A sample synthesis method for remote sensing image interpretation based on diffusion model

Through the remote sensing image interpretation sample synthesis method based on diffusion model, the problems of high computational complexity and insufficient interpretation accuracy in high resolution and multispectral image processing are solved, and image detail enhancement and noise suppression are achieved, which improves the The interpretation accuracy and efficiency of remote sensing images improve the robustness of the algorithm.

CN118521914BActive Publication Date: 2024-11-15ZHONGKAN MAIPU (JIANGSU) TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410861132.7
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-06-28
Publication Date
2024-11-15
Estimated Expiration
2044-06-28

AI Technical Summary

Technical Problem

Traditional remote sensing image interpretation methods have high computational complexity and insufficient interpretation accuracy when processing high-resolution, multi-spectral or hyperspectral images, which cannot meet the high requirements of modern remote sensing technology for image interpretation accuracy and efficiency.

Method used

Using the remote sensing image interpretation sample synthesis method based on diffusion model, the original remote sensing image data is received and preprocessed, features are extracted, the diffusion model is set, the diffusion process is dynamically adjusted, and multi-scale fusion processing and robustness optimization are carried out, verification and iterative optimization are carried out. , and finally output high-quality synthetic samples.

Benefits of technology

It realizes accurate capture and noise suppression of image details, improves image interpretation accuracy and efficiency, improves the robustness of the algorithm, and maintains high accuracy under complex conditions.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118521914B_ABST
    Figure CN118521914B_ABST
Patent Text Reader

Abstract

The remote sensing image interpretation sample synthesis method based on the diffusion model provided by the present invention relates to the technical field of remote sensing image processing, and includes the following steps: receiving and preprocessing original remote sensing image data; extracting remote sensing image data after feature preprocessing; setting diffusion model initialization; dynamically adjusting the adaptive diffusion process according to image features; multi-scale fusion processing to obtain synthetic samples; performing robustness optimization processing on the synthetic samples; performing sample verification and iterative optimization on the synthetic samples; and outputting verified and optimized high-quality synthetic samples for use in remote sensing image interpretation tasks. By combining the diffusion model, the present invention can accurately capture image details, effectively suppress noise, provide clearer ground features and boundaries, achieve detail enhancement and noise suppression, and optimize image interpretation by adaptively adjusting multi-scale features, thereby improving the quality and applicability of synthetic samples and the interpretation accuracy and efficiency of remote sensing images.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to the technical field of remote sensing image processing, and in particular to a remote sensing image interpretation sample synthesis method based on a diffusion model. Background Art

[0002] As an important means of obtaining information about the earth's surface, remote sensing technology has undergone a transformation from early aerial photography to modern satellite remote sensing. With the development of satellite technology, the acquisition of remote sensing images has become more convenient and efficient, and the image resolution has also been greatly improved. Currently, remote sensing technology can provide image data in a variety of spectral ranges, from visible light to infrared, microwaves, etc.

[0003] With the widespread application of remote sensing technology in environmental monitoring, disaster assessment, urban planning and other fields, higher requirements are placed on the accuracy and efficiency of image interpretation. Diffusion model, as an emerging image processing technology, improves image quality or enhances image features by simulating the physical diffusion process.

[0004] Traditional remote sensing image interpretation methods usually rely on manual interpretation or simple image processing techniques, such as threshold segmentation, edge detection, etc. These methods often face problems such as high computational complexity and insufficient interpretation accuracy when processing high-resolution, multispectral or hyperspectral images. In order to overcome the shortcomings of the existing technology, it is necessary to provide a remote sensing image interpretation sample synthesis method based on a diffusion model to solve the above technical problems. Summary of the invention

[0005] The present invention provides a remote sensing image interpretation sample synthesis method based on a diffusion model, which solves the problems raised in the above background technology.

