A method and apparatus for correcting remote sensing images based on big data statistical camera distortion

Abstract

This application provides a method and apparatus for correcting remote sensing images based on statistical camera distortion using big data. Its purpose is to eliminate reliance on ground calibration fields and, through control point matching and model fitting, utilize an image distortion correction model to correct the remote sensing image to be corrected, ensuring the accuracy of the remote sensing image's location. The method for correcting remote sensing images based on statistical camera distortion using big data includes: acquiring historical remote sensing images and existing digital orthophoto products of the target area, and the remote sensing image to be corrected; matching feature points between the historical remote sensing images and the digital orthophotos to obtain control points; calculating the image-square residuals of the control points using affine transformation formulas; identifying points with image-square residuals less than a preset threshold as sample points; establishing an image distortion correction model using the image-square residuals and image-square coordinates of the sample points; inputting the image-square coordinates of the remote sensing image to be corrected into the image distortion correction model to obtain the corresponding image-square correction value; and generating the target remote sensing image based on the image-square correction value.

Description

Technical Field

[0001] This application relates to the field of remote sensing image technology, and in particular to a method and apparatus for correcting remote sensing image distortion based on big data statistical camera distortion. Background Technology

[0002] With the continuous advancement of optical remote sensing satellite technology, optical remote sensing imagery, with its high spatial resolution, is increasingly widely used in fields such as land surveys and disaster monitoring. During image acquisition, factors such as temperature changes and attitude jitter during satellite operation can cause distortion in the charge-coupled device (CCD) detection units of the satellite's optical camera, resulting in offsets of ground features in the image. Therefore, it is necessary to correct the images acquired by the satellite's optical camera to ensure the accuracy of image positioning.

[0003] In existing technologies, image geometric inconsistencies caused by azimuth element distortion within satellite optical cameras are typically corrected using a ground-based calibration field method. This method requires the pre-deployment of a dedicated ground calibration field. Using the high-resolution digital orthophoto and digital elevation model of the calibration field as reference data, control points are obtained by matching the reference data with the image. The control point data and ground calibration field parameter data are then used to construct an imaging geometric model to achieve correction, thereby eliminating image distortion.

[0004] However, this method requires the prior selection and construction of a dedicated ground calibration field. In areas such as oceans, plateaus, and deserts, geographical limitations prevent the establishment of ground calibration fields, thus restricting the calibration range. Therefore, there is an urgent need for a remote sensing image correction method that does not rely on a ground calibration field. Summary of the Invention

[0005] This application provides a remote sensing image correction method and apparatus based on big data statistical camera distortion. Its purpose is to eliminate the dependence on ground calibration fields, and at the same time, through precise matching of control points and model fitting, use the image distortion correction model to correct remote sensing images and ensure the accuracy of the location of ground features in remote sensing images.

[0006] To address the aforementioned technical problems, the embodiments of this application provide the following technical solutions: The first aspect of this application provides a remote sensing image correction method based on big data statistical camera distortion, including: Acquire historical remote sensing images and existing digital orthophoto products of the target area, as well as remote sensing images to be corrected; Control points are obtained by matching feature points in digital orthophotos obtained from historical remote sensing images and existing digital orthophoto products of the target area. The control points have image-side coordinates and object-side coordinates. Based on the image-side coordinates and object-side coordinates, the image-side residual of the control points is calculated using the affine transformation formula, and points whose image-side residuals are less than a preset threshold are determined as sample points. Using the image-square residuals and image-square coordinates of the sample points, B-spline fitting is performed to obtain an image distortion correction model. The image distortion correction model is used to determine the image-square correction value corresponding to the image-square coordinates in the remote sensing image. After inputting the image-side coordinates of the remote sensing image to be corrected into the image distortion correction model, the image-side correction value corresponding to the image-side coordinates of the remote sensing image to be corrected is obtained, and the target remote sensing image is generated based on the image-side correction value.

[0007] A second aspect of this application provides a remote sensing image correction device based on big data statistical camera distortion, comprising: The acquisition unit is used to acquire historical remote sensing images and existing digital orthophoto products of the target area, as well as remote sensing images to be corrected. The determination unit is used to match feature points of the digital orthophoto obtained by acquiring historical remote sensing images of the target area in the acquisition unit and existing digital orthophoto products to obtain control points. The control points have image-side coordinates and object-side coordinates. The determining unit is used to calculate the image-side residual of the control points based on the image-side coordinates and the object-side coordinates using the affine transformation formula, and to determine the points whose image-side residuals are less than a preset threshold as sample points. The fitting unit is used to perform B-spline fitting using the image-square residuals and image-square coordinates of the sample points in the determining unit to obtain an image distortion correction model. The image distortion correction model is used to determine the image-square correction value corresponding to the image-square coordinates in the remote sensing image. The generation unit is used to obtain the image-square correction value corresponding to the image-square coordinates of the remote sensing image to be corrected by inputting the image-square coordinates of the remote sensing image to be corrected in the acquisition unit into the image distortion correction model in the fitting unit, and to generate the target remote sensing image based on the image-square correction value.

[0008] A third aspect of this application provides a storage medium comprising a stored program, wherein, when the program is executed, it controls the device containing the storage medium to perform the aforementioned image correction method for satellite camera distortion based on big data statistics.

[0009] A fourth aspect of this application provides an electronic device, the device including at least one processor, at least one memory and a bus connected to the processor; wherein the processor and the memory communicate with each other through the bus; the processor is used to call program instructions in the memory to execute the above-described image correction method for satellite camera distortion based on big data statistics.

[0010] Compared to existing technologies, this application presents a remote sensing image correction method based on big data statistical camera distortion. Firstly, this method fundamentally eliminates the reliance on fixed ground calibration fields by matching control points using historical remote sensing images and digital orthophotos of the target area. This allows correction to be applied to satellite images acquired in any geographical region, such as oceans, plateaus, and deserts, greatly expanding the method's versatility. Secondly, the method determines sample points through the image-space residuals of control points. After B-spline fitting based on the image-space residuals and image-space coordinates of the sample points, the resulting image distortion correction model accurately determines the distortion of the remote sensing image, ultimately generating the target remote sensing image. Therefore, the above-mentioned technical method achieves a dual breakthrough in cost and applicability while ensuring the accuracy of the corrected remote sensing image. Attached Figure Description

[0011] The above and other objects, features, and advantages of exemplary embodiments of this application will become readily understood by reading the following detailed description with reference to the accompanying drawings. In the drawings, several embodiments of this application are illustrated by way of example and not limitation, with the same or corresponding reference numerals denoteing the same or corresponding parts, wherein: Figure 1 A flowchart illustrating a remote sensing image correction method based on big data statistical camera distortion is shown. Figure 2 A flowchart illustrating another remote sensing image correction method based on big data statistical camera distortion is shown schematically. Figure 3 A schematic diagram of a remote sensing image correction device based on big data statistical camera distortion is shown. Figure 4 The diagram schematically illustrates the structure of another remote sensing image correction device based on big data statistical camera distortion. Detailed Implementation

[0012] Exemplary embodiments of this application will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of this application are shown in the drawings, it should be understood that this application may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of this application and to fully convey the scope of this application to those skilled in the art.

[0013] It should be noted that, unless otherwise stated, the technical or scientific terms used in this application shall have the ordinary meaning as understood by one of ordinary skill in the art to which this application pertains.

