructed based on the pixel intensity relationship between the fusion result and the main feature component of the optical image to obtain the first constraint, as shown in formula (9):

[0049]

[0050] At the same time, the gradient information of the main feature component of the SAR image also contains a large amount of information such as the main contour and surface cover type. Therefore, the image gradient difference between the fusion result and the main feature component of the SAR image can be used to construct the second constraint term (10):

[0051]

[0052] Where x represents the main feature fusion result, u o 、u s Represent the main characteristic components of optical image and SAR image respectively, denote the p and q norms respectively, is the gradient operator.

[0053] Combining the two constraints, we can get the target functional model (11):

[0054]

[0055] where λ0 is a positive parameter that controls the trade-off between the two terms.

[0056] Let y = xu s , formula (11) can be rewritten as a typical total variation regularization model optimization problem, as shown in the following formulas (12) and (13):

[0057]

[0058] Among them D h and D vare the discrete differential operators in the horizontal and vertical directions of the pixel, respectively. Formula (12) is a convex function, and there exists an optimal solution y. The first term can be regarded as a data fidelity term, and the second term can be regarded as a regularization term. * It is the minimum value operation of the functional of y, as shown in formula (14):

[0059]

[0060] Step S402, using the IRN (Iterative Weighted Norm) algorithm to solve the variational model equations (12) and (13) proposed in step S401, specifically includes:

[0061] S421, given the standard total variation regularization model (15) and its IRN algorithm solution:

[0062]

[0063] After multiple iterations of formula (15), the k+1th solution for y is obtained, as shown in formula (16):

[0064]

[0065] Where diag is the constructed diagonal matrix, f F,ε is a control operator.

[0066] S422, perform the following transformation on the standard total variation regularization model (15) given in step S421, setting A = E, b = u oi -u si , we get the formula (12) and (13) given in step S401, and use the IRN algorithm to iteratively solve to get the optimal solution y. Then the main feature fusion result u after the first fusion is obtained by formula (17):

[0067] u=y+u s (17).

[0068] Step S05 , performing a second fusion operation on the detail feature components of the optical image and the SAR image based on feature similarity, so that the detail feature fusion result after fusion based on the detail feature components retains richer detail image information in the optical image and the SAR image.

[0069] Step S501: Construct feature descriptors for detail feature components of optical and SAR images. Since Gabor filters have significant advantages in extracting image detail texture features and can express detail features in different directions and at different scales, this embodiment uses Gabor filters as feature descriptors for detail feature components, specifically including:

[0070] Step S511: Generate a Gabor filter. The Gabor filter is a complex sine function modulated by a Gaussian kernel function. It can extract spatial frequency (scale) and local detail features in multiple directions within a local area of ​​the image. The image is convolved with Gabor filters at five scales and eight directions. Gabor filters of different scales and directions are calculated according to the following formula (18):

[0071]

[0072] Among them, λ g is the wavelength parameter of the cosine function, θ is the strip direction, The phase parameter of the cosine function, γ is the spatial aspect ratio, and δ is the standard deviation of the Gaussian function.

[0073] Step S512, generating Gabor feature descriptors: performing convolution operations on the image using Gabor filters of different directions and scales generated in step S511 to form a multi-dimensional feature description vector, and performing downsampling operations on the image features to effectively reduce feature description redundancy. Then, performing normalization processing to form Gabor feature descriptors corresponding to the image detail feature components.

[0074] Step S502: Perform a similarity measurement on the Gabor feature descriptors of the detail feature components. Since the Gabor feature descriptors generated in step 501 reflect the pixel-by-pixel feature information of the image detail feature components, the differences between the Gabor feature descriptors can be used to measure the similarity between the detail feature components of the optical image and the SAR image, thereby adopting an appropriate strategy for the fusion of the detail feature components.

[0075] The KL (Kullback-Leibler) divergence, also known as relative entropy, is an asymmetric measure of the difference between two probability distributions. In information theory, the KL divergence is the difference between the information entropies of the two probability distributions. In this embodiment, the KL divergence is used to measure the difference between Gabor feature descriptions. By considering the probability distribution of the Gabor feature description vector at a pixel x of the detail feature component of the optical image and the SAR image as P and Q, respectively, the KL divergence value is obtained as follows (19):

[0076]

[0077] Since the KL divergence value does not satisfy the symmetry, that is, KL[P||Q]≠KL[Q||P], a new similarity metric SMV is constructed by calculating the mean of KL[P||Q] and KL[Q||P]. The smaller the SMV value, the more similar the detail features at the pixel are. The specific calculation of SMV is shown in formula (20):

