A method for analyzing the geolocation accuracy of medium-resolution satellite imagery

Abstract

This invention provides a method for geolocation accuracy analysis of medium-resolution satellite imagery, relating to the field of geolocation technology. The method includes: acquiring control point images and geographic location information corresponding to a ground truth image; acquiring a satellite imagery reference channel image and geographic location information; segmenting the satellite image using the Otsu thresholding method to obtain cloud and non-cloud area images; remapping the ground truth image using Gaussian filtering; extracting feature points from the cloud and non-cloud area images and the remapped image, and performing feature matching using the SIFT feature point matching algorithm; removing outlier matching point pairs using the RANSAC algorithm based on the matching point pairs; extracting the location information of precise matching point pairs based on the geographic location information of the ground truth image and the satellite image, and converting this information into positioning error; and verifying the positioning accuracy based on the positioning error to obtain the positioning accuracy verification result. This invention solves the technical problem that current positioning accuracy mainly depends on the coastline.

Description

Technical Field

[0001] This invention relates to the field of geolocation technology, and in particular to a method for analyzing the geolocation accuracy of medium-resolution satellite imagery. Background Technology

[0002] Remote sensing images are the foundation of remote sensing and many other applications. As spatial data, remote sensing images have the concept of spatial geographic location. Regardless of the application problem, it is necessary to determine the correspondence between image information and ground location. Geographic positioning of satellite images is an important prerequisite for satellite data preprocessing, and obtaining accurate geographic positioning is of great importance.

[0003] Geolocation of remote sensing images is the process of calculating the coordinates of images observed by spaceborne remote sensing instruments in a ground-based coordinate system. Existing technical methods for evaluating the geolocation accuracy of remote sensing images mainly determine the geolocation error of the satellite under test based on the remote sensing image data of the satellite under test, the land and sea mask, and the reference coastline landmarks. The coastline identification method depends on the accuracy of the land and sea template data, and the number of landmarks that can be used for identification is limited, making it impossible to evaluate the geolocation accuracy in areas without coastlines. Therefore, there is an urgent need to develop an image geolocation accuracy evaluation method that does not depend on the coastline. Summary of the Invention

[0004] This invention provides a method for analyzing the geolocation accuracy of medium-resolution satellite imagery, solving the technical problem that current positioning accuracy mainly depends on the coastline.

[0005] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows:

[0006] In a first aspect, a method for analyzing the geolocation accuracy of medium-resolution satellite imagery, the method comprising:

[0007] Obtain the control point images corresponding to the ground truth images and the geographic location information of the ground truth images;

[0008] Acquire satellite imagery reference channel images and satellite imagery geographic location information;

[0009] The Otsu thresholding method was used to segment the satellite image to obtain images of cloud areas and non-cloud areas;

[0010] The ground truth image is remapped using a Gaussian filter to obtain the remapped image.

[0011] Extract feature points from cloud and non-cloud areas of the image, and from the remapped image;

[0012] The SIFT feature point matching algorithm is used to perform feature matching to obtain matching point pairs;

[0013] Based on the matching point pairs, the RANSAC algorithm is used to remove outlier matching point pairs in order to obtain accurate matching point pairs;

[0014] Based on the geographic location information of the ground truth image and the geographic location information of the satellite image, the location information of the exact matching point pairs is extracted and converted into positioning error;

[0015] The positioning accuracy is checked based on the positioning error to obtain the positioning accuracy check result.

[0016] Furthermore, the ground truth image corresponding to the control point image and the ground truth image's geographic location information are obtained from the high-precision control point image database, including:

[0017] Obtain ground truth images from a high-precision control point image library:

[0018] Based on the high-precision control point image database, retrieve the corresponding control point image of the ground truth image to obtain the control point image corresponding to the ground truth image;

[0019] Geographic location information is extracted from the ground truth image and the corresponding control point image to obtain the geographic location information of the ground truth image.

[0020] Furthermore, satellite imagery reference channel images and satellite imagery geographic location information are obtained from the payload database to be evaluated, including:

[0021] Obtain satellite images from the database of loads to be evaluated;

[0022] Based on the payload database to be evaluated, retrieve the reference channel imagery of the satellite imagery to obtain the reference channel imagery of the satellite imagery;

[0023] Geographic location information is extracted from the satellite imagery reference channel to obtain the satellite imagery geographic location information.

[0024] Furthermore, the Otsu thresholding method is used to segment the satellite imagery to obtain images of cloud and non-cloud regions, including:

[0025] Acquire satellite imagery reference channel images;

[0026] Convert the satellite imagery reference channel to a grayscale image;

[0027] Calculate the histogram of gray levels for each pixel in a grayscale image to obtain the number of pixels at each gray level in the image;

[0028] Divide the number of pixels at each gray level by the total number of pixels to obtain the pixel probability distribution at each gray level;

[0029] Using Otsu's thresholding method, a threshold is set to segment the grayscale image into two categories. The variance between the two categories after segmentation is calculated to obtain the inter-class variance.

[0030] Between-class variance = w0 × w1 × (μ0 - μ1) 2

[0031] Where w0 and w1 are the sum of pixel probabilities of category 0 and category 1, respectively, and μ0 and μ1 are the average gray values ​​of category 0 and category 1, respectively;

[0032] Based on the inter-class variance, select the threshold that maximizes the inter-class variance as the optimal threshold.

