Hyperspectral image super-resolution method with alternative diffusion band selection and degradation self-adaption

Abstract

The invention relates to a hyperspectral image super-resolution method based on alternative diffusion band selection and degradation self-adaption. The method comprises the following steps of: constructing a blind image restoration target function and decomposing the blind image restoration target function into a high-quality image and a degradation matrix solving sub-problem; then constructing a hyperspectral image super-resolution model containing three modules: selecting a wave band with a large amount of information by a wave band selection and Gaussian dimension reduction module according to variance distribution and a Gaussian attenuation model; the sampling prior and degradation kernel estimation module generates a random degradation matrix through Monte Carlo sampling, and combines a data fidelity loss function to optimize parameters so as to dynamically estimate the degradation matrix; the frequency band projection and diffusion guide image super-division module solves a high-quality image by means of a pre-training diffusion model, band selection and a degradation matrix, and optimizes an intermediate high-quality image through a data fidelity loss function and total variation regularization. And finally, the three modules are iterated to preset times, and an obtained middle high-quality image is a recovered result. By adopting the method, the image recovery performance and efficiency can be improved.

Description

Technical Field

[0001] This application relates to the field of image processing technology, and in particular to a hyperspectral image super-resolution method with alternating diffusion band selection and degradation adaptation. Background Technology

[0002] Hyperspectral super-resolution (HSI-SR) technology aims to reconstruct high-resolution (HR) hyperspectral images from low-resolution (LR) hyperspectral images. This process not only significantly improves the spatial resolution of images but also preserves the rich spectral information of hyperspectral images, providing a foundation for more accurate image analysis and applications. In remote sensing, HSI-SR technology enables images acquired from space or high-altitude platforms to more clearly reveal the details and features of ground objects, thereby improving the accuracy of tasks such as land use classification, crop growth monitoring, and mineral resource exploration. In environmental monitoring, hyperspectral images processed with super-resolution can more sensitively capture subtle differences in phenomena such as environmental pollution, water quality changes, and forest degradation, providing a scientific basis for environmental assessment and the formulation of protection measures. In agriculture, this technology contributes to the development of precision agriculture by analyzing crop health status, pest and disease distribution, and soil nutrient content through high-resolution spectral information, optimizing agricultural production management decisions. In geological exploration, HSI-SR technology helps geologists more accurately identify rock types, mineral distribution, and geological structures, providing crucial data support for mineral resource development and geological disaster prevention.

[0003] However, the HSI-SR task faces numerous severe challenges. First, the HSI-SR problem itself is ill-posed, meaning that the mapping from LR images to HR images has multiple possible solutions, making it extremely difficult to find a unique and accurate HR image. Second, the complexity of hyperspectral data further exacerbates this problem. Hyperspectral images typically have hundreds of bands, each corresponding to different spectral information. These bands exhibit rich correlations and contain unique detailed features. How to fully utilize this spectral information during super-resolution reconstruction while avoiding excessive computational costs and overfitting is a key challenge that current research needs to address. Furthermore, degradation models in practical applications are often complex and variable, involving the combined effects of multiple factors such as blurring, noise, and downsampling. Accurately modeling and estimating these degradation processes is crucial for achieving effective super-resolution reconstruction. In conclusion, although HSI-SR technology has shown great application potential in multiple fields, achieving efficient and accurate super-resolution reconstruction still requires overcoming many technical obstacles and challenges.

[0004] Initially, hyperspectral image super-resolution (HSI-SR) techniques primarily relied on model-driven methods, but now most have shifted to learning-based approaches. Learning-based HSI-SR methods are mainly divided into supervised and unsupervised categories. Supervised HSI-SR techniques have achieved significant results using deep neural networks (DNNs) trained on large LR-HR paired datasets. Network architectures include spectral-spatial networks, attention mechanisms, and Transformer-based models. However, these methods typically require large amounts of labeled HSI data, which is extremely costly to obtain and often impractical in certain real-world scenarios. Some methods attempt to mitigate the data dependency problem by utilizing readily available RGB datasets as auxiliary information, but in testing scenarios, performance still significantly degrades when encountering conditions different from those during training (such as different HSI datasets or kernel settings).

