A spatial context-enhanced multi-temporal super-resolution reconstruction method
The multi-temporal super-resolution reconstruction method enhanced by spatial context solves the problems of insufficient differentiation of land cover categories and inadequate modeling of regional differences in multi-temporal remote sensing images, achieving high-precision image reconstruction and scene understanding, and improving the detail restoration and structural consistency of remote sensing images.
Patent Information
- Application Number
- CN202610771090.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-06-01
- Publication Date
- 2026-06-30
Smart Images
Figure CN122312385A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of remote sensing image processing and intelligent interpretation technology, specifically relating to a method for super-resolution reconstruction of multi-temporal remote sensing images with spatial context enhancement. Background Technology
[0002] With the continuous development of remote sensing technology, high-resolution remote sensing images are playing an increasingly important role in resource surveys, environmental monitoring, disaster assessment, land change analysis, and urban planning. However, limited by factors such as satellite platform payload capacity, imaging distance, sensor size, and transmission bandwidth, remote sensing observation systems often struggle to simultaneously achieve high spatial and temporal resolution, resulting in acquired remote sensing images that are often characterized by insufficient spatial resolution and relatively dense temporal sampling. Therefore, reconstructing high-resolution images from multi-temporal low-resolution remote sensing image sequences has become an important direction in remote sensing image super-resolution research.
[0003] While existing multi-temporal super-resolution methods can utilize complementary information between images from different time phases to some extent, most focus on low-level texture and temporal correlation modeling, neglecting the utilization of land cover categories, scene structure, and regional semantic information. Remote sensing scenes typically contain various types of land cover, such as buildings, roads, vegetation, water bodies, and bare land, with significant differences in texture frequency, edge structure, and spatial distribution among different categories. Using a uniform feature enhancement and fusion approach can easily lead to representational confusion between different categories, resulting in problems such as blurred building edges, decreased road connectivity, distorted water body boundaries, and insufficient vegetation texture restoration.
[0004] Furthermore, some existing methods only use pixel reconstruction error as the optimization objective, lacking constraints on semantic structure consistency. This results in improved reconstruction results in numerical metrics, but they still struggle to fully realize their potential in subsequent tasks such as land cover classification, target recognition, and scene understanding. Therefore, there is an urgent need to propose a multi-temporal remote sensing image super-resolution reconstruction method that can explicitly incorporate semantic prior information, perform differentiated reconstruction based on different land cover categories, and improve spatial detail and semantic structure consistency through multi-task collaborative optimization. Summary of the Invention
[0005] The purpose of this invention is to provide a spatial context-enhanced multi-temporal remote sensing image super-resolution reconstruction method to effectively solve the problems of insufficient differentiation of land cover categories, insufficient modeling of regional differences, and weak structural details and semantic consistency of reconstruction results in existing multi-temporal remote sensing image super-resolution technologies.
[0006] To achieve the above objectives, the present invention provides a spatial context-enhanced multi-temporal super-resolution reconstruction method, comprising the following steps:
[0007] S1. Acquire multi-temporal low-resolution remote sensing image sequences, and perform radiometric calibration, atmospheric correction, geometric correction and spatial registration processing on the image sequences to construct a standardized time-series dataset;
[0008] S2. Perform spatial context feature extraction processing on multi-temporal images to extract local texture information, neighborhood structure information and regional symbiotic relationship features from multi-temporal images to form a multi-scale spatial context representation.
[0009] S3. Perform context-enhanced multi-temporal feature fusion processing on multi-temporal observation features and spatial context representation, and jointly utilize multi-temporal observation features and spatial context representation to adaptively aggregate features from different time frames and different regions to obtain a fusion representation with regional structural consistency.
[0010] S4. Perform super-resolution reconstruction and structural constraint co-optimization processing. In the reconstruction process, introduce constraints such as spectral consistency, local structure preservation, spatial continuity and texture authenticity. Supervised training and network parameter optimization are performed on the super-resolution reconstruction model composed of feature encoder, spatial context feature extraction processing, multi-temporal feature fusion processing and super-resolution reconstruction network.
[0011] S5. Input the multi-temporal low-resolution remote sensing images to be reconstructed into the trained model, and output the high-resolution reconstruction results.