[0078]

[0079] For image pixels with similar detail features, the fusion result should be obtained by combining the detail feature components of the optical image and the SAR image. For image pixels with large differences in detail features, the image component with richer detail feature information should be selected. Therefore, the similarity measure SMV of all pixels in the image is averaged to obtain SMV. mean , SMV mean It is used as a threshold to determine the size of the feature difference between the image detail feature components: when the detail feature similarity of a certain pixel is less than the threshold, the value of the detail feature fusion result at that pixel is obtained by the average value of the optical image and the SAR image pixels. Otherwise, the gradient amplitude of the two images is calculated, and the grayscale value of the image pixel with the larger gradient amplitude is taken as the grayscale value of the pixel at the corresponding position of the detail feature fusion result. The detail feature component fusion process based on similarity measurement is shown in formulas (21) and (22):

[0080]

[0081]

[0082] Where F(m,n) represents the detail feature component at the midpoint (m, n) of the detail feature fusion result v, V O (m,n) and V S (m, n) represent the pixel grayscale values ​​of the optical image and SAR image at point (m, n), respectively. O (m,n) and G S (m, n) represents the gradient amplitude of the optical image and the SAR image at point (m, n), respectively.

[0083] In step S06, the main feature fusion result after the main feature component fusion and the detail feature fusion result after the detail feature component fusion are added together to obtain a preliminary feature decomposition fusion image, and the feature decomposition fusion image is further fast-fused with the original optical image to obtain a final fusion image.

[0084] Steps S04 and S05 respectively give the calculation process of the fusion results u and v of the main feature component and the detail feature component. The fusion results u and v of the two components are added together to obtain the preliminary fusion result I of the optical image intensity component and the simulated SAR image. f , as shown in formula (23):

[0085] I f =u+v(23),

[0086] The intensity component of the optical image is combined with the preliminary fusion image I of the simulated SAR image fPerform fast IHS fusion transformation with the original optical image I according to formula (24) to obtain the final fused image:

[0087]

[0088] In formula (24), R f , G f 、B f The final fused image I fuse The three band components are combined to obtain the final fused image I fuse :

[0089] I fuse =R f +G f +B f .

[0090] In another embodiment, the present invention further provides a multi-source image fusion system based on subject and detail features, wherein the image fusion system is implemented based on the steps of any of the aforementioned multi-source image fusion method embodiments, and the system includes:

[0091] A preprocessing module, used for acquiring optical images and SAR images, and preprocessing the optical images and SAR images;

[0092] The spectrum correction module is used to generate an optical image intensity component based on the multi-band mean characteristics of each pixel in the optical image, and adjust the mean and standard deviation of the SAR image based on the mean and standard deviation of the optical image intensity component to generate a simulated SAR image;

[0093] A decomposition module is used to decompose the intensity component of the optical image and the grayscale image of the simulated SAR image into main features and detail features, respectively, to obtain the main feature component and detail feature component of each image;

[0094] The main feature fusion module is used to fuse the main feature component of the optical image intensity component and the main feature component of the simulated SAR image to obtain a main feature fusion result;

[0095] A detail feature fusion module is used to fuse the main feature component of the optical image intensity component and the detail feature component of the simulated SAR image to obtain a detail feature fusion result;

[0096] The final fusion module is used to add the main feature fusion results and the detail feature fusion results to obtain a preliminary feature decomposition fusion image, and further quickly fuse the feature decomposition fusion image with the original optical image to obtain the final fusion image.

[0097] As an experimental comparison, the present invention further compares the image fusion method based on subject and detail features in this application with the six classic fusion methods (IHS, LP, HPF, CVT, DWT, NSCT) to obtain the fusion result comparison diagram as shown in the figure below. Figure 4 It can be seen that the fusion method based on subject and detail features of the present invention not only retains important ground objects (such as power towers) in SAR images, but also does not cause distortion of the color information of optical images, and effectively integrates the image feature information of the two.

[0098] The above describes a specific embodiment of the present invention. The multi-source image fusion method and system constructed by the present invention can perform a rapid IHS transform on an optical image to obtain its intensity component image. The optical intensity component image and simulated SAR image are then decomposed into principal and detail features. Different fusion rules are then designed for different feature components. Principal feature components are fused based on a variational model, preserving both the pixel intensity distribution of the optical image and structural information such as principal edges and contours of the SAR image. Detail feature components are fused based on feature similarity, preserving richer image information representing detailed texture features in both optical and SAR images. Finally, the resulting decomposed feature-fused image is rapidly fused with the original optical image through IHS, resulting in a fused image with more vivid color information, achieving more detailed and effective fusion processing for optical and SAR images. By employing different fusion strategies for the principal and detail feature components of an image, the present invention reduces fusion discrepancies caused by different image feature components and improves the overall image fusion effect.