[0005] Recently, unsupervised HSI-SR methods have gained increasing popularity. These methods typically design an objective function that uses a known degenerate kernel and observed LR HSI to guide the updates of an untrained network. Major unsupervised methods include Deep Image Prior (DIP) methods and diffusion model methods. However, existing unsupervised methods suffer from two main limitations: first, insufficient utilization of band information; the spectral redundancy of hyperspectral images leads to inefficient bandwidth utilization, as commonly addressed by methods such as full-band iteration or fixed-bandwidth compression (e.g., SVD-based methods); second, oversimplification of the degenerate model; current unsupervised methods use fixed degenerate priors (such as fixed blur kernels) to constrain the solution space of HSI-SR, while degenerate priors in real-world scenarios are often unknown and complex. Summary of the Invention

[0006] Therefore, it is necessary to provide a hyperspectral image super-resolution method with alternating diffusion band selection and degradation adaptation that can improve image restoration performance and efficiency, addressing the aforementioned technical problems.

[0007] A hyperspectral image super-resolution method with alternating diffusion band selection and degradation adaptation, the method comprising: Obtain the target image to be restored; construct the objective function for the blind image restoration problem of the target image; The objective function is decomposed into two sub-problems: solving for high-quality images and the degradation matrix. A hyperspectral image super-resolution model is constructed. The hyperspectral image super-resolution model includes a band selection and Gaussian dimensionality reduction module, a sampling prior and degradation kernel estimation module, and a frequency band projection and diffusion-guided image super-resolution module. A band selection and Gaussian dimensionality reduction module is used to dynamically select bands with high information content through variance distribution and Gaussian attenuation model; A sampling prior and degradation kernel estimation module is adopted. Monte Carlo sampling is used to generate a random degradation matrix. The neural network parameters are optimized by combining the data fidelity loss function to achieve dynamic estimation of the degradation matrix. A frequency band projection and diffusion-guided image super-resolution module is adopted. High-quality images are solved by using a pre-trained diffusion model, band selection and degradation matrix. Intermediate high-quality images are optimized by data fidelity loss function and total variation regularization. The high-resolution model of hyperspectral images is based on the three modules of iterative recovery until the number of iterations meets the preset number. The intermediate high-quality image obtained in the current iteration is the high-quality image after the target image is recovered.

[0008] The aforementioned hyperspectral image super-resolution method with alternating diffusion band selection and degradation adaptation decomposes the objective function of the blind image restoration problem into two sub-problems: solving for a high-quality image and solving for a degradation matrix. These sub-problems are solved using an alternating iterative approach. In each iteration, the degradation matrix sub-problem is solved first, using a neural network combined with random prior guidance and a data-fidelity loss function to update and optimize the estimated value of the degradation matrix. Then, using the updated degradation matrix and a pre-trained diffusion model, the high-quality image sub-problem is solved. Through the combined constraints of the data-fidelity loss function and total variation regularization, a new intermediate high-quality image is obtained. This process continuously alternates between the two sub-problems, with each iteration further improving the quality and accuracy of image restoration based on the previous result. Attached Figure Description

[0009] Figure 1 This is a flowchart illustrating a hyperspectral image super-resolution method with alternating diffusion band selection and degradation adaptation in one embodiment. Detailed Implementation

[0010] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0011] In one embodiment, such as Figure 1 As shown, a hyperspectral image super-resolution method with alternating diffusion band selection and degradation adaptation is provided, including the following steps: Step 102: Obtain the target image to be restored; construct the objective function for the blind image restoration problem of the target image; decompose the objective function into two sub-problems that solve for the high-quality image and the degradation matrix.

[0012] Obtain the target image for super-resolution of the image to be performed. The target image is a low-resolution (LR) degraded image. Substitute the target image into the objective function of the blind image restoration problem.

[0013] For the HSI super-resolution problem, the degradation process of the low-resolution image is generally as shown in the following formula: (1) In formula (1), Indicates a degraded image. This indicates the corresponding high-quality image, that is, the image after restoration. Represents the degeneracy matrix, This indicates additive noise.