[0006] In order to solve the above technical problems, the present invention provides a remote sensing image interpretation sample synthesis method based on a diffusion model, comprising the following steps:

[0007] Step 1: receiving and preprocessing original remote sensing image data; the original remote sensing image data includes multispectral images and high-resolution images;

[0008] Step 2, feature extraction of preprocessed remote sensing image data;

[0009] Step 3, setting the diffusion model initialization; initializing the diffusion parameters of the diffusion model, including diffusion rate and diffusion direction;

[0010] Step 4: dynamically adjust the adaptive diffusion process according to the image features;

[0011] Step 5: Multi-scale fusion processing to obtain synthetic samples;

[0012] Step 6: Perform robustness optimization processing on the synthetic samples;

[0013] Step 7: Perform sample verification and iterative optimization on the synthetic samples;

[0014] Step 8: Output verified and optimized high-quality synthetic samples for use in remote sensing image interpretation tasks.

[0015] As a preferred method, the original remote sensing image data is preprocessed, and the specific steps of preprocessing are:

[0016] Preprocessing includes denoising and contrast enhancement. In denoising, wavelet transform or Gaussian filtering is used to remove noise in the image to obtain a denoised remote sensing image.

[0017] During contrast enhancement processing, contrast-limited adaptive histogram equalization is applied to improve the local contrast of the image and obtain a contrast-enhanced remote sensing image.

[0018] As a preferred embodiment, the remote sensing image data after feature extraction preprocessing is specifically:

[0019] A convolutional neural network is used to extract the texture, shape and spectral features of the preprocessed remote sensing image data. A convolutional neural network is established, which automatically learns image features through multiple convolutional layers and pooling layers. Each convolutional layer applies a set of learnable convolution kernels to the input data and generates a feature map through an activation function. The pooling layer further reduces the spatial dimension of the feature map while retaining important feature information.

[0020] Preferably, based on step three, the diffusion model captures the change of the distribution of matter or energy in space over time through differential equations.

[0021] Preferably, based on step 4, the specific diffusion process of the adaptive diffusion process is expressed using an iterative formula: ;in, represents the image intensity at time step t, represents the image intensity at the next time step t+1, λ represents the intensity of the diffusion process, represents the diffusion coefficient that depends on the image feature F, Indicates that the Laplace operator acts on the current image The above results.

[0022] Preferably, based on step seven, sample verification and iterative optimization are performed on the synthetic sample, and the specific method is as follows:

[0023] Perform quality verification on synthetic samples, evaluate the quality of synthetic samples, set the corresponding real samples in the synthetic samples, unify the format and scale of real samples and synthetic samples; set quantitative evaluation indicators for synthetic samples; quantitative evaluation indicators include mean square error, peak signal-to-noise ratio, and structural similarity index; process each synthetic sample according to the quantitative evaluation indicators, perform weighted calculation on the mean square error, peak signal-to-noise ratio, and structural similarity index to obtain the evaluation value of the synthetic sample.

[0024] An evaluation threshold is set. If the evaluation value is greater than or equal to the evaluation threshold, it indicates that the quality of the synthesized sample meets the requirements, and the synthesized sample is marked as a high-quality synthesized sample output; if the evaluation value is less than the evaluation threshold, it indicates that the quality of the synthesized sample does not meet the requirements, and steps one to seven are repeated.

[0025] As a preferred embodiment, the acquisition logic of the synthetic sample is expressed as:

[0026] Decompose the original image into sub-images of different scales, each scale contains multiple sub-images;

[0027] Feature extraction is performed on the sub-image of each scale to obtain features; weights are assigned to the sub-image of each scale; and fusion calculations are performed using a preset formula to obtain synthetic samples.

[0028] Preferably, based on the evaluation value being less than its evaluation threshold, a feedback adjustment operation is performed according to the quantitative evaluation index, specifically:

[0029] Set the indicator threshold of any indicator in the quantitative evaluation indicators, calculate the difference between any indicator in the quantitative evaluation indicators and its corresponding indicator threshold, and obtain the indicator difference; set the indicator difference threshold, and take the quantitative evaluation indicator corresponding to the indicator difference greater than the indicator difference threshold as the indicator to be regulated; adjust the corresponding parameters according to the indicator to be regulated, specifically:

[0030] If the indicator to be regulated is the indicator corresponding to the mean square error, the parameters in the preprocessing and feature extraction in steps one and two are adjusted according to the evaluation value;

[0031] If the indicator to be regulated is an indicator corresponding to the peak signal-to-noise ratio, the parameters of the diffusion model are adjusted according to the evaluation value;

[0032] If the indicator to be regulated is the indicator corresponding to the structural similarity index, the weight in the fusion calculation is adjusted according to the evaluation value;

[0033] Rerun steps one to seven with the adjusted parameters until the evaluation value of the synthetic sample meets its metric threshold.