[0099] The present invention is not limited to the aforementioned specific embodiments, but extends to any new features or any new combination disclosed in this specification, as well as any new method or process steps or any new combination disclosed.

Claims

1. A multi-source image fusion method based on subject and detail features, characterized in that: The following steps are involved: Step S01, acquiring an optical image and a SAR image, and preprocessing the optical image and the SAR image; Step S02, calculating the multi-band mean feature of each pixel of the optical image to generate the optical image intensity component; The mean and standard deviation of the optical image intensity component and the SAR image are calculated respectively, and the mean and standard deviation of the SAR image are adjusted based on the mean and standard deviation of the optical image intensity component to generate a simulated SAR image. Step S03, performing main feature decomposition and detail feature decomposition on the optical image intensity component and the simulated SAR image respectively to obtain main feature components and detail feature components of the corresponding images; Step S04, performing a first fusion of the main feature components of the optical image and the main feature components of the SAR image based on a variational method to obtain a fused main feature fusion result; Step S05, performing a second fusion of the detail feature components of the optical image and the detail feature components of the SAR image based on feature similarity to obtain a fused detail feature fusion result; In step S06, the main feature fusion result after the main feature component fusion and the detail feature fusion result after the detail feature component fusion are added together to obtain a preliminary feature decomposition fusion image, and the feature decomposition fusion image is further quickly fused with the original optical image to obtain a final fusion image.

2. The multi-source image fusion method based on subject and detail features according to claim 1, characterized in that: The optical image and the SAR image are different types of multi-source image pairs acquired by using different types of sensors for the same scene.

3. The multi-source image fusion method based on subject and detail features according to claim 1, characterized in that: The preprocessing is to process the optical image and the SAR image so that the resolution and size of the two are the same.

4. The multi-source image fusion method based on subject and detail features according to claim 1, characterized in that: The step S02 includes: According to formula (1), the multi-band mean feature of each pixel of the optical image is calculated to generate the optical image intensity component I: Where I is the intensity component of the optical image, R is the red channel of the optical image, G is the green channel of the optical image, and B is the blue channel of the optical image; Adjust the mean and standard deviation of the SAR image according to formula (2): where f sar is the original SAR image, is the adjusted simulated SAR image, μ sar , σ sar is the mean and standard deviation of the original SAR image, μ I , σ I are the mean and standard deviation of the optical image intensity component I.

5. The multi-source image fusion method based on subject and detail features according to claim 1, characterized in that: The process of decomposing the main body and detail features in step S03 is as follows: Step S301: define the local total variation LTV of the grayscale image f at pixel x σ (f)(x) and its relative attenuation rate λ(x) are shown in formulas (4) and (5): Among them L σ is a nonlinear low-pass filter, is the gradient operator; Step S302: The relative attenuation rate λ(x) is applied to the original image f and the low-pass filtered image L. σ ×f is weighted averaged to obtain a set of fast low-pass filter and high-pass filter pairs, as shown in formulas (6) and (7), to obtain the main feature component u(x) and detail feature component v(x) corresponding to the grayscale image: u(x)=w(λ(x))L σ ×f+(1-w(λ(x)))f(6) v(x)=f(x)-u(x)(7) Where w(λ(x)) is the weighted average function, which is defined as shown in formula (8):

