Satellite hyperspectral and multispectral spectral response matching optimization method, system and medium
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- ZHUHAI ORBIT SATELLITE BIG DATA CO LTD
- Filing Date
- 2026-07-07
- Publication Date
- 2026-08-07
AI Technical Summary
[0006]本发明所要解决的技术问题是克服现有技术的不足，提供了一种面相异源传感器的卫星高光谱与多光谱光谱响应匹配优化方法、系统及介质，以解决现有技术中跨传感器波段对应不准确、非重叠谱段难以处理、边缘响应缺失、光谱匹配矩阵不连续以及融合训练不稳定等问题，从而提高真实卫星高光谱—多光谱融合场景中的光谱一致性、建模鲁棒性与工程实用性
[0016] The beneficial effects of the present invention are: 1. The present invention constructs a hybrid response matrix by combining the spectral response function of a real hyperspectral sensor with the spectral response function of a multispectral sensor, which can more accurately characterize the physical response relationship between heterogeneous sensors and avoid the mapping error caused by simply relying on empirical band pairing or fixed linear projection.
Smart Images
Figure CN122529977A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the fields of remote sensing image processing, satellite heterogeneous sensor modeling, spectral response analysis and fusion reconstruction, and particularly to a satellite hyperspectral and multispectral spectral response matching optimization method, system and medium for heterogeneous sensors. It can be applied to tasks such as real satellite hyperspectral-multispectral fusion super-resolution reconstruction, cross-sensor spectral mapping, spectral consistency modeling and preprocessing before fusion. Background Technology
[0002] Hyperspectral remote sensing imagery is characterized by continuous spectral bands, narrow spectral bands, and high spectral resolution, enabling precise characterization of ground cover spectral features. It holds significant application value in fields such as mineral identification, vegetation monitoring, environmental analysis, fine classification, and target detection. Multispectral remote sensing imagery typically possesses high spatial resolution, providing rich texture and structural details. To simultaneously obtain high spatial resolution and hyperspectral resolution information, it is usually necessary to fuse and reconstruct low-spatial-resolution hyperspectral imagery with high-spatial-resolution multispectral imagery.
[0003] However, in real-world satellite scenarios, hyperspectral and multispectral sensors typically originate from different satellite platforms, payload systems, or imaging modes, resulting in significant differences in their center wavelength, bandwidth, band coverage, and response curve shape. For example... Figure 1 As shown, the spectral response functions of the Zhuhai-1 hyperspectral and multispectral sensors exhibit significant differences in band coverage, response width, and response curve shape. Existing technologies commonly employ empirical band pairing, simple interpolation, or fixed linear projection methods to address the spectral mapping relationship between hyperspectral and multispectral sensors, which fail to accurately reflect the differences in imaging response between different sensors.
[0004] Especially in real heterogeneous satellite data, the following problems often occur: First, multispectral widebands usually correspond to multiple hyperspectral narrowbands, forming a one-to-many hybrid relationship; Second, there is only weak overlap or even no overlap between hyperspectral edge bands and multispectral bands; Third, due to the influence of discrete sampling, calibration errors, and on-orbit drift, the theoretically existing weak response is easily truncated to zero in the discrete response matrix; Fourth, in real satellite fusion super-resolution reconstruction, if spectral mapping is established only based on strictly overlapping areas, some bands will lack effective constraints in subsequent models, leading to problems such as edge band distortion, local spectral anomalies, and collapse of some output bands.
[0005] Therefore, it is urgent to propose a spectral response matching and optimization method that can be adapted to real satellite heterogeneous sensor conditions, comprehensively utilize spectral response functions, extract the main action region, compensate for non-overlapping spectral band expansion, and optimize continuity, so as to provide stable, interpretable, and physically meaningful spectral priors for subsequent fusion reconstruction. Summary of the Invention
[0006] The technical problem to be solved by this invention is to overcome the shortcomings of the prior art and provide a method, system and medium for matching and optimizing the hyperspectral and multispectral spectral responses of satellites with heterogeneous phase sensors. This invention addresses the problems in the prior art, such as inaccurate cross-sensor band correspondence, difficulty in handling non-overlapping spectral bands, missing edge responses, discontinuous spectral matching matrices and unstable fusion training. This improves the spectral consistency, modeling robustness and engineering practicality in real satellite hyperspectral-multispectral fusion scenarios.
