An ultraprecision machining method, device and electronic equipment

Abstract

The application provides a kind of super-resolution imaging method, device and electronic equipment, wherein the method comprises: obtaining a low-resolution hyperspectral image;The low-resolution hyperspectral image is nonlinearly initialized, and a spectral response matrix is obtained;The low-resolution hyperspectral image is estimated for spectral degradation, and the data after spectral degradation estimation is obtained;The spectral response matrix and the data after spectral degradation estimation are respectively blocked in spatial dimension, and a plurality of corresponding matrix blocks are obtained;The plurality of matrix blocks are reconstructed, and a target high-resolution hyperspectral image is obtained. By implementing the present application, under the condition of no accurate prior information, low-resolution hyperspectral image and high-resolution low-spectrum image can be efficiently fused to generate high-quality high-resolution low-spectrum image, optimize the calculation efficiency, accurately capture the nonlinear characteristics of the image, and realize stable, efficient and optimized solution process. Meet a variety of use requirements.

Description

Technical Field

[0001] This application relates to the field of hyperspectral image super-resolution technology, and more specifically, to a super-resolution imaging method, apparatus, and electronic device. Background Technology

[0002] Existing super-resolution imaging methods mainly include the following approaches: ① Algorithms based on Coupled Nonnegative Matrix Factorization (CNMF); ② Algorithms based on sparse representations (such as K-SVD, OMP, SOMP, etc.); ③ Optimization model algorithms that introduce regularization (such as HySure, NSSR, CO-CNMF, etc.).

[0003] The above methods are based on the matrix factorization framework and introduce various regularization terms to enhance spatial smoothness, preserve edges, or improve spectral fidelity. They all attempt to estimate the decomposed matrix by solving a complex optimization problem of an observation model that links low-resolution hyperspectral images and high-resolution low-spectral images, and then reconstruct the target high-resolution hyperspectral image.

[0004] The above methods have many drawbacks, such as: strong dependence on prior information, which can easily lead to decreased algorithm robustness and deterioration of reconstruction performance; low computational efficiency, making it difficult to meet real-time requirements and integrate into spatial platforms or online processing systems with high real-time requirements; insufficient modeling of the nonlinear structure of high-dimensional data, which may lead to inaccurate estimation of spectral information, thereby affecting the spectral fidelity of the reconstructed image; and high limitations of the optimization strategy, which restricts the full utilization of computing resources and results in long computation time. Summary of the Invention

[0005] The purpose of this application is to provide a super-resolution imaging method, device, and electronic device that can efficiently fuse low-resolution hyperspectral images and high-resolution low-spectral images to generate high-quality high-resolution low-spectral images without precise prior information, optimize computational efficiency, accurately capture the nonlinear features of images, achieve a stable, efficient, and optimized solution process, and meet various application requirements.

[0006] In a first aspect, embodiments of this application provide a super-resolution imaging method, the method comprising: Acquire low-resolution hyperspectral images; The low-resolution hyperspectral image is nonlinearly initialized to obtain the spectral response matrix; The low-resolution hyperspectral image is subjected to spectral degradation estimation to obtain the data after spectral degradation estimation; The spectral response matrix and the data after spectral degradation estimation are respectively divided into blocks in the spatial dimension to obtain multiple corresponding matrix blocks; The target high-resolution hyperspectral image is obtained by reconstructing multiple matrix blocks.

[0007] In the above implementation process, by performing spectral degradation estimation on the low-resolution hyperspectral image and dividing and reconstructing the spectral response matrix and the data after spectral degradation estimation, a high-quality high-resolution low-spectral image can be generated by efficiently fusing low-resolution hyperspectral images and high-resolution low-spectral images without precise prior information. This optimizes computational efficiency, accurately captures the nonlinear features of the image, and achieves a stable, efficient, and optimized solution process, meeting various application requirements.

[0008] Further, the step of performing nonlinear initialization on the low-resolution hyperspectral image to obtain the spectral response matrix includes: Obtain the nearest neighbor node of each spectral vector in the spectral space of the low-resolution hyperspectral image; Based on the Euclidean distance of the nearest neighbor nodes and the adjustable parameters, the joint probability of the low-resolution hyperspectral image in the high-dimensional space after symmetry is obtained; Construct a loss function between the symmetric joint probabilities in the high-dimensional space and the corresponding symmetric joint probabilities in the low-dimensional space; The loss function is optimized and iterated to obtain the spectral response matrix.

[0009] In the above implementation process, nonlinear initialization of low-resolution hyperspectral images can improve the spectral fidelity of reconstructed images, accurately capture spectral nonlinear mixing characteristics, correct deviations in low-resolution hyperspectral images, improve initialization stability, and further enhance the utilization rate of low-resolution hyperspectral images, thereby increasing practicality and compatibility and providing fundamental support for spectral fidelity and accuracy improvement.