[0014] In the problem of non-blind image restoration, high-quality images Unknown, and degenerate matrix It is known. For example, in the image super-resolution problem, the degradation matrix... For the downsampling matrix, in the image completion problem, the degradation matrix is... Let be the mask matrix, and in the image deblurring problem, the degradation matrix. The given fuzzy matrix is ​​known.

[0015] In the problem of blind image restoration, high-quality images Unknown, and degenerate matrix The relevant prior information is also unknown. For example, in the blind image deblurring problem, the degradation matrix... Given an unknown fuzzy matrix, consider the degradation matrix in image deraining, dehazing, and shadow removal problems. This refers to the corresponding rain, fog, and shadow matrices. In practical applications, real-world image degradation may not be a single type of degradation, but rather a combination of various different degradation matrices. This application, however, focuses on image restoration in the context of blind image restoration problems.

[0016] Furthermore, the non-blind image restoration problem can be viewed as a special form of the blind image restoration problem. This is because the degradation matrix in the non-blind image restoration problem... It is known that only high-quality images are available. It needs to be solved. Therefore, the objective function of the non-blind image restoration problem can be written as: (2) The objective function for the blind image restoration problem is shown below: (3) The objective function aims to recover a high-quality image by minimizing the difference between the high-quality image and the degraded image, while taking into account the image's regularization term.

[0017] In formula (3), and Indicates high-quality image and degeneracy matrix Priors, and This represents the corresponding weight. The non-blind image restoration problem can be viewed as a special form of the blind image restoration problem. Due to the degradation matrix in the non-blind image restoration problem... As is known, for ease of description, the subsequent image restoration problem will be introduced in the form of the blind image restoration problem in formula (3).

[0018] Since there are two unknown variables in formula (3) that need to be solved, the problem is further decomposed into two sub-problems for high-quality images. and degeneracy matrix Solve them separately: (4) In this embodiment, for the two unknown variables in equation (4), a blind hyperspectral image super-resolution method based on alternating diffusion dynamic band selection and degradation adaptation is proposed to solve them. That is, for the degenerate matrix To solve this problem, a plug-and-play approach using pre-trained, readily available diffusion models is proposed. That is, high-quality images Solve the problem.

[0019] Step 104: Construct a hyperspectral image super-resolution model; the hyperspectral image super-resolution model includes a band selection and Gaussian dimensionality reduction module, a sampling prior and degradation kernel estimation module, and a frequency band projection and diffusion-guided image super-resolution module.

[0020] Step 106: Using the band selection and Gaussian dimensionality reduction module, the bands with high information content are dynamically selected through variance distribution and Gaussian attenuation model.

[0021] The core of the band selection and Gaussian dimensionality reduction module lies in dynamically selecting information-rich bands through variance distribution and a Gaussian attenuation model. This avoids the high computational cost of full-band processing, which involves calculating the variance of each band first; bands with higher variance typically contain more information. A Gaussian attenuation model is used to adjust band selection based on differences between bands, ensuring that the selected bands with the highest information content are both flexible and sensitive to changes, thus avoiding the computational cost of full-band hyperspectral data analysis found in current methods. This guarantees the diversity of selected bands and improves computational efficiency. In each iteration, the most suitable band combination is adaptively and dynamically selected based on the current image features and restoration requirements, providing strong support for subsequent image restoration.

[0022] Specifically, given the observed degraded image The objective of this application is to... Select one band from the available bands. A subset of each band ,in , so that in a given In this case, the posterior probability of the selected band is maximized. This can be expressed as a maximum a posteriori (MAP) problem, as shown in Equation (5): (5) However, directly calculating this problem becomes infeasible due to the combinatorial explosion of possible band combinations. Therefore, this application employs a sampling-based efficient approximation method to approximate the MAP solution. According to Bayes' theorem, the posterior distribution... Through the likelihood term and prior terms The proportional relationship is sampled as shown in formula (6): (6) For the likelihood term This application assumes that bands with higher variance contain more information and are therefore better able to interpret the observation data. This application approximates the likelihood as the product of the variances of the selected bands, as shown in formula (7): (7) in It is a band The variance is calculated using formula (8): (8) here It is a band The mean.