[0007] The technical solution adopted in this invention is: a satellite hyperspectral and multispectral spectral response matching optimization method, characterized by the following steps: S100: Obtain the spectral response function of the hyperspectral sensor and the spectral response function of the multispectral sensor, and unify them to the same wavelength sampling grid; S200, based on the hyperspectral spectral response function and multispectral spectral response function after unified wavelength sampling, the initial hybrid response matrix from hyperspectral to multispectral is constructed by discrete integration; S300. Based on the non-zero response positions or response positions greater than the preset threshold in the initial mixed response matrix, extract the hyperspectral main action region corresponding to each multispectral band. S400. For hyperspectral bands outside the main action range but in the vicinity, perform spectral band expansion compensation and attenuation weighting, establish the compensation mapping relationship between weakly overlapping spectral bands and non-overlapping neighboring spectral bands and corresponding multispectral bands, and obtain the response matrix after expansion compensation. S500, perform continuity optimization, non-negativity constraint and normalization on the response matrix after expansion compensation to obtain the optimized spectral matching matrix; S600, use the optimized spectral matching matrix as a spectral prior or initialization mapping matrix or linear anchor branch or spectral consistency constraint in the hyperspectral and multispectral fusion reconstruction.
[0008] Furthermore, in step S100, if the original wavelength sampling intervals of the hyperspectral sensor and the multispectral sensor are different, a unified wavelength sampling grid is first established by interpolation, resampling, or wavelength alignment, so that the spectral response function of the hyperspectral sensor and the spectral response function of the multispectral sensor are located in the same wavelength coordinate system.
[0009] Furthermore, in step S200, if the unified wavelength sampling grid is non-equally spaced, the initial mixed response matrix is obtained by first calculating the multispectral band spectral response function and the hyperspectral band spectral response function on the unified wavelength sampling grid through a weighted overlapping integral with wavelength sampling interval weights, and then through discrete integral.
[0010] Furthermore, in step S300, the hyperspectral main active region is determined by the region consisting of the minimum and maximum indices in the hyperspectral band index that satisfy the condition that the response coefficient is greater than zero or greater than a preset threshold.
[0011] Furthermore, in step S400, the spectral band expansion compensation adopts fixed-width expansion, adaptive-width expansion, or asymmetric-width expansion; the adaptive-width expansion determines the expansion width based on at least one of the following in the multispectral band spectral response function: half-peak width, edge slope, local energy attenuation degree, and corresponding hyperspectral non-zero response distribution; in step S400, the attenuation weighting is constructed using a distance attenuation function, which includes at least one of exponential attenuation function and linear attenuation function.
[0012] Furthermore, in step S500, the continuity optimization is achieved using first-order differential smoothing constraints, second-order differential smoothing constraints, convolutional smoothing filtering, or iterative optimization methods along the spectral band direction; in step S500, the non-negativity constraint is achieved by truncating the negative values in the response matrix after expansion compensation, and the normalization process is achieved by normalizing the hyperspectral response weights corresponding to each multispectral band; the optimized spectral matching matrix is used to characterize the spectral degradation relationship from hyperspectral to multispectral; when used for spectral upsampling or initial mapping from multispectral to hyperspectral, it is implemented using the transpose or pseudo-inverse matrix of the optimized spectral matching matrix or a learnable linear layer constrained by the optimized spectral matching matrix.
[0013] Furthermore, in step S600, the fusion reconstruction includes a linear anchor branch and a residual refinement branch. The linear anchor branch uses an optimized spectral matching matrix to generate an initial hyperspectral estimate, and the residual refinement branch is used to compensate for spectral details in the initial hyperspectral estimate.