[0012] Furthermore, the specific steps in S1 for acquiring multi-temporal low-resolution image sequences, performing radiometric calibration, atmospheric correction, and spatial alignment to construct a standardized time-series dataset include:
[0013] S1.1 Acquire multiple low-resolution remote sensing images of the same area at different time phases, and use sensor calibration parameters to convert the original observation values into radiance values or apparent reflectance.
[0014] S1.2 Perform atmospheric correction, noise reduction and geometric fine correction on the multi-temporal images to eliminate the effects of atmospheric disturbance, sensor noise and imaging geometric differences during the imaging process;
[0015] S1.3. Using geographic reference information, control point information, or resampling methods, images from different time phases are unified to the same geographic coordinate system and spatial resolution to achieve pixel-level or sub-pixel-level spatial alignment and construct a standardized time-series dataset.
[0016] Furthermore, in step S2, spatial context feature extraction processing is performed on the multi-temporal images to extract local texture information, neighborhood structure information, and regional symbiotic relationship features from the multi-temporal images, forming a multi-scale spatial context representation, specifically including the following:
[0017] S2.1 Multi-temporal feature extraction: Representing multi-temporal low-resolution remote sensing image sequences as... The low-resolution images from each time phase are then input into a feature encoder with shared parameters to obtain the corresponding deep feature representations.
[0018]
[0019] in, This represents the low-resolution input image at time phase t; T represents the total number of time phases. For feature encoder; For the set of trainable parameters of the feature encoder; This represents the deep feature map extracted at time phase t;
[0020] S2.2 Local texture feature extraction: Local texture information is extracted from each temporal image through convolution operations or other feature extraction methods, as follows:
[0021]
[0022] in, Indicates the first Texture features of each phase, This indicates a convolution operation or other texture extraction method.
[0023] S2.3, Neighborhood Structure Information Extraction: Based on pixel neighborhood relationships, the neighborhood structure features of the image region are extracted, expressed by the formula:
[0024]
[0025] in, Indicates the first Neighborhood structure characteristics of each phase, For convolution kernel operations;
[0026] S2.4, Extraction of Regional Symbiotic Relationships: Extracting symbiotic relationship features between adjacent regions, expressed by the formula:
[0027]
[0028] in, Indicates the first Regional symbiotic characteristics of each temporal phase; Represents a set of adjacent regions;
[0029] S2.5. The above local texture, neighborhood structure, and region co-occurrence features are fused to obtain a multi-scale spatial context feature representation:
[0030]
[0031] in, , and These are the trainable weighting coefficients.
[0032] Furthermore, in S3, context-enhanced multi-temporal feature fusion processing is performed on multi-temporal observation features and spatial context representation. By jointly utilizing multi-temporal observation features and spatial context representation, adaptive aggregation of features from different time frames and different regions is performed to obtain a fused representation with regional structural consistency. Specifically, this includes the following:
[0033] S3.1. Jointly model the multi-temporal context features extracted from step S2 with the multi-temporal low-resolution features to obtain a joint feature representation:
[0034]
[0035] in, Indicates the first The fusion characteristics of each phase; Indicates feature concatenation operation;
[0036] S3.2 Calculate the importance weight of each time phase. Adjust the feature weights of each time phase according to its importance. The formula is as follows:
[0037]
[0038] in, For the first The weight of each phase, For the first Characteristics of each phase, These are trainable attention weights;
[0039] S3.3. Weighted aggregation of multi-temporal features yields a fused representation with regional structural consistency:
[0040]
[0041] in, This represents the final fusion feature representation.
[0042] Furthermore, in step S4, super-resolution reconstruction and structural constraint co-optimization are performed. Constraints on spectral consistency, local structure preservation, spatial continuity, and texture realism are introduced during the reconstruction process. Supervised training and optimization of the super-resolution reconstruction model, composed of a feature encoder, spatial context feature extraction, multi-temporal feature fusion, and a super-resolution reconstruction network, are conducted. Specifically, this includes the following steps:
[0043] S4.1 Spectral Consistency Constraint Construction: By calculating the difference between the reconstructed result and the real image, a spectral consistency loss is constructed.