6. The multi-source image fusion method based on subject and detail features according to claim 1, characterized in that: The step S04 includes: Step S401, constructing a variational fusion model of the main feature components of the optical image and the SAR image; Utilize l p The norm is constructed based on the pixel intensity relationship between the fusion result and the main feature component of the optical image to obtain the first constraint term, as shown in formula (9): The second constraint term (10) is constructed using the image gradient difference between the fusion result and the main feature component of the SAR image: Where x represents the main feature fusion result, u o 、u s Represent the main characteristic components of optical image and SAR image respectively, denote the p and q norms respectively, is the gradient operator; Combining the two constraints, we can get the target functional model (11): Where λ0 is a positive parameter that controls the trade-off between the two terms; Let y = xu s , rewrite formula (11) into a typical total variation regularization model optimization problem, as shown in the following formulas (12) and (13): Among them D h and D v are the discrete differential operators in the horizontal and vertical directions of the pixel respectively; Formula (12) is a convex function, and there exists an optimal solution y. The first term can be regarded as a data fidelity term, and the second term can be regarded as a regularization term; y * It is the minimum value operation of the functional of y, as shown in formula (14): Step S402, using the iterative weighted norm method to solve the variational model equations (12) and (13), specifically includes: S421, given the standard total variation regularization model (15) and its solution method: After multiple iterations of formula (15), the k+1th solution for y is obtained, as shown in formula (16): Where diag is the constructed diagonal matrix, f F,ε is a control operator; S422, make the following transformation on the standard total variation regularization model (15), let A = E, b = u oi -u si , we get the formula (12) and (13) given in step S401, and we can get the optimal solution y by iterative solution. Then the main feature fusion result u after the first fusion is obtained by formula (17): u=y+u s (17) 7. The multi-source image fusion method based on subject and detail features according to claim 5, characterized in that: The step S05 includes: Step S501, constructing feature descriptors of detail feature components of optical images and SAR images; Step S502: perform similarity measurement on the feature descriptors of the detail feature components.

8. The multi-source image fusion method based on subject and detail features according to claim 7, characterized in that: The step S501 specifically further includes: Step S511, generate Gabor filters: perform convolution operations on the image at 5 scales and 8 directions of Gabor filters. Gabor filters of different scales and directions are calculated according to the following formula (18): Among them, λ g is the wavelength parameter of the cosine function, θ is the strip direction, The phase parameter of the cosine function, γ is the spatial aspect ratio, and δ is the standard deviation of the Gaussian function; Step S512, generating Gabor feature descriptors: performing convolution operations on the image using Gabor filters of different directions and scales generated in step S511 to form a multi-dimensional feature description vector, and performing downsampling operations on the image features to effectively reduce feature description redundancy. Then, performing normalization processing to form Gabor feature descriptors corresponding to the image detail feature components.

9. The multi-source image fusion method based on subject and detail features according to claim 7, characterized in that: The step S502 specifically includes: The KL divergence is used to measure the difference between feature descriptions. The probability distribution of the feature description vector at a certain pixel x of the detail feature component of the optical image and the SAR image is regarded as P and Q respectively, and the KL divergence value is obtained as follows (19): Since the KL divergence value does not satisfy the symmetry, that is, KL[P||Q]≠KL[Q||P], a new similarity measure SMV is constructed by calculating the mean of KL[P||Q] and KL[Q||P], as shown in formula (20): The similarity measure SMV of all pixels in the image is averaged to obtain SMV mean , SMV mean It is used as a threshold to judge the size of the feature difference between the image detail feature components: when the detail feature similarity of a certain pixel is less than the threshold, the value of the detail feature fusion result at this pixel is obtained by the average value of the optical image and the SAR image pixels; otherwise, the gradient amplitude of the two images is calculated, and the grayscale value of the image pixel with the larger gradient amplitude is taken as the grayscale value of the pixel at the corresponding position of the detail feature fusion result; The fusion process of detail feature components based on similarity measurement is shown in formulas (21) and (22): Where F(m,n) represents the detail feature component at the midpoint (m,n) of the detail feature fusion result v, V O (m,n) and V S (m,n) represent the pixel grayscale values ​​of the optical image and SAR image at point (m,n), respectively. O (m,n) and G S (m,n) represents the gradient amplitude of the optical image and SAR image at point (m,n), respectively.

10. A multi-source image fusion system based on subject and detail features, characterized in that: The system comprises: A preprocessing module is used to acquire optical images and SAR images and preprocess the optical images and SAR images; The spectrum correction module is used to generate an optical image intensity component based on the multi-band mean characteristics of each pixel in the optical image, and adjust the mean and standard deviation of the SAR image based on the mean and standard deviation of the optical image intensity component to generate a simulated SAR image; A decomposition module is used to decompose the intensity component of the optical image and the grayscale image of the simulated SAR image into main features and detail features, respectively, to obtain the main feature component and detail feature component of each image; The main feature fusion module is used to fuse the main feature component of the optical image intensity component and the main feature component of the simulated SAR image to obtain a main feature fusion result; A detail feature fusion module is used to fuse the detail feature components of the optical image intensity component and the detail feature components of the simulated SAR image to obtain a detail feature fusion result; The final fusion module is used to add the main feature fusion results and the detail feature fusion results to obtain a preliminary feature decomposition fusion image, and further quickly fuse the feature decomposition fusion image with the original optical image to obtain the final fusion image.

Images