A remote sensing image fusion method based on multi-scale morphological gradient adaptive simplified pulse coupled neural network model

Abstract

The invention belongs to the field of remote sensing image fusion, and specifically relates to a remote sensing image fusion method based on a multi-scale morphological gradient adaptive simplified pulse coupled neural network model; the method comprises: acquiring a synthetic aperture radar image and a multi-spectral image; performing Gaussian curvature filter decomposition and Gaussian filter decomposition on the synthetic aperture radar image respectively to obtain a small-scale image, a large-scale image and a base-scale image; performing color saturation intensity transformation on the multi-spectral image to obtain an intensity image; fusing the base-scale image and the intensity image to obtain an approximate image; fusing the approximate image, the small-scale image and the large-scale image to obtain a fused intensity image; reconstructing the fused intensity image to obtain a fused remote sensing image; the invention combines the scale morphological gradient with the adaptive simplified pulse coupled neural network to avoid the block effect and the phenomenon of local blurring of the image.

Description

Technical Field

[0001] The invention belongs to the field of remote sensing image fusion, and in particular relates to a remote sensing image fusion method based on a multi-scale morphological gradient adaptive simplified pulse coupled neural network model. Background Art

[0002] With the continuous development of remote sensing application technology, the requirements for remote sensing image quality in fields such as environmental protection, weather forecasting, military reconnaissance, smart agriculture, and marine applications are constantly increasing. Remote sensing images from a single sensor can no longer meet current application needs. Remote sensing image fusion is to generate multi-modal images by integrating single-modal images acquired by multi-source sensors, removing redundant information, adding complementary information, and making up for the lack of information features of single-modal images, thereby improving the integrity and richness of information details of the fused image, reflecting the different image features of heterogeneous sensors in the fused image, providing higher quality remote sensing images, and helping to better interpret the image. It can be seen that remote sensing image fusion technology has great research value. Therefore, remote sensing image fusion occupies an important position in the current field of remote sensing application technology.

[0003] In the field of remote sensing image fusion, most of the current fusion algorithms can be divided into two types: algorithms based on spatial domain transformation and algorithms based on transform domain transformation.

[0004] In the image fusion process, most of the algorithms based on spatial domain transformation use grayscale values ​​and weighting coefficients as fusion strategies, which have low time complexity. However, if the pixel difference between the two images is too large, the image contrast will be reduced, causing image edge blur. For example, the component replacement algorithm, which mostly separates the components to be fused from the multispectral image, such as replacing the content of the corresponding part of the synthetic aperture radar image with the content of the intensity space or the main component, and then reconstructs the image using the inverse transformation. Due to its relatively simple model structure, it can achieve spatial resolution fusion for remote sensing images, but the band width of the synthetic aperture radar image cannot be completely covered by the multispectral image, resulting in spectral distortion.

[0005] The core idea of ​​transform domain-based transform algorithms is to choose different decomposition strategies and different fusion strategies. For example, the multi-scale transform algorithm decomposes the image into sub-images of different scales, merges the sub-images of corresponding scales according to the characteristics of different bands using different fusion rules, and then reconstructs the image using inverse transform. This type of algorithm can achieve effective fusion in the spectral domain, but the spatial resolution of the fused image is low. Summary of the invention

[0006] In view of the shortcomings of the prior art, the present invention proposes a remote sensing image fusion method based on a multi-scale morphological gradient adaptive simplified pulse coupled neural network model, the method comprising:

[0007] S1: Acquire synthetic aperture radar images and multispectral images;

[0008] S2: Perform Gaussian curvature filter decomposition and Gaussian filter decomposition on the synthetic aperture radar image to obtain a small-scale image, a large-scale image and a base-scale image;

[0009] S3: Perform color saturation intensity transformation on the multispectral image to obtain an intensity image;

[0010] S4: An adaptive simplified pulse coupled neural network model based on multi-scale morphological gradients is used to fuse the base scale image and the intensity image to obtain an approximate image;

[0011] S5: Fusing the approximate image, the small-scale image and the large-scale image to obtain a fused intensity image;

[0012] S6: Reconstruct the fused intensity image using the inverse color saturation intensity transform to obtain a fused remote sensing image.

