A heterogeneous remote sensing satellite image full-scene high-precision matching and multi-dimensional evaluation method and system

Abstract

The embodiment of the application provides a kind of heterogeneous remote sensing satellite image full scene high-precision matching and multi-dimensional evaluation method and system, the method comprises: obtaining the image to be registered and reference image of target area, by time-space pairing and quality control, establish heterogeneous remote sensing image dataset;Multi-level screening is carried out to the heterogeneous remote sensing image dataset, and the optimal control point set of the same name is extracted;Parameter combination is preset, and the candidate same name point is searched from the optimal control point set based on parameter combination, arithmetic average fusion is carried out after threshold screening, to generate fusion point set;Obtain transformation model, solve the transformation model using fusion point set, and perform iterative matching until convergence;Based on transformation model, obtain matched image, and carry out qualitative evaluation, after qualitative evaluation passes, quantitative evaluation is calculated in combination with independent verification point set, and the matching precision result is output.

Description

Technical Field

[0001] This invention relates to the field of aerospace data processing technology, and in particular to a method and system for high-precision matching and multi-dimensional evaluation of heterogeneous remote sensing satellite images across all scenes. Background Technology

[0002] With the continuous iteration, network improvement, and large-scale application of the space-based remote sensing Earth observation system, heterogeneous remote sensing satellite imagery, represented by visible light, infrared, and synthetic aperture radar (SAR), has been deeply integrated into core application scenarios such as precise target identification, fine classification of ground features, detection of multi-temporal changes, construction of deep learning sample libraries, emergency remote sensing monitoring, and land spatial mapping, becoming a key data support for integrated space-air-ground Earth observation. Constrained by the inherent differences in sensor physical imaging mechanisms, observation geometry, and radiative transfer characteristics, single-modality remote sensing observation suffers from significant limitations such as limited information dimensions, insufficient environmental adaptability, and incomplete ground feature representation. However, collaborative acquisition of heterogeneous images can achieve geometric information complementarity, enhanced radiative features, and redundant verification of spatiotemporal information, significantly improving the robustness of target identification and the reliability of scene monitoring in complex environments. This represents a core technological path to overcome the bottlenecks in the application of single-modality satellite imagery.

[0003] High-precision matching is a crucial preliminary step in the application of heterogeneous remote sensing satellite imagery. It plays a key role in establishing geometric correspondences between images and eliminating observational biases. The reliability and accuracy of the selection of corresponding points directly determine the usability of subsequent target identification, image fusion, and sample library creation results, and is a significant technical bottleneck restricting the engineering application of heterogeneous remote sensing data. Currently, remote sensing satellite data sources exhibit complex characteristics of multiple platforms, payloads, resolutions, perspectives, and time phases. Different satellites show significant inconsistencies in spatial resolution, imaging time windows, observation elevation and azimuth angles, radiometric calibration systems, and spectral response functions. In addition, infrared images are affected by the coupling of surface temperature and emissivity, and SAR images suffer from speckle noise and side-look geometric distortion. Both types of data generally suffer from inherent problems such as nonlinear radiometric distortion, topographic perspective contraction and overlay, geometric projection differences, and grayscale distribution mismatch, further exacerbating the characteristic heterogeneity between heterogeneous images and significantly increasing the technical difficulty of high-precision matching.

[0004] However, to address the aforementioned issues, existing traditional remote sensing image matching algorithms are mostly designed for images from the same source. They primarily rely on gray-level correlation coefficients, local gradient extrema, and single-point low-level visual features to complete feature detection and matching. These algorithms have stringent requirements for radiometric consistency and geometric similarity, making it difficult to adapt to the nonlinear radiometric differences and complex geometric distortions of cross-modal data such as infrared-SAR, infrared-visible light, and SAR-visible light. In heterogeneous scenarios, they commonly exhibit problems such as low feature repetition rates, scarce correct matching points, high mismatch rates, and matching failures. Furthermore, in complex scenarios such as mountainous areas, hills, deserts, Gobi, oceans, and glaciers, terrain undulations cause local nonlinear geometric deformations, weak / textured areas have scarce features, and ground object scattering / radiative characteristics tend to converge. This easily leads to matching points concentrating in high-textured areas, unbalanced spatial distribution of control points, excessive matching residuals, and overall model distortion, making it difficult to meet the high-precision registration requirements across all scenarios.

[0005] Furthermore, existing matching technologies lack comprehensive, scientific, hierarchical, and global precision quality control and multidimensional evaluation methods. Relying solely on macroscopic indicators such as global root mean square error (RMSE) and single-point geometric residuals to evaluate matching results fails to achieve refined quality assessment by region, terrain, and level. This easily masks significant deviations and systematic errors in sensitive local areas such as mountainous regions, deserts, and coastal zones, resulting in insufficient usability and poor stability of matching results in actual production. In summary, for the application of multimodal heterogeneous remote sensing satellite data in complex scenarios across the entire Earth's surface, there is an urgent need to develop a robust, versatile, and high-precision heterogeneous remote sensing satellite image matching and multidimensional evaluation method. Summary of the Invention

[0006] To address the problems existing in the prior art, embodiments of the present invention provide a method and system for high-precision matching and multi-dimensional evaluation of heterogeneous remote sensing satellite imagery across all scenes.

[0007] Firstly, embodiments of this specification provide a method for high-precision matching and multi-dimensional evaluation of heterogeneous remote sensing satellite imagery across all scenes, the method comprising:

[0008] Acquire the images to be registered and reference images of the target area, and establish a heterogeneous remote sensing image dataset through spatiotemporal pairing and quality control;

[0009] The heterogeneous remote sensing image dataset is subjected to multi-level filtering to extract the optimal set of control points with the same name;

[0010] A preset parameter combination is used to search for candidate points with the same name from the optimal control point set. After threshold screening, the points are fused by arithmetic average to generate a fused point set.

[0011] Obtain the transformation model, solve the transformation model using the fusion point set, and perform iterative matching until convergence;

[0012] The matched images are obtained based on the transformation model and a qualitative evaluation is performed. After the qualitative evaluation is passed, a quantitative evaluation is calculated by combining the independent verification point set, and the matching accuracy result is output.

[0013] Secondly, embodiments of this specification provide a high-precision matching and multi-dimensional evaluation system for all scenes of heterogeneous remote sensing satellite imagery, the system comprising:

[0014] The acquisition module is used to acquire the images to be registered and the reference images of the target area, and to establish a heterogeneous remote sensing image dataset through spatiotemporal pairing and quality control.