[0010] Further, the step of performing spectral degradation estimation on the low-resolution hyperspectral image to obtain the spectral degradation-estimated data includes: Obtain the mapping relationship between the low-resolution hyperspectral image and the high-resolution low-spectral image; Based on the mapping relationship, the high-resolution low-spectral image is initialized to obtain the initialized high-resolution low-spectral image; Applying regularization constraints to the initialized high-resolution low-spectral image yields a spectral degradation matrix. The initialized high-resolution low-spectral image and the spectral degradation matrix are optimized and iterated to obtain an optimized high-resolution low-spectral image and an optimized spectral degradation matrix. The optimized high-resolution low-spectral image and the optimized spectral degradation matrix are used as the data after spectral degradation estimation.

[0011] In the above implementation process, the spectral degradation estimation operation can improve the accuracy of spectral degradation estimation, accurately capture the degradation law of the spectrum, broaden the applicable scenarios and applicability, simplify the operation process, improve the estimation efficiency, and enhance the stability and anti-interference ability of degradation estimation.

[0012] Further, the step of obtaining the mapping relationship between the low-resolution hyperspectral image and the high-resolution low-spectral image includes: Obtain the physical imaging model for hyperspectral imaging; Based on the physical imaging model of the hyperspectral imaging, an observation model is constructed between the low-resolution hyperspectral image, the high-resolution low-spectral image, and the target high-resolution hyperspectral image. Based on the observation model, the mapping relationship between the low-resolution hyperspectral image and the high-resolution low-spectral image is obtained.

[0013] In the above implementation process, by mapping the low-resolution hyperspectral image to the high-resolution low-spectral image, the spatial-spectral dual precision of the low-resolution hyperspectral image can be improved, and the corresponding rules between spatial detail features can be accurately captured. This effectively mines spectral coupling information, optimizes the spectral information transfer and fusion effect, and improves the stability and reliability of subsequent hyperspectral image processing.

[0014] Further, the step of dividing the spectral response matrix and the spectral degradation-estimated data into spatial blocks to obtain corresponding multiple matrix blocks includes: The low-resolution hyperspectral image is upsampled to the same spatial size as the high-resolution low-spectral image to obtain the upsampled low-resolution hyperspectral image. Obtain the spectral response matrix and spatial response matrix used to characterize the high-resolution hyperspectral image of the target; The upsampled low-resolution hyperspectral image, the optimized high-resolution low-spectral image, the spectral response matrix, and the spatial response matrix are uniformly divided in the spatial dimension to obtain multiple matrix blocks.

[0015] In the above implementation process, the block-based approach can effectively solve the technical pain points of insufficient spatial resolution in low-resolution hyperspectral images, such as loss of spatial details and blurred spatial structure. While improving the spatial dimension quality of the image, it maximizes the preservation of the original spectral information, providing reliable support for the subsequent processing and practical application of hyperspectral images, and significantly improving the practicality of the technology.

[0016] Further, the step of obtaining the spectral response matrix and spatial response matrix for characterizing the high-resolution hyperspectral image of the target includes: The target high-resolution hyperspectral image is decomposed and characterized into the spectral response matrix and the spatial response matrix.

[0017] In the above implementation process, the processing strategy for high-dimensional data can be optimized and simplified by adopting the above method. It does not rely on precise prior information, improves the applicability and robustness of the algorithm during operation, and provides data support for the accuracy of the target high-resolution hyperspectral image.

[0018] Furthermore, prior to the step of reconstructing the plurality of matrix blocks to obtain the target high-resolution hyperspectral image, the method further includes: The optimization of the matrix blocks of spatial response matrices and spectral response matrices in the multiple matrix blocks is specifically as follows: Based on the alternating direction multiplier method, the matrix blocks of the spatial response matrix and the matrix blocks of the spectral response matrix are optimized according to the matrix blocks of low-resolution hyperspectral images and high-resolution low-spectral images in the multiple matrix blocks, to obtain optimized matrix blocks of the spatial response matrix and the spectral response matrix.

[0019] In the above implementation process, by optimizing the matrix blocks of the spatial response matrix and the matrix blocks of the spectral response matrix, the defects of existing matrix blocks such as bias, poor adaptability and weak anti-interference ability can be effectively solved, the representation accuracy of the spatial response matrix and the spectral response matrix on the spatial and spectral features of hyperspectral images can be improved, and the performance of hyperspectral technology can be further enhanced.

[0020] Further, the step of reconstructing the plurality of matrix blocks to obtain a target high-resolution hyperspectral image includes: The optimized spatial response matrix and spectral response matrix blocks are reorganized according to their original spatial positions to obtain the global spatial response matrix and global spectral response matrix. By performing matrix multiplication, the spatial response matrix is ​​replaced with the global spatial response matrix, and the spectral response matrix is ​​replaced with the global spectral response matrix, to obtain a high-resolution hyperspectral image of the target.

[0021] In the above implementation process, by reorganizing the optimized matrix blocks, the problems of scattered and independent optimized matrix blocks, insufficient coordination and poor correlation are effectively solved, realizing the integration and efficient coordination of optimized matrix blocks, providing complete and coordinated matrix support for hyperspectral image processing, and improving the practicality of the technology.