[0023] For prior terms This application aims to select diverse and non-adjacent bands to avoid redundancy, as adjacent bands in hyperspectral images are typically highly correlated. This application employs a Gaussian function to reduce the probability of bands near the previously selected bands, as shown in equation (9): (9) This approach is conceptually similar to the use of Gaussian kernels in spectral clustering or the Gaussian function in non-maximum suppression (NMS) in object detection, where the Gaussian function models the similarity or influence between nearby entities. By combining likelihood and prior terms, this application can achieve similarity through a repeated sampling process. This is used to obtain an approximate MAP solution. Ultimately, from the selected band Select hyperspectral images of the channel It was then used for subsequent kernel estimation and hyperspectral image restoration.

[0024] Step 108: Using the sampling prior and degradation kernel estimation module, a random degradation matrix is ​​generated by Monte Carlo sampling, and the degradation matrix is ​​dynamically estimated by combining the data fidelity loss function to optimize the neural network parameters.

[0025] The main function of the sampling prior and degradation kernel estimation module is to generate a random degradation matrix through Monte Carlo sampling and optimize the parameters of the neural network by combining it with a data-fidelity loss function, thereby achieving dynamic estimation of the degradation matrix. This includes using the Monte Carlo method to randomly sample and generate a series of possible degradation matrices, which serve as labels to guide the learning of the neural network. The predicted degradation matrix output by the neural network is compared with these labels, and the loss function is calculated. Based on the predicted degradation matrix, the intermediate high-quality image obtained in the previous iteration is degraded, and the resulting degraded image is compared with the target image to calculate the data-fidelity loss function. By comprehensively considering these two types of loss information, the sampling prior and degradation kernel estimation module can effectively update the parameters of the neural network, making the predicted degradation matrix more accurate, thus providing a reliable basis for the restoration of high-quality images.

[0026] Specifically, a sampling kernel prior is first performed. Based on the Gaussian approximation of the fuzzy kernel, this application constructs an unsupervised but effective pre-empirical model. For any size... The Gaussian kernel can be defined by formula (10): (10) in, It is a precision matrix. These are spatial coordinates. Once the size of the kernel is determined, the fuzzy kernel... Completely composed of the precision matrix Decisions, such as variance and rotation angle Gauss's prior experience It can be expressed by the following integral formula (11): (11) However, collect all possible It is very difficult. Therefore, this application uses Monte Carlo (MC) simulation to approximate the above integral, as shown in Equation (12): (12) in, Indicates the sampling series, Indicates the first The precision matrix of the random sampling. Indicates the corresponding fuzzy kernel. Indicates the first The weights of the subsamples can be obtained by calculating the posterior probability formula (13): (13) in, This indicates the band selected from the high-resolution HSI.

[0027] For the estimation of the fuzzy kernel, a lightweight network is used. To estimate the kernel, the network uses fixed random noise. As input, the formula is as follows (14): (14) in, These are network parameters. The update includes two items: data consistency items. and random prior terms Given a kernel randomly sampled through Monte Carlo simulation. Network parameters The optimization can be expressed as formula (15): (15) in, (16) (17) Implemented using the well-known Adam algorithm. Optimization. The optimized version. It can be used to generate kernels via formula (14) .

[0028] Step 110: A frequency band projection and diffusion-guided image super-resolution module is used to solve for high-quality images by means of a pre-trained diffusion model, band selection and degradation matrix, and intermediate high-quality images are optimized by data fidelity loss function and total variation regularization.

[0029] The Band Projection and Diffusion Guided Image Super-Resolution (DHR) module optimizes intermediate high-quality images through a data fidelity loss function and total variation regularization to ensure that the recovered image is both faithful to the original image and has good visual quality. First, a pre-trained diffusion model is used, which can progressively recover image details and structure from noise. Combined with the intermediate degradation matrix, the DHR module can better understand the image degradation, thus enabling more targeted image restoration.

[0030] To ensure consistency between the restored image and the target image, the band-projection and diffusion-guided image super-resolution module calculates a data fidelity loss function to measure the difference between the restored and target images. To preserve image edges and details, the module also introduces a total variation regularization term to constrain image smoothness and detail. Through the combined effect of these two mechanisms, the band-projection and diffusion-guided image super-resolution module effectively optimizes intermediate high-quality images, ultimately yielding a clear and accurate restoration result.