[0015] The filtering module is used to perform multi-level filtering on the heterogeneous remote sensing image dataset and extract the optimal control point set with the same name;

[0016] The fusion module is used to preset parameter combinations, search for candidate points with the same name from the optimal control point set based on the parameter combinations, perform arithmetic mean fusion after threshold screening, and generate a fused point set.

[0017] The transformation module is used to obtain the transformation model, solve the transformation model using the fusion point set, and perform iterative matching until convergence.

[0018] The evaluation module is used to obtain the matched image based on the transformation model and perform a qualitative evaluation. After the qualitative evaluation is passed, a quantitative evaluation is calculated by combining the independent verification point set, and the matching accuracy result is output.

[0019] Thirdly, embodiments of this specification provide an electronic device, including a processor and a memory;

[0020] The processor is connected to the memory;

[0021] The memory is used to store executable program code;

[0022] The processor runs a program corresponding to the executable program code stored in the memory to perform the methods described in one or more embodiments.

[0023] Fourthly, embodiments of this specification provide a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the above-described method for high-precision matching and multi-dimensional evaluation of heterogeneous remote sensing satellite imagery across all scenes.

[0024] In view of the above, the beneficial effects of the technical solutions provided by some embodiments of this specification include at least the following:

[0025] In one or more embodiments of this specification, a target area image to be registered and a reference image are acquired. A heterogeneous remote sensing image dataset is established through spatiotemporal pairing and quality control. The heterogeneous remote sensing image dataset undergoes multi-level filtering to extract the optimal set of control points with the same name. Preset parameter combinations are used to search for candidate points with the same name from the optimal control point set. After threshold filtering, arithmetic mean fusion is performed to generate a fused point set. A transformation model is obtained, and the transformation model is solved using the fused point set, performing iterative matching until convergence. The matched image is obtained based on the transformation model, and a qualitative evaluation is performed. After the qualitative evaluation is passed, a quantitative evaluation is calculated using an independent validation point set, and the matching accuracy result is output. This embodiment employs a progressive approach, employing heterogeneous remote sensing satellite data acquisition and quality control, adaptive extraction of the optimal set of corresponding control points, iterative fine matching across different scenarios, and multi-dimensional accuracy evaluation based on terrain and scene collaboration. Each step demonstrates strong cross-modal compatibility and scene adaptability, enabling unified processing of heterogeneous remote sensing images from visible light, infrared, and SAR sources with vastly different imaging mechanisms, eliminating the need for individual adjustments to core strategies for different sensors. Furthermore, it covers a complete control point extraction and optimization mechanism, comprehensively improving matching quality. Moreover, it establishes a hierarchical, multi-dimensional accuracy evaluation and quality control system, resolving issues such as inflated accuracy and distorted evaluation inherent in traditional self-inspection and evaluation methods, ensuring that matching results meet the requirements of high accuracy, robustness, and availability across the entire domain. Attached Figure Description

[0026] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0027] Figure 1 This is a flowchart of a method for high-precision matching and multi-dimensional evaluation of heterogeneous remote sensing satellite imagery across all scenes, provided in one embodiment of this specification.

[0028] Figure 2 This is a flowchart of a multidimensional accuracy evaluation provided in one embodiment of this specification.

[0029] Figure 3 This is a schematic diagram of the structure of a heterogeneous remote sensing satellite image full-scene high-precision matching and multi-dimensional evaluation system provided in one embodiment of this specification.

[0030] Figure 4 This is a schematic diagram of the structure of an electronic device provided in one embodiment of this specification. Detailed Implementation

[0031] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed merely to enable those skilled in the art to better understand and implement the subject matter described herein, and are not intended to limit the scope, applicability, or examples set forth in the claims. The function and arrangement of the elements discussed may be changed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the various examples. For example, the described methods may be performed in a different order than described, and steps may be added, omitted, or combined. Furthermore, features described in some examples may be combined in other examples.

[0032] As used herein, the term "comprising" and its variations are open terms meaning "including but not limited to". The term "based on" means "at least partially based on". The terms "one embodiment" and "an embodiment" mean "at least one embodiment". The term "another embodiment" means "at least one other embodiment". The terms "first", "second", etc., may refer to different or the same objects. Other definitions, whether explicit or implicit, may be included below. Unless explicitly indicated by the context, the definition of a term shall remain consistent throughout the specification.

[0033] Please see Figure 1 , Figure 1 This document presents an overall flowchart of a method for high-precision matching and multi-dimensional evaluation of heterogeneous remote sensing satellite imagery across all scenes, as provided in an embodiment of this specification.

[0034] like Figure 1 As shown, this method for high-precision matching and multi-dimensional evaluation of heterogeneous remote sensing satellite imagery across all scenes includes at least the following steps:

[0035] Step S102: Obtain the image to be registered and the reference image of the target area, and establish a heterogeneous remote sensing image dataset through spatiotemporal pairing and quality control.

[0036] Specifically, the purpose of this step is to construct a heterogeneous remote sensing image dataset with unified spatiotemporal references, complete modal types, and consistent quality standards, thereby addressing issues such as cloud and snow occlusion, excessive temporal differences, radiometric distortion, and missing positioning information inherent in the input data. This includes:

[0037] When collecting data for the target area, corresponding acquisition rules should be set. These include, but are not limited to, spatiotemporal consistency, modal completeness, and data availability rules. Spatiotemporal consistency ensures that the reference image and the image to be registered are similar in imaging time, season, and day / night conditions, reducing differences caused by mismatch factors such as variations in ground radiation, shadows, and vegetation phenology. Modal completeness ensures the simultaneous availability of multiple modal data, including visible light, thermal infrared, and SAR, covering the main combinations of heterogeneous matching. Data availability, while meeting spatiotemporal consistency, guarantees complete spatial coverage of the image and avoids severe cloud and snow obstruction. Specific data sources for the target area may include: thermal infrared data: ZY1-02E thermal infrared band imagery; SAR data: RADARSAT-2 and GF-3 satellite imagery; visible light reference data: Sentinel-2 satellite imagery; and terrain-aided data: Copernicus 30m resolution DEM, etc. Among them, thermal infrared and SAR are used as images to be registered, visible light is used as reference images, and DEM is used as auxiliary information for subsequent geometric correction and terrain co-evaluation.

[0038] Based on the imaging range and geographical coverage boundaries of thermal infrared and SAR (images to be registered) images, spatiotemporal pairing and coverage control are performed. For example, visible light reference images with the same season, similar time phase, cloudless or low-cloud, and seamless full spatial coverage are rigorously selected to construct a basic dataset for pairing reference images with images to be registered. In addition, pairing is differentiated by various time periods, such as spring, summer, autumn, and winter, and day / night dual time periods, and covers various terrains, including plains, urban built-up areas, deserts / Gobi, mountains / hills, and other landforms, taking into account both natural and man-made areas. This ensures spatial consistency, avoiding subsequent spatial range misalignment; temporal consistency; and scene diversity.