[0022] Secondly, embodiments of this application also provide a super-resolution imaging device, the device comprising: The acquisition module is used to acquire low-resolution hyperspectral images; A nonlinear initialization module is used to perform nonlinear initialization on the low-resolution hyperspectral image to obtain a spectral response matrix; The spectral degradation estimation module is used to perform spectral degradation estimation on the low-resolution hyperspectral image to obtain the data after spectral degradation estimation. The block segmentation module is used to divide the spectral response matrix and the spectral degradation estimation data into blocks in the spatial dimension to obtain multiple corresponding matrix blocks; The reconstruction module is used to reconstruct multiple matrix blocks to obtain a high-resolution hyperspectral image of the target.

[0023] In the above implementation process, by performing spectral degradation estimation on the low-resolution hyperspectral image and dividing and reconstructing the spectral response matrix and the data after spectral degradation estimation, a high-quality high-resolution low-spectral image can be generated by efficiently fusing low-resolution hyperspectral images and high-resolution low-spectral images without precise prior information. This optimizes computational efficiency, accurately captures the nonlinear features of the image, and achieves a stable, efficient, and optimized solution process, meeting various application requirements.

[0024] Thirdly, an electronic device provided in this application includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the method as described in any of the first aspects.

[0025] Fourthly, embodiments of this application provide a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform the method described in any of the first aspects.

[0026] Other features and advantages of this disclosure will be set forth in the following description, or some features and advantages may be inferred from the description or determined without doubt, or may be learned by practicing the techniques described above.

[0027] It can be implemented in accordance with the contents of the specification. The preferred embodiments of this application are described in detail below with reference to the accompanying drawings. Attached Figure Description

[0028] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments of this application will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation on the range. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0029] Figure 1A flowchart illustrating a super-resolution imaging method is provided for embodiments of this application; Figure 2 A schematic diagram of the structure of a super-resolution imaging device is provided for embodiments of this application; Figure 3 This is a schematic diagram of the structural composition of the electronic device provided in the embodiments of this application. Detailed Implementation

[0030] The technical solutions in the embodiments of this application will now be described with reference to the accompanying drawings.

[0031] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this application, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0032] The specific embodiments of this application will be described in further detail below with reference to the accompanying drawings and examples. The following examples are used to illustrate this application, but are not intended to limit the scope of this application.

[0033] Existing methods rely heavily on prior information, have low computational efficiency, are difficult to meet real-time requirements, lack sufficient modeling of nonlinear structures in high-dimensional data, and have high limitations in optimization strategies, which restricts the full utilization of computing resources and fails to meet the diverse needs of imaging.

[0034] This application provides a fast, robust, and high-fidelity hyperspectral image super-resolution method that efficiently fuses low-resolution hyperspectral images with high-resolution low-spectral images without requiring precise prior information, thus meeting the urgent needs of remote sensing, medical, security and other fields for real-time and accurate imaging analysis.

[0035] Example 1 Figure 1 This is a flowchart illustrating the super-resolution imaging method provided in an embodiment of this application, as shown below. Figure 1 As shown, the method includes: S1, acquire low-resolution hyperspectral images; S2, perform nonlinear initialization on the low-resolution hyperspectral image to obtain the spectral response matrix; S3, perform spectral degradation estimation on the low-resolution hyperspectral image to obtain the data after spectral degradation estimation; S4, divide the spectral response matrix and the data after spectral degradation estimation into blocks in the spatial dimension to obtain multiple corresponding matrix blocks; S5 reconstructs multiple matrix blocks to obtain a high-resolution hyperspectral image of the target.

[0036] In the above implementation process, by performing spectral degradation estimation on low-resolution hyperspectral images and then dividing and reconstructing the spectral response matrix and the data after spectral degradation estimation, high-quality high-resolution low-spectral images can be generated by efficiently fusing low-resolution hyperspectral images and high-resolution low-spectral images without precise prior information. This optimizes computational efficiency, accurately captures the nonlinear features of the images, and achieves a stable, efficient, and optimized solution process, meeting various application requirements.

[0037] The core of this application lies in using an innovative algorithm framework to efficiently fuse low-resolution hyperspectral images (LR-HSI) and high-resolution low-spectral images (HR-LSI) to reconstruct a high-resolution hyperspectral image (HR-HSI, which in this application embodiment corresponds to the target high-resolution hyperspectral image).

[0038] Furthermore, S2 includes: Obtain the nearest neighbor node of each spectral vector in the spectral space of a low-resolution hyperspectral image; Based on the Euclidean distance between nearest neighbors and adjustable parameters, the joint probability of low-resolution hyperspectral images in high-dimensional space after symmetry transformation is obtained. Construct a loss function between the symmetric joint probabilities in the high-dimensional space and the corresponding symmetric joint probabilities in the low-dimensional space; The loss function is optimized iteratively to obtain the spectral response matrix.

[0039] In the above implementation process, nonlinear initialization of low-resolution hyperspectral images can improve the spectral fidelity of reconstructed images, accurately capture spectral nonlinear mixing characteristics, correct deviations in low-resolution hyperspectral images, improve initialization stability, and further enhance the utilization rate of low-resolution hyperspectral images, thereby increasing practicality and compatibility and providing fundamental support for spectral fidelity and accuracy improvement.