[0014] Furthermore, a satellite hyperspectral and multispectral spectral response matching optimization system includes: The SRF acquisition module is used to acquire the spectral response function of the hyperspectral sensor and the spectral response function of the multispectral sensor. A wavelength unification module is used to unify the spectral response function of the hyperspectral sensor and the spectral response function of the multispectral sensor to the same wavelength sampling grid. The hybrid response matrix construction module is used to construct an initial hybrid response matrix from hyperspectral to multispectral based on the spectral response function after sampling at a unified wavelength. The main active region extraction module is used to extract the hyperspectral main active region corresponding to each multispectral band based on the initial mixing response matrix. The expansion compensation module is used to perform expansion compensation and attenuation weighting on weakly overlapping spectral bands and non-overlapping neighboring spectral bands outside the main action range; The continuous optimization module is used to perform continuous optimization, non-negativity constraint and normalization on the response matrix after expansion compensation to obtain the optimized spectral matching matrix; The fusion prior output module is used to use the optimized spectral matching matrix as a spectral prior, initialization mapping matrix, linear anchor branch, or spectral consistency constraint in the fusion reconstruction.
[0015] Furthermore, a computer-readable storage medium storing a computer program that, when executed by a processor, implements a method for matching and optimizing the hyperspectral and multispectral spectral responses of a phase-dependent sensor.
[0016] The beneficial effects of the present invention are: 1. The present invention constructs a hybrid response matrix by combining the spectral response function of a real hyperspectral sensor with the spectral response function of a multispectral sensor, which can more accurately characterize the physical response relationship between heterogeneous sensors and avoid the mapping error caused by simply relying on empirical band pairing or fixed linear projection.
[0017] 2. This invention improves the interpretability of cross-sensor band matching by extracting the hyperspectral main active region corresponding to the multispectral band and transforming the continuous spectral response relationship into a structured "multispectral band - hyperspectral region" correspondence.
[0018] 3. This invention performs outward expansion compensation and attenuation weighting on weakly overlapping spectral bands and non-overlapping neighboring spectral bands outside the main operating range, so that the edge spectral bands obtain weak but non-zero response constraints, which can alleviate the problem of missing edge band response caused by strict overlap modeling.
[0019] 4. This invention reduces abrupt changes in the spectral matching matrix along the spectral band direction through continuous optimization, making the final matching matrix more consistent with the law of continuous change of the spectral response of the real sensor with wavelength.
[0020] 5. The optimized spectral matching matrix obtained in this invention can be used as a spectral prior or initialization mapping matrix or linear anchor branch or spectral consistency constraint in the fusion reconstruction model, which helps to improve the stability of real satellite hyperspectral-multispectral fusion reconstruction and suppress partial output band collapse and all-zero output. Attached Figure Description
[0021] Figure 1 A schematic diagram of the hyperspectral and multispectral spectral response functions of Zhuhai-1. Figure 2 This is a flowchart of the overall process of the present invention; Figure 3 This is a schematic diagram of SRF uniform wavelength sampling; Figure 4 A schematic diagram of the main action region extraction; Figure 5A schematic diagram of constructing a hybrid response matrix; Figure 6 This is a schematic diagram illustrating the expansion compensation and continuity optimization. Figure 7 A schematic diagram of the linear anchor point branch structure for fusion and reconstruction. Detailed Implementation
[0022] In this embodiment, the present invention includes a method, system, and medium for optimizing satellite hyperspectral and multispectral spectral response matching of a surface-phase heterogeneous sensor. The overall flow of the optimization method is as follows: Figure 2 As shown, the main steps include unified SRF modeling, hybrid response matrix construction, principal action region extraction, weakly overlapping and non-overlapping spectral band expansion compensation, continuity optimization, and fusion reconstruction of prior output. Specifically, it includes: S100: Obtain the spectral response function of the hyperspectral sensor and the spectral response function of the multispectral sensor, and unify them to the same wavelength sampling grid; Obtain the spectral response function matrix of the hyperspectral sensor Spectral response function matrix of multispectral sensors ; , in, This indicates the number of sampling points for a uniform wavelength. Indicates the number of hyperspectral bands. This indicates the number of multispectral bands, and R is an abbreviation for spectral response, representing the response values of each band of the sensor at different wavelengths. If the original wavelength sampling intervals of the two types of sensors are different, the spectral response function of the hyperspectral sensor and the spectral response function of the multispectral sensor are first transformed to the same wavelength sampling grid through interpolation, resampling, or wavelength alignment. That is, the spectral response function matrix of the hyperspectral sensor is transformed using interpolation, resampling, or wavelength alignment. Spectral response function matrix of multispectral sensors Both spectral response curves are mapped to the same wavelength grid. , The unified wavelength sampling process is as follows Figure 3 As shown. This process ensures that the spectral response functions of different sensors can be integrated and calculated in the same wavelength coordinate system.