[0044]
[0045] in, This indicates a loss of spectral uniformity. The corresponding high-resolution real image when referring to a reference; This represents the corresponding high-resolution reconstruction result when referring to the reference. This represents the total number of elements in a high-resolution image;
[0046] S4.2 Texture consistency constraint: By calculating the gradient difference between the reconstructed image and the real image, a texture realism loss is constructed.
[0047]
[0048] in, This indicates a loss of texture realism; Represents the spatial gradient operator; A gradient map representing the reconstructed image; A gradient map representing a real image;
[0049] S4.3 Spatial Consistency Constraint: Construct a spatial consistency loss by comparing the spatial structure of the reconstructed image with that of the real image. The formula is expressed as:
[0050]
[0051] in, For spatial consistency loss, These are the pixel values at the location in the reconstructed image and the real image, respectively. The number of pixels in the image;
[0052] S4.4 Introducing adaptive spectral constraints, spectral consistency is enhanced by color mapping correction of the reconstructed image, expressed by the formula:
[0053]
[0054] in, For spectral adjustment loss, In the color channels of the reconstructed image and the real image, respectively. The value on, This represents the number of color channels.
[0055] Furthermore, in step S5, the multi-temporal low-resolution remote sensing images to be reconstructed are input into the trained model, and high-resolution reconstruction results are output. Specific steps include:
[0056] S5.1 Input the image sequence to be reconstructed into the trained model to obtain high-resolution reconstruction results:
[0057]
[0058] in, To reconstruct high-resolution images, This is represented as a trained super-resolution model; This is represented as a sequence of low-resolution images as input.
[0059] S5.2 Perform detail restoration and result optimization processing on the initial high-resolution reconstruction result, refine and enhance the edge regions, texture regions, and structural transition regions in the reconstructed image, and eliminate reconstruction artifacts and local blurring through spatial consistency constraints to obtain the final high-resolution reconstruction result:
[0060]
[0061] in, This represents the final high-resolution reconstructed image output. This represents the initial high-resolution reconstruction result. The detail restoration and result optimization module is used to enhance image edge texture details and improve spatial structure consistency.
[0062] Beneficial Effects: This invention effectively improves the overall performance of multi-temporal remote sensing image super-resolution reconstruction by introducing a semantic prior extraction module, a category-aware attention mechanism, and a multi-task joint optimization strategy. By performing pixel-level semantic parsing on the input scene, it explicitly distinguishes different types of land features such as buildings, roads, vegetation, water bodies, and bare land, enabling the reconstruction network to adopt differentiated feature enhancement and fusion strategies based on regional semantic attributes. This significantly improves the edge blurring, texture confusion, and structural distortion problems caused by unified modeling in existing methods. By constructing a spatial context-enhanced super-resolution reconstruction module, it fully utilizes redundant and complementary information between multi-temporal observations and combines a category-aware attention mechanism to highlight key regional features, improving the detail recovery capability and reconstruction stability in complex scenes. By introducing constraints such as spectral consistency, spatial structure preservation, semantic consistency, and texture realism, it strengthens the reliability of the reconstruction results in terms of radiometric response, spatial layout, and semantic expression, avoiding the over-smoothing problem caused by relying solely on pixel regression. By establishing a multi-task collaborative optimization framework for super-resolution reconstruction and semantic parsing, it achieves mutual promotion between image reconstruction and scene understanding, further improving the model's reconstruction accuracy, structural fidelity, and applicability to subsequent tasks. This invention has significant application value in fields such as resource surveys, feature identification, environmental monitoring, disaster assessment, and urban planning. Attached Figure Description
[0063] Figure 1 This is a schematic diagram of the overall process of the present invention;
[0064] Figure 2 This is a flowchart of the spatial context feature extraction process of the present invention;
[0065] Figure 3 This is the multi-temporal feature fusion process of the present invention;
[0066] Figure 4 This is a super-resolution reconstruction and optimization diagram of the present invention. Detailed Implementation
[0067] The invention will now be further described with reference to the accompanying drawings.