[0013] Preferably, the formula for performing Gaussian curvature filtering decomposition on synthetic aperture radar image data is:

[0014] R(i,j)=I(i,j)+min{|d n |,n=1,…,8}

[0015] d1=(I(i-1,j)+I(i+1,j)) / 2-I(i,j)

[0016] d2=(I(i,j-1)+I(i,j+1)) / 2-I(i,j)

[0017] d3=(I(i-1,j-1)+I(i+1,j+1)) / 2-I(i,j)

[0018] d4=(I(i-1,j+1)I(i+1,j-1)) / 2-I(i,j)

[0019] d5=(I(i-1,j)+I(i,j-1)+I(i-1,j-1)) / 3-I(i,j)

[0020] d6=(I(i-1,j)+I(i,j+1)+I(i-1,j+1)) / 3-I(i,j)

[0021] d7=(I(i,j-1)+I(i+1,j)+I(i+1,j-1)) / 3-I(i,j)

[0022] d8=(I(i,j+1)+I(i+1,j)+I(i+1,j+1)) / 3-I(i,j)

[0023] Where R(i,j) represents the Gaussian curvature filter decomposition result of pixel point (i,j), I(i,j) represents the pixel point (i,j) in the synthetic aperture radar image I, and d n Represents the filtering result of the nth window.

[0024] Preferably, the formula for performing Gaussian filtering decomposition on synthetic aperture radar image data is:

[0025]

[0026] Where I(a,b) represents the Gaussian filter decomposition result of the pixel point (a,b), and σ represents the standard deviation.

[0027] Preferably, the process of obtaining the small-scale image, the large-scale image and the base-scale image includes: performing Gaussian curvature filtering decomposition on the synthetic aperture radar image to obtain a first filtered image; performing Gaussian filtering decomposition on the synthetic aperture radar image to obtain a second filtered image and use it as a base-scale image; performing a subtraction operation on the first filtered image and the second filtered image to obtain a large-scale image; and performing a subtraction operation on the synthetic aperture radar image and the first filtered image to obtain a small-scale image.

[0028] Preferably, the process of constructing the adaptive simplified pulse coupled neural network model based on multi-scale morphological gradient includes:

[0029] Setting adaptive parameters, including standard deviation relationship value of input image, connection strength, amplitude of dynamic threshold and attenuation factor of dynamic threshold; obtaining adaptive simplified pulse coupled neural network model according to the adaptive parameters;

[0030] A multi-scale morphological gradient model is used to extract the gradient features of the base scale image and the gradient features of the intensity image respectively;

[0031] According to the gradient features of the base scale image and the gradient features of the intensity image, a multi-scale morphological gradient operator of the base scale image and a multi-scale morphological gradient operator of the intensity image are obtained respectively;

[0032] The adaptive simplified pulse coupled neural network model is modulated by using the multi-scale morphological gradient operator of the base scale image and the multi-scale morphological gradient operator of the intensity image, and an adaptive simplified pulse coupled neural network model based on multi-scale morphological gradient is obtained.

[0033] Furthermore, the adaptive parameters are:

[0034] α F =log(1 / σ(S))

[0035]

[0036]

[0037]

[0038] Among them, α F represents the standard deviation relationship value of the input image, σ(S) represents the standard deviation of the input image, S represents the pixel value of the input image, β ij ′ represents the connection strength, m represents the number of neurons in the region, and x kl represents the gray value of the neighborhood neuron of the central pixel, represents the mean gray value of neurons in the region, (k, l) represents the neighborhood pixel of the central pixel, V E represents the amplitude of the dynamic threshold, β′ represents the standard deviation of the input image, and α E represents the attenuation factor of the dynamic threshold, and S′ represents the normalized Otsu threshold.

[0039] Furthermore, the formula for extracting the gradient features of an image using a multi-scale morphological gradient model is:

[0040]

[0041] Among them, G t (x,y) represents the gradient feature, t represents the number of scales, I)x,y) represents the input image, g t (x,y) represents the structural element, represents the dilation operation, and Θ represents the erosion operation.