[0023] S200, based on the hyperspectral spectral response function and multispectral spectral response function after unified wavelength sampling, the initial hybrid response matrix from hyperspectral to multispectral is constructed by discrete integration; When wavelength sampling is at equal intervals, we can set: , When the uniform wavelength sampling grid is not equally spaced, the initial mixed response matrix is obtained by first calculating the multispectral band spectral response function and the hyperspectral band spectral response function on the uniform wavelength sampling grid through a weighted overlap integral with wavelength sampling interval weights, and then through discrete integration, i.e., based on the wavelength sequence. Calculate its discrete gradient to obtain the wavelength weight vector: , For any multispectral band With any hyperspectral band Calculate its mixed response coefficients in discrete integral form: , Transform it into discrete subscript equivalence , in, Indicates the first The multispectral band and the first Mixed response coefficients between hyperspectral bands; This represents the spectral response function value of the m-th multispectral band at wavelength λ. This represents the spectral response function value of the h-th hyperspectral band at wavelength λ. Indicates the wavelength sampling interval or wavelength integration weight; This is the numerically stable term. It can be written in matrix form (initial mixed response) as follows: , in, This represents a diagonal matrix composed of wavelength integral weights. This represents the transpose of the multispectral response matrix. To ensure physical plausibility, a nonnegativity constraint is imposed on the initial mixed response matrix: , Then, the hyperspectral response weights corresponding to each multispectral band are normalized: , This yields an initial spectral matching matrix that is non-negative and whose hyperspectral weights for each spectral band sum to 1. The initial mixed response matrix construction process is as follows: Figure 5 As shown.
[0024] S300. Based on the non-zero response positions or response positions greater than the preset threshold in the initial mixed response matrix, extract the hyperspectral main action region corresponding to each multispectral band. Obtaining the initial matching matrix Then, for each multispectral band m, search for hyperspectral band indices that satisfy the following conditions: , in, Alternatively, a preset threshold can be used, or the response coefficient being greater than zero can be directly taken as the criterion. If If not empty, then the primary scope consists of the minimum and maximum indices that satisfy the conditions: , in, , The main active region reflects the primary coverage area of the m-th multispectral band over the hyperspectral band. Through this process, the continuous spectral response relationship can be transformed into a structured "multispectral band-hyperspectral region" correspondence, providing a foundation for subsequent expansion compensation of non-overlapping and weakly overlapping spectral bands. The main active region extraction process is as follows: Figure 4 As shown.
[0025] S400. For hyperspectral bands outside the main action range but in the vicinity, perform spectral band expansion compensation and attenuation weighting, establish the compensation mapping relationship between weakly overlapping spectral bands and non-overlapping neighboring spectral bands and corresponding multispectral bands, and obtain the response matrix after expansion compensation. For the principal action region For hyperspectral bands located outside but near their neighborhood boundaries, this invention does not directly regard them as irrelevant spectral bands, but establishes a compensation response relationship for them through interval spectral band expansion compensation and attenuation weighting.
[0026] Construct the outer extension range: , in, and These represent the left and right expansion widths, respectively. The expansion range is limited to... Within the designated area to avoid crossing the boundary.
[0027] The expansion width of the spectral band extension compensation can be achieved using one of the following methods: (1) Fixed width expansion, i.e., preset fixed number of bands ,make ; (2) Adaptive width expansion, that is, the expansion width is automatically determined based on at least one of the following in the multispectral band response function: half-peak width, edge slope, local energy attenuation degree, or corresponding hyperspectral non-zero response distribution; (3) Asymmetric width expansion, i.e., for cases where the center wavelength shift is significant or the edges are asymmetrical, adopts Expand outwards in this manner.