[0068] Example
[0069] Furthermore, such as Figure 1 As shown, a spatial context-enhanced multi-temporal super-resolution reconstruction method includes the following steps:
[0070] S1. Acquire multi-temporal low-resolution remote sensing image sequences, and perform radiometric calibration, atmospheric correction, geometric correction and spatial registration processing on the image sequences to construct a standardized time-series dataset;
[0071] S2. Perform spatial context feature extraction processing on multi-temporal images to extract local texture information, neighborhood structure information and regional symbiotic relationship features from multi-temporal images to form a multi-scale spatial context representation.
[0072] S3. Perform context-enhanced multi-temporal feature fusion processing on multi-temporal observation features and spatial context representation, and jointly utilize multi-temporal observation features and spatial context representation to adaptively aggregate features from different time frames and different regions to obtain a fusion representation with regional structural consistency.
[0073] S4. Perform super-resolution reconstruction and structural constraint co-optimization processing. In the reconstruction process, introduce constraints such as spectral consistency, local structure preservation, spatial continuity and texture authenticity. Supervised training and network parameter optimization are performed on the super-resolution reconstruction model composed of feature encoder, spatial context feature extraction processing, multi-temporal feature fusion processing and super-resolution reconstruction network.
[0074] S5. Input the multi-temporal low-resolution remote sensing images to be reconstructed into the trained model, and output the high-resolution reconstruction results.
[0075] Furthermore, such as Figure 2 As shown, the specific steps in S1 for acquiring multi-temporal low-resolution image sequences, performing radiometric calibration, atmospheric correction, and spatial alignment, and constructing a standardized time-series dataset include:
[0076] S1.1 Acquire multiple low-resolution remote sensing images of the same area at different time phases, and use sensor calibration parameters to convert the original observation values into radiance values or apparent reflectance.
[0077] S1.2 Perform atmospheric correction, noise reduction and geometric fine correction on the multi-temporal images to eliminate the effects of atmospheric disturbance, sensor noise and imaging geometric differences during the imaging process;
[0078] S1.3. Using geographic reference information, control point information, or resampling methods, images from different time phases are unified to the same geographic coordinate system and spatial resolution to achieve pixel-level or sub-pixel-level spatial alignment and construct a standardized time-series dataset.
[0079] Furthermore, such as Figure 2 As shown, in step S2, spatial context feature extraction processing is performed on multi-temporal images to extract local texture information, neighborhood structure information, and regional symbiotic relationship features from the multi-temporal images, forming a multi-scale spatial context representation. The specific content is as follows:
[0080] S2.1 Multi-temporal feature extraction: Representing multi-temporal low-resolution remote sensing image sequences as... The low-resolution images from each time phase are then input into a feature encoder with shared parameters to obtain the corresponding deep feature representations.
[0081]
[0082] in, This represents the low-resolution input image at time phase t; T represents the total number of time phases. For feature encoder; For the set of trainable parameters of the feature encoder; This represents the deep feature map extracted at time phase t;
[0083] S2.2 Local texture feature extraction: Local texture information is extracted from each temporal image through convolution operations or other feature extraction methods, as follows:
[0084]
[0085] in, Indicates the first Texture features of each phase, This indicates a convolution operation or other texture extraction method.
[0086] S2.3, Neighborhood Structure Information Extraction: Based on pixel neighborhood relationships, the neighborhood structure features of the image region are extracted, expressed by the formula:
[0087]
[0088] in, Indicates the first Neighborhood structure characteristics of each phase, For convolution kernel operations;
[0089] S2.4, Extraction of Regional Symbiotic Relationships: Extracting symbiotic relationship features between adjacent regions, expressed by the formula:
[0090]
[0091] in, Indicates the first Regional symbiotic characteristics of each temporal phase; Represents a set of adjacent regions;
[0092] S2.5. The above local texture, neighborhood structure, and region co-occurrence features are fused to obtain a multi-scale spatial context feature representation:
[0093]
[0094] in, , and These are the trainable weighting coefficients.