[0031] Specifically, the Denoising Diffusion Probability Model (DDPM) defines a T-step forward process that progressively adds noise to the data, and a T-step backward process that reconstructs the target data from the noise. When using readily available DDPM... When solving the image restoration problem, the inverse process is used as an inference process to estimate the high-quality image, as shown in formula (18): (18) in, Indicates the first HR image estimated step These are hyperparameters. Given an estimate... Observed LR HSI and estimated kernel The data consistency term can be expressed as formula (19): (19) To constrain the estimation of HSI The additional noise, to achieve TV regularization of the HSI prior, can be expressed as formula (20): (20) Combining data consistency terms and TV regularization terms, HR HSI The prediction can be obtained by solving the following optimization problem, formula (21): (twenty one) Network-based full-band hyperspectral image projection employs a lightweight network. study and Mapping between It can be expressed as formula (22): (twenty two) Among them, network input It is fixed random noise. Parameters Optimize by minimizing the data consistency loss formula (23): (twenty three) in, (twenty four) Optimized Used to generate mapping matrix . Obtained With mapping matrix Multiply to obtain the full-band HRHSI : (25) It should be noted that this application uses a shared mapping matrix. express arrive and arrive The transformation.

[0032] Step 112: According to the three modules in the hyperspectral image super-resolution model, iterative recovery continues until the number of iterations meets the preset number. Then, the intermediate high-quality image obtained by the current iteration is the high-quality image after the target image is recovered.

[0033] In each iteration, the solution process for the degradation matrix subproblem is first invoked. A neural network, combined with random prior guidance and a data-fidelity loss function, updates and optimizes the estimated value of the degradation matrix. Then, using the newly updated degradation matrix and a pre-trained diffusion model, the high-quality image subproblem is solved. Through the combined constraints of the data-fidelity loss function and total variation regularization, a new intermediate high-quality image is obtained. This process alternates between the two iterations, with each iteration further improving the quality and accuracy of image restoration based on the previous result.

[0034] Specifically, band selection and the Gaussian dimensionality reduction module work by analyzing the posterior distribution. Sampling to select the most informative spectral bands Using formulas (6)-(9), we can obtain the result through formula (11). HSI of the band To ensure that subsequent calculations are performed The calculation is performed on a band-by-band basis to avoid full-band computation. The sampling prior and degenerate kernel estimation module obtains the kernel prior by randomly sampling from a Gaussian distribution. A lightweight kernel estimator combines the kernel prior and a data consistency term to estimate the kernel in an unsupervised manner. In the band projection and diffusion-guided image super-resolution module, a data consistency term and a total variation regularization term are used to guide the existing diffusion model for high-resolution HSI recovery. By alternately performing band selection, kernel estimation, and HSI recovery, high-resolution HSI recovery can be obtained. HSI of the band Ultimately, a lightweight network is learned to generate the mapping matrix. ,Will Projected onto full-band HSI It is worth noting that the entire DASR algorithm operates in an unsupervised manner, requiring no prior training and can be applied in a plug-and-play manner, significantly improving flexibility and generalization ability.

[0035] In the aforementioned hyperspectral image super-resolution method with alternating diffusion band selection and degradation adaptation, this application decomposes the objective function of the blind image restoration problem into two sub-problems: solving for a high-quality image and solving for a degradation matrix. These sub-problems are solved using an alternating iterative approach. In each iteration, the solution process for the degradation matrix sub-problem is first invoked. A neural network combined with random prior guidance and a data fidelity loss function updates and optimizes the estimated value of the degradation matrix. Then, using the newly updated degradation matrix and a pre-trained diffusion model, the high-quality image sub-problem is solved. Through the combined constraints of the data fidelity loss function and total variation regularization, a new intermediate high-quality image is obtained. This process continuously alternates between the two, with each iteration further improving the quality and accuracy of image restoration based on the previous result.

[0036] In one embodiment, the objective function of the blind image restoration problem is: ; in, Indicates a degraded image. This refers to a high-quality image, that is, an image after restoration. Represents the degeneracy matrix, and Indicates high-quality image and degeneracy matrix Priors, and This indicates its corresponding weight.