[0014] With the continuous advancement of optical remote sensing satellite technology, optical remote sensing imagery, with its high spatial resolution, is increasingly widely used in fields such as land surveys and disaster monitoring. During image acquisition, satellites are affected by factors such as temperature changes and attitude jitter during on-orbit operation, leading to distortion in the charge-coupled device (CCD) detection units of the satellite's optical camera, which in turn causes offsets in the image's ground feature positions. Therefore, it is necessary to correct the images acquired by the satellite's optical camera to ensure the accuracy of image positions. In existing technologies, image correction for geometric inconsistencies caused by azimuth element distortion within the satellite's optical camera is typically performed using a ground calibration field-based method. This method requires the pre-deployment of a dedicated ground calibration field. Using the high-resolution digital orthophoto and digital elevation model (DEM) of the calibration field as reference data, control points are obtained by matching the reference data with the remote sensing image. The control point data and ground calibration field parameter data are then used to construct an imaging geometric model to achieve correction, thereby eliminating remote sensing image distortion.

[0015] To address this issue, the applicant proposes a remote sensing image correction method that does not rely on a ground calibration field. This method utilizes historical remote sensing and digital orthophotos of the target area, obtains control points through matching, calculates image-side residuals, and performs B-spline fitting to obtain an image distortion correction model. This model is then used to correct the remote sensing image to be corrected, thus solving the problem of limited correction in special areas. Specifically, as follows... Figure 1 As shown: Step 101: Acquire historical remote sensing images and existing digital orthophoto products of the target area, as well as remote sensing images to be corrected.

[0016] In this step, historical remote sensing imagery refers to high-resolution black-and-white images and multispectral images acquired by satellite optical cameras in the panchromatic band. Historical digital orthophoto maps (DOMs) are image maps that have undergone geometric correction (orthorectification) and been assigned geographic coordinates, eliminating displacement caused by terrain undulations and sensor distortion; that is, they are distortion-free images. The target area refers to the geographic region defined for camera distortion analysis and correction, typically composed of multiple regular sub-regions. The remote sensing image to be corrected is the current image captured by the satellite optical camera and is the core processing object of this correction process. The target area can consist of multiple regions or a single region.

[0017] Specifically, historical remote sensing images are acquired through satellite data sharing platforms, requiring a spatial resolution of at least GB. DOMThe cloud cover should be ≤5%, and imaging should be concentrated in the same season to reduce the impact of illumination differences. A complete data package containing image RPC parameters should be collected. The target remote sensing image must come from the same satellite sensor as historical images to ensure consistency in imaging geometry. After acquisition, radiometric calibration preprocessing is performed to remove residual atmospheric correction errors and ensure data quality meets subsequent matching requirements. In this step, when acquiring remote sensing images and digital orthophotos of the target area, the minimum resolution of the digital orthophoto is determined by a preset formula, where the preset formula is: G DOM *k≤G L1 Among them, G DOM G is the lowest resolution of the digital orthophoto, k is the matching algorithm precision, and G is the lowest resolution of the digital orthophoto. L1 Let be the resolution of the remote sensing image. The above formula ensures high-precision control points even with low-resolution digital orthophotos, significantly reducing data costs and reliance on specific high-resolution calibration fields.

[0018] Step 102: Match the feature points of the digital orthophotos obtained by combining the historical remote sensing images of the target area with the existing digital orthophoto products to obtain control points.

[0019] Specifically, in implementation, for large-scale historical remote sensing images, a block processing strategy is adopted, dividing each image into multiple small blocks. Using global low-resolution historical digital orthophotos and open-source DEMs as references, control points are obtained by feature point matching between historical remote sensing images and digital orthophotos of the target area. The matching window in the coarse matching stage can be determined based on object space coordinates: first, the pixel coordinate range of the small blocks of historical remote sensing images is inversely calculated to object space using an RFM model, and the corresponding window on the digital orthophoto is located using this obtained geographic coordinate range to extract control points. Subsequently, the points matched on all image blocks are merged to form an initial control point set. This method avoids the dependence of traditional methods on a single high-precision calibration field, enabling automatic and batch acquisition of high-quality control points over a wide geographical area. A control point is a "location reference point" that exists simultaneously in both the "original remote sensing image to be calibrated" and "reference data with known precise geographic coordinates (such as DOM, DEM)," and corresponds to the same real ground feature (such as fixed building corners, road intersections, landmark facility bases, etc.).

[0020] In this step, the control points have image-side coordinates and object-side coordinates. A control point is a uniquely identifiable feature point on both remote sensing imagery and reference data (such as DOM), used to establish the geometric relationship between the two types of data. This feature point is a pixel point corresponding to the same ground feature, existing simultaneously in historical remote sensing imagery and digital orthophoto imagery, and is the core element for coordinate matching. Image-side coordinates refer to the coordinate values ​​of an image point in its corresponding image coordinate system (usually in pixels). Object-side coordinates refer to its geographic coordinates (latitude, longitude, and elevation) on the corresponding image.

[0021] Step 103: Based on the image-side coordinates and object-side coordinates, calculate the image-side residual of the control points using the affine transformation formula, and determine the points whose image-side residuals are less than a preset threshold as sample points.

[0022] In this step, an error equation is established based on the image-side coordinates and object-side coordinates using the affine transformation formula. The object-side coordinates of the control points are calculated using the image-side projection coordinates of the remote sensing image surface using the rational function model: X=NumS(u,v,w) / DenS(u,v,w) Y = NumL(u,v,w) / DenL(u,v,w) NumS(.), DenS(.), NumL(.), and DenL(.) represent polynomial functions, and u, v, and w represent standardized object-space coordinates. The error equation is calculated using a point-by-point method to obtain the normal equation. Based on the principle of least squares, the affine transformation parameters are calculated using the normal equation. Based on the affine transformation parameters, the image-space coordinates, and the object-space coordinates, the image-space residuals of the control points are calculated using the affine transformation formula. The image-space residuals include the image-space residuals perpendicular to the track and the image-space residuals along the track.

[0023] The affine transformation formula is as follows: x = f0 + f1*X + f2*Y +X y = e0 + e1*X + e2*Y +X Where (x, y) represents the image-side coordinates of the control point on the remote sensing image, (X, Y) is the image-side projection coordinate of the control point object-side coordinates projected onto the remote sensing image surface using a rational function model, and f0, f1, f2, e0, e1, e2 are affine transformation parameters, where f0 is the vertical-track translation, f1 is the scaling and rotation coefficient of the vertical-track X coordinate, f2 is the cross-coupling coefficient of the vertical-track Y coordinate, e0 is the translation along the track direction, e1 is the cross-coupling coefficient of the track direction X coordinate, and e2 is the scaling and rotation coefficient of the track direction Y coordinate.

[0024] In this embodiment, the construction of the normal equation is a centralized processing of the error equation. Since the number of control points far exceeds the number of affine transformation parameters to be determined (e.g., 20 million control points corresponding to 6 unknown parameters), multiple error equations need to be integrated into a normal equation through point-by-point normalization. The mathematical essence of the normal equation is derived based on the least squares criterion of "minimizing the sum of squared residuals," by multiplying both sides of the error equation by the transpose of the coefficient matrix. Taking the x-direction (perpendicular) error equation as an example, let the coefficient matrix be A, the residual vector be v, the unknown parameter vector be Δ, and the observation deviation vector be l, then the error equation can be expressed as: AΔ + l = v Multiplying both sides by Aᵀ (the transpose of A) on the left, we get the normal equation: AᵀAΔ + Aᵀl = 0.

[0025] The simplified formula is: AᵀAΔ = -Aᵀl.

[0026] Where AᵀA is the product of the transpose of the coefficient matrix and itself (a symmetric positive definite matrix, ensuring the uniqueness of the solution), Δ is the correction value of the affine transformation parameter to be found (used to optimize the initial parameters), and -Aᵀl is the constant term vector.