[0040] In this embodiment, considering the inherent nonlinear structure of hyperspectral data in the spectral dimension, the Unified Manifold Approximation and Projection (UMAP) algorithm is used to perform nonlinear dimensionality reduction and decomposition on the image matrix (denoted as Y) of the input low-resolution hyperspectral image, resulting in a spectral response matrix that better conforms to the inherent manifold structure of the data. UMAP captures local and global structure by constructing and optimizing a graph representation of high-dimensional spatial data, as follows: For each spectral vector (pixel) of LR-HSI, its k nearest neighbors are calculated in the spectral space. The joint probability after symmetry in the high-dimensional space is calculated using a distance metric (Euclidean distance in this embodiment) and an adjustable parameter (min_dist in this embodiment). ).

[0041] The low-dimensional embedding is obtained by minimizing the loss function (such as cross-entropy) defined by minimizing the attraction between the joint probability after symmetry transformation and the corresponding points in the low-dimensional representation, and then performing gradient descent. .

[0042] Where k is the number of nearest neighbor nodes, 5 <k<100,0.001<min_dist<0.5。

[0043] The above techniques can quickly obtain a high-quality initial estimate of spectral primitives, reducing the number of subsequent optimization iterations.

[0044] This application introduces UMAP for nonlinear decomposition initialization. Its graph-based manifold learning method better preserves the local and global structure of the high-dimensional spectral space, thus obtaining an initial estimate that more closely approximates the real spectral characteristic primitives. Further optimization on this basis enables more accurate separation and reconstruction of spectral information. Simultaneously, ADMM optimization of the spatial response matrix C with a sparsity regularization term effectively promotes the sparse representation of the spatial structure, helping to preserve edge and texture details.

[0045] Furthermore, S3 includes: Obtain the mapping relationship between low-resolution hyperspectral images and high-resolution low-spectral images; Based on the mapping relationship, the high-resolution low-spectral image is initialized to obtain the initialized high-resolution low-spectral image; The initial high-resolution low-spectral image is subjected to regularization constraints to obtain the spectral degradation matrix; The initialized high-resolution low-spectral image and spectral degradation matrix are optimized iteratively to obtain the optimized high-resolution low-spectral image and the optimized spectral degradation matrix. The optimized high-resolution low-spectral image and the optimized spectral degradation matrix are used as the data after spectral degradation estimation.

[0046] In the above implementation process, the spectral degradation estimation operation can improve the accuracy of spectral degradation estimation, accurately capture the degradation law of the spectrum, broaden the applicable scenarios and applicability, simplify the operation process, improve the estimation efficiency, and enhance the stability and anti-interference ability of degradation estimation.

[0047] In obtaining the spectral response matrix Subsequently, the embodiments of this application do not rely on a preset spectral response function (SRF) for prior calculation, but instead analyze the intrinsic relationship (mapping relationship) between LR-HSI (Y) and HR-LSI (in this application, the image matrix corresponding to the high-resolution low-spectral image is denoted as Z).

[0048] Furthermore, the step of obtaining the mapping relationship between the low-resolution hyperspectral image and the high-resolution low-spectral image includes: Obtain the physical imaging model for hyperspectral imaging; Based on the physical imaging model of hyperspectral imaging, an observation model is constructed between low-resolution hyperspectral images, high-resolution low-spectral images, and high-resolution hyperspectral images of the target. Based on the observation model, the mapping relationship between low-resolution hyperspectral images and high-resolution low-spectral images is obtained.

[0049] In the above implementation process, by mapping the low-resolution hyperspectral image to the high-resolution low-spectral image, the spatial-spectral dual precision of the low-resolution hyperspectral image can be improved, and the corresponding rules between spatial detail features can be accurately captured. This effectively mines spectral coupling information, optimizes the spectral information transfer and fusion effect, and improves the stability and reliability of subsequent hyperspectral image processing.

[0050] The physical imaging model is: Y=XBS and Z=RX, where B is the spatial blur matrix, S is the downsampling matrix (which can be initialized using Gaussian blur and a fixed downsampling rate), and X is HR-HSI.

[0051] Based on the above formula, the relationship between the low-resolution hyperspectral image, the high-resolution low-spectral image and the target high-resolution hyperspectral image (i.e., the observation model) can be constructed. After combining the equations and eliminating X, the mapping relationship between the low-resolution hyperspectral image and the high-resolution low-spectral image can be obtained.

[0052] The formula for initializing a high-resolution low-spectral image is as follows:

[0053] in, This is the initial high-resolution low-spectral image, i.e., the low-resolution low-spectral image after spatial degradation of the high-resolution low-spectral image.

[0054] Direct estimation of the spectral degradation matrix (denoted as R): This application estimates the spectral degradation matrix by solving a regularized optimization problem:

[0055] in, This is the regularization coefficient, which can be set to 0.1.

[0056] The initialized high-resolution low-spectral image is iteratively optimized using an optimization problem with non-negativity constraints. And spectral degradation matrix:

[0057] Among them, the left side of the equation For the optimized high-resolution low-spectral image, the right side of the equation is... This is the initialized high-resolution low-spectral image.