[0095] Furthermore, such as Figure 3 As shown, S3 involves context-enhanced multi-temporal feature fusion processing of multi-temporal observation features and spatial context representation. It jointly utilizes multi-temporal observation features and spatial context representation to adaptively aggregate features from different time frames and regions, obtaining a fused representation with regional structural consistency. Specifically, this includes the following:
[0096] S3.1. Jointly model the multi-temporal context features extracted from step S2 with the multi-temporal low-resolution features to obtain a joint feature representation:
[0097]
[0098] in, Indicates the first The fusion characteristics of each phase; Indicates feature concatenation operation;
[0099] S3.2 Calculate the importance weight of each time phase. Adjust the feature weights of each time phase according to its importance. The formula is as follows:
[0100]
[0101] in, For the first The weight of each phase, For the first Characteristics of each phase, These are trainable attention weights;
[0102] S3.3. Weighted aggregation of multi-temporal features yields a fused representation with regional structural consistency:
[0103]
[0104] in, This represents the final fusion feature representation.
[0105] Furthermore, such as Figure 4 As shown, in step S4, super-resolution reconstruction and structural constraint co-optimization are performed. Constraints of spectral consistency, local structure preservation, spatial continuity, and texture realism are introduced during the reconstruction process. Supervised training and optimization of the super-resolution reconstruction model, composed of a feature encoder, spatial context feature extraction, multi-temporal feature fusion, and a super-resolution reconstruction network, are conducted. Specifically, the steps include:
[0106] S4.1 Spectral Consistency Constraint Construction: By calculating the difference between the reconstructed result and the real image, a spectral consistency loss is constructed.
[0107]
[0108] in, This indicates a loss of spectral uniformity. The corresponding high-resolution real image when referring to a reference; This represents the corresponding high-resolution reconstruction result when referring to the reference. This represents the total number of elements in a high-resolution image;
[0109] S4.2 Texture consistency constraint: By calculating the gradient difference between the reconstructed image and the real image, a texture realism loss is constructed.
[0110]
[0111] in, This indicates a loss of texture realism; Represents the spatial gradient operator; A gradient map representing the reconstructed image; A gradient map representing a real image;
[0112] S4.3 Spatial Consistency Constraint: Construct a spatial consistency loss by comparing the spatial structure of the reconstructed image with that of the real image. The formula is expressed as:
[0113]
[0114] in, For spatial consistency loss, These are the pixel values at the location in the reconstructed image and the real image, respectively. The number of pixels in the image;
[0115] S4.4 Introducing adaptive spectral constraints, spectral consistency is enhanced by color mapping correction of the reconstructed image, expressed by the formula:
[0116]
[0117] in, For spectral adjustment loss, In the color channels of the reconstructed image and the real image, respectively. The value on, This represents the number of color channels.
[0118] Furthermore, in step S5, the multi-temporal low-resolution remote sensing images to be reconstructed are input into the trained model, and high-resolution reconstruction results are output. Specific steps include:
[0119] S5.1 Input the image sequence to be reconstructed into the trained model to obtain high-resolution reconstruction results:
[0120]
[0121] in, To reconstruct high-resolution images, This is represented as a trained super-resolution model; This is represented as a sequence of low-resolution images as input.
[0122] S5.2 Perform detail restoration and result optimization processing on the initial high-resolution reconstruction result, refine and enhance the edge regions, texture regions, and structural transition regions in the reconstructed image, and eliminate reconstruction artifacts and local blurring through spatial consistency constraints to obtain the final high-resolution reconstruction result:
[0123]
[0124] in, This represents the final high-resolution reconstructed image output. This represents the initial high-resolution reconstruction result. The detail restoration and result optimization module is used to enhance image edge texture details and improve spatial structure consistency.
[0125] To illustrate the practical implementation of the method of this invention, this embodiment selects the BreizhSR multi-temporal remote sensing super-resolution dataset as validation data. The input data consists of images from the four commonly used bands B2, B3, B4, and B8 in Sentinel-2, with a reconstruction magnification of 4x. Each input sample uses 13 low-resolution remote sensing images of the same region as the standard, but can be expanded to 5, 9, 17, or 25 frames depending on the actual data completeness; the corresponding high-resolution images serve as supervision labels. The input images are first radiometrically calibrated, geometrically corrected, normalized, and invalid pixels are removed. Then, the frame with the highest temporal quality and the most central location is used as the reference temporal phase, and the remaining temporal images are unified to the same geographic coordinate system and spatial resolution, completing the multi-temporal sequence organization and standardization process.