[0027] Furthermore, the image-side residual includes the image-side residual in the perpendicular direction and the image-side residual along the track. Points whose image-side residual is less than a preset threshold are determined as sample points. The method further includes: calculating the first average value and the first mean error of the image-side residual in the perpendicular direction; calculating the second average value and the second mean error of the image-side residual along the track; and, based on the three sigma principle, determining the control points whose image-side residual in the perpendicular direction is located in the interval [first average value - 3 * first mean error, first average value + 3 * first mean error] and whose image-side residual along the track is located in the interval [second average value - 3 * second mean error, second average value + 3 * second mean error] as sample points.

[0028] Specifically, affine transformation is a linear geometric transformation that includes translation, rotation, and scaling, used to describe the transformation relationship between two two-dimensional coordinate systems. Image-side residual refers to the difference between the actual measured image-side coordinates of a control point and its theoretical coordinates calculated using a geometric model (here, the affine transformation formula), reflecting the model's fitting error or the system's geometric distortion. Vertical / along-track direction refers to the direction perpendicular to the satellite's flight trajectory (image x-axis); along-track direction refers to the direction parallel to the satellite's flight trajectory (image y-axis).

[0029] Step 104: Use the image-side residuals and image-side coordinates of the sample points to perform B-spline fitting to obtain the image distortion correction model.

[0030] Specifically, step 104 aims to construct a correction model that can accurately describe the variation of image distortion with image plane position. The image-side coordinates (specifically, the x-coordinate in the perpendicular direction) of the sample points obtained after refinement in step 103 are used as input variables, and the corresponding image-side residuals in the perpendicular direction and along the track direction are used as observed values. B-spline curves are chosen as the basis functions of the correction model because of their advantages of local support and continuous differentiability, allowing for flexible fitting of complex nonlinear errors. The unknown parameters of the B-spline function are solved using a least-squares fitting algorithm, thereby establishing the correction values ​​V from the image-side coordinates x to the image-side values ​​in the perpendicular direction. x Image-side correction value V along the track direction y The mapping relationship is known as the image distortion correction model. This model is essentially a smooth, continuous curve describing the systematic distortion correction amount of the linear charge-coupled device (CCD sensor) at different positions in the x-direction (linear array direction). The model, obtained by fitting large datasets (such as over 20 million control points), can capture subtle nonlinear distortion features caused by factors such as CCD stitching and temperature variations, providing a foundation for subsequent pixel-level precise correction. This image distortion correction model is used to determine the image-side correction value corresponding to the image-side coordinates in the remote sensing image.

[0031] In this step, the B-spline function is a mathematical model that uses B-spline basis functions to fit data variations. Here, it is used to fit the complex nonlinear relationship between image-space coordinates and image-space correction values. The image distortion correction model refers to a mathematical model obtained through fitting that can output geometric distortion correction values ​​based on image point positions, used to correct distortion errors in images captured by satellite optical cameras. The image-space correction value refers to the correction amount (including the vertical orbital correction V) that needs to be applied to the original image-space coordinates to correct image geometric distortion. x and the correction amount V along the track direction y .

[0032] Step 105: Input the image-side coordinates of the remote sensing image to be corrected into the image distortion correction model to obtain the image-side correction value corresponding to the image-side coordinates of the remote sensing image to be corrected, and generate the target remote sensing image based on the image-side correction value.

[0033] In this step, the corrected remote sensing image is kept to the same size as the original remote sensing image to ensure no missing ground feature information. For the remote sensing image to be corrected, processed in blocks, the image-side coordinates (x0, y0) of each pixel are extracted pixel by pixel. The x0 coordinates are then input into the image distortion correction model, and the corresponding vertical track correction value V is obtained based on the output of the image distortion correction model. x and track correction value V y Based on the coordinate transformation formula (x, y) 校正后 = (x0, y0) 原影像 +(Vx V y The corrected coordinates of each pixel are calculated. A bilinear interpolation resampling method is used to extract the corresponding pixel grayscale value from the remote sensing image to be corrected based on the corrected coordinates, and assign it to the corresponding pixel position in the target remote sensing image. After all pixels are resampled, the geometric accuracy of the corrected image is verified by comparing the ground feature stitching accuracy before and after correction to ensure that misalignment errors at the CCD stitching points are eliminated, ultimately generating a new target remote sensing image with satisfactory geometric quality.

[0034] In this step, the coordinate transformation relationship is a mathematical mapping formula describing the pixel coordinates of the original image and the corrected coordinates. The core is to superimpose image-side correction values ​​to achieve distortion correction. Bilinear interpolation resampling is an image resampling method that calculates the target location pixel value by weighting the gray values ​​of four adjacent pixels, ensuring image smoothness. Target remote sensing image: After correction by the image distortion correction model, the final output image achieves the required geometric accuracy, eliminating the distortion errors of the original image. Pixel resampling is the process of recalculating pixel values ​​based on the corrected coordinates, ensuring the sharpness and feature integrity of the corrected image.

[0035] Furthermore, embodiments of this application provide more specific methods for correcting remote sensing images, such as... Figure 2 As shown, the details are as follows: Step 201: Acquire remote sensing images, digital orthophotos, open-source digital elevation models, and remote sensing images to be calibrated for the target area.

[0036] Specifically, the core of this step is data preparation and quality control. The remote sensing imagery of the target area is historical data acquired by satellite optical cameras, while the digital orthorectified imagery is an image map that has undergone geometric correction (orthorectification) and been assigned geographic coordinates. The remote sensing imagery to be corrected is the imagery acquired by the aforementioned satellite optical cameras, and this imagery requires correction. The remote sensing imagery of the target area should be historically accumulated L1-level data, requiring clear imaging, cloud cover below a threshold (e.g., ≤5%), and originating from the same satellite sensor to ensure consistent imaging geometry; simultaneously, a complete RPC parameter package must be obtained for subsequent geometric processing. The digital orthorectified imagery (DOM) must meet the minimum resolution requirement, which must be determined by formula G. DOM ×k≤G L1 Verification is performed to ensure control point quality is maintained even when using a low-resolution DOM, thanks to a high-precision matching algorithm. An open-source DEM is used to determine the matching window between the remote sensing imagery and the digital orthophoto imagery for the target area. The remote sensing imagery to be corrected must be from the same source as historical data and undergo radiometric calibration preprocessing to eliminate atmospheric errors. This step, through multi-source data collaboration and resolution adaptation, eliminates the reliance on high-cost ground calibration fields, laying the foundation for subsequent control point extraction.

[0037] Step 202: Match control points by feature points of historical remote sensing images and digital orthophotos of the target area. An open-source digital elevation model is used to determine the matching window.

[0038] This step aims to obtain high-precision control points through a segmentation strategy and refined screening.

[0039] First, the large-size remote sensing image is divided into multiple regular small blocks to reduce the amount of data processed per batch and improve matching efficiency. The pixel coordinate range of the historical remote sensing image blocks is inversely calculated to the object space using an RFM model. This obtained geographic coordinate range is then used to locate the corresponding window on the digital orthophoto image and extract control points. Subsequently, the matching results of all blocks are merged to form an initial control point set. Control points must possess both image-space and object-space coordinates. This process, combined with block-based parallel processing, enables automatic and batch acquisition of control points over a large area, where each control point has both image-space and object-space coordinates.

[0040] Step 203: Based on the image-side coordinates and object-side coordinates, calculate the image-side residual of the control points using the affine transformation formula. Determine the points whose image-side residuals are less than a preset threshold as sample points, and perform B-spline fitting using the image-side residuals and image-side coordinates of the sample points to obtain the image distortion correction model.