[0058] The two equations above are optimized alternately, with one variable fixed and the other variable updated, until convergence is achieved, ultimately yielding the optimized high-resolution low-spectral image and spectral degradation matrix.

[0059] None of the above techniques require precise point spread function (PSF) or spectral response function (SRF) prior information, which enhances robustness and effectively reduces dependence on physical priors. UMAP adaptively extracts the intrinsic structure of the data through manifold learning, while the estimation of R is entirely based on the intrinsic relationships between the input image pairs.

[0060] Furthermore, S4 includes: A low-resolution hyperspectral image is upsampled to the same spatial size as a high-resolution hyperspectral image to obtain an upsampled low-resolution hyperspectral image. Obtain the spectral response matrix and spatial response matrix used to characterize the high-resolution hyperspectral image of the target; The upsampled low-resolution hyperspectral image, the optimized high-resolution low-spectral image, the spectral response matrix, and the spatial response matrix are uniformly divided into multiple matrix blocks in the spatial dimension.

[0061] In the above implementation process, the block-based approach can effectively solve the technical pain points of insufficient spatial resolution in low-resolution hyperspectral images, such as loss of spatial details and blurred spatial structure. While improving the spatial dimension quality of the image, it maximizes the preservation of the original spectral information, providing reliable support for the subsequent processing and practical application of hyperspectral images, and significantly improving the practicality of the technology.

[0062] This application embodiment combines the matrix block parallel optimization strategy of ADMM. By upsampling the low-resolution hyperspectral image and the corresponding matrix block, the optimization problem of each sub-block becomes smaller and simpler. All matrix blocks can be solved completely independently and in parallel, ensuring the stability and efficiency of solving sub-problems.

[0063] Further, the steps of obtaining the spectral response matrix and spatial response matrix used to characterize the high-resolution hyperspectral image of the target include: The high-resolution hyperspectral image of the target is decomposed and characterized into a spectral response matrix and a spatial response matrix.

[0064] In the above implementation process, the processing strategy for high-dimensional data can be optimized and simplified by adopting the above method. It does not rely on precise prior information, improves the applicability and robustness of the algorithm during operation, and provides data support for the accuracy of the target high-resolution hyperspectral image.

[0065] In this application embodiment, the target high-resolution hyperspectral image is decomposed and characterized into spectral response matrix and spatial response matrix using the following formula:

[0066] in, Let C be the spectral response matrix and C be the spatial response matrix.

[0067] LR-HSI primarily provides spectral information, while HR-LSI primarily provides spatial information. This approach reasonably decouples the complex global coupled optimization problem into two relatively independent and more easily parallelizable subproblems, which are then optimized for the spectral response matrix and the spatial response matrix, respectively. This sacrifices a small amount of coupling information in exchange for a significant improvement in computational efficiency.

[0068] Upsampling the LR-HSI matrix Y to the same spatial size as HR-LSI yields an upsampled low-resolution hyperspectral image. .

[0069] The upsampled low-resolution hyperspectral image, the optimized high-resolution low-spectral image, the spectral response matrix, and the spatial response matrix are uniformly divided into blocks in the spatial dimension, i.e. Z (representing the optimized high-resolution low-spectral image) Dividing C into uniform blocks along the spatial dimension yields multiple matrix blocks, represented as follows:

[0070] Furthermore, before the step of reconstructing multiple matrix blocks to obtain a high-resolution hyperspectral image of the target, the following steps are also included: The optimization is performed on the matrix blocks of spatial response matrices and spectral response matrices within multiple matrix blocks, specifically as follows: Based on the alternating direction multiplier method, the matrix blocks of the spatial response matrix and the matrix blocks of the spectral response matrix are optimized according to the matrix blocks of low-resolution hyperspectral images and high-resolution low-spectral images in multiple matrix blocks, so as to obtain the optimized matrix blocks of the spatial response matrix and the matrix blocks of the spectral response matrix.

[0071] In the above implementation process, by optimizing the matrix blocks of the spatial response matrix and the matrix blocks of the spectral response matrix, the defects of existing matrix blocks such as bias, poor adaptability and weak anti-interference ability can be effectively solved, the representation accuracy of the spatial response matrix and the spectral response matrix on the spatial and spectral features of hyperspectral images can be improved, and the performance of hyperspectral technology can be further enhanced.

[0072] For each block i, the matrix block of the spatial response matrix The optimization subproblems are solved using the Alternating Direction Multiplier Method (ADMM) framework. ADMM decomposes the original problem into a series of simpler subproblems by introducing auxiliary variables and Lagrange multipliers, and solves them iteratively, exhibiting good convergence.

[0073] The formula is as follows:

[0074] in, is the sparsity regularization coefficient.

[0075] For the matrix block of the spectral response matrix The solution can be obtained in parallel using standard optimization tools (such as MATLAB's fmincon function) or a framework similar to ADMM, as shown in the following formula:

[0076] Furthermore, S5 includes: The optimized spatial response matrix and spectral response matrix blocks are reorganized according to their original spatial positions to obtain the global spatial response matrix and global spectral response matrix. By performing matrix multiplication, the spatial response matrix is ​​replaced with the global spatial response matrix, and the spectral response matrix is ​​replaced with the global spectral response matrix, thus obtaining a high-resolution hyperspectral image of the target.