[0028] Hyperspectral bands within the extended range but not within the primary effective range are assigned a distance-dependent attenuation compensation weight. For hyperspectral bands... Its closest distance to the boundary of the principal action interval is defined as: , An exponential decay function can be used: , in, This is the attenuation parameter. A linear attenuation function can also be used: , in, This represents the maximum outward expansion width. The compensated outward expansion response coefficient can be expressed as: , in, Either exponentially decaying weights or linearly decaying weights can be used. The mean of the response within the main action interval, the boundary mean, or the local interpolation amount. This is the scaling factor for the compensation region. Typically, the response intensity in the compensation region is made smaller than that in the main action region to ensure that the compensation mapping does not alter the original main response relationship. Through this method, weakly overlapping spectral bands and non-overlapping neighboring spectral bands can obtain weak but non-zero response constraints, thus preventing them from completely losing their mapping basis in subsequent fusion and reconstruction.
[0029] S500, perform continuity optimization, non-negativity constraint and normalization on the response matrix after expansion compensation to obtain the optimized spectral matching matrix; To avoid abrupt changes in the spectral direction after expansion, this invention further optimizes the continuity of the expansion matrix using first-order difference smoothing constraints, second-order difference smoothing constraints, convolutional smoothing filters, or iterative optimization methods in the spectral direction. In this case, a smooth continuity constraint (first-order difference smoothing constraint / second-order difference smoothing constraint) is applied. The continuity is constructed as follows: , In practical implementation, the continuously optimized matrix can be obtained by minimizing the following objective function: , in, For continuity constraint weights, Let Frobenius norm be represented. After continuity optimization, a nonnegativity constraint is reapplied to the matrix, which is achieved by truncating the negative values in the outwardly compensated response matrix: , Then, row normalization is performed, which is achieved by normalizing the hyperspectral response weights corresponding to each multispectral band: , This yields the final optimized spectral matching matrix that satisfies non-negativity constraints, row normalization constraints, and exhibits stronger continuity in spectral directions. The optimized spectral matching matrix is used to characterize the spectral degradation relationship from hyperspectral to multispectral. When used for spectral upsampling or initialization mapping from multispectral to hyperspectral, it is implemented using the transpose or pseudo-inverse matrix of the optimized spectral matching matrix, or a learnable linear layer constrained by the optimized spectral matching matrix. The expansion compensation and continuity optimization process is as follows: Figure 6 As shown.
[0030] S600, Use the optimized spectral matching matrix as a spectral prior or initialization mapping matrix or linear anchor branch or spectral consistency constraint in the hyperspectral and multispectral fusion reconstruction. The optimized spectral matching matrix obtained in this invention It can be used not only for preprocessing before fusion but also as a spectral prior in the fusion reconstruction model, and can be used for: 1. As a spectral degradation matrix from hyperspectral to multispectral; 2. Spectral consistency constraint matrix; 3. Priors for multispectral to hyperspectral initialization mapping; 4. Prior weights for linear anchor branch; 5. Incorporate spectral regularization terms in super-resolution reconstruction.
[0031] The linear anchor branch generates an initial hyperspectral estimate using an optimized spectral matching matrix, and the residual refinement branch is used to compensate for spectral details in the initial hyperspectral estimate. In this implementation, the optimized spectral matching matrix is embedded in the multispectral-to-hyperspectral upsampling module, constructing a fusion reconstruction structure combining the linear anchor branch and the residual refinement branch, as shown below. Figure 7 As shown, the linear anchor branch first provides a stable spectral basis mapping, and the residual refinement branch further compensates for hyperspectral details, thereby improving the training stability of the model under real satellite single-scene, weak supervision or no high-resolution hyperspectral ground truth conditions, and suppressing partial output band collapse and all-zero output.
[0032] Example: Response reconstruction of real satellite 3-band multispectral and 32-band hyperspectral imaging; The following uses Zhuhai-1 hyperspectral and multispectral satellite data as an example to illustrate the implementation of the present invention.