[0041] This step constructs an image distortion correction model through error modeling and nonlinear fitting. First, the image-side coordinates (measured values) and object-side coordinates of the control points are substituted into the affine transformation formula to establish the error equation. The error equation is then transformed into a normal equation using a point-by-point method, and the affine parameters are solved based on the least squares method to calculate the vertical and along-track image-side residuals for each control point.

[0042] Furthermore, the three-sigma principle is used to eliminate abnormal points in the image-side residuals (such as residuals exceeding the mean ± 3 times the standard error range), which are then identified as sample points. Subsequently, based on the push-broom imaging principle of linear CCDs, the distribution pattern of the residuals along the vertical track direction (x-axis) is analyzed to capture the nonlinear distortion characteristics at the CCD stitching points. Using the image-side coordinates (x-direction) of the sample points as input, and the vertical and along-track residuals as observed values, the unknown parameters of the B-spline function are calculated using the least squares method. The formula for the B-spline function is: f(x) = Among them, c i Let represent the coefficients to be determined, n represent the number of coefficients to be determined, and K represent the order. The basis function is represented by a model that accurately describes the variation of distortion with the image plane position through big data fitting (such as tens of millions of control points), providing core support for pixel-level correction.

[0043] Before using the image-side residuals and image-side coordinates of the sample points to perform B-spline fitting and obtain the image distortion correction model, the method further includes: based on the push-broom imaging principle of the linear array charge-coupled device for remote sensing images, determining the distribution of the image-side residuals in the vertical and horizontal directions of the sample points in the linear array direction; based on the distribution in the linear array direction, analyzing the image distortion errors of the satellite optical camera acquiring the remote sensing images in the vertical and horizontal directions, so as to determine the internal orientation stability of the charge-coupled device detection unit of the satellite optical camera.

[0044] Specifically, the method for determining the interior orientation stability is as follows: Before performing B-spline fitting using the image-side residuals and image-side coordinates of the sample points, a systematic analysis of the image-side residual distribution of the sample points needs to be conducted based on the push-broom imaging principle of the linear charge-coupled device (CCD) in remote sensing imagery, in order to determine the interior orientation stability of the CCD detection unit of the satellite optical camera. The specific implementation method for this step is as follows: First, based on the pushbroom imaging mechanism of the satellite optical camera's linear CCD array, the CCD elements of the optical camera are linearly arranged in the vertical orbit direction (x-axis), and the size of all images captured by the same optical camera is fixed in the x-axis direction. Based on this physical characteristic, the sample points, after being filtered according to the three-sigma principle, are plotted and statistically analyzed according to their vertical orbit coordinates (x-values) in the image-side coordinate system. This plotting and statistical analysis along the linear array direction can visually reveal the variation pattern of the residuals on the x-axis, such as whether there are abrupt changes at the CCD chip splicing points or whether there are periodic fluctuations.

[0045] Secondly, based on the distribution results along the linear array direction, the distortion error characteristics of the satellite optical camera are analyzed in depth. The distribution of the image-side residuals in the vertical-track direction directly reflects the geometric stability of the CCD elements in the x-direction: if the residuals show a significant shift or a sudden increase in dispersion in a specific x-coordinate interval, it indicates that there is a systematic distortion in the corresponding CCD element; while the distribution of the residuals along the track direction is related to the cumulative effect of timing errors during the push-broom process. Through big data statistical methods (such as kernel density estimation or trend surface fitting), the degree of deviation between the residual distribution and the ideal linear model can be quantitatively evaluated, thereby judging the stability of the interior orientation elements. For example, if the residual distribution curve is smooth overall and has a small fluctuation range, it indicates that the CCD element arrangement is stable; if multiple peaks or faults appear, it suggests that there is misalignment error or temperature-sensitive distortion at the CCD splicing point. This analysis process needs to be verified in conjunction with the physical design parameters of the satellite optical camera (such as the number of CCD elements and the spacing between elements). By correlating the residual distribution with the physical location of the CCD, the source of distortion can be clearly identified: for example, if the x-coordinate of the residual abnormal interval matches the CCD stitching boundary, it confirms stitching error; if the residual shows a gradual trend with the x-coordinate, it may originate from lens distortion or thermal deformation. Finally, by generating distribution maps (such as scatter plots or fitted curves) of image distortion errors along the vertical and track directions, the camera stability results are visualized, providing a physical basis and parameter constraints for the subsequent calculation of unknown parameters of the B-spline function, ensuring that the model can accurately capture the nonlinear distortion characteristics of the CCD probe. Through big data statistical analysis of the image distortion error distribution pattern, the reliance on ground calibration fields is completely eliminated, overcoming the limitation of not being able to set up calibration fields in special geographical areas. This method can adapt to various imaging modes such as single-line array, multi-line array stereo observation, and agile imaging satellites, significantly improving the universality and timeliness of the method. Simultaneously, error distribution analysis based on physical imaging principles provides an accurate description of the distortion characteristics of the B-spline function, ensuring the accuracy of subsequent pixel-level correction and fundamentally guaranteeing the accuracy of image feature locations.

[0046] In this embodiment, an image distortion correction model is obtained, including: using the image-side coordinates of the sample points as input variables, and the perpendicular-to-rail image-side residuals and along-rail image-side residuals corresponding to the image-side coordinates of the sample points as observed values, determining the unknown parameters of the B-spline function using a least-squares fitting algorithm; and establishing mapping relationships from the image-side coordinates to the perpendicular-to-rail image-side correction values ​​and the along-rail image-side correction values ​​based on the unknown parameters of the B-spline function. The B-spline function uses a uniform node vector configuration and obtains a smooth and continuous image distortion correction curve by fitting the sample point data. The image distortion correction curve describes the functional relationship between the image-side coordinates and the perpendicular-to-rail and along-rail image-side correction values.

[0047] Specifically, in the implementation of the image distortion correction model, the input variables and observed values ​​must first be clearly defined. The model uses the perpendicular-to-orbit image-side coordinates of the refined sample points as input variables, and their corresponding perpendicular-to-orbit image-side residuals and along-orbit image-side residuals as observed values. To ensure the model's stability and fitting ability, the B-spline function is configured with uniform node vectors, which ensures that the basis functions are uniformly distributed within the domain, better adapting to complex nonlinear distortions.

[0048] In practice, the unknown parameters of the B-spline function are solved using a least-squares fitting algorithm. The core of this process is constructing an error equation and minimizing the sum of squared residuals between observed and predicted values. The algorithm iteratively adjusts the coefficients of the B-spline function until the optimal solution is found, ensuring that the generated correction curve closely matches the distribution of the actual residuals. Based on the solved unknown parameters of the B-spline function, mapping relationships are established from image-side coordinates to image-side correction values ​​in the vertical and along-track directions. This essentially generates two smooth and continuous image distortion correction curves: one describing the variation of distortion in the vertical direction with image-side coordinates, and the other describing the variation in the along-track direction. For example, if significant nonlinear distortion exists at the CCD element stitching point, the correction curve will exhibit obvious fluctuations within the corresponding coordinate range, thus accurately capturing subtle error characteristics at the element level. The final image distortion correction model can output corresponding image-side correction values ​​in the vertical and along-track directions based on the input image-side coordinates, achieving pixel-level geometric distortion correction. This model effectively overcomes the dependence of traditional methods on specific ground calibration fields by fitting tens of millions of high-quality control points. The image distortion correction model obtained through big data fitting can accurately describe the nonlinear distortion of the CCD detector elements of satellite optical cameras, achieving pixel-level geometric correction. Ultimately, the internal geometric error of the image is controlled within a preset pixel limit (0.3), significantly improving the accuracy of ground feature positioning and the internal geometric quality of remote sensing images. Simultaneously, it eliminates the dependence on high-cost ground calibration fields and is applicable to optical remote sensing satellites with various imaging modes.