[0034] Compared with related technologies, the remote sensing image interpretation sample synthesis method based on diffusion model provided by the present invention has the following beneficial effects:

[0035] 1. By combining the diffusion model, the present invention can accurately capture image details, effectively suppress noise, provide clearer ground features and boundaries, achieve detail enhancement and noise suppression, and optimize image interpretation by adaptively adjusting multi-scale features, thereby improving the quality and applicability of synthetic samples and the interpretation accuracy and efficiency of remote sensing images.

[0036] 2. The present invention utilizes the iterative characteristics of the diffusion model to significantly improve the robustness of the algorithm, and can maintain high-accuracy interpretation performance even under complex and changeable remote sensing image conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 A flowchart of a remote sensing image interpretation sample synthesis method based on a diffusion model provided by the present invention. DETAILED DESCRIPTION

[0038] 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.

[0039] The terms used in this disclosure are for the purpose of describing specific embodiments only and are not intended to limit the disclosure. The singular forms of "the", "group" and "the" used in this disclosure and the appended claims are also intended to include plural forms unless the context clearly indicates other meanings. It should also be understood that the term "and / or" used herein refers to and includes any or all possible combinations of one or more associated listed items.

[0040] It should be understood that although the terms first, second, third, etc. may be used in the present disclosure to describe various information, such information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of the present disclosure, the first information may also be referred to as the second information, and similarly, the second information may also be referred to as the first information. Depending on the context, the word "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining".

[0041] Please refer to Figure 1 The remote sensing image interpretation sample synthesis method based on the diffusion model includes the following steps:

[0042] Step 1: receiving and preprocessing original remote sensing image data; the original remote sensing image data includes multispectral images and high-resolution images;

[0043] Step 2, feature extraction of preprocessed remote sensing image data;

[0044] Step 3, setting the diffusion model initialization; initializing the diffusion parameters of the diffusion model, including diffusion rate and diffusion direction;

[0045] Step 4: dynamically adjust the adaptive diffusion process according to the image features;

[0046] Step 5: Multi-scale fusion processing to obtain synthetic samples;

[0047] Step 6: Perform robustness optimization processing on the synthetic samples. Specifically, use the total variation regularization technique to smooth the image. The total variation regularization formula is expressed as: , Represents the gradient at point j in the remote sensing image;

[0048] Step 7: Perform sample verification and iterative optimization on the synthetic samples;

[0049] Step 8: Output verified and optimized high-quality synthetic samples for use in remote sensing image interpretation tasks.

[0050] In the present invention, the original remote sensing image data is preprocessed, and the specific steps of preprocessing are:

[0051] Preprocessing includes denoising and contrast enhancement. In denoising, wavelet transform or Gaussian filtering is used to remove noise from the image. The formula is: , get the denoised remote sensing image I f ; Among them, I o ,I f They represent the image in the original remote sensing image data and the filtered image, respectively, and σ represents the filtering parameter;

[0052] In contrast enhancement processing, contrast-limited adaptive histogram equalization is applied to improve the local contrast of the image. The formula is: , and the contrast-enhanced remote sensing image Ie is obtained; where α and β represent the scaling and translation parameters determined by histogram equalization, and (m, n) represents the position of a pixel at one point in the image coordinate system.

[0053] It should be noted that, in addition to the denoising and contrast enhancement processing mentioned above, preprocessing also includes radiation correction, color space conversion, edge enhancement, texture analysis, feature normalization, cloud occlusion processing, shadow length correction, seasonal effect correction, etc., which are all existing technologies and constitute a comprehensive remote sensing image preprocessing process for the original remote sensing image data, aiming to improve the quality of remote sensing images and provide a more accurate and reliable data basis for subsequent interpretation and analysis; through these preprocessing steps, the present invention can adapt to different remote sensing image data and application scenarios, and improve the accuracy and reliability of the interpretation results.