[0077] In the above implementation process, by reorganizing the optimized matrix blocks, the problems of scattered and independent optimized matrix blocks, insufficient coordination and poor correlation are effectively solved, realizing the integration and efficient coordination of optimized matrix blocks, providing complete and coordinated matrix support for hyperspectral image processing, and improving the practicality of the technology.

[0078] The optimization processes for the spatial response matrix block and the spectral response matrix block are independent of each other and can be executed in parallel in a multi-core CPU, GPU or distributed computing environment, greatly reducing the overall optimization time.

[0079] After the above optimization process is completed, the matrix blocks of the optimized spatial response matrix and the matrix blocks of the spectral response matrix are reorganized according to their original spatial positions, and the global spatial response matrix and the global spectral response matrix are substituted into the formula. Matrix multiplication is performed to obtain the target high-resolution hyperspectral image (HR-HSI) matrix X. After obtaining matrix X, X is converted from matrix form to three-dimensional image cube format. X is an M*N matrix, and the three-dimensional image cube format is H*W*L, thus realizing the conversion from matrix form back to three-dimensional image cube format.

[0080] Example 2 To implement the method corresponding to Embodiment 1 above and achieve the corresponding functional and technical effects, a super-resolution imaging device is provided below, such as... Figure 2 As shown, the device includes: Module 1 is used to acquire low-resolution hyperspectral images; Nonlinear initialization module 2 is used to perform nonlinear initialization on low-resolution hyperspectral images to obtain the spectral response matrix; Spectral degradation estimation module 3 is used to perform spectral degradation estimation on low-resolution hyperspectral images to obtain data after spectral degradation estimation; Blocking module 4 is used to divide the spectral response matrix and the data after spectral degradation estimation into blocks in the spatial dimension to obtain multiple corresponding matrix blocks; Reconstruction module 5 is used to reconstruct multiple matrix blocks to obtain a high-resolution hyperspectral image of the target.

[0081] In the above implementation process, by performing spectral degradation estimation on low-resolution hyperspectral images and then dividing and reconstructing the spectral response matrix and the data after spectral degradation estimation, high-quality high-resolution low-spectral images can be generated by efficiently fusing low-resolution hyperspectral images and high-resolution low-spectral images without precise prior information. This optimizes computational efficiency, accurately captures the nonlinear features of the images, and achieves a stable, efficient, and optimized solution process, meeting various application requirements.

[0082] Furthermore, the nonlinear initialization module 2 is also used for: Obtain the nearest neighbor node of each spectral vector in the spectral space of a low-resolution hyperspectral image; Based on the Euclidean distance between nearest neighbors and adjustable parameters, the joint probability of low-resolution hyperspectral images in high-dimensional space after symmetry transformation is obtained. Construct a loss function between the symmetric joint probabilities in the high-dimensional space and the corresponding symmetric joint probabilities in the low-dimensional space; The loss function is optimized iteratively to obtain the spectral response matrix.

[0083] In the above implementation process, nonlinear initialization of low-resolution hyperspectral images can improve the spectral fidelity of reconstructed images, accurately capture spectral nonlinear mixing characteristics, correct deviations in low-resolution hyperspectral images, improve initialization stability, and further enhance the utilization rate of low-resolution hyperspectral images, thereby increasing practicality and compatibility and providing fundamental support for spectral fidelity and accuracy improvement.

[0084] Furthermore, the spectral degradation estimation module 3 is also used for: Obtain the mapping relationship between low-resolution hyperspectral images and high-resolution low-spectral images; Based on the mapping relationship, the high-resolution low-spectral image is initialized to obtain the initialized high-resolution low-spectral image; The initial high-resolution low-spectral image is subjected to regularization constraints to obtain the spectral degradation matrix; The initialized high-resolution low-spectral image and spectral degradation matrix are optimized iteratively to obtain the optimized high-resolution low-spectral image and the optimized spectral degradation matrix. The optimized high-resolution low-spectral image and the optimized spectral degradation matrix are used as the data after spectral degradation estimation.

[0085] In the above implementation process, the spectral degradation estimation operation can improve the accuracy of spectral degradation estimation, accurately capture the degradation law of the spectrum, broaden the applicable scenarios and applicability, simplify the operation process, improve the estimation efficiency, and enhance the stability and anti-interference ability of degradation estimation.

[0086] Furthermore, the spectral degradation estimation module 3 is also used for: Obtain the physical imaging model for hyperspectral imaging; Based on the physical imaging model of hyperspectral imaging, an observation model is constructed between low-resolution hyperspectral images, high-resolution low-spectral images, and high-resolution hyperspectral images of the target. Based on the observation model, the mapping relationship between low-resolution hyperspectral images and high-resolution low-spectral images is obtained.