[0049] Step 204: After inputting the image-side coordinates of the remote sensing image to be corrected into the image distortion correction model, the image-side correction value corresponding to the image-side coordinates of the remote sensing image to be corrected is obtained, and the target remote sensing image is generated based on the image-side correction value.

[0050] In this step, the step of inputting the image-side coordinates of the remote sensing image to be corrected into the image distortion correction model to obtain the image-side correction value corresponding to the image-side coordinates, and generating the target remote sensing image based on the image-side correction value corresponding to the image-side coordinates, includes: performing coordinate transformation on the image-side coordinates in the remote sensing image according to the coordinate transformation relationship to obtain the corrected remote sensing image coordinates, wherein the coordinate transformation relationship is (x, y). 校正后= (x0, y0) 原影像 +(V x V y ), where (x, y) represents the image-side coordinates in the corrected remote sensing image, and (x0, y0) represents the image-side coordinates in the original remote sensing image. V x V is the image-side correction value in the vertical rail direction. y The image-side correction value along the track direction is used; a bilinear interpolation resampling method is adopted to extract the pixel value at the corresponding position from the original remote sensing image based on the corrected remote sensing image coordinates, and the pixel value is assigned to the pixel position corresponding to the corrected image-side coordinates in the target remote sensing image to complete the generation of the target remote sensing image.

[0051] Specifically, the remote sensing image to be corrected is first preprocessed to ensure it has a consistent geometric reference with the data used to establish the image distortion correction model. Then, the image is traversed pixel-by-pixel to extract the original image-side coordinates (x0, y0) for each pixel. The vertical alignment coordinate x0 is then input into the fitted image distortion correction model, and the model automatically outputs the corresponding vertical alignment image-side correction value V. x Image-side correction value V along the track direction y Next, the corrected image-side coordinates of each pixel are calculated based on the coordinate transformation relationship. This step essentially involves precisely correcting the geometric position of each pixel in the original image by superimposing the image-side correction values, eliminating positional deviations caused by distortions of orientation elements within the optical camera. After coordinate transformation, a bilinear interpolation resampling method is used to extract the corresponding pixel value from the original remote sensing image based on the corrected image-side coordinates. Specifically, for each corrected coordinate, its four adjacent pixels in the original image are found, and the grayscale value of the target position is calculated using a distance-weighted average, which is then assigned to the corresponding pixel in the target remote sensing image. This process ensures the smoothness and continuity of the image after geometric transformation, avoiding pixel holes or jagged edges. Finally, the quality of the generated target remote sensing image is verified to ensure that the internal geometric error is ≤0.3 pixels, and the seamless connection of ground features at the CCD stitching points is checked. This step achieves pixel-level geometric distortion correction through an image distortion correction model, controlling the internal geometric error of the image within a preset pixel (0.3), significantly improving the accuracy of ground feature positioning. This method eliminates the reliance on ground calibration fields and is applicable to single-line array, multi-line array, and agile imaging satellites, exhibiting strong versatility and high timeliness.

[0052] Based on this, the technical effects of this remote sensing image correction method are mainly reflected in the following aspects: In terms of operational efficiency and timeliness, this method achieves batch extraction of control points and B-spline fitting through an automated process, completely changing the traditional operational mode of manually setting up calibration fields and waiting for satellite revisits. Based on a nationwide low-resolution DOM and open-source DEM, the method can process multiple images in parallel, completing the screening and modeling of tens of millions of control points in a short time (e.g., processing 2000 images to obtain over 20 million control points), shortening the traditional calibration cycle of several months to several days, significantly improving the timeliness of remote sensing image processing. In terms of cost control and applicability, this method utilizes historical remote sensing imagery and publicly available geographic data to replace dedicated calibration fields, overcoming geographical limitations. Simultaneously, the applicability of low-resolution DOM is verified through formulas, significantly reducing data costs while ensuring accuracy, achieving a low-investment, wide-coverage technical effect. In terms of correction accuracy and versatility, pixel-level geometric distortion correction is achieved based on B-spline functions. By inputting the vertical alignment coordinate x0 into the fitted image distortion correction model, the image-side correction values ​​Vx and Vy are automatically output. Combined with coordinate transformation and bilinear interpolation resampling, the internal geometric error of the image is controlled within 0.3 pixels. This method is applicable to various imaging modes, including single-line arrays, multi-line arrays, and agile imaging satellites, demonstrating significant versatility. Regarding quality controllability, sample points are obtained by selecting control points using the three-sigma principle, ensuring the reliability of the correction results. The corrected image exhibits an internal geometric error ≤0.3 pixels, and ground features are seamlessly integrated at CCD stitching points, verifying the stability and reliability of this method in engineering applications.

[0053] Furthermore, the applicant provided specific examples of the above methods, as follows: Taking the correction of a certain satellite panchromatic remote sensing image as an example, the specific data and parameters are given below, and the actual application process of this correction method is explained in detail below: First, reference data and image data to be processed for the target area are acquired. The reference data consists of pre-processed low-resolution digital orthophoto images (DOM) and open-source digital elevation models (DEMs) covering the entire country. The DOM has a ground resolution of 1m, and the DEM has a ground resolution of 30m. The images to be processed are 2000 panchromatic remote sensing images of a certain satellite distributed across the country from January to June 2024. The ground resolution of these images is 0.5m, cloud cover is controlled within 5%, and each image contains complete RPC parameters. Radiometric calibration preprocessing has also been performed in advance to remove residual atmospheric correction errors. During the data preparation stage, the applicability of the DOM resolution also needs to be verified according to the preset formula "G...". DOM ×k≤G L1 (where G)DOM G represents the minimum resolution of the DOM, k represents the matching algorithm precision, and G represents the minimum resolution of the DOM. L1 Given the resolution of the remote sensing image, and the matching algorithm accuracy k = 0.2 pixels, 1m (G DOM ), 0.2pixel (k), 0.5m (G L1 Substituting into the formula, we calculate that 1m × 0.2pixel = 0.2m ≤ 0.5m, which meets the minimum resolution requirement and ensures the accuracy of subsequent control point matching.

[0054] Secondly, control point extraction and refinement were performed. Considering the large size of the 2000 panchromatic remote sensing images, a block processing strategy was adopted, dividing each image into multiple standardized small blocks. Using the national 1m resolution DOM and 30m resolution DEM as reference data, the pixel coordinate range of each small image block was back-calculated to the object space using an RFM model. This obtained geographic coordinate range was then used to locate the corresponding windows on the digital orthophoto image, extracting control points. After matching, the coarse matching points of all small blocks in a single image were merged to form an initial control point set. Subsequently, control point refinement was performed. The image-space coordinates (x, y) of the initial control points and the image-space projection coordinates (X, Y) obtained through forward calculation using the RFM model were substituted into the affine transformation formula to establish an error equation. The affine transformation formula is: x = f0 + f1*X + f2*Y +X y = e0 + e1*X + e2*Y +X, The error equation is: Where F(x0) and F(y0) are approximate values ​​of the image-side coordinates. , , , , , , where represents the correction values ​​for the affine transformation parameters of the image plane. The normal equations are established using a point-by-point method, and the affine transformation parameters are calculated using the least squares principle, thereby obtaining the image-side residuals in the perpendicular and orbital directions for each control point.