[0054] In the present invention, the remote sensing image data after feature extraction preprocessing is specifically:

[0055] Use convolutional neural networks to extract texture, shape, and spectral features of preprocessed remote sensing image data; establish a convolutional neural network that automatically learns image features through multiple convolutional layers and pooling layers; each convolutional layer applies a set of learnable convolution kernels to the input data and generates a feature map through an activation function; the pooling layer further reduces the spatial dimension of the feature map while retaining important feature information;

[0056] Among them, the mathematical expression of the convolution layer is: , y represents the output feature map of the convolution operation, Represents the convolution operation, k represents the convolution kernel or filter, x represents the input remote sensing image data, f represents the activation function, and b represents the bias term.

[0057] In the present invention, based on step three, the diffusion model captures the distribution of matter or energy in space over time through differential equations, which is expressed by the formula: ,in, represents the partial derivative of image intensity I with respect to time t, where I represents image intensity, t represents image acquisition time, D(F) represents the feature-dependent diffusion coefficient, represents the Laplace operator.

[0058] In the present invention, based on step 4, the specific diffusion process of the adaptive diffusion process is expressed using an iterative formula: ;in, represents the image intensity at time step t, represents the image intensity at the next time step t+1, λ represents the intensity of the diffusion process, represents the diffusion coefficient that depends on the image feature F, Indicates that the Laplace operator acts on the current image The above results.

[0059] In the present invention, based on step seven, sample verification and iterative optimization are performed on the synthetic samples. The specific method is as follows:

[0060] Verify the quality of synthetic samples, evaluate the quality of synthetic samples, set the corresponding real samples in the synthetic samples, unify the format and scale of real samples and synthetic samples; set quantitative evaluation indicators for synthetic samples; quantitative evaluation indicators include mean square error, peak signal-to-noise ratio, and structural similarity index; process each synthetic sample according to the quantitative evaluation indicators, specifically:

[0061] The calculation expression of mean square error is: ; Where Ii and Si represent the pixel values ​​of the real sample and the synthetic sample respectively, i represents the number of any pixel on the synthetic sample, and N represents the total number of pixels on the synthetic sample;

[0062] The calculation expression of peak signal-to-noise ratio is: ; Among them, 255 is the maximum possible value of the pixel value, and log10 represents the logarithm with base 10;

[0063] The expression of the structural similarity index is: ; Where uI and uS represent the means of the real samples and the synthetic samples respectively, σI and σS represent the standard deviations of the real samples and the synthetic samples respectively, σIS represents the covariance of the real samples and the synthetic samples, c1 and c2 are small constants used to avoid the denominator being zero;

[0064] The mean square error, peak signal-to-noise ratio and structural similarity index are weighted and calculated, and the evaluation value PG of the synthetic sample is obtained using the formula PG=MSE×g1+PSNR×g2+SSIM×g3; among them, g1, g2, and g3 represent the weights corresponding to the mean square error, peak signal-to-noise ratio and structural similarity index, respectively.

[0065] An evaluation threshold is set. If the evaluation value is greater than or equal to the evaluation threshold, it indicates that the quality of the synthesized sample meets the requirements, and the synthesized sample is marked as a high-quality synthesized sample output; if the evaluation value is less than the evaluation threshold, it indicates that the quality of the synthesized sample does not meet the requirements, and steps one to seven are repeated.

[0066] In the present invention, based on step 5, the acquisition logic of the synthetic sample is expressed as:

[0067] The original image is decomposed into sub-images of different scales, each scale contains multiple sub-images, represented as ; Where a represents the scale, and b represents the number of the sub-image at that scale;

[0068] For each scale sub-image Perform feature extraction and obtain feature representation ; Assign weights to sub-images of each scale ; Weight Determining based on the importance or feature saliency of sub-images;

[0069] Use the preset formula for fusion calculation, the formula is expressed as , and obtain the synthetic sample S; where the first sum represents the accumulation of all scales, and the second sum represents the accumulation of all sub-images in each scale.

[0070] It should be noted that the feature representation in practice is, for example, represented as a set of pixel intensity values, which is used to represent the grayscale or color information of the remote sensing image; or represented as a set of values ​​obtained from the filter response, for example, edge information extracted using a Gaussian filter or a Sobel operator; or represented as high-level features obtained from a deep learning model, such as the intermediate layer activation of a convolutional neural network.