[0037] In one embodiment, the objective function is decomposed into two sub-problems: solving for the high-quality image and the degradation matrix. The objective function is decomposed into two subproblems: solving for the high-quality image and the degradation matrix. ; in, Indicates a degraded image. This refers to a high-quality image, that is, an image after restoration. Represents the degeneracy matrix, and Indicates high-quality image and degeneracy matrix Priors, and This indicates its corresponding weight.

[0038] In another embodiment, when using a neural network to solve the degenerate matrix subproblem, the optimization problem of the degenerate matrix is ​​transformed into a parameter optimization problem of the neural network. The transformed degenerate matrix subproblem is expressed as: ; In the above formula, and These are the blur kernel and the image data fidelity loss function, respectively, used to measure the difference between the estimated degradation matrix and the actual degradation; It is a random prior loss function. It is the TV regularization loss function, which, based on prior knowledge of the degradation matrix, helps the network learn a more accurate degradation model; and It is a hyperparameter that balances the weights between the two; It is a neural network Parameters; What is the amount of fixed noise input? It is the estimated image output by the diffusion model at step t.

[0039] In one embodiment, a band selection and Gaussian dimensionality reduction module is used to dynamically select bands with high information content through variance distribution and Gaussian attenuation model, including: Given the observed degraded image Later from Select one band from the available bands. The subset representation of each band is a maximum a posteriori probability problem, where, Indicates the image height. Indicates the image width; Efficient approximation methods based on sampling are used to approximate the maximum a posteriori probability problem. According to Bayes' theorem, the posterior distribution... Through the likelihood term and prior terms The proportional relationship is sampled, the likelihood term is approximated as the product of the variances of the selected bands, and the Gaussian function is used to reduce the probability of bands near the previously selected bands to design the prior term; Based on the prior terms and the likelihood terms, through a repeated sampling process This will allow us to obtain an approximate solution to the maximum a posteriori probability problem. ; From the selected band Select hyperspectral images of the channel It is used for subsequent kernel estimation and hyperspectral image restoration.

[0040] In one embodiment, approximating the likelihood term as the product of the variances of the selected band includes: The likelihood term is approximated as the product of the variances of the selected bands as follows: ; in, It is a band variance Indicates the selected band.

[0041] In one embodiment, a Gaussian function is used to reduce the probability design of bands near the previously selected band, including: The prior term is designed by using a Gaussian function to reduce the probability of bands near the previously selected band: ; in, Indicates the selected band. Indicates the band number. and Indicates different frequency bands.

[0042] In one embodiment, a sampling prior and degradation kernel estimation module is employed to generate a random degradation matrix using Monte Carlo sampling. The degradation matrix is ​​dynamically estimated by combining a data fidelity loss function with neural network parameters. The method also includes: The kernel prior, i.e., the random degeneracy matrix, is obtained by randomly sampling from the Gaussian distribution. The random degeneracy matrix is ​​used as a label to guide the learning of the neural network. The data fidelity loss function and the loss function for random prior terms are designed by comparing the predicted degradation matrix output by the neural network with the label. The network parameters of the lightweight network are optimized using the data fidelity loss function and the loss function of the random prior term. The kernel, i.e. the degradation matrix, is estimated based on the optimized lightweight network.

[0043] In one embodiment, the network parameters of the lightweight network are optimized using a data fidelity loss function and a loss function with random prior terms, including: Given a kernel randomly sampled by Monte Carlo simulation Network parameters The optimization is expressed as: ; in, ; ; The data fidelity loss function represents the network parameters. The loss function represents the random prior term. This refers to a high-quality image, that is, an image after restoration. Indicates a lightweight network. Indicates fixed random noise. Indicates a degraded image. Indicates the corresponding fuzzy kernel. Indicates the first Weights for each sample.