[0033] In this embodiment, the hyperspectral data covers approximately 400 nm to 1000 nm, comprising 32 bands; the multispectral data consists of 3 bands, primarily covering the range of approximately 400 nm to 700 nm. Since a wide multispectral band typically covers multiple narrow hyperspectral bands, and some hyperspectral edge bands only weakly overlap with or even do not overlap with the multispectral response, such as... Figure 1 and Figure 6 As shown, if strict overlap modeling is used directly, the coefficients of several hyperspectral edge bands in the matching matrix will be zero.
[0034] When using the method of this invention, a unified wavelength sampling grid is first constructed based on the SRF of hyperspectral and multispectral sensors; then, an initial hybrid response matrix is constructed based on discrete integrals; subsequently, the corresponding hyperspectral main action region is extracted according to the non-zero response position or the response position greater than the threshold of each multispectral band; then, fixed-width or adaptive-width expansion is performed on adjacent edge spectral bands, and a compensation response is constructed through a distance attenuation function; finally, the final optimized spectral matching matrix is obtained by using continuity optimization, non-negativity constraints, and normalization processing. .
[0035] The resulting optimized spectral matching matrix It can be directly used for hyperspectral-multispectral fusion modeling of real satellites, which can improve the problems of insufficient response of edge spectral bands and insufficient constraints of weakly overlapping spectral bands, and provide stable spectral priors for subsequent fusion reconstruction.
[0036] System Implementation Examples
[0037] This invention also provides a satellite hyperspectral and multispectral spectral response matching optimization system for heterogeneous sensors, comprising: The SRF acquisition module is used to acquire the spectral response function of the hyperspectral sensor and the spectral response function of the multispectral sensor. A wavelength unification module is used to unify the spectral response function of the hyperspectral sensor and the spectral response function of the multispectral sensor to the same wavelength sampling grid. The hybrid response matrix construction module is used to construct an initial hybrid response matrix from hyperspectral to multispectral based on the spectral response function after sampling at a unified wavelength. The main active region extraction module is used to extract the hyperspectral main active region corresponding to each multispectral band based on the initial mixing response matrix. The expansion compensation module is used to perform expansion compensation and attenuation weighting on weakly overlapping spectral bands and non-overlapping neighboring spectral bands outside the main action range; The continuous optimization module is used to perform continuous optimization, non-negativity constraint and normalization on the response matrix after expansion compensation to obtain the optimized spectral matching matrix; The fusion prior output module is used to use the optimized spectral matching matrix as a spectral prior, initialization mapping matrix, linear anchor branch, or spectral consistency constraint in the fusion reconstruction.
[0038] Storage Media Examples The present invention also provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, it implements the satellite hyperspectral and multispectral spectral response matching optimization method for heterogeneous sensors described in this invention.
[0039] The computer-readable storage medium may be a read-only memory, random access memory, magnetic disk, optical disk, flash memory, or other media capable of storing program code.
[0040] Although the embodiments of the present invention are described with reference to actual solutions, they do not constitute a limitation on the meaning of the present invention. Modifications to the embodiments and combinations with other solutions based on this specification will be obvious to those skilled in the art.
Claims
1. A method for matching and optimizing the hyperspectral and multispectral responses of satellites, characterized in that: It includes the following steps: S100: Obtain the spectral response function of the hyperspectral sensor and the spectral response function of the multispectral sensor, and unify them to the same wavelength sampling grid; S200, based on the hyperspectral spectral response function and multispectral spectral response function after unified wavelength sampling, the initial hybrid response matrix from hyperspectral to multispectral is constructed by discrete integration; S300. Based on the non-zero response positions or response positions greater than the preset threshold in the initial mixed response matrix, extract the hyperspectral main action region corresponding to each multispectral band. S400. For hyperspectral bands outside the main action range but in the vicinity, perform spectral band expansion compensation and attenuation weighting, establish the compensation mapping relationship between weakly overlapping spectral bands and non-overlapping neighboring spectral bands and corresponding multispectral bands, and obtain the response matrix after expansion compensation. S500, perform continuity optimization, non-negativity constraint and normalization on the response matrix after expansion compensation to obtain the optimized spectral matching matrix; S600, use the optimized spectral matching matrix as a spectral prior or initialization mapping matrix or linear anchor branch or spectral consistency constraint in the hyperspectral and multispectral fusion reconstruction.