[0039] In addition, standardized multi-level quality control is performed on the acquired raw images, checking them scene by scene and band by band. Quality control indicators include, but are not limited to, cloud cover percentage (removing images severely obscured by thick clouds or snow), map sheet integrity (checking for data loss, stripe noise, missing borders, etc.), temporal consistency (checking whether the reference image and the image to be registered belong to the same or similar season, and whether day and night are consistent), radiometric calibration accuracy (checking whether radiometric calibration parameters are valid and whether radiometric distortion is within acceptable limits), and the validity of the location file. Unqualified data with problems such as data loss, stripe noise, radiometric distortion, excessively large temporal spans, or missing location information are removed, ultimately forming a standardized heterogeneous remote sensing image dataset with unified spatiotemporal references, complete modal types, and consistent quality standards.

[0040] Step S104: Perform multi-level filtering on the heterogeneous remote sensing image dataset to extract the optimal control point set with the same name.

[0041] Specifically, this step employs a multi-level screening strategy to extract high-quality control points between the image to be registered and the reference image from heterogeneous remote sensing image datasets. The process includes:

[0042] 1. Initial screening of points with the same name (RANSAC algorithm can be used):

[0043] Based on the image size of the heterogeneous remote sensing image dataset, the image is adaptively divided into multiple non-overlapping uniform blocks to ensure that each block is of moderate size and evenly distributed. The number of blocks is dynamically adjusted to ensure that each block is of moderate size. When the image is large, more blocks are used to avoid diluting local features due to excessively large blocks; when the image is small, fewer blocks are used to avoid insufficient feature points. This ensures that feature points are spatially uniformly distributed from the source. Then, the FAST feature response value is calculated pixel by pixel within each block, and the top few pixels with the largest FAST response values ​​(e.g., the first [100–3000]) are selected as the initial feature points for that block. Finally, a globally uniformly distributed initial set of corresponding points is formed by cross-image stitching (stitching between the reference image and the image to be registered).

[0044] Then, based on the initial set of corresponding points, a minimum sample subset is randomly selected (e.g., generating corresponding point pairs based on feature descriptors or template matching). The number of sample points strictly meets the minimum degrees of freedom requirement for solving the matching transformation model, such as 3 or more sets, to solve the model with the fewest point pairs, ensuring the efficient operation of the RANSAC algorithm. A temporary matching transformation model is calculated based on this minimum subset, completing the initial solution of the model parameters. The temporary model is applied to all initial matching points, and a forward transformation is performed on each reference image feature point to obtain the predicted coordinates on the image to be registered. The Euclidean distance between the predicted coordinates and the measured matching coordinates of the point on the image to be registered is calculated. A global error threshold is set; points with a distance less than the threshold are considered reliable points supporting the current model, otherwise they are considered suspicious points. Iterative optimization is performed, repeating random sampling and model solving, retaining the model with the most reliable points as the global optimal model. Based on the optimal model, all reliable corresponding points are retained, and mismatched points are eliminated, completing the initial screening of corresponding points.

[0045] 2. Refinement of points with the same name:

[0046] Based on the initial screening, slope stratification constraints are implemented using DEM (Digital Equation Model), and a two-level error threshold differentiation judgment mechanism is established. This mechanism removes gross errors while preserving key control points in terrain-sensitive areas to the maximum extent possible, avoiding the loss of true terrain information. This includes:

[0047] By employing a cross-validation strategy and selecting an appropriate matching transformation model, the abscissa residual Δx, ordinate residual Δy, and single-point comprehensive residual SPPR of each pair of corresponding points in the initial screening point set are calculated, thereby realizing the quantitative expression of single-point error.

[0048] ,

[0049] By dividing the point set into a training set and a validation set, the model is fitted using the training set and the residuals are calculated using the validation set, thus avoiding model overfitting that could lead to inaccurate residual estimates.

[0050] Then, the single-point composite residuals are compared sequentially with strict error thresholds (e.g., 0.5–1 pixels, which can be adaptively adjusted) and lenient error thresholds (e.g., 1–2 pixels, which can also be adaptively adjusted), and the terrain attributes of the point are determined in conjunction with the DEM: If the point is located in a flat area, high residuals are mainly caused by positioning gross errors and mismatches; if the SPPR exceeds the strict threshold, it is directly removed. If the point is located in a region with drastic undulations such as mountains or steep slopes, high residuals may be caused by real surface factors such as terrain nonlinear distortion and projection differences; it is only removed when the residual exceeds the lenient threshold, otherwise it is retained. This allows the DEM to be used to determine the degree of terrain undulation at the point, providing a basis for threshold selection. This approach can remove gross errors while maximizing the retention of key control points in terrain-sensitive areas, avoiding the loss of true terrain information.

[0051] After multiple levels of iterative screening—that is, by repeatedly estimating the model, calculating residuals, and removing outliers—a stable and reliable set of control points is gradually approximated. This results in a set of corresponding points that is simplified in number, reliable in accuracy, and has a more reasonable spatial distribution. This set effectively suppresses noise interference and fully preserves the true geometric features of the land surface, and is defined as the basic control point set.

[0052] 3. Control Point Supplementation: To address the problem of insufficient and sparsely distributed basic control points in complex terrain scenes, which cannot support high-precision model solving, an interactive manual supplementation algorithm with line feature constraints is introduced. In matching weak areas and abrupt terrain changes, stable linear terrain features such as ridgelines and valley lines are identified through visual interpretation and edge detection algorithms. Feature points are extracted uniformly along the target lines, and corresponding points are picked at the corresponding locations in the reference image. After point-by-point confirmation, these points are incorporated into the control point set, forming a final complete, robust, and evenly distributed optimal control point set. Ridgelines, valley lines, and other terrain lines typically exhibit good geometric stability in different modal images; extracting points along these lines ensures that the supplemented points have a clear geometric correspondence. Supplementing the basic control point set with control points in weak areas ensures that the point set meets engineering requirements in terms of quantity, accuracy, and distribution. In this step, the initial set of corresponding points refers to the set of points obtained after initial screening of corresponding points; the basic control point set refers to the set of points obtained after refining the initial set of corresponding points; and the optimal control point set refers to the final set of points obtained by supplementing the basic control point set with control points.

[0053] Step S106: Preset parameter combination, search for candidate points with the same name from the optimal control point set based on the parameter combination, perform arithmetic mean fusion after threshold screening, and generate a fused point set;

[0054] Step S108: Obtain the transformation model, solve the transformation model using the fusion point set, and perform iterative matching until convergence.