[0071] In the present invention, based on the evaluation value being less than its evaluation threshold, a feedback adjustment operation is performed according to the quantitative evaluation index, specifically:

[0072] Set the indicator threshold of any indicator in the quantitative evaluation indicators, calculate the difference between any indicator in the quantitative evaluation indicators and its corresponding indicator threshold, and obtain the indicator difference; set the indicator difference threshold, and take the quantitative evaluation indicator corresponding to the indicator difference greater than the indicator difference threshold as the indicator to be regulated; adjust the corresponding parameters according to the indicator to be regulated, specifically:

[0073] If the indicator to be regulated is the indicator corresponding to the mean square error, the parameters in the preprocessing and feature extraction in steps one and two are adjusted according to the evaluation value;

[0074] If the indicator to be regulated is an indicator corresponding to the peak signal-to-noise ratio, the parameters of the diffusion model are adjusted according to the evaluation value;

[0075] If the indicator to be regulated is the indicator corresponding to the structural similarity index, the weight in the fusion calculation is adjusted according to the evaluation value;

[0076] Rerun steps one to seven with the adjusted parameters until the evaluation value of the synthetic sample meets its metric threshold.

[0077] Those skilled in the art will readily appreciate other embodiments of the present invention after considering the specification and practicing the invention disclosed herein. This application is intended to cover any variations, uses or adaptations of the present invention that follow the general principles of the present invention and include common knowledge or customary techniques in the art that are not disclosed in this disclosure. The specification and examples are to be considered exemplary only, and the true scope and spirit of the present invention are indicated by the following claims.

[0078] It should be understood that the present invention is not limited to the exact construction that has been described above and shown in the drawings and that various modifications and changes may be made without departing from the scope thereof. The scope of the present invention is limited only by the appended claims.

Claims

1. A remote sensing image interpretation sample synthesis method based on a diffusion model, characterized in that: The following steps are involved: Step 1: receiving and preprocessing original remote sensing image data; the original remote sensing image data includes multispectral images and high-resolution images; Step 2, feature extraction of preprocessed remote sensing image data; Step 3, setting the diffusion model initialization; initializing the diffusion parameters of the diffusion model, including diffusion rate and diffusion direction; Step 4: dynamically adjust the adaptive diffusion process according to the image features; the specific diffusion process of the adaptive diffusion process is expressed by an iterative formula: ;in, represents the image intensity at time step t, represents the image intensity at the next time step t+1, λ represents the intensity of the diffusion process, represents the diffusion coefficient that depends on the image feature F, Indicates that the Laplace operator acts on the current image The above results; Step 5: Multi-scale fusion processing to obtain synthetic samples; the acquisition logic of synthetic samples is expressed as: The original image is decomposed into sub-images of different scales, each scale contains multiple sub-images, represented as ; Where a represents the scale, b represents the number of the sub-image at that scale; for each sub-image of each scale Perform feature extraction and obtain feature representation ; Assign weights to sub-images of each scale ; Use the preset formula for fusion calculation, the formula is expressed as , and obtain the synthetic sample S; where the first sum represents the accumulation of all scales, and the second sum represents the accumulation of all sub-images in each scale; Step 6: Perform robustness optimization processing on the synthetic samples; Step 7: Perform sample verification and iterative optimization on the synthetic samples. The specific method is as follows: Verify the quality of synthetic samples, evaluate the quality of synthetic samples, set the corresponding real samples in the synthetic samples, unify the format and scale of real samples and synthetic samples; set quantitative evaluation indicators for synthetic samples; quantitative evaluation indicators include mean square error, peak signal-to-noise ratio, and structural similarity index; process each synthetic sample according to the quantitative evaluation indicators, specifically: The calculation expression of mean square error is: ; Where Ii and Si represent the pixel values ​​of the real sample and the synthetic sample respectively, i represents the number of any pixel on the synthetic sample, and N represents the total number of pixels on the synthetic sample; The calculation expression of peak signal-to-noise ratio is: ; Among them, 255 is the maximum possible value of the pixel value, and log10 represents the logarithm with base 10; The expression of the structural similarity index is: ; Where uI and uS represent the means of the real samples and the synthetic samples respectively, σI and σS represent the standard deviations of the real samples and the synthetic samples respectively, σIS represents the covariance of the real samples and the synthetic samples, c1 and c2 are small constants used to avoid the denominator being zero; The mean square error, peak signal-to-noise ratio and structural similarity index are weighted and calculated, and the evaluation value PG of the synthetic sample is obtained using the formula PG=MSE×g1+PSNR×g2+SSIM×g3; where g1, g2 and g3 represent the weights corresponding to the mean square error, peak signal-to-noise ratio and structural similarity index respectively; An evaluation threshold is set. If the evaluation value is greater than or equal to the evaluation threshold, it indicates that the quality of the synthesized sample meets the requirements, and the synthesized sample is marked as a high-quality synthesized sample output; if the evaluation value is less than the evaluation threshold, it indicates that the quality of the synthesized sample does not meet the requirements, and steps one to seven are repeated; Step 8: Output verified and optimized high-quality synthetic samples for use in remote sensing image interpretation tasks.