[0042] Furthermore, the formula of the multi-scale morphological gradient operator is obtained as follows:

[0043]

[0044] Among them, M(x,y) represents the multi-scale morphological gradient operator, w t represents the weight of the tth scale, G t (x, y) represents the gradient feature of the tth scale, and n represents the total number of scales.

[0045] Furthermore, the formula for modulating the adaptive simplified pulse coupled neural network model is:

[0046]

[0047]

[0048] Where M1 represents the multi-scale morphological gradient operator of the base scale image, represents the connection strength of the base scale image, M2 represents the multi-scale morphological gradient operator of the intensity image, Represents the connection strength of the intensity image.

[0049] Furthermore, the formula of the adaptive simplified pulse coupled neural network model is:

[0050]

[0051]

[0052] Among them, E ij (n) represents the dynamic threshold, V E Represents the amplitude of the dynamic threshold, Y ij (n) represents the neuron state judgment parameter, U ij (n) represents an internal activity item, represents the pixel value of the base scale image at coordinate (i, j), represents the connection strength of the base scale image, W ijkl represents the inverse of the Euclidean distance between adjacent neurons, represents the pixel value of the intensity image at coordinate (i, j), represents the connection strength of the intensity image, N ij represents the local area centered at coordinate (i, j), Y kl (n-1) represents the state judgment parameter of the neighboring neuron, and n represents the number of iterations.

[0053] The beneficial effects of the present invention are as follows: the present invention performs multi-scale decomposition of an image through a hybrid decomposition algorithm, extracts small-scale images, large-scale images and base-scale images while retaining edge detail information of the image; adopts an adaptive simplified pulse coupled neural network to reduce coupling, simplify the model structure and the number of parameters, and facilitates parameter acquisition through the setting of adaptive parameters, and no longer requires a large amount of experimental experience to acquire, thereby improving the efficiency of image fusion; combines scale morphological gradients with an adaptive simplified pulse coupled neural network, and utilizes multi-scale morphological gradients to extract multi-scale image features, thereby avoiding the block effect and local image blurring caused by the insensitivity of the adaptive simplified pulse neural network to image edge information when used alone. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 It is a flow chart of the remote sensing image fusion method based on the adaptive simplified pulse coupled neural network model of multi-scale morphological gradient in the present invention;

[0055] Figure 2 It is a schematic diagram of the fusion process of synthetic aperture radar image and multispectral image in the present invention. DETAILED DESCRIPTION

[0056] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0057] The present invention proposes a remote sensing image fusion method based on a multi-scale morphological gradient adaptive simplified pulse coupled neural network model. Figure 1 As shown, the method includes:.

[0058] S1: Acquire synthetic aperture radar images and multispectral images;

[0059] S2: Perform Gaussian curvature filter decomposition and Gaussian filter decomposition on the synthetic aperture radar image to obtain a small-scale image, a large-scale image and a base-scale image;

[0060] S3: Perform color saturation intensity transformation on the multispectral image to obtain an intensity image;

[0061] S4: An adaptive simplified pulse coupled neural network model based on multi-scale morphological gradients is used to fuse the base scale image and the intensity image to obtain an approximate image;

[0062] S5: Fusing the approximate image, the small-scale image and the large-scale image to obtain a fused intensity image;

[0063] S6: Reconstruct the fused intensity image using inverse color saturation intensity transform to obtain a fused remote sensing image.

[0064] like Figure 2 As shown in the figure, the specific process of fusing synthetic aperture radar images and multispectral images using the remote sensing image fusion method based on the adaptive simplified pulse coupled neural network model of multi-scale morphological gradient is as follows:

[0065] Gaussian curvature filtering is an effective edge-preserving filter, which is superior to rolling guided filtering in image edge preservation and can preserve small-scale image edge structure information. The parameter-free characteristics of Gaussian curvature filtering are superior to bilateral filtering and image guided filtering. The hybrid decomposition algorithm combining Gaussian curvature filtering and Gaussian filtering uses the edge preservation of Gaussian curvature waves and the smoothness of Gaussian filtering to keep the edge information in the local area unchanged when smoothing the image. The hybrid decomposition algorithm combining Gaussian curvature filtering and Gaussian filtering is used to decompose the image at multiple scales, extract small-scale images, large-scale images and base-scale images, and retain the edge detail information of the image. The process is as follows:

[0066] Acquire a synthetic aperture radar image and a multispectral image; perform Gaussian curvature filtering decomposition on the synthetic aperture radar image data to obtain a first filtered image; perform Gaussian filtering decomposition on the synthetic aperture radar image data to obtain a second filtered image and use it as a base scale image; perform a subtraction operation on the first filtered image and the second filtered image to obtain a large-scale image; perform a subtraction operation on the synthetic aperture radar image and the first filtered image to obtain a small-scale image.

[0067] Gaussian curvature filtering does not need to calculate the Gaussian curvature of the image by adjusting each pixel value. It uses a sliding window to filter from the upper left corner to the lower right corner of the image. The formula for Gaussian curvature filtering decomposition of synthetic aperture radar image data is:

[0068] R(i,j)=I(i,j)+min{|d n |,n=1,…,8}

[0069] d1=(I(i-1,j)+I(i+1,j)) / 2-I(i,j)

[0070] d2=(I(i,j-1)+I(i,j+1)) / 2-I(i,j)

[0071] d3=(I(i-1,j-1)+I(i+1,j+1)) / 2-I(i,j)

[0072] d4=(I(i-1,j+1)+I(i+1,j-1)) / 2-I(i,j)

[0073] d5=(I(i-1,j)+I(i,j-1)+I(i-1,j-1)) / 3-I(i,j)

[0074] d6=(I(i-1,j)+I(i,j+1)+I(i-1,j+1)) / 3-I(i,j)

[0075] d7=(I(i,j-1)+I(i+1,j)+I(i+1,j-1)) / 3-I(i,j)

[0076] d8=(I(i,j+1)+I(i+1,j)+I(i+1,j+1)) / 3-I(i,j)

[0077] Where R(i,j) represents the Gaussian curvature filter decomposition result of pixel point (i,j), I(i,j) represents the pixel point (i,j) in the synthetic aperture radar image I, and d n Represents the filtering result of the nth window.

[0078] The formula for Gaussian filtering decomposition of synthetic aperture radar image data is:

[0079]

[0080] Where I(a,b) represents the Gaussian filter decomposition result of the pixel point (a,b), and σ represents the standard deviation.

[0081] By transforming the multispectral image from RGB space to HSI color space using color saturation intensity transformation, the intensity image of the multispectral image can be extracted. The formula is:

[0082]

[0083]

[0084]

[0085] Among them, H represents the wavelength of the image color, that is, the hue component, I represents the intensity component of the image, S represents the depth of the image color, that is, the saturation component, v1 represents the first intermediate variable, v2 represents the second intermediate variable, R, G, and B represent the first color channel, the second color channel, and the third color channel, respectively.

[0086] Each neuron of the adaptive simplified pulse coupled neural network represents a pixel point, and does not require learning or training process. It can better extract effective information from complex backgrounds and has the synchronous pulse emission and global coupling of neurons. The construction process of the adaptive simplified pulse coupled neural network model based on multi-scale morphological gradient is as follows:

[0087] The simplified pulse coupled neural network model has 5 adjustable parameters, and the value of each parameter will affect the experimental results. The optimal parameter setting of the model needs to be manually obtained through a large number of experiments, and the characteristics of different images are also different. A large number of repeated experiments are required to obtain the optimal parameters, which takes a lot of time. The adaptive parameters set by the present invention can avoid wasting time by repeated experiments to find the optimal parameters and improve efficiency. The adaptive parameters include the standard deviation relationship value of the input image, the connection strength, the amplitude of the dynamic threshold, and the attenuation factor of the dynamic threshold; the adaptive parameters of the present invention are set as follows:

[0088] α F =log(1 / σ(S))

[0089]

[0090]

[0091]

[0092] Among them, α Frepresents the standard deviation relationship value of the input image, σ(S) represents the standard deviation of the input image, S represents the pixel value of the input image, β ij ′ represents the connection strength, m represents the number of neurons in the region, (k, l) represents the neighborhood pixels of the central pixel, V E represents the amplitude of the dynamic threshold, β′ represents the standard deviation of the input image, and α E represents the attenuation factor of the dynamic threshold, S′ represents the normalized Otsu threshold, and x kl represents the gray value of the neighborhood neuron of the central pixel, Represents the mean gray value of neurons in the region.