[0055] Specifically, although the obtained optimal control point set is relatively reliable and evenly distributed, it is still a discrete set of corresponding points with certain residuals. This step processes these discrete corresponding points, estimating a high-precision, scene-adaptive geometric mapping model to accurately correct the image to be registered to the reference image coordinate system. This includes:

[0056] Several sets of parameters are preset, including the number of control points, search window, and template window, to determine the "granularity" of matching corresponding points. The number of control points is the number of corresponding points used for matching or model estimation in the current parameter set. This is achieved by randomly selecting points proportionally from the optimal control point set each round, or by selecting the top N points based on their FAST response values. The search window is the size of the rectangular region on the image to be registered, centered on the predicted location of a feature point in the reference image, to search for corresponding points. The template window is the size of a local image patch centered on the feature point, used for similarity measurement. A coarse-to-fine matching strategy is then adopted, with gradually increasing point density and gradually decreasing matching window, to adapt to multi-level distortion correction needs from global to local. The reason for this matching strategy is that in the coarse matching stage, fewer control points, a larger search window, and a larger template window are used. The aim is to quickly obtain the global geometric transformation trend between images, such as overall translation, rotation, and scale differences. A larger search range is allowed at this stage to avoid failing to find corresponding points due to large initial deviations. In the fine matching stage, the control point density is increased, and the search window and template window are reduced. At this point, the global geometric relationship has been roughly determined. Reducing the window can reduce the amount of computation, lower the false matching rate, and better capture local nonlinear distortions (such as local projection differences in mountains and local deformations caused by SAR overlay).

[0057] Then, iterate through several sets of parameters using a template-matching local refinement method to generate corresponding candidate points. The specific process can be as follows: Take a pair of corresponding points from the optimal control point set; extract a reference template A (centered on the initial coordinates) of the template window size from the reference image; then slide the template within the search window (centered on the initial coordinates) on the image to be registered; at each position, extract a candidate image block B of the same size; calculate the similarity measure between A and B. Based on the similarity measure, select the optimal position C on the image to be registered; the refined coordinates of C and the initial coordinates of A generate candidate points.

[0058] Calculate quantitative indicators such as residuals and overlap for each candidate point. The smaller the residual, the more consistent the candidate point is with the current geometric model, and the higher the matching reliability. Retain candidate points with residuals less than a preset threshold. The threshold can be set with reference to the strict error threshold (e.g., 0.5–1 pixel) and the lenient error threshold (e.g., 1–2 pixels) in step S104, targeting areas with strong local distortion such as mountains and steep slopes, as well as flat areas.

[0059] After several sets of parameter matching (according to the matching strategy, from coarse matching to fine matching), several batches of candidate points with the same name are generated. These points are distributed in different locations in the image and correspond to different matching granularities. Greedy clustering and arithmetic mean fusion are performed on the candidate point sets obtained from multiple rounds of screening. Greedy clustering uses the first point as the cluster center, traverses the remaining points, calculates the Euclidean distance between them, and if the distance is less than a threshold, they are assigned to the same cluster; otherwise, a new cluster is created, and this process is repeated until all points are assigned. Arithmetic mean fusion involves collecting the coordinates of all candidate points within each cluster, calculating their arithmetic mean, and using this as the fusion coordinates for that cluster to generate a fusion point set. In this way, each cluster outputs a point, representing the best estimate of the ground location.

[0060] Because the geometric distortion properties of images differ under different terrain conditions, for example, in areas with significant terrain undulations and strong local distortions, the local nonlinear deformation is strong and cannot be accurately described by a single global model. Local models such as irregular triangular mesh interpolation are needed to adapt. In areas with gentle changes in ground features and a need to balance global consistency and local accuracy, the overall changes are gentle, but there are some local curvatures. Thin plate spline interpolation models are more suitable for balancing global consistency and local accuracy. In areas with flat terrain and no obvious undulations, the distortion is mainly global translation, rotation, and scaling, which can be well described by a global polynomial model.

[0061] Therefore, automatically selecting an appropriate transformation model based on the scene type is to ensure that the model complexity matches the degree of terrain distortion, avoiding underfitting due to an overly simple model or overfitting noise due to an overly complex model. Specifically, it automatically selects the optimal matching transformation model based on the current image scene type, solves for the model parameters using the fusion point set, performs iterative matching, and continuously optimizes the matching accuracy until convergence. The transformation model selection method can be based on scene classification as described above, or it can be based on DEM data to calculate the slope of each point in the fusion point set. The number or proportion of control points within each slope interval is then statistically analyzed; if a certain type of area is dominant, the corresponding model is selected. Alternatively, a global multinomial model can be fitted, and the spatial distribution of the residuals of each control point can be analyzed to select the corresponding model based on the residuals.

[0062] Furthermore, the iterative matching process can involve fitting a selected geometric transformation model to the current fused point set; performing a forward transformation on the reference image control points using this model to obtain predicted coordinates; calculating the residual for each point and discarding points with residuals exceeding limits; refitting the model with the updated point set; and repeating this process until the model converges. Convergence conditions can include, for example, an average residual change of <0.01 pixels or reaching the maximum number of iterations. This results in a high-precision, highly robust geometric transformation model between the reference image and the image to be registered.

[0063] Step S110: Obtain the matched image based on the transformation model and perform a qualitative evaluation. After the qualitative evaluation is passed, calculate the quantitative evaluation by combining the independent verification point set and output the matching accuracy result.

[0064] Specifically, based on the converged transformation model, the image obtained after correcting and resampling the image to be registered, and spatially aligned with the reference image, is the matched image. This step assesses the quality of the transformation model and the matched image, including qualitative and quantitative evaluations. The qualitative evaluation aims to quickly identify obvious matching errors through visual inspection, including:

[0065] Image overlay: The matched image is divided into regions, and a sufficient number of typical sub-regions are selected within each region. For example, the entire image is divided into a 3×3 grid. One sub-region is selected in the center, one at each of the four corners (4 sub-regions), and one near the midpoint of each of the four sides (4 sub-regions). This covers the center, corners, and edges of the image, avoiding checking only the center while ignoring edge distortion. For each sub-region, the size is 3km×3km, serving as a local window for visual inspection. At least 6 sub-regions are selected per region to ensure sufficient sampling and avoid randomness. Typical sub-regions can prioritize clearly defined features such as roads, embankments, riverbanks, and building edges.

[0066] Roll-up comparison: Overlay the reference image and the matched image in the same window, set a dividing line, display the reference image on one side of the dividing line and the matched image on the other side; check whether the corresponding features on both sides of the dividing line are continuous and aligned. If they are aligned, the image is properly fitted; otherwise, there is a deviation.