[0055] After obtaining the image-side residuals, control points are selected according to the three sigma principle. First, the first average value m of the image-side residuals in the perpendicular direction is calculated. + and the first standard error σ + And the second average value m of the image-side residual along the orbital direction. - Second mean error σ - Then retain the image-side residual in the vertical orbit direction located at [m] + -3σ + m ++3σ + Within the interval, the image-side residual along the track direction is located at [m]. - -3σ - m - +3σ - The control points within the interval were ultimately selected from 2,000 images to obtain at least 20 million high-precision sample points.

[0056] Next, we proceed to the B-spline fitting stage. Based on the pushbroom imaging principle of satellite optical camera linear CCD arrays, the CCD detectors of the optical camera are linearly arranged in the vertical direction (x-axis), and the x-axis dimensions of all images captured by the same optical camera are fixed. Based on this characteristic, we conducted big data statistical analysis on the vertical and along-track image distortion errors of 20 million sample points, observed the distribution pattern of errors along the x-axis, and clarified the interior orientation stability characteristics of the satellite optical camera CCD detectors. We found that the errors exhibit a non-linear distribution, and the error changes are more significant at the CCD stitching points. Based on the error distribution pattern, we selected B-spline curves to construct an image distortion correction model. The vertical image-side coordinates (x) of the sample points are used as input variables, and the corresponding vertical image-side residuals and along-track image-side residuals are used as observed values. We adopted a uniform node vector configuration for the B-spline basis function and solved the unknown parameters of the B-spline function using a least-squares fitting algorithm. We then established the correction values ​​V from the image-side coordinates x to the vertical image-side correction values ​​V. x From image-side coordinate x to image-side correction value V along the track direction y The mapping relationship ultimately yields an image distortion correction model that can accurately describe the distortion correction amount at different x-coordinate positions. This model is a smooth and continuous curve that can cover the error correction requirements of the entire image x-axis direction.

[0057] Finally, the steps of processing the remote sensing image to be corrected and generating the target remote sensing image are performed. A panchromatic remote sensing image to be corrected, from the same satellite and with the same resolution as the aforementioned 2000 images, is selected to ensure that the corrected target remote sensing image and the image to be corrected are of the same size, avoiding the loss of ground feature information. The image to be corrected is then processed pixel-by-pixel. First, the original image-side coordinates (x0, y0) of each pixel are extracted. Then, x0 is substituted into the fitted image distortion correction model to calculate the vertical orbital direction image-side correction value V corresponding to that pixel. x Image-side correction value V along the track direction yNext, based on the coordinate transformation relationship, the corrected image-side coordinates of each pixel are calculated. Then, using a bilinear interpolation resampling method, based on the corrected image-side coordinates, the four adjacent pixels around the coordinates in the image to be corrected are found, and the pixel grayscale value corresponding to the corrected coordinates is obtained through weighted calculation. Finally, the calculated pixel grayscale value is assigned to the pixel position corresponding to the corrected image-side coordinates in the target remote sensing image. After all pixels have been processed, the geometric accuracy of the generated target remote sensing image is verified. The results show that the internal geometric error is ≤0.3 pixels. At the same time, comparing the ground feature positions at the CCD stitching point of the images before and after correction, it is found that the stitching misalignment error is completely eliminated, meeting the geometric quality requirements of the remote sensing image. Thus, the generation of the target remote sensing image is completed.

[0058] Furthermore, as a response to the above Figure 1-2 The present invention also provides a remote sensing image correction device based on big data statistical camera distortion, which is used to correct remote sensing images, in addition to the implementation of the method embodiments shown. The embodiments of this device correspond to the foregoing method embodiments. For ease of reading, this embodiment will not repeat the details of the foregoing method embodiments one by one, but it should be clear that the device in this embodiment can implement all the contents of the foregoing method embodiments. Specifically, as shown... Figure 3 As shown, the device includes: The acquisition unit 31 is used to acquire historical remote sensing images and existing digital orthophoto products of the target area, as well as remote sensing images to be corrected. The determination unit 32 is used to perform feature point matching on the digital orthophoto obtained from the historical remote sensing images of the target area in the acquisition unit 31 and the existing digital orthophoto products to obtain control points. The control points have image-side coordinates and object-side coordinates. The determining unit 32 is used to calculate the image-side residual of the control points based on the image-side coordinates and the object-side coordinates using the affine transformation formula, and to determine the points whose image-side residuals are less than a preset threshold as sample points. The fitting unit 33 is used to perform B-spline fitting using the image-square residuals and image-square coordinates of the sample points in the determining unit 32 to obtain an image distortion correction model. The image distortion correction model is used to determine the image-square correction value corresponding to the image-square coordinates in the remote sensing image. The generation unit 34 is used to obtain the image-square correction value corresponding to the image-square coordinates of the remote sensing image to be corrected by inputting the image-square coordinates of the remote sensing image to be corrected in the acquisition unit 31 into the image distortion correction model in the fitting unit 33, and generate the target remote sensing image based on the image-square correction value.

[0059] Furthermore, such as Figure 4 As shown, the determining unit 32 includes: Module 321 is used to establish an error equation based on the image-space coordinates and object-space coordinates using an affine transformation formula, wherein the affine transformation formula is: x = f0 + f1*X + f2*Y +X y = e0 + e1*X + e2*Y +X (x, y) represents the image-side coordinates of the control point on the remote sensing image, (X, Y) represents the image-side projection coordinates of the control point object-side coordinates projected onto the remote sensing image surface using a rational function model, f0, f1, f2, e0, e1, e2 are affine transformation parameters, f0 is the vertical-track translation, f1 is the scaling and rotation coefficient of the vertical-track direction relative to the X coordinate, f2 is the cross-coupling coefficient of the vertical-track direction relative to the Y coordinate, e0 is the translation along the track direction, e1 is the cross-coupling coefficient of the along the track direction relative to the X coordinate, and e2 is the scaling and rotation coefficient of the along the track direction relative to the Y coordinate. The formula for calculating the image-side projection coordinates of control point object coordinates onto the remote sensing image surface using the rational function model is as follows: X=Num S (u,v,w) / Den S (u,v,w) Y = Num L (u,v,w) / Den L (u,v,w) Num S (.), Den S (.), Num L (.), Den L (.) denotes a polynomial function, and u, v, w denote the standardized object-space coordinates; The determining module 321 is used to calculate the error equation using a point-by-point method to obtain the normal equation; Calculation module 322 is used to calculate affine transformation parameters based on the principle of least squares and using the normal equations in determination module 321; The calculation module 322 is used to calculate the image-side residual of the control point based on the affine transformation parameters, the image-side coordinates and the object-side coordinates, where the image-side residual includes the image-side residual in the perpendicular direction and the image-side residual along the track direction.

[0060] Furthermore, such as Figure 4 As shown, the image-side residual includes the image-side residual perpendicular to the track and the image-side residual along the track. After calculating the image-side residual of the control points using the affine transformation formula, the calculation module 322 includes: Calculate the first mean and first mean error of the image-side residuals in the vertical direction; Calculate the second mean and second mean error of the image-side residuals along the track direction; Based on the three sigma principle, control points whose image-side residuals in the vertical direction are within the interval of [first average value - 3 * first mean error, first average value + 3 * first mean error] and whose image-side residuals along the track direction are within the interval of [second average value - 3 * second mean error, second average value + 3 * second mean error] are determined as sample points.