[0093] Calculate the connection strength of the base scale image and the connection strength of the intensity image: if x kl is the gray value of the neighboring pixel points in a certain area of ​​the base scale image, is the mean value of pixels in a certain area of ​​the base-scale image, then the connection strength of the base-scale image can be calculated; if x kl is the gray value of the neighboring pixel in a certain area of ​​the intensity image, is the mean grayscale value of pixels in a certain area of ​​the intensity image, then the connection strength of the intensity image can be calculated.

[0094] According to the adaptive parameters, the adaptive simplified pulse coupled neural network model is obtained, which is defined as:

[0095]

[0096]

[0097] Among them, E ij (n) represents the dynamic threshold, V E Represents the amplitude of the dynamic threshold, Y ij (n) represents the neuron state judgment parameter, U ij (n) represents an internal activity item, represents the pixel value of the base scale image at coordinate (i, j), represents the connection strength of the base scale image, W ijkl represents the inverse of the Euclidean distance between adjacent neurons, represents the pixel value of the intensity image at coordinate (i, j), represents the connection strength of the intensity image, N ij represents the local area centered at coordinate (i, j), Y kl (n-1) represents the state judgment parameter of the neighboring neuron, and n represents the number of iterations.

[0098] When U ij (n)>E ij (n-1), Yij (n)=1, otherwise, Y ij (n)=0; when (i,j)≠(k,l), W ijkl =1 / (S(i,j)-S(k,l)) 2 , when (i,j)=(k,l), W ijkl =0.

[0099] Multiscale morphological gradient can effectively extract the gradient information of the image, and can easily change the size of the structure to adapt to the scale of the structure. Combining multiscale morphological gradient and adaptive simplified pulse coupled neural network model can enhance the sensitivity of image edge information when using adaptive simplified pulse coupled neural network model alone, thereby extracting edge information and avoiding block effect and local blurring of fused image; Multiscale morphological gradient model is used to extract the gradient features of the image, and its formula is:

[0100]

[0101] Among them, G t (x, y) represents the gradient feature, t represents the number of scales, I(x, y) represents the input image, g t (x,y) represents the structural element, represents the dilation operation, and Θ represents the erosion operation.

[0102] The image multi-scale morphological gradient operator is calculated based on the gradient features of the image; its formula is:

[0103]

[0104] Among them, M(x,y) represents the multi-scale morphological gradient operator, w t represents the weight of the tth scale, G t (x, y) represents the gradient feature of the tth scale, and n represents the total number of scales.

[0105] By inputting the base-scale image and the intensity image into the multi-scale morphological gradient model respectively, a multi-scale morphological gradient operator M1 of the base-scale image and a multi-scale morphological gradient operator M2 of the intensity image can be obtained.

[0106] Since one pixel of the image corresponds to one neuron in the adaptive simplified pulse coupled neural network model, it is difficult to find the edge information of the image when the model is used alone. In this case, the fused image will produce block effects and local blurring. Therefore, it is necessary to introduce feature operations, and the multiscale morphological gradient operator can produce good effects on multiscale edge extraction of images. The present invention uses a multiscale morphological gradient operator to modulate an adaptive simplified pulse coupled neural network to solve the block effects and local blurring. The adaptive simplified pulse coupled neural network model is modulated by the multiscale morphological gradient operator of the base scale image and the multiscale morphological gradient operator of the intensity image to obtain an adaptive simplified pulse coupled neural network model based on the multiscale morphological gradient; the modulation formula is:

[0107]

[0108]

[0109] Where M1 represents the multi-scale morphological gradient operator of the base scale image, represents the connection strength of the base scale image, M2 represents the multi-scale morphological gradient operator of the intensity image, Represents the connection strength of the intensity image.