[0067] Zonal sampling verification: If the visual matching error of all sub-regions is less than 1 pixel, the region is considered qualified; only if all 9 regions are qualified can the quantitative evaluation proceed. That is, if even one region fails the visual verification, the matching result is considered to have a local problem and is directly rejected, not proceeding to the quantitative stage. This rapid qualitative initial screening verifies the matching results.

[0068] Quantitative evaluation can calculate geometric error indices using independent verification points, and set different thresholds according to terrain slope and scene texture type to ultimately determine whether the entire scene is acceptable, including:

[0069] Construct a validation point set. The validation point set is used to validate the model and cannot be the same as the control point set used to build the transformation model. Therefore, the same feature extraction algorithm as the control point set in step S104 is used, but the parameters are adjusted (e.g., changing the FAST response threshold, block size, etc.) so that the extracted feature points are different from the control points, forming an independent test set.

[0070] Given the wide coverage, significant terrain differences, and uneven texture of remote sensing images, a single global indicator is prone to misjudgment. Therefore, a three-level evaluation mechanism of terrain zoning, scene recognition, and differentiated thresholds is adopted.

[0071] Slope Classification: The slope is then calculated using a DEM (Digital Elevation Model), and the verification points are divided into three evaluation sub-regions based on terrain undulation: flat areas, gentle slope areas, and steep slope areas. Specific slope thresholds can be set as follows: flat areas < 5°, gentle slope areas 5°~15°, and steep slope areas > 15°. If the number of verification points in a certain sub-region is less than 10, the sample size is insufficient and the points are not included in the final judgment.

[0072] Scene feature classification: Based on terrain zoning, additional scene attributes are added to divide each sub-region into strongly textured scenes (urban building areas, road networks, farmland interspersed areas) and weakly textured scenes (desert, Gobi, water areas, snowfields). Texture richness is an important factor affecting matching accuracy and is independent of terrain slope. Specific classification methods can be derived from land cover classification data or through manual visual interpretation.

[0073] Based on the terrain-scene classification described above, six sub-region types can be obtained. Different sub-regions are assigned appropriate accuracy standards.

[0074] Calculate the three-times standard deviation (3σ) independently for the global region and each effective subregion:

[0075] ,

[0076] in, This represents the number of verification points. The SPPR (Single-Point Composite Residue) reflects the location error of each validation point. The standard deviation σ measures the dispersion of the SPPR. A small standard deviation indicates concentrated and uniform error; a large standard deviation indicates the presence of some outliers with significant errors. 3σ indicates that almost all (99.7%) samples are within the mean ± 3σ range. In accuracy evaluation, 3σ reflects the upper limit of error, which is more stringent than the mean error and can capture large local errors. Perform partition threshold determination:

[0077] Strict standard (3σ≤1 pixel): Applicable to flat areas and scenes with strong texture features. These areas have gentle terrain and rich features, achieving extremely high matching accuracy at the sub-pixel level;

[0078] The tolerance standard (3σ≤2 pixels) is applicable to both gentle and steep slopes, as well as scenes with weak texture features at any slope. This ensures that the accuracy standard matches the matching difficulty, avoiding complete rejection due to objective limitations, while guaranteeing high accuracy requirements in flat, strongly textured areas.

[0079] Final qualification principle (weakest link determination principle):

[0080] The entire scene is considered qualified when all valid partitions (number of points ≥ 10) have 3σ values ​​that meet their corresponding qualification thresholds.

[0081] If the 3σ of any valid partition exceeds the limit, it is deemed unqualified and returned to the process of refining the corresponding points and subsequent steps.

[0082] Once all valid partitions are deemed acceptable, multiple metrics are further calculated to comprehensively quantify the matching accuracy, resulting in the final accuracy report:

[0083] Root mean square error: assesses the overall degree of deviation between the predicted point and the actual location;

[0084] ,

[0085] Mean error: Calculate the arithmetic mean of the residuals in the X and Y directions to determine whether there is a uniform systematic deviation. If the mean error is significantly non-zero, it indicates that there is a systematic shift (such as overall translation or rotation), and it is necessary to check whether there is a systematic bias in the model.

[0086] ,

[0087] ,

[0088] Maximum error: The maximum absolute value of the residuals in the statistical validation set, used to evaluate the worst local case of the match.

[0089] ,

[0090] ,

[0091] Structural similarity: Based on the shared structural and contour features of heterogeneous images, the structural overlap within the neighborhood of corresponding pixels is calculated by extracting high-frequency edge maps from the matched images for further evaluation. Higher similarity indicates a more precise alignment of contours of features such as rivers, roads, and building edges within the images. The similarity calculation is as follows:

[0092] ,

[0093] in, and These are the reference and matched binary images obtained from edge detection, respectively. This is a convolution operation; For a neighborhood window, an odd number of windows, such as [3-9] × [3-9], can be used. This indicates the number of pixels in the neighborhood that simultaneously have an edge. This indicates the number of pixels in the neighborhood where at least one image has an edge. ; A value ≥0.85 indicates good contour overlay and high matching accuracy; conversely, a value lower indicates poor performance. ESS evaluates matching accuracy at the structural contour level, compensating for the limitation of point error indices in reflecting the overlay of linear features. For heterogeneous imagery, although there are large grayscale differences, the edge structure is relatively stable, making ESS well-suited for these applications.

[0094] In this step, the flowchart for multidimensional accuracy evaluation is as follows: Figure 2 As shown, the multidimensional accuracy evaluation includes two parts: qualitative evaluation and quantitative evaluation, ultimately outputting a multidimensional accuracy evaluation report. The qualitative evaluation process is as follows: Paired image data undergoes manual roll-up comparison verification, assessing the degree of match in terms of spatial distribution, land cover type, and feature points. If the match is good, random sampling is performed within each zone to determine the degree of match within that zone. If the match within each zone is good, the verification is passed, and the process proceeds to quantitative evaluation. If the match or the match within each zone is poor, re-registration is required. The quantitative evaluation process is as follows: It is determined whether new points are generated independently; if so, an independent validation set is used; otherwise, a cross-validation set is used. Core evaluation indicators are calculated; terrain and scene partitioning is performed to generate effective evaluation sub-regions; it is determined whether the core indicators for each zone meet the standards; if they do, the indicators are output; if they do not, re-registration is required. This step involves not only qualitative and quantitative evaluations, but also the output of accuracy report indicators such as root mean square error, mean error, maximum error, and structural similarity. If there are any non-compliance issues, the process can return to step S104 or S106 for optimization. For example, if the quality or quantity of control points is insufficient, the process can return to step S104 for supplementation or re-screening. If the model selection or matching accuracy is insufficient, the process can return to step S108 to adjust parameters or the model.