[0087] In the above implementation process, by mapping the low-resolution hyperspectral image to the high-resolution low-spectral image, the spatial-spectral dual precision of the low-resolution hyperspectral image can be improved, and the corresponding rules between spatial detail features can be accurately captured. This effectively mines spectral coupling information, optimizes the spectral information transfer and fusion effect, and improves the stability and reliability of subsequent hyperspectral image processing.

[0088] Furthermore, the block module 4 is also used for: A low-resolution hyperspectral image is upsampled to the same spatial size as a high-resolution hyperspectral image to obtain an upsampled low-resolution hyperspectral image. Obtain the spectral response matrix and spatial response matrix used to characterize the high-resolution hyperspectral image of the target; The upsampled low-resolution hyperspectral image, the optimized high-resolution low-spectral image, the spectral response matrix, and the spatial response matrix are uniformly divided into multiple matrix blocks in the spatial dimension.

[0089] In the above implementation process, the block-based approach can effectively solve the technical pain points of insufficient spatial resolution in low-resolution hyperspectral images, such as loss of spatial details and blurred spatial structure. While improving the spatial dimension quality of the image, it maximizes the preservation of the original spectral information, providing reliable support for the subsequent processing and practical application of hyperspectral images, and significantly improving the practicality of the technology.

[0090] Furthermore, the block module 4 is also used for: The high-resolution hyperspectral image of the target is decomposed and characterized into a spectral response matrix and a spatial response matrix.

[0091] In the above implementation process, the processing strategy for high-dimensional data can be optimized and simplified by adopting the above method. It does not rely on precise prior information, improves the applicability and robustness of the algorithm during operation, and provides data support for the accuracy of the target high-resolution hyperspectral image.

[0092] Furthermore, the device also includes an optimization module for: The optimization is performed on the matrix blocks of spatial response matrices and spectral response matrices within multiple matrix blocks, specifically as follows: Based on the alternating direction multiplier method, the matrix blocks of the spatial response matrix and the matrix blocks of the spectral response matrix are optimized according to the matrix blocks of low-resolution hyperspectral images and high-resolution low-spectral images in multiple matrix blocks, so as to obtain the optimized matrix blocks of the spatial response matrix and the matrix blocks of the spectral response matrix.

[0093] In the above implementation process, by optimizing the matrix blocks of the spatial response matrix and the matrix blocks of the spectral response matrix, the defects of existing matrix blocks such as bias, poor adaptability and weak anti-interference ability can be effectively solved, the representation accuracy of the spatial response matrix and the spectral response matrix on the spatial and spectral features of hyperspectral images can be improved, and the performance of hyperspectral technology can be further enhanced.

[0094] Furthermore, refactoring module 5 is also used for: The optimized spatial response matrix and spectral response matrix blocks are reorganized according to their original spatial positions to obtain the global spatial response matrix and global spectral response matrix. By performing matrix multiplication, the spatial response matrix is ​​replaced with the global spatial response matrix, and the spectral response matrix is ​​replaced with the global spectral response matrix, thus obtaining a high-resolution hyperspectral image of the target.

[0095] In the above implementation process, by reorganizing the optimized matrix blocks, the problems of scattered and independent optimized matrix blocks, insufficient coordination and poor correlation are effectively solved, realizing the integration and efficient coordination of optimized matrix blocks, providing complete and coordinated matrix support for hyperspectral image processing, and improving the practicality of the technology.

[0096] Example 3 This application provides an electronic device, including a memory and a processor. The memory stores a computer program, and the processor runs the computer program to enable the electronic device to perform the super-resolution imaging method of Embodiment 1.

[0097] Alternatively, the aforementioned electronic device may be a server.

[0098] Please see Figure 3 , Figure 3 This is a schematic diagram illustrating the structural composition of an electronic device provided in an embodiment of this application. The electronic device may include a processor 31, a communication interface 32, a memory 33, and at least one communication bus 34. The communication bus 34 is used to enable direct communication between these components.

[0099] Optionally, the electronic device may also include a storage controller and an input / output unit. The memory 33, storage controller, processor 31, peripheral interface, and input / output unit are electrically connected to each other directly or indirectly to realize data transmission or interaction.

[0100] Input / output units are used to enable users to create tasks and set optional start periods or preset execution times for those tasks, facilitating user-server interaction. Input / output units can be, but are not limited to, a mouse and keyboard.

[0101] Understandable. Figure 3 The structure shown is for illustrative purposes only; the electronic device may also include components that are more advanced than those shown. Figure 3 The more or fewer components shown, or having the same Figure 3 Different configurations are shown. Additionally, embodiments of this application also provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the super-resolution imaging method of Embodiment 1.

[0102] This application also provides a computer program product that, when run on a computer, causes the computer to perform the method described in the method embodiment.

[0103] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application. It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0104] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of protection of the claims.

Claims

1. A super-resolution imaging method, characterized in that, The method includes: Acquire low-resolution hyperspectral images; The low-resolution hyperspectral image is nonlinearly initialized to obtain the spectral response matrix; The low-resolution hyperspectral image is subjected to spectral degradation estimation to obtain the data after spectral degradation estimation; The spectral response matrix and the data after spectral degradation estimation are respectively divided into blocks in the spatial dimension to obtain multiple corresponding matrix blocks; The target high-resolution hyperspectral image is obtained by reconstructing multiple matrix blocks.