[0110] The base-scale image and the intensity image are input into an adaptive simplified pulse-coupled neural network model based on multi-scale morphological gradient for fusion to obtain an approximate image.

[0111] The approximate image obtained by fusing the base-scale image with the intensity image replaces the synthetic aperture radar base-scale image and fuses it with the small-scale image and the large-scale image to obtain a fused intensity image; preferably, the fusion can be performed by weighted summing the grayscale values ​​of corresponding pixels in the approximate image, the small-scale image and the large-scale image.

[0112] The fused intensity image is reconstructed by using the inverse color saturation intensity transform to obtain the remote sensing image fused from the synthetic aperture radar image and the multispectral image. The reconstruction formula is:

[0113]

[0114] The present invention performs multi-scale decomposition of an image through a hybrid decomposition algorithm, extracts a small-scale image, a large-scale image and a base-scale image while retaining the edge detail information of the image; adopts an adaptive simplified pulse coupled neural network to reduce coupling, simplify the model structure and the number of parameters, and facilitates parameter acquisition through the setting of adaptive parameters, and no longer requires a large amount of experimental experience to acquire, thereby improving the efficiency of image fusion; combines the scale morphological gradient with the adaptive simplified pulse coupled neural network, and utilizes the multi-scale morphological gradient to extract multi-scale image features, thereby avoiding the block effect and local image blurring caused by the insensitivity of the adaptive simplified pulse neural network to the image edge information when used alone.

[0115] The above embodiments further illustrate the purpose, technical solutions and advantages of the present invention in detail. It should be understood that the above embodiments are only preferred implementation modes of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made to the present invention within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A remote sensing image fusion method based on a multi-scale morphological gradient adaptive simplified pulse coupled neural network model, characterized in that: include: S1: Acquire synthetic aperture radar images and multispectral images; S2: Perform Gaussian curvature filter decomposition and Gaussian filter decomposition on the synthetic aperture radar image to obtain a small-scale image, a large-scale image and a base-scale image; S3: Perform color saturation intensity transformation on the multispectral image to obtain an intensity image; S4: An adaptive simplified pulse coupled neural network model based on multi-scale morphological gradients is used to fuse the base scale image and the intensity image to obtain an approximate image; The construction process of the adaptive simplified pulse coupled neural network model based on multi-scale morphological gradient includes: Setting adaptive parameters, including standard deviation relationship value of input image, connection strength, amplitude of dynamic threshold and attenuation factor of dynamic threshold; obtaining adaptive simplified pulse coupled neural network model according to the adaptive parameters; The adaptive parameters are: a F =log(1 / σ(S)) Among them, α F represents the standard deviation relationship value of the input image, σ(S) represents the standard deviation of the input image, S represents the pixel value of the input image, β ij ′ represents the connection strength, m represents the number of neurons in the region, and x kl represents the gray value of the neighborhood neuron of the central pixel, represents the mean gray value of neurons in the region, (k, l) represents the neighborhood pixel of the central pixel, V E represents the amplitude of the dynamic threshold, β′ represents the standard deviation of the input image, and α E represents the attenuation factor of the dynamic threshold, S′ represents the normalized Otsu threshold; The multi-scale morphological gradient model is used to extract the gradient features of the base scale image and the gradient features of the intensity image respectively; the formula for extracting the gradient features of the image using the multi-scale morphological gradient model is: Among them, G t (x, y) represents the gradient feature, t represents the number of scales, I(x, y) represents the input image, g t (x,y) represents the structural element, represents the expansion operation, Θ represents the erosion operation; According to the gradient features of the base scale image and the gradient features of the intensity image, a multi-scale morphological gradient operator of the base scale image and a multi-scale morphological gradient operator of the intensity image are obtained respectively; The adaptive simplified pulse coupled neural network model is modulated by using the multi-scale morphological gradient operator of the base scale image and the multi-scale morphological gradient operator of the intensity image, so as to obtain an adaptive simplified pulse coupled neural network model based on the multi-scale morphological gradient. S5: Fusing the approximate image, the small-scale image and the large-scale image to obtain a fused intensity image; S6: Reconstruct the fused intensity image using the inverse color saturation intensity transform to obtain a fused remote sensing image.