[0095] In this embodiment, a method for high-precision matching and multi-dimensional evaluation of heterogeneous remote sensing satellite imagery across all scenes includes: acquiring images to be registered and reference images of the target area; establishing a heterogeneous remote sensing image dataset through spatiotemporal pairing and quality control; performing multi-level screening on the heterogeneous remote sensing image dataset to extract the optimal control point set with the same name; pre-setting parameter combinations; searching for candidate points with the same name from the optimal control point set based on the parameter combinations; performing arithmetic mean fusion after threshold screening to generate a fused point set; acquiring a transformation model; solving the transformation model using the fused point set; performing iterative matching until convergence; acquiring the matched image based on the transformation model and performing qualitative evaluation; after passing the qualitative evaluation, calculating a quantitative evaluation in conjunction with an independent verification point set, and outputting the matching accuracy result. This embodiment, through heterogeneous remote sensing satellite data acquisition and quality control, adaptive extraction of the optimal control point set with the same name, scene-specific iterative fine matching, and terrain-scene collaborative multi-dimensional accuracy evaluation, has the following advantages as each step is progressively advanced:

[0096] This embodiment possesses strong cross-modal compatibility and scene adaptability, and can uniformly process remote sensing images from different sources such as visible light, infrared, and SAR with vastly different imaging mechanisms, without requiring separate adjustments to the core strategy for different sensors. At the same time, it can operate stably in all types of surface scenes, including plains and towns, Gobi deserts, high mountains and hills, glaciers, and water bodies, automatically overcoming multiple adverse factors such as radiation distortion, noise interference, terrain undulations, and weak textures, and continuously outputting high-precision and highly consistent matching results, demonstrating high robustness and stability.

[0097] This embodiment establishes a comprehensive control point extraction and optimization mechanism to improve matching quality in all aspects. In the gross error removal stage, significantly mismatched points with obvious deviations are first removed using the RANSAC global model. Then, DEM terrain constraints are combined to achieve slope zoning and layered threshold determination, effectively distinguishing between real terrain offset and matching noise, preserving the true geometric information of the land surface to the maximum extent while removing gross errors. In the control point refinement stage, multiple sets of progressive parameter combinations from coarse to fine are used. Through multiple batches of extraction and cluster mean fusion, a highly robust point set is formed, significantly offsetting random errors in single matching. For areas with scarce features such as mountains and hills, control points are supplemented using an interactive supplementation method based on line feature constraints, utilizing stable line features such as ridges and valleys to compensate for the shortcomings of undulating terrain features, greatly reducing local registration errors and achieving high-precision uniform matching across the entire region.

[0098] This embodiment establishes a hierarchical, multi-dimensional accuracy evaluation and quality control system to solve the problems of traditional self-inspection and self-evaluation, inflated accuracy, and distorted evaluation. In terms of verification strategy, it prioritizes independent verification point sets to avoid overfitting caused by model self-validation; in areas with complex terrain or scarce textures where independent samples are difficult to obtain, it automatically switches to cross-validation to ensure the validity of evaluation samples; in terms of evaluation dimensions, it integrates multi-level indicators such as geometric error, structural consistency, zonal statistics, and visual verification to achieve quantitative and qualitative complementarity and a balance between global and local considerations; the evaluation results have a strict closed-loop mechanism, with unqualified data automatically substituted back into the refinement and iterative registration stages for reprocessing to ensure that the final matching results are authentic, reliable, and usable, meeting the quality requirements of high-precision remote sensing applications.

[0099] Please refer to the following. Figure 3 , Figure 3 This diagram illustrates the structure of a high-precision matching and multi-dimensional evaluation system for all scenes of heterogeneous remote sensing satellite imagery, as provided in an embodiment of this specification. It should be noted that... Figure 3 The heterogeneous remote sensing satellite imagery full-scene high-precision matching and multi-dimensional evaluation system shown is used to execute this specification. Figure 1 The methods shown in the embodiments are illustrated for ease of explanation, showing only the parts related to the embodiments of this specification. For specific technical details not disclosed, please refer to this specification. Figure 1 The example shown.

[0100] like Figure 3 As shown, this heterogeneous remote sensing satellite imagery full-scene high-precision matching and multi-dimensional evaluation system can include at least:

[0101] The acquisition module S302 is used to acquire the image to be registered and the reference image of the target area, and to establish a heterogeneous remote sensing image dataset through spatiotemporal pairing and quality control.

[0102] The filtering module S304 is used to perform multi-level filtering on the heterogeneous remote sensing image dataset and extract the optimal control point set with the same name.

[0103] The fusion module S306 is used to preset parameter combinations, search for candidate points with the same name from the optimal control point set based on the parameter combinations, perform arithmetic mean fusion after threshold screening, and generate a fused point set.

[0104] Transformation module S308 is used to obtain a transformation model, solve the transformation model using the fusion point set, and perform iterative matching until convergence.

[0105] The evaluation module S310 is used to acquire the matched image based on the transformation model and perform a qualitative evaluation. After the qualitative evaluation is passed, it calculates the quantitative evaluation by combining the independent verification point set and outputs the matching accuracy result.

[0106] Those skilled in the art will clearly understand that the technical solutions of the embodiments in this specification can be implemented by means of software and / or hardware. In this specification, "unit" and "module" refer to software and / or hardware capable of independently performing or cooperating with other components to perform a specific function. The hardware may be, for example, a Field-Programmable Gate Array (FPGA), an Integrated Circuit (IC), etc.

[0107] Each processing unit and / or module in the embodiments of this specification can be implemented by an analog circuit that implements the functions described in the embodiments of this specification, or by software that executes the functions described in the embodiments of this specification.

[0108] See Figure 4 It shows a schematic diagram of the structure of an electronic device according to an embodiment of this specification, which can be used to implement... Figure 1 The method in the illustrated embodiment. (As shown) Figure 4 As shown, the electronic device 400 may include: at least one central processing unit 401, at least one network interface 404, user interface 403, memory 405, and at least one communication bus 402.

[0109] The communication bus 402 is used to enable communication between these components.

[0110] The user interface 403 may include a display screen and a camera. Optionally, the user interface 403 may also include a standard wired interface and a wireless interface.

[0111] The network interface 404 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface).