[0126] In terms of model structure, standardized multi-temporal low-resolution images are input into a feature encoder with shared parameters, and the basic feature dimension is set to 48. The feature encoder consists of convolutional layers, normalization layers, and activation layers, used to extract basic features from images at each temporal stage. Subsequently, local texture branches, neighborhood structure branches, and region symbiotic relationship branches are set, where the local texture branch is used to extract edge, thin line, and high-frequency texture information; the neighborhood structure branch is used to extract spatial relationships such as building outlines, road directions, and waterfronts; and the region symbiotic relationship branch is used to describe the spatial organization relationship between adjacent land feature areas. After the three types of features are fused with learnable weights, a multi-scale spatial context representation is formed.
[0127] In the multi-temporal feature fusion process, spatial context representation and basic features of each temporal phase are jointly modeled, and temporal importance weights and spatial context gating graphs are calculated to adaptively aggregate features from different temporal phases. Image frames with clear structure, high quality, and strong consistency with the reference temporal phase are assigned higher fusion weights; image frames with cloud or fog obstruction, noise perturbation, local misalignment, or temporal missing information have their fusion weights reduced, thereby minimizing the impact of unreliable temporal features on the reconstruction results.
[0128] During training, the fused multi-temporal features are input into the super-resolution reconstruction network, and high-resolution reconstruction results are obtained through upsampling and detail restoration. During training, spectral consistency loss, local structure preservation loss, spatial continuity loss, and texture realism loss are used to jointly constrain the model parameters. Spectral consistency loss is used to maintain stable radiative responses across bands; local structure preservation loss is used to enhance edge and texture responses; spatial continuity loss is used to reduce road breaks and regional fragmentation; and texture realism loss is used to reduce over-smoothing and pseudo-texture generation. The model is trained end-to-end using the Adam optimizer, with parameters β1 set to 0.9, β2 set to 0.999, and an initial learning rate of [missing information]. In the later stages, a cosine decay strategy is adopted. During inference, the multi-temporal low-resolution remote sensing images to be reconstructed are input into the trained model to obtain the initial high-resolution reconstructed image. Then, detail restoration and result optimization are performed on the edge regions, texture regions, and structural transition regions to output the final high-resolution reconstruction result.
[0129] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention. The scope of protection of the present invention should be determined by the scope of protection of the appended claims.
Claims
1. A spatial context-enhanced multi-temporal super-resolution reconstruction method, characterized in that, Includes the following steps: S1. Acquire multi-temporal low-resolution remote sensing image sequences, and perform radiometric calibration, atmospheric correction, geometric correction and spatial registration processing on the image sequences to construct a standardized time-series dataset; S2. Perform spatial context feature extraction processing on multi-temporal images to extract local texture information, neighborhood structure information and regional symbiotic relationship features from multi-temporal images to form a multi-scale spatial context representation. S3. Perform context-enhanced multi-temporal feature fusion processing on multi-temporal observation features and spatial context representation, and jointly utilize multi-temporal observation features and spatial context representation to adaptively aggregate features from different time frames and different regions to obtain a fusion representation with regional structural consistency. S4. Perform super-resolution reconstruction and structural constraint co-optimization processing. In the reconstruction process, introduce constraints such as spectral consistency, local structure preservation, spatial continuity and texture authenticity. Supervised training and network parameter optimization are performed on the super-resolution reconstruction model composed of feature encoder, spatial context feature extraction processing, multi-temporal feature fusion processing and super-resolution reconstruction network. S5. Input the multi-temporal low-resolution remote sensing images to be reconstructed into the trained model, and output the high-resolution reconstruction results.
2. The spatial context-enhanced multi-temporal super-resolution reconstruction method according to claim 1, characterized in that, S1 includes the following steps: S1.1 Acquire multiple low-resolution remote sensing images of the same area at different time phases, and use sensor calibration parameters to convert the original observation values into radiance values or apparent reflectance. S1.2 Perform atmospheric correction, noise reduction and geometric fine correction on the multi-temporal images to eliminate the effects of atmospheric disturbance, sensor noise and imaging geometric differences during the imaging process; S1.
3. Using geographic reference information, control point information, or resampling methods, images from different time phases are unified to the same geographic coordinate system and spatial resolution to achieve pixel-level or sub-pixel-level spatial alignment and construct a standardized time-series dataset.