[0044] In one embodiment, a band projection and diffusion-guided image super-resolution module is employed. High-quality images are solved using a pre-trained diffusion model, band selection, and degradation matrix. Intermediate high-quality images are optimized using a data fidelity loss function and total variation regularization, including: Using a pre-trained diffusion model and intermediate degradation matrix to solve for high-quality images, the details and structure of the image are gradually recovered from the noise, as shown below: ; in, Indicates the first High-quality images estimated step by step, It's a hyperparameter; Calculate the data fidelity loss function and total variation regularization for high-quality images, and optimize intermediate high-quality images using the data fidelity loss function and total variation regularization.

[0045] In one embodiment, calculating the data fidelity loss function and total variation regularization for the high-quality image, and optimizing the intermediate high-quality image using the data fidelity loss function and total variation regularization, includes: Given an estimate Degraded images observed And the estimated kernel K, The data fidelity loss function is expressed as: ; The total variation is regularized as follows: ; Combining data consistency terms and TV regularization terms, high-quality images The prediction is obtained by solving the following optimization problem: ; Network-based full-band hyperspectral image projection employs a lightweight network. study and Mapping between , represented as: ; Among them, network input It is fixed random noise; parameter Optimize by minimizing the formula for data consistency loss: ; in, ; Optimized Used to generate mapping matrix ; Obtained With mapping matrix Multiplying yields a high-quality image across the entire band. : .

[0046] In one embodiment, pseudocode is provided to implement the above-described blind hyperspectral image super-resolution method based on alternating diffusion dynamic band selection and degradation adaptation, as shown below: Input: Low-quality image Pre-trained diffusion model ; initialization: ; for to0do; % DSB module; Calculate the likelihood term using formulas (7) and (8). ; Calculate the prior terms according to formula (9) ; According to formula (6), from the posterior distribution Mid-sampling ; according to ,get ; % DSK module; calculate ; Sampling using the Monte Carlo method ; calculate ; ; % DHR module; ; ; ; ; ; ; Output: Estimated high-resolution HSI Fuzzy kernel .

[0047] It should be understood that, although Figure 1 The steps in the flowchart are shown sequentially as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order in which these steps are executed, and they can be performed in other orders. Figure 1 At least some of the steps in the process may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least some of the sub-steps or stages of other steps.

[0048] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0049] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these modifications and improvements all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A hyperspectral image super-resolution method with alternating diffusion band selection and degradation adaptation, characterized in that, The method includes: Obtain the target image to be restored; construct the objective function for the blind image restoration problem of the target image; decompose the objective function into two sub-problems: solving for the high-quality image and the degradation matrix; A hyperspectral image super-resolution model is constructed; the hyperspectral image super-resolution model includes a band selection and Gaussian dimensionality reduction module, a sampling prior and degradation kernel estimation module, and a frequency band projection and diffusion-guided image super-resolution module; A band selection and Gaussian dimensionality reduction module is used to dynamically select bands with high information content through variance distribution and Gaussian attenuation model; A sampling prior and degradation kernel estimation module is adopted. Monte Carlo sampling is used to generate a random degradation matrix. The neural network parameters are optimized by combining the data fidelity loss function to achieve dynamic estimation of the degradation matrix. A frequency band projection and diffusion-guided image super-resolution module is adopted. High-quality images are solved by using a pre-trained diffusion model, band selection and degradation matrix. Intermediate high-quality images are optimized by data fidelity loss function and total variation regularization. The high-resolution model of hyperspectral images is based on the three modules of iterative recovery until the number of iterations meets the preset number. The intermediate high-quality image obtained in the current iteration is the high-quality image after the target image is recovered.

2. The method according to claim 1, characterized in that, The objective function for the blind image restoration problem is: in, Indicates a degraded image. This refers to a high-quality image, that is, an image after restoration. Represents the degeneracy matrix, and Indicates high-quality image and degeneracy matrix Priors, and This indicates its corresponding weight.

3. The method according to claim 1, characterized in that, The objective function is decomposed into two sub-problems: solving for the high-quality image and the degradation matrix. The objective function is decomposed into two subproblems: solving for the high-quality image and the degradation matrix. in, Indicates a degraded image. This refers to a high-quality image, that is, an image after restoration. Represents the degeneracy matrix, and Indicates high-quality image and degeneracy matrix Priors, and This indicates its corresponding weight.