[0112] The processor 401 may include one or more processing cores. The processor 401 connects to various parts within the electronic device 400 using various interfaces and lines, and performs various functions and processes data by running or executing instructions, programs, code sets, or instruction sets stored in the memory 405, and by calling data stored in the memory 405. Optionally, the processor 401 may be implemented using at least one hardware form of Digital Signal Processing (DSP), Field-Programmable Gate Array (FPGA), or Programmable Logic Array (PLA). The processor 401 may integrate one or a combination of several of the following: a Central Processing Unit (CPU), a Graphics Processing Unit (GPU), and a modem. The CPU primarily handles the operating system, user interface, and applications; the GPU is responsible for rendering and drawing the content required for display; and the modem handles wireless communication. It is understood that the modem may also be implemented as a separate chip, without being integrated into the processor 401.

[0113] The memory 405 may include random access memory (RAM) or read-only memory. Optionally, the memory 405 may include a non-transitory computer-readable storage medium. The memory 405 may be used to store instructions, programs, code, code sets, or instruction sets. The memory 405 may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for at least one function (such as touch function, sound playback function, image playback function, etc.), instructions for implementing the above-described method embodiments, etc.; the data storage area may store data involved in the above-described method embodiments, etc. Optionally, the memory 405 may also be at least one storage system located remotely from the aforementioned processor 401. Figure 4 As shown, the memory 405, which serves as a computer storage medium, may include an operating system, a network communication module, a user interface module, and program instructions.

[0114] exist Figure 4In the illustrated electronic device 400, the user interface 403 is mainly used to provide an input interface for the user and to acquire user input data; while the processor 401 can be used to call the image-based interactive application stored in the memory 405 and specifically perform the following operations:

[0115] Acquire the images to be registered and reference images of the target area, and establish a heterogeneous remote sensing image dataset through spatiotemporal pairing and quality control;

[0116] The heterogeneous remote sensing image dataset is subjected to multi-level filtering to extract the optimal set of control points with the same name;

[0117] A preset parameter combination is used to search for candidate points with the same name from the optimal control point set. After threshold screening, the points are fused by arithmetic average to generate a fused point set.

[0118] Obtain the transformation model, solve the transformation model using the fusion point set, and perform iterative matching until convergence;

[0119] The matched images are obtained based on the transformation model and a qualitative evaluation is performed. After the qualitative evaluation is passed, a quantitative evaluation is calculated by combining the independent verification point set, and the matching accuracy result is output.

[0120] As an optional embodiment of this specification, the multi-level screening of the heterogeneous remote sensing image dataset to extract the optimal set of control points with the same name includes initial screening of control points with the same name, refinement of control points with the same name, and supplementation of control points. The initial screening of control points with the same name includes:

[0121] Based on the image size of heterogeneous remote sensing images, the images are divided into blocks, and initial feature points are selected within each block. Through cross-image matching, an initial set of corresponding points is formed.

[0122] Based on the sample subset pairs of the initial set of points with the same name, the initial matching model is solved using the sample subsets.

[0123] The initial feature points are transformed in a positive direction using the initial matching model to obtain the predicted coordinates. The Euclidean distance between the predicted coordinates and the corresponding measured matching coordinates on the image to be registered is calculated, and the corresponding points are initially screened based on the Euclidean distance.

[0124] The refinement of the corresponding points includes:

[0125] Calculate the comprehensive residual of the set of corresponding points after initial screening, compare it with the preset threshold, and refine the corresponding points based on the comparison results.

[0126] As an optional embodiment of this specification, the step of calculating the comprehensive residual of the set of corresponding points after initial screening, comparing it with a preset threshold, and refining the corresponding points based on the comparison results includes...

[0127] The single-point comprehensive residual is calculated based on the x-coordinate residual and y-coordinate residual of each pair of corresponding points in the set of corresponding points after the initial screening.

[0128] Obtain strict error thresholds and lenient error thresholds;

[0129] The location terrain attributes of the corresponding points of the single-point comprehensive residual are detected. If the location terrain belongs to a flat area, the single-point comprehensive residual is compared with the strict error threshold. If the single-point comprehensive residual is greater than the strict error threshold, the corresponding points are directly removed.

[0130] If the terrain of the point is in a region with drastic undulations, then the single-point comprehensive residual is compared with the leniency error threshold. If the single-point comprehensive residual is greater than the leniency error threshold, the corresponding point is directly removed.

[0131] As an optional embodiment of this specification, the preset parameter combination includes control point count, search window, and template window;

[0132] The preset parameter combination, based on which candidate corresponding points are searched from the optimal control point set, includes:

[0133] Take out pairs of corresponding points from the set of optimal control points, and crop a reference template of the size of the template window from the reference image. The number of control points controls the number of pairs of corresponding points. During the multi-level correction process, the number of control points gradually increases and the search window and template window gradually decrease.

[0134] Using the initial predicted position on the image to be registered as the center, slide within the search window range, and capture candidate image blocks each time you slide;

[0135] By combining the reference template and the candidate image blocks, a similarity measure between them is calculated. Based on the similarity measure, the optimal position on the image to be registered is selected, and candidate corresponding points are generated by combining the initial coordinates of the reference template.

[0136] As an optional embodiment of this specification, the step of obtaining the transformation model, solving the transformation model using the fusion point set, and performing iterative matching until convergence includes:

[0137] Based on the scene type of the target area, a transformation model is selected, and the transformation model is fitted using the fusion point set to obtain the residuals before and after fitting. Points in the fusion point set whose residuals exceed the threshold are removed.

[0138] Refit the model with the updated fusion point set, repeating until the model converges.

[0139] As an optional embodiment of this specification, the qualitative evaluation process includes:

[0140] The reference image and the matched image are overlaid and displayed, a roll-up comparison is performed, and a zonal sampling verification is conducted. When all zonal sampling verifications pass, quantitative evaluation is initiated.

[0141] As an optional embodiment of this specification, the quantitative evaluation process includes:

[0142] Construct a set of verification points;

[0143] Based on the slope and scene attributes of each verification point, the verification points are divided into zones and categories, and the appropriate accuracy standard is determined based on the slope classification and scene feature classification.

[0144] Calculate three times the standard deviation of the entire image and each zone, compare the three times the standard deviation with the corresponding accuracy standard, and determine the quantitative evaluation result based on the comparison result.

[0145] This specification also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described method. The computer-readable storage medium may include, but is not limited to, any type of disk, including floppy disks, optical disks, DVDs, CD-ROMs, microdrives, as well as magneto-optical disks, ROMs, RAMs, EPROMs, EEPROMs, DRAMs, VRAMs, flash memory devices, magnetic cards or optical cards, nanosystems (including molecular memory ICs), or any type of medium or device suitable for storing instructions and / or data.

[0146] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this specification is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this specification. Furthermore, those skilled in the art should also understand that the embodiments described in this specification are preferred embodiments, and the actions and modules involved are not necessarily essential to this specification.

[0147] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0148] In the embodiments provided in this specification, it should be understood that the disclosed system can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some service interface; the indirect coupling or communication connection between systems or units may be electrical or other forms.