2. The satellite hyperspectral and multispectral spectral response matching optimization method according to claim 1, characterized in that: In step S100, if the original wavelength sampling intervals of the hyperspectral sensor and the multispectral sensor are different, a unified wavelength sampling grid is first established by interpolation, resampling, or wavelength alignment so that the spectral response function of the hyperspectral sensor and the spectral response function of the multispectral sensor are located in the same wavelength coordinate system.
3. The satellite hyperspectral and multispectral spectral response matching optimization method according to claim 1, characterized in that: In step S200, if the uniform wavelength sampling grid is non-equally spaced, the initial mixed response matrix is obtained by first calculating the multispectral band spectral response function and the hyperspectral band spectral response function on the uniform wavelength sampling grid through a weighted overlapping integral with wavelength sampling interval weights, and then through discrete integral.
4. The satellite hyperspectral and multispectral spectral response matching optimization method according to claim 1, characterized in that: In step S300, the hyperspectral main active region is determined by the region consisting of the minimum and maximum indices in the hyperspectral band index that satisfy the response coefficient being greater than zero or greater than a preset threshold.
5. The satellite hyperspectral and multispectral spectral response matching optimization method according to claim 1, characterized in that: In step S400, the spectral band expansion compensation adopts fixed-width expansion, adaptive-width expansion, or asymmetric-width expansion; the adaptive-width expansion determines the expansion width based on at least one of the following in the multispectral band spectral response function: half-peak width, edge slope, local energy attenuation degree, and corresponding hyperspectral non-zero response distribution; in step S400, the attenuation weighting is constructed using a distance attenuation function, which includes at least one of exponential attenuation function and linear attenuation function.
6. The satellite hyperspectral and multispectral spectral response matching optimization method according to claim 1, characterized in that: In step S500, continuity optimization is achieved using first-order differential smoothing constraints, second-order differential smoothing constraints, convolutional smoothing filtering, or iterative optimization methods along the spectral band direction. In step S500, the non-negativity constraint is achieved by truncating the negative values in the response matrix after expansion compensation, and the normalization process is achieved by normalizing the hyperspectral response weights corresponding to each multispectral band. The optimized spectral matching matrix is used to characterize the spectral degradation relationship from hyperspectral to multispectral. When used for spectral upsampling or initialization mapping from multispectral to hyperspectral, the transpose or pseudo-inverse matrix of the optimized spectral matching matrix or a learnable linear layer constrained by the optimized spectral matching matrix is used.
7. The satellite hyperspectral and multispectral spectral response matching optimization method according to claim 1, characterized in that: In step S600, the fusion reconstruction includes a linear anchor branch and a residual refinement branch. The linear anchor branch uses an optimized spectral matching matrix to generate an initial hyperspectral estimate, and the residual refinement branch is used to compensate for spectral details in the initial hyperspectral estimate.
8. A satellite hyperspectral and multispectral spectral response matching optimization system, characterized in that, The system includes: The SRF acquisition module is used to acquire the spectral response function of the hyperspectral sensor and the spectral response function of the multispectral sensor. A wavelength unification module is used to unify the spectral response function of the hyperspectral sensor and the spectral response function of the multispectral sensor to the same wavelength sampling grid. The hybrid response matrix construction module is used to construct an initial hybrid response matrix from hyperspectral to multispectral based on the spectral response function after sampling at a unified wavelength. The main active region extraction module is used to extract the hyperspectral main active region corresponding to each multispectral band based on the initial mixing response matrix. The expansion compensation module is used to perform expansion compensation and attenuation weighting on weakly overlapping spectral bands and non-overlapping neighboring spectral bands outside the main action range; The continuous optimization module is used to perform continuous optimization, non-negativity constraint and normalization on the response matrix after expansion compensation to obtain the optimized spectral matching matrix; The fusion prior output module is used to use the optimized spectral matching matrix as a spectral prior, initialization mapping matrix, linear anchor branch, or spectral consistency constraint in the fusion reconstruction.
9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method described in any one of claims 1 to 7.