2. The super-resolution imaging method according to claim 1, characterized in that, The step of performing nonlinear initialization on the low-resolution hyperspectral image to obtain the spectral response matrix includes: Obtain the nearest neighbor node of each spectral vector in the spectral space of the low-resolution hyperspectral image; Based on the Euclidean distance of the nearest neighbor nodes and the adjustable parameters, the joint probability of the low-resolution hyperspectral image in the high-dimensional space after symmetry is obtained; Construct a loss function between the symmetric joint probabilities in the high-dimensional space and the corresponding symmetric joint probabilities in the low-dimensional space; The loss function is optimized and iterated to obtain the spectral response matrix.

3. The super-resolution imaging method according to claim 1, characterized in that, The step of performing spectral degradation estimation on the low-resolution hyperspectral image to obtain the spectral degradation-estimated data includes: Obtain the mapping relationship between the low-resolution hyperspectral image and the high-resolution low-spectral image; Based on the mapping relationship, the high-resolution low-spectral image is initialized to obtain the initialized high-resolution low-spectral image; Applying regularization constraints to the initialized high-resolution low-spectral image yields a spectral degradation matrix. The initialized high-resolution low-spectral image and the spectral degradation matrix are optimized and iterated to obtain an optimized high-resolution low-spectral image and an optimized spectral degradation matrix. The optimized high-resolution low-spectral image and the optimized spectral degradation matrix are used as the data after spectral degradation estimation.

4. The super-resolution imaging method according to claim 3, characterized in that, The step of obtaining the mapping relationship between the low-resolution hyperspectral image and the high-resolution low-spectral image includes: Obtain the physical imaging model for hyperspectral imaging; Based on the physical imaging model of the hyperspectral imaging, an observation model is constructed between the low-resolution hyperspectral image, the high-resolution low-spectral image, and the target high-resolution hyperspectral image. Based on the observation model, the mapping relationship between the low-resolution hyperspectral image and the high-resolution low-spectral image is obtained.

5. The super-resolution imaging method according to claim 1, characterized in that, The step of dividing the spectral response matrix and the spectral degradation estimated data into spatial blocks to obtain multiple corresponding matrix blocks includes: The low-resolution hyperspectral image is upsampled to the same spatial size as the high-resolution low-spectral image to obtain the upsampled low-resolution hyperspectral image. Obtain the spectral response matrix and spatial response matrix used to characterize the high-resolution hyperspectral image of the target; The upsampled low-resolution hyperspectral image, the optimized high-resolution low-spectral image, the spectral response matrix, and the spatial response matrix are uniformly divided into blocks in the spatial dimension to obtain multiple matrix blocks.

6. The super-resolution imaging method according to claim 5, characterized in that, The step of obtaining the spectral response matrix and spatial response matrix used to characterize the high-resolution hyperspectral image of the target includes: The target high-resolution hyperspectral image is decomposed and characterized into the spectral response matrix and the spatial response matrix.

7. The super-resolution imaging method according to claim 1, characterized in that, Before the step of reconstructing the plurality of matrix blocks to obtain the target high-resolution hyperspectral image, the method further includes: The optimization of the matrix blocks of spatial response matrices and spectral response matrices in the multiple matrix blocks is specifically as follows: Based on the alternating direction multiplier method, the matrix blocks of the spatial response matrix and the matrix blocks of the spectral response matrix are optimized according to the matrix blocks of low-resolution hyperspectral images and high-resolution low-spectral images in the multiple matrix blocks, to obtain optimized matrix blocks of the spatial response matrix and the spectral response matrix.

8. The super-resolution imaging method according to claim 6, characterized in that, The step of reconstructing multiple matrix blocks to obtain a target high-resolution hyperspectral image includes: The optimized spatial response matrix and spectral response matrix blocks are reorganized according to their original spatial positions to obtain the global spatial response matrix and global spectral response matrix. By performing matrix multiplication, the spatial response matrix is ​​replaced with the global spatial response matrix, and the spectral response matrix is ​​replaced with the global spectral response matrix, to obtain a high-resolution hyperspectral image of the target.

9. A super-resolution imaging device, characterized in that, The device includes: The acquisition module is used to acquire low-resolution hyperspectral images; A nonlinear initialization module is used to perform nonlinear initialization on the low-resolution hyperspectral image to obtain a spectral response matrix; The spectral degradation estimation module is used to perform spectral degradation estimation on the low-resolution hyperspectral image to obtain spectral degradation-estimated data. The block segmentation module is used to divide the spectral response matrix and the data after spectral degradation estimation into blocks in the spatial dimension to obtain multiple corresponding matrix blocks; The reconstruction module is used to reconstruct multiple matrix blocks to obtain a high-resolution hyperspectral image of the target.

10. An electronic device, characterized in that, include: A memory, a processor, and a computer program stored in the memory and executable on the processor, causing the computer to perform the method as described in claims 1-8.

Images