[0061] Furthermore, before performing B-spline fitting using the image-side residuals and image-side coordinates of the sample points and obtaining the image distortion correction model, the fitting unit 33 includes: The image-side residual module 331 is used to determine the image-side residuals in the vertical direction and the image-side residuals along the track direction of the sample points, based on the push-broom imaging principle of the linear array charge-coupled device based on remote sensing images, and their distribution in the linear array direction. Analysis module 332 is used to analyze the image distortion error of the satellite optical camera acquiring remote sensing images in the vertical and along-track directions based on the distribution of the linear array direction determined by the image residual module 331, so as to determine the internal orientation stability of the charge-coupled device detection unit of the satellite optical camera.

[0062] Furthermore, such as Figure 4 As shown, the analysis module 332 further includes: Using the image-side coordinates of the sample points as input variables, and the image-side residuals in the vertical and along-track directions corresponding to the image-side coordinates of the sample points as observed values, the unknown parameters of the B-spline function are determined by the least squares fitting algorithm. Based on the unknown parameters of the B-spline function, a mapping relationship is established from the image-side coordinates to the image-side correction value in the vertical direction and the image-side correction value in the track direction.

[0063] Furthermore, such as Figure 4 As shown, the generation unit 34 includes: The coordinate acquisition module 341 is used to perform coordinate transformation on the image-side coordinates in the remote sensing image according to the coordinate transformation relationship, so as to obtain the corrected remote sensing image coordinates. The coordinate transformation relationship is (x, y). 校正后 = (x0, y0) 原影像 +(V x V y ) Where (x, y) 校正后 This represents the image-side coordinates (x0, y0) in the corrected remote sensing image. 原影像 V represents the image-side coordinates in the original remote sensing image. x V is the image-side correction value in the vertical rail direction. y This is the image-side correction value along the track direction; The generation module 342 is used to extract the pixel value of the corresponding position from the original remote sensing image based on the corrected remote sensing image coordinates in the acquisition coordinate module 341 using the bilinear interpolation resampling method, and assign the pixel value to the pixel position corresponding to the corrected image coordinates in the target remote sensing image, thereby completing the generation of the target remote sensing image.

[0064] Furthermore, such as Figure 4 As shown, the acquisition unit 31 further includes: When acquiring historical remote sensing images and digital orthophotos of the target area, the minimum resolution of the digital orthophotos is determined by a preset formula, wherein the preset formula is: G DOM *k≤G L1 Among them, G DOM G is the lowest resolution of the digital orthophoto, k is the matching algorithm precision, and G is the lowest resolution of the digital orthophoto. L1 This refers to the resolution of the remote sensing image.

[0065] Furthermore, embodiments of the present invention also provide a readable storage medium for storing a computer program, wherein the computer program, when running, controls the device where the storage medium is located to perform the above-described actions. Figure 1-2 The remote sensing image correction method based on big data statistical camera distortion as described in any one of the following.

[0066] Furthermore, embodiments of the present invention also provide an electronic device, the electronic device including a storage medium; and one or more processors, the storage medium being coupled to the processors, the processors being configured to execute program instructions stored in the storage medium; the program instructions, when executed, perform as described above. Figure 1-2 The remote sensing image correction method based on big data statistical camera distortion as described in any one of the following.

[0067] 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.

[0068] It is understood that the relevant features in the above methods and apparatus can be referenced interchangeably. Furthermore, the terms "first," "second," etc., in the above embodiments are used to distinguish between embodiments and do not represent the superiority or inferiority of any particular embodiment.

[0069] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0070] The algorithms and displays provided herein are not inherently related to any particular computer, virtual system, or other device. Various general-purpose systems can also be used in conjunction with the teachings herein. The required structure for constructing such systems is apparent from the above description. Furthermore, this invention is not directed to any particular programming language. It should be understood that the contents of the invention described herein can be implemented using various programming languages, and the above description of specific languages ​​is for the purpose of disclosing the best mode of implementation of the invention. Additionally, the memory may include non-persistent memory in computer-readable media, random access memory (RAM), and / or non-volatile memory, such as read-only memory (ROM) or flash RAM, and the memory may include at least one memory chip.

[0071] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0072] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0073] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0074] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0075] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0076] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0077] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0078] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A remote sensing image correction method based on big data statistical camera distortion, characterized in that, include: Acquire historical remote sensing images and existing digital orthophoto products of the target area, as well as remote sensing images to be corrected; Control points are obtained by matching feature points in digital orthophotos derived from historical remote sensing images and existing digital orthophoto products of the target area. The control points have image-side coordinates and object-side coordinates. The control points are feature points that can be uniquely identified and whose locations correspond to each other in both remote sensing images and reference data. They are used to establish the geometric relationship between the two types of data. The feature points are pixels that exist simultaneously in historical remote sensing images and digital orthophotos and correspond to the same ground feature. They are the core elements for achieving coordinate matching. The step of matching control points by feature points of historical remote sensing images of the target area and digital orthophotos obtained from existing digital orthophoto products includes: for large-size historical remote sensing images of the target area, a block processing strategy is adopted to divide each historical remote sensing image of the target area into multiple image blocks, and using global low-resolution historical digital orthophotos and open-source DEMs as references, feature point matching is performed between the historical remote sensing images of the target area and the digital orthophotos to obtain control points. The matching window in the matching stage is determined based on object space coordinates, including: calculating the pixel coordinate range of image patches of the target area's historical remote sensing images to the object space using the RFM model to obtain the corresponding geographic coordinate range; locating the corresponding window on the digital orthophoto based on the geographic coordinate range and extracting control points; and merging all the control points extracted from the image patches to form a control point set. Based on the image-side coordinates and object-side coordinates, the image-side residual of the control points is calculated using the affine transformation formula. Points with image-side residuals less than a preset threshold are identified as sample points, including: An error equation is established based on the image-space coordinates and object-space coordinates using the affine transformation formula, where the affine transformation formula is: x = f0 + f1*X + f2*Y +X y = e0 + e1*X + e2*Y +X (x, y) are the image-side coordinates of the control point on the remote sensing image, (X, Y) are the image-side projection coordinates of the control point object-side coordinates projected onto the remote sensing image surface using a rational function model, f0, f1, f2, e0, e1, e2 are affine transformation parameters, f0 is the vertical-track translation, f1 is the scaling and rotation coefficient of the vertical-track direction relative to the X coordinate, f2 is the cross-coupling coefficient of the vertical-track direction relative to the Y coordinate, e0 is the translation along the track direction, e1 is the cross-coupling coefficient of the track direction relative to the X coordinate, and e2 is the scaling and rotation coefficient of the track direction relative to the Y coordinate. The formula for calculating the image-side projection coordinates of control point object coordinates onto the remote sensing image surface using the rational function model is as follows: X=Num S (u,v,w) / Den S (u,v,w) Y= Num L (u,v,w) / Den L (u,v,w) Num S (.), Den S (.), Num L (.), Den L (.) denotes a polynomial function, and u, v, w denote the standardized object-space coordinates; The error equation is calculated using a point-by-point method to obtain the normal equation; Based on the principle of least squares, the affine transformation parameters are calculated using the aforementioned normal equations; Based on the affine transformation parameters, the image-side coordinates and the object-side coordinates, the image-side residual of the control point is calculated using the affine transformation formula. The image-side residual includes the image-side residual in the direction perpendicular to the rail and the image-side residual along the rail. Using the image-square residuals and image-square coordinates of the sample points, B-spline fitting is performed to obtain an image distortion correction model. The image distortion correction model is used to determine the image-square correction value corresponding to the image-square coordinates in the remote sensing image. After inputting the image-side coordinates of the remote sensing image to be corrected into the image distortion correction model, the image-side correction value corresponding to the image-side coordinates of the remote sensing image to be corrected is obtained, and the target remote sensing image is generated based on the image-side correction value.