2. The remote sensing image fusion method based on the adaptive simplified pulse coupled neural network model of multi-scale morphological gradient according to claim 1 is characterized in that: The formula for Gaussian curvature filtering decomposition of synthetic aperture radar image data is: R(i,j)=I(i,j)+min{|d n |,n=1,…,8} d1=(I(i-1,j)+I(i+1,j)) / 2-I(i,j) d2=(I(i,j-1)+I(i,j+1)) / 2-I(i,j) d3=(I(i-1,j-1)+I(i+1,j+1)) / 2-I(i,j) d4=(I(i-1,j+1)+I(i+1,j-1)) / 2-I(i,j) d5=(I(i-1,j)+I(i,j-1)+I(i-1,j-1)) / 3-I(i,j) d6=(I(i-1,j)+I(i,j+1)+I(i-1,j+1)) / 3-I(i,j) d7=(I(i,j-1)+I(i+1,j)+I(i+1,j-1)) / 3-I(i,j) d8=(I(i,j+1)+I(i+1,j)+I(i+1,j+1)) / 3-I(i,j) Where R(i,j) represents the Gaussian curvature filter decomposition result of pixel point (i,j), I(i,j) represents the pixel point (i,j) in the synthetic aperture radar image I, and d n Represents the filtering result of the nth window.

3. The remote sensing image fusion method based on the adaptive simplified pulse coupled neural network model of multi-scale morphological gradient according to claim 1 is characterized in that: The formula for Gaussian filtering decomposition of synthetic aperture radar image data is: Where I(a,b) represents the Gaussian filter decomposition result of the pixel point (a,b), and σ represents the standard deviation.

4. The remote sensing image fusion method based on the adaptive simplified pulse coupled neural network model of multi-scale morphological gradient according to claim 1 is characterized in that: The process of obtaining a small-scale image, a large-scale image and a base-scale image includes: performing Gaussian curvature filtering decomposition on a synthetic aperture radar image to obtain a first filtered image; performing Gaussian filtering decomposition on the synthetic aperture radar image to obtain a second filtered image and use it as a base-scale image; performing a subtraction operation on the first filtered image and the second filtered image to obtain a large-scale image; and performing a subtraction operation on the synthetic aperture radar image and the first filtered image to obtain a small-scale image.

5. The remote sensing image fusion method based on the adaptive simplified pulse coupled neural network model of multi-scale morphological gradient according to claim 1 is characterized in that: The formula for the multi-scale morphological gradient operator is: Among them, M(x,y) represents the multi-scale morphological gradient operator, w t represents the weight of the tth scale, G t (x, y) represents the gradient feature of the tth scale, and n represents the total number of scales.

6. The remote sensing image fusion method based on a multi-scale morphological gradient adaptive simplified pulse coupled neural network model according to claim 1, characterized in that: The formula for modulating the adaptive simplified pulse coupled neural network model is: Where M1 represents the multi-scale morphological gradient operator of the base scale image, represents the connection strength of the base scale image, M2 represents the multi-scale morphological gradient operator of the intensity image, Represents the connection strength of the intensity image.

7. The remote sensing image fusion method based on a multi-scale morphological gradient adaptive simplified pulse coupled neural network model according to claim 1, characterized in that: The formula of the adaptive simplified pulse coupled neural network model is: Among them, E ij (n) represents the dynamic threshold, V E Represents the amplitude of the dynamic threshold, Y ij (n) represents the neuron state judgment parameter, U ij (n) represents an internal activity item, represents the pixel value of the base scale image at coordinate (i, j), represents the connection strength of the base scale image, W ijkl represents the inverse of the Euclidean distance between adjacent neurons, represents the pixel value of the intensity image at coordinate (i, j), represents the connection strength of the intensity image, N ij represents the local area centered at coordinate (i, j), Y kl (n-1) represents the state judgment parameter of the neighboring neuron, and n represents the number of iterations.

Images