3. The spatial context-enhanced multi-temporal super-resolution reconstruction method according to claim 1, characterized in that, S2 includes the following steps: S2.1 Multi-temporal feature extraction: Representing multi-temporal low-resolution remote sensing image sequences as... The low-resolution images from each time phase are then input into a feature encoder with shared parameters to obtain the corresponding deep feature representations. in, This represents the low-resolution input image at time phase t; T represents the total number of time phases. For feature encoder; For the set of trainable parameters of the feature encoder; This represents the deep feature map extracted at time phase t; S2.2 Local texture feature extraction: Local texture information is extracted from each temporal image through convolution operations or other feature extraction methods, as follows: in, Indicates the first Texture features of each phase, This indicates a convolution operation or other texture extraction method. S2.3, Neighborhood Structure Information Extraction: Based on pixel neighborhood relationships, the neighborhood structure features of the image region are extracted, expressed by the formula: in, Indicates the first Neighborhood structure characteristics of each phase, For convolution kernel operations; S2.4, Extraction of Regional Symbiotic Relationships: Extracting symbiotic relationship features between adjacent regions, expressed by the formula: in, Indicates the first Regional symbiotic characteristics of each temporal phase; Represents a set of adjacent regions; S2.
5. The above local texture, neighborhood structure, and region co-occurrence features are fused to obtain a multi-scale spatial context feature representation: in, , and These are the trainable weighting coefficients.
4. The spatial context-enhanced multi-temporal super-resolution reconstruction method according to claim 1, characterized in that, S3 includes the following steps: S3.
1. Jointly model the multi-temporal context features extracted from step S2 with the multi-temporal low-resolution features to obtain a joint feature representation: in, Indicates the first The fusion characteristics of each phase; Indicates feature concatenation operation; S3.2 Calculate the importance weight of each time phase. Adjust the feature weights of each time phase according to its importance. The formula is as follows: in, For the first The weight of each phase, For the first Characteristics of each phase, These are trainable attention weights; S3.
3. Weighted aggregation of multi-temporal features yields a fused representation with regional structural consistency: in, This represents the final fusion feature representation.
5. The spatial context-enhanced multi-temporal super-resolution reconstruction method according to claim 1, characterized in that, S4 includes the following steps: S4.1 Spectral Consistency Constraint Construction: By calculating the difference between the reconstructed result and the real image, a spectral consistency loss is constructed. in, This indicates a loss of spectral uniformity. The corresponding high-resolution real image when referring to a reference; This represents the corresponding high-resolution reconstruction result when referring to the reference. This represents the total number of elements in a high-resolution image; S4.2 Texture consistency constraint: By calculating the gradient difference between the reconstructed image and the real image, a texture realism loss is constructed. in, This indicates a loss of texture realism; Represents the spatial gradient operator; A gradient map representing the reconstructed image; A gradient map representing a real image; S4.3 Spatial Consistency Constraint: Construct a spatial consistency loss by comparing the spatial structure of the reconstructed image with that of the real image. The formula is expressed as: in, For spatial consistency loss, These are the pixel values at the location in the reconstructed image and the real image, respectively. The number of pixels in the image; S4.4 Introducing adaptive spectral constraints, spectral consistency is enhanced by color mapping correction of the reconstructed image, expressed by the formula: in, For spectral adjustment loss, In the color channels of the reconstructed image and the real image, respectively. The value on, This represents the number of color channels.
6. The spatial context-enhanced multi-temporal super-resolution reconstruction method according to claim 1, characterized in that, S5 includes the following steps: S5.1 Input the image sequence to be reconstructed into the trained model to obtain high-resolution reconstruction results: in, To reconstruct high-resolution images, This is represented as a trained super-resolution model; This is represented as a sequence of low-resolution images as input. S5.2 Perform detail restoration and result optimization processing on the initial high-resolution reconstruction result, refine and enhance the edge regions, texture regions, and structural transition regions in the reconstructed image, and eliminate reconstruction artifacts and local blurring through spatial consistency constraints to obtain the final high-resolution reconstruction result: in, This represents the final high-resolution reconstructed image output. This represents the initial high-resolution reconstruction result. The detail restoration and result optimization module is used to enhance image edge texture details and improve spatial structure consistency.