2. The method according to claim 1, characterized in that, Points with image-square residuals less than a preset threshold are identified as sample points. The method includes: Calculate the first mean and first mean error of the image-side residuals in the vertical direction; Calculate the second mean and second mean error of the image-side residuals along the track direction; Based on the three sigma principle, control points whose image-side residuals in the vertical direction are located within the interval of [first average value - 3 * first mean error, first average value + 3 * first mean error] and whose image-side residuals along the track direction are located within the interval of [second average value - 3 * second mean error, second average value + 3 * second mean error] are determined as sample points.

3. The method according to claim 2, characterized in that, Before performing B-spline fitting using the image-side residuals and image-side coordinates of the sample points to obtain the image distortion correction model, the method further includes: Based on the push-broom imaging principle of linear charge-coupled devices based on remote sensing images, the distribution of image-side residuals in the vertical and along-track directions of sample points in the linear array direction is determined. Based on the distribution of the linear array direction, the image distortion error of the satellite optical camera acquiring remote sensing images in the vertical and along-track directions is analyzed to determine the internal orientation stability of the charge-coupled device detection unit of the satellite optical camera.

4. The method according to claim 3, characterized in that, The resulting image distortion correction model includes: Using the image-side coordinates of the sample points as input variables, and the image-side residuals in the vertical and along-track directions corresponding to the image-side coordinates of the sample points as observed values, the unknown parameters of the B-spline function are determined by the least squares fitting algorithm. Based on the unknown parameters of the B-spline function, a mapping relationship is established from the image-side coordinates to the image-side correction value in the vertical direction and the image-side correction value in the track direction.

5. The method according to claim 4, characterized in that, The process of inputting the image-space coordinates of the remote sensing image to be corrected into the image distortion correction model to obtain the image-space correction value corresponding to the image-space coordinates of the remote sensing image to be corrected, and generating the target remote sensing image based on the image-space correction value, includes: Based on the coordinate transformation relationship, the image-side coordinates in the remote sensing image are transformed to obtain the corrected remote sensing image coordinates. The coordinate transformation relationship is (x, y). 校正后 = (x0, y0) 原影像 +(V x V y ) Where (x, y) 校正后 This represents the image-side coordinates (x0, y0) in the corrected remote sensing image. 原影像 V represents the image-side coordinates in the original remote sensing image. x V is the image-side correction value in the vertical rail direction. y Image-side correction value along the track direction; The bilinear interpolation resampling method is used to extract the pixel value at the corresponding position from the original remote sensing image based on the corrected remote sensing image coordinates, and then assign the pixel value to the pixel position corresponding to the corrected image coordinates in the target remote sensing image to complete the generation of the target remote sensing image.

6. The method according to any one of claims 1-5, characterized in that, The method further includes: When acquiring historical remote sensing and digital orthophoto images of the target area, the minimum resolution of the digital orthophoto image is determined by a preset formula, which is: G DOM *k≤G L1 Among them, G DOM G is the lowest resolution of the digital orthophoto, k is the matching algorithm precision, and G is the lowest resolution of the digital orthophoto. L1 This refers to the resolution of the remote sensing image.

7. A remote sensing image correction device based on big data statistical camera distortion, characterized in that, include: The acquisition unit is used to acquire historical remote sensing images and existing digital orthophoto products of the target area, as well as remote sensing images to be corrected. The determination unit is used to match control points by feature points of historical remote sensing images and digital orthophotos obtained from existing digital orthophoto products of the target area in the acquisition unit. The control points have image-side coordinates and object-side coordinates. The control points refer to feature points that can be uniquely identified and whose positions correspond to each other in both remote sensing images and reference data. They are used to establish the geometric relationship between the two types of data. The feature points are pixels that exist simultaneously in historical remote sensing images and digital orthophotos and correspond to the same ground feature. They are the core elements for achieving coordinate matching. For obtaining control points by matching feature points of historical remote sensing images of the target area and digital orthophotos obtained from existing digital orthophoto products in the acquisition unit, the determination unit is specifically used to: for large-size historical remote sensing images of the target area, adopt a block processing strategy to divide each historical remote sensing image of the target area into multiple image blocks, and use global low-resolution historical digital orthophotos and open-source DEMs as references to perform feature point matching between the historical remote sensing images of the target area and the digital orthophotos to obtain control points; The matching window in the matching stage is determined based on object space coordinates, including: calculating the pixel coordinate range of image patches of historical remote sensing images of the target area to the object space through the RFM model to obtain the corresponding geographic coordinate range; locating the corresponding window on the digital orthophoto based on the geographic coordinate range and extracting control points; and merging all the control points extracted from the image patches to form a control point set. The determining unit is used to calculate the image-side residual of the control points based on the image-side coordinates and object-side coordinates using an affine transformation formula, and to determine points whose image-side residuals are less than a preset threshold as sample points; the determining unit includes: a determining module and a calculation module. The determining module is used to establish an error equation based on the image-space coordinates and object-space coordinates using an affine transformation formula, wherein the affine transformation formula is: x = f0 + f1*X + f2*Y +X y = e0 + e1*X + e2*Y +X (x, y) represents the image-side coordinates of the control point on the remote sensing image, (X, Y) represents the image-side projection coordinates of the control point object-side coordinates projected onto the remote sensing image surface using a rational function model, f0, f1, f2, e0, e1, e2 are affine transformation parameters, f0 is the vertical-track translation, f1 is the scaling and rotation coefficient of the vertical-track direction relative to the X coordinate, f2 is the cross-coupling coefficient of the vertical-track direction relative to the Y coordinate, e0 is the translation along the track direction, e1 is the cross-coupling coefficient of the track direction relative to the X coordinate, and e2 is the scaling and rotation coefficient of the track direction relative to the Y coordinate. The formula for calculating the image-side projection coordinates of control point object coordinates onto the remote sensing image surface using the rational function model is as follows: X=Num S (u,v,w) / Den S (u,v,w) Y= Num L (u,v,w) / Den L (u,v,w) Num S (.), Den S (.), Num L (.), Den L (.) denotes a polynomial function, and u, v, w denote the standardized object-space coordinates; The determining module is used to calculate the error equation using a point-by-point method to obtain the normal equation; The calculation module is used to calculate the affine transformation parameters based on the principle of least squares and using the normal equations in the determination module. The calculation module is used to calculate the image-side residual of the control point based on the affine transformation parameters, the image-side coordinates and the object-side coordinates, using the affine transformation formula. The image-side residual includes the image-side residual in the direction perpendicular to the rail and the image-side residual along the rail. The fitting unit is used to perform B-spline fitting using the image-square residuals and image-square coordinates of the sample points in the determining unit to obtain an image distortion correction model. The image distortion correction model is used to determine the image-square correction value corresponding to the image-square coordinates in the remote sensing image. The generation unit is used to obtain the image-square correction value corresponding to the image-square coordinates of the remote sensing image to be corrected by inputting the image-square coordinates of the remote sensing image to be corrected in the acquisition unit into the image distortion correction model in the fitting unit, and to generate the target remote sensing image based on the image-square correction value.

8. A computer-readable storage medium, characterized in that, The storage medium includes a stored program, wherein, when the device containing the storage medium runs the program, the remote sensing image correction method based on big data statistical camera distortion as described in any one of claims 1-6 is executed.

9. An electronic device, characterized in that, The device includes at least one processor, at least one memory connected to the processor, and a bus; wherein the processor and the memory communicate with each other through the bus; the processor is used to call program instructions in the memory to execute the method for correcting remote sensing images based on big data statistical camera distortion as described in any one of claims 1-6.

Images