[0149] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0150] Furthermore, the functional units in the various embodiments of this specification can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0151] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage device (CMD). Based on this understanding, the technical solution of this specification, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a memory and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this specification. The aforementioned memory includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0152] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, which may include: a flash drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, etc.

[0153] The foregoing has described specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

Claims

1. A method for high-precision matching and multi-dimensional evaluation of heterogeneous remote sensing satellite imagery across all scenes, characterized in that, The method includes: Acquire the images to be registered and reference images of the target area, and establish a heterogeneous remote sensing image dataset through spatiotemporal pairing and quality control; The heterogeneous remote sensing image dataset is subjected to multi-level filtering to extract the optimal set of control points with the same name; A preset parameter combination is used to search for candidate points with the same name from the optimal control point set. After threshold screening, the points are fused by arithmetic average to generate a fused point set. Obtain the transformation model, solve the transformation model using the fusion point set, and perform iterative matching until convergence; The matched images are obtained based on the transformation model and a qualitative evaluation is performed. After the qualitative evaluation is passed, a quantitative evaluation is calculated by combining the independent verification point set, and the matching accuracy result is output.

2. The method for high-precision matching and multi-dimensional evaluation of heterogeneous remote sensing satellite imagery across all scenes according to claim 1, characterized in that, The process of performing multi-level screening on the heterogeneous remote sensing image dataset to extract the optimal set of control points with the same name includes initial screening of control points with the same name, refinement of control points with the same name, and supplementation of control points. The initial screening of control points with the same name includes: Based on the image size of heterogeneous remote sensing images, the images are divided into blocks, and initial feature points are selected within each block. Through cross-image matching, an initial set of corresponding points is formed. Based on the sample subset pairs of the initial set of points with the same name, the initial matching model is solved using the sample subsets. The initial feature points are transformed in a positive direction using the initial matching model to obtain the predicted coordinates. The Euclidean distance between the predicted coordinates and the corresponding measured matching coordinates on the image to be registered is calculated, and the corresponding points are initially screened based on the Euclidean distance. The refinement of the corresponding points includes: Calculate the comprehensive residual of the set of corresponding points after initial screening, compare it with the preset threshold, and refine the corresponding points based on the comparison results.

3. The method for high-precision matching and multi-dimensional evaluation of heterogeneous remote sensing satellite imagery across all scenes according to claim 2, characterized in that, The calculation of the comprehensive residual of the set of corresponding points after initial screening, and comparison with a preset threshold, followed by refinement of the corresponding points based on the comparison results, includes: The single-point comprehensive residual is calculated based on the x-coordinate residual and y-coordinate residual of each pair of corresponding points in the set of corresponding points after the initial screening. Obtain strict error thresholds and lenient error thresholds; The location terrain attributes of the corresponding points of the single-point comprehensive residual are detected. If the location terrain belongs to a flat area, the single-point comprehensive residual is compared with the strict error threshold. If the single-point comprehensive residual is greater than the strict error threshold, the corresponding points are directly removed. If the terrain of the point is in a region with drastic undulations, then the single-point comprehensive residual is compared with the leniency error threshold. If the single-point comprehensive residual is greater than the leniency error threshold, the corresponding point is directly removed.

4. The method for high-precision matching and multi-dimensional evaluation of heterogeneous remote sensing satellite imagery across all scenes according to claim 1, characterized in that, The preset parameter combination includes the number of control points, the search window, and the template window; The preset parameter combination, based on which candidate corresponding points are searched from the optimal control point set, includes: Take out pairs of corresponding points from the set of optimal control points, and crop a reference template of the size of the template window from the reference image. The number of control points controls the number of pairs of corresponding points. During the multi-level correction process, the number of control points gradually increases and the search window and template window gradually decrease. Using the initial predicted position on the image to be registered as the center, slide within the search window range, and capture candidate image blocks each time you slide; By combining the reference template and the candidate image blocks, a similarity measure between them is calculated. Based on the similarity measure, the optimal position on the image to be registered is selected, and candidate corresponding points are generated by combining the initial coordinates of the reference template.

5. The method for high-precision matching and multi-dimensional evaluation of heterogeneous remote sensing satellite imagery across all scenes according to claim 1, characterized in that, The process of obtaining the transformation model, solving the transformation model using the fusion point set, and performing iterative matching until convergence includes: Based on the scene type of the target area, a transformation model is selected, and the transformation model is fitted using the fusion point set to obtain the residuals before and after fitting. Points in the fusion point set whose residuals exceed the threshold are removed. Refit the model with the updated fusion point set, repeating until the model converges.

6. The method for high-precision matching and multi-dimensional evaluation of heterogeneous remote sensing satellite imagery across all scenes according to claim 1, characterized in that, The qualitative evaluation process includes: The reference image and the matched image are overlaid and displayed, a roll-up comparison is performed, and a zonal sampling verification is conducted. When all zonal sampling verifications pass, quantitative evaluation is initiated.

7. The method for high-precision matching and multi-dimensional evaluation of heterogeneous remote sensing satellite imagery across all scenes according to claim 6, characterized in that, The quantitative evaluation process includes: Construct a set of verification points; Based on the slope and scene attributes of each verification point, the verification points are divided into zones and categories, and the appropriate accuracy standard is determined based on the slope classification and scene feature classification. Calculate three times the standard deviation of the entire image and each zone, compare the three times the standard deviation with the corresponding accuracy standard, and determine the quantitative evaluation result based on the comparison result.

8. A high-precision matching and multi-dimensional evaluation system for all scenes of heterogeneous remote sensing satellite imagery, characterized in that, The system includes: The acquisition module is used to acquire the images to be registered and the reference images of the target area, and to establish a heterogeneous remote sensing image dataset through spatiotemporal pairing and quality control. The filtering module is used to perform multi-level filtering on the heterogeneous remote sensing image dataset and extract the optimal control point set with the same name; The fusion module is used to preset parameter combinations, search for candidate points with the same name from the optimal control point set based on the parameter combinations, perform arithmetic mean fusion after threshold screening, and generate a fused point set. The transformation module is used to obtain the transformation model, solve the transformation model using the fusion point set, and perform iterative matching until convergence. The evaluation module is used to obtain the matched image based on the transformation model and perform a qualitative evaluation. After the qualitative evaluation is passed, a quantitative evaluation is calculated by combining the independent verification point set, and the matching accuracy result is output.

9. An electronic device, comprising a processor and a memory; The processor is connected to the memory; The memory is used to store executable program code; The processor runs a program corresponding to the executable program code stored in the memory to perform the method as described in any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method as described in any one of claims 1-7.

Images