2. The remote sensing image interpretation sample synthesis method based on diffusion model according to claim 1 is characterized in that: Preprocess the original remote sensing image data. The specific steps of preprocessing are: Preprocessing includes denoising and contrast enhancement. In denoising, wavelet transform or Gaussian filtering is used to remove noise from the image. The formula is: , get the denoised remote sensing image I f ; Among them, I o ,I f They represent the image in the original remote sensing image data and the filtered image, respectively, and σ represents the filtering parameter; In contrast enhancement processing, contrast-limited adaptive histogram equalization is applied to improve the local contrast of the image. The formula is: , and the contrast-enhanced remote sensing image Ie is obtained; where α and β represent the scaling and translation parameters determined by histogram equalization, and (m, n) represents the position of a pixel at one point in the image coordinate system.

3. The remote sensing image interpretation sample synthesis method based on diffusion model according to claim 1 is characterized in that: The remote sensing image data after feature extraction preprocessing is as follows: Use convolutional neural networks to extract texture, shape, and spectral features of preprocessed remote sensing image data; establish a convolutional neural network that automatically learns image features through multiple convolutional layers and pooling layers; each convolutional layer applies a set of learnable convolution kernels to the input data and generates a feature map through an activation function; the pooling layer further reduces the spatial dimension of the feature map while retaining important feature information; Among them, the mathematical expression of the convolution layer is: , y represents the output feature map of the convolution operation, Represents the convolution operation, k represents the convolution kernel or filter, x represents the input remote sensing image data, f represents the activation function, and b represents the bias term.

4. The remote sensing image interpretation sample synthesis method based on diffusion model according to claim 1, characterized in that: Based on step three, the diffusion model captures the distribution of matter or energy in space over time through differential equations, which are expressed as: ,in, represents the partial derivative of image intensity I with respect to time t, where I represents image intensity, t represents image acquisition time, D(F) represents the feature-dependent diffusion coefficient, represents the Laplace operator.

5. The remote sensing image interpretation sample synthesis method based on diffusion model according to claim 1, characterized in that: Based on the evaluation value being less than its evaluation threshold, feedback adjustment operations are performed according to the quantitative evaluation indicators, specifically: Set the indicator threshold of any indicator in the quantitative evaluation indicators, calculate the difference between any indicator in the quantitative evaluation indicators and its corresponding indicator threshold, and obtain the indicator difference; set the indicator difference threshold, and take the quantitative evaluation indicator corresponding to the indicator difference greater than the indicator difference threshold as the indicator to be regulated; adjust the corresponding parameters according to the indicator to be regulated, specifically: If the indicator to be regulated is the indicator corresponding to the mean square error, the parameters in the preprocessing and feature extraction in steps one and two are adjusted according to the evaluation value; If the indicator to be regulated is an indicator corresponding to the peak signal-to-noise ratio, the parameters of the diffusion model are adjusted according to the evaluation value; If the indicator to be regulated is the indicator corresponding to the structural similarity index, the weight in the fusion calculation is adjusted according to the evaluation value; Rerun steps one to seven with the adjusted parameters until the evaluation value of the synthetic sample meets its metric threshold.