4. The method according to claim 1, characterized in that, A band selection and Gaussian dimensionality reduction module is used to dynamically select bands with high information content through variance distribution and Gaussian attenuation models, including: Given the observed degraded image Later from Select one band from the available bands. The subset representation of each band is a maximum a posteriori probability problem, where, Indicates the image height. Indicates the image width; Efficient approximation methods based on sampling are used to approximate the maximum a posteriori probability problem. According to Bayes' theorem, the posterior distribution... Through the likelihood term and prior terms The proportional relationship is sampled, the likelihood term is approximated as the product of the variances of the selected bands, and the Gaussian function is used to reduce the probability of bands near the previously selected bands to design the prior term; Based on the prior terms and likelihood terms, through a repeated sampling process This will allow us to obtain an approximate solution to the maximum a posteriori probability problem. ; From the selected band Select Hyperspectral images of the channel It is used for subsequent kernel estimation and hyperspectral image restoration.

5. The method according to claim 4, characterized in that, The likelihood term is approximated as the product of the variances of the selected bands, including: The likelihood term is approximated as the product of the variances of the selected bands as follows: in, It is a band variance Indicates the selected band.

6. The method according to claim 1, characterized in that, A Gaussian function is used to reduce the probability of bands near the previously selected band in the design of prior terms, including: The prior term is designed by using a Gaussian function to reduce the probability of bands near the previously selected band: in, Indicates the selected band. Indicates the band number. and Indicates different frequency bands.

7. The method according to claim 1, characterized in that, The system employs a sampling prior and degradation kernel estimation module, utilizing Monte Carlo sampling to generate a random degradation matrix. It then combines this with a data-fidelity loss function to optimize neural network parameters, achieving dynamic estimation of the degradation matrix. Other features include: The kernel prior, i.e., the random degeneracy matrix, is obtained by randomly sampling from a Gaussian distribution. This random degeneracy matrix is ​​used as a label to guide the learning of the neural network. The predicted degradation matrix output by the neural network is compared with the label to design a data fidelity loss function and a loss function for random prior terms; The network parameters of the lightweight network are optimized using the data fidelity loss function and the loss function of the random prior term. The kernel, i.e. the degradation matrix, is estimated based on the optimized lightweight network.

8. The method according to claim 7, characterized in that, The network parameters of lightweight networks are optimized using a data fidelity loss function and a loss function with random prior terms, including: Given a kernel randomly sampled by Monte Carlo simulation Network parameters The optimization is expressed as: in, The data fidelity loss function represents the network parameters. The loss function represents the random prior term. This refers to a high-quality image, that is, an image after restoration. Indicates a lightweight network. Indicates fixed random noise. Indicates a degraded image. Indicates the corresponding fuzzy kernel. Indicates the first Weights for each sample.

9. The method according to claim 1, characterized in that, A frequency band projection and diffusion-guided image super-resolution module is employed. High-quality images are obtained using a pre-trained diffusion model, band selection, and degradation matrix. Intermediate high-quality images are optimized through a data fidelity loss function and total variation regularization, including: Using a pre-trained diffusion model and intermediate degradation matrix to solve for high-quality images, the details and structure of the image are gradually recovered from the noise, as shown below: in, Indicates the first High-quality images estimated step by step, It's a hyperparameter; Calculate the data fidelity loss function and total variation regularization for high-quality images, and optimize intermediate high-quality images using the data fidelity loss function and total variation regularization.

10. The method according to claim 9, characterized in that, Calculate the data fidelity loss function and total variation regularization for high-quality images, and optimize intermediate high-quality images using the data fidelity loss function and total variation regularization, including: Given an estimate Degraded images observed And the estimated kernel K, The data fidelity loss function is expressed as: The total variation is regularized as follows: Combining data consistency terms and TV regularization terms, high-quality images The prediction is obtained by solving the following optimization problem: Network-based full-band hyperspectral image projection employs a lightweight network. study and Mapping between , represented as: ； Among them, network input It is fixed random noise; parameter Optimize by minimizing the formula for data consistency loss: in, ; Optimized Used to generate mapping matrix ; Obtained With mapping matrix Multiplying yields a high-quality image across the entire band. : 。

Images