[0033] Based on the optimal threshold, the grayscale image is binarized. Pixels with grayscale levels greater than the threshold are designated as cloud regions, and pixels with grayscale levels less than or equal to the threshold are designated as non-cloud regions.

[0034] Furthermore, Gaussian filtering is used to remap the ground truth image to obtain the remapped image, including:

[0035] Obtain the control point image corresponding to the ground truth image;

[0036] Based on the control point image corresponding to the true image, read the latitude and longitude range of the true image and find the corresponding region in the latitude and longitude coordinate map of the corresponding load;

[0037] Within the region corresponding to the satellite payload, find the corresponding coordinate position on the ground truth image pixel by pixel;

[0038] Based on the corresponding coordinates on the ground truth image, Gaussian filtering is used to extract the pixel mean, and the pixel mean is projected onto the coordinates of the corresponding position on the satellite payload to obtain the remapped image.

[0039] Furthermore, feature points are extracted from both cloud and non-cloud areas of the image, as well as from the remapped image, including:

[0040] Acquire images of cloud and non-cloud areas, and the remapped images;

[0041] Based on cloud and non-cloud images and remapped images, interest points in the image that are invariant to scale and rotation are detected at different scales using the difference of Gaussian function to obtain extreme points;

[0042] Based on the extreme points, the location and scale of key points are determined by fitting a fine-grained model;

[0043] Based on the gradient direction of the local image, a principal direction is assigned to each key point;

[0044] Within the neighborhood of the keypoint, the local gradient of the image is measured at a selected scale. The descriptor of the keypoint is generated by calculating the local gradient of the image to obtain the feature points of the cloud region and non-cloud region images, as well as the remapped image.

[0045] Furthermore, the SIFT feature point matching algorithm is used to perform feature matching to obtain matching point pairs, including:

[0046] Acquire the location, scale, principal orientation, and descriptor of feature points in cloud and non-cloud images, and the remapped image;

[0047] Calculate the similarity between each feature point in the cloud region and the non-cloud region images and all feature points in the remapped image;

[0048] Feature points are filtered based on a similarity threshold to obtain matching point pairs.

[0049] Furthermore, based on the matching point pairs, the RANSAC algorithm is used to remove outlier matching point pairs to obtain exact matching point pairs, including:

[0050] Get the matching point pairs and set them as the dataset;

[0051] Randomly select the minimum required number of data points from the dataset as the initial model;

[0052] Based on the initial model, the parameters of the model are estimated using selected data points to obtain the estimated model;

[0053] Calculate the fitting error between the remaining matching point pairs in the dataset and the estimated model;

[0054] Based on a preset threshold, points with a fitting error less than the estimated model are marked as inliers, and points with a fitting error greater than the threshold are marked as outliers.

[0055] Compare the number of interior points of each estimated model and select the model with the most interior points as the best model;

[0056] Repeat the above steps until the preset maximum number of iterations is reached or the number of interior points meets the preset conditions. Select the model with the most interior points as the final exact matching model.

[0057] Using the final exact matching model, outliers are removed to obtain exact matching point pairs.

[0058] Furthermore, based on the geographic location information of the ground truth image and the geographic location information of the satellite image, the location information of the exact matching point pairs is extracted and converted into positioning error, including:

[0059] Obtain the precise matching points between the ground truth image and the satellite image, and obtain the latitude and longitude location information of the matching points;

[0060] Based on the latitude and longitude location information of the matching point, find four grid points in the neighborhood of the matching point and obtain the latitude and longitude coordinates of these four grid points;

[0061] Based on the latitude and longitude location information of the matching point and the latitude and longitude grid of the payload image, the position of the matching point in sub-pixel coordinates is calculated;

[0062] Based on the sub-pixel coordinates of the matching points, the elevation information of the matching points in the sub-pixel coordinates is fitted.

[0063] Calculate the latitude and longitude coordinates of the matching point in the subpixel coordinates.

[0064] The difference between the grid coordinates of the matching point and its position in the ground truth image is used to obtain the positioning error along the track and along the track direction.

[0065] Furthermore, the positioning accuracy is checked based on the positioning error to obtain the positioning accuracy check results, including:

[0066] Obtain the positioning error along the track and in the track direction, and calculate the statistics of the positioning error;

[0067] The positioning accuracy test results are obtained based on the statistics of the positioning error.

[0068] The above-described solution of the present invention has at least the following beneficial effects:

[0069] The above-mentioned solution of the present invention addresses the problem that the current positioning accuracy mainly depends on the coastline. It uses ground-value image reprojection based on Gaussian filter for remapping, SIFT algorithm for feature point matching, and RANSAC algorithm for outlier removal, thus realizing an automated evaluation method for geolocation accuracy of satellite remote sensing images based on a high-precision control point image library. Attached Figure Description

[0070] Figure 1 This is a flowchart illustrating the geolocation accuracy analysis method for medium-resolution satellite imagery provided in an embodiment of the present invention. Detailed Implementation

[0071] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0072] like Figure 1 As shown, an embodiment of the present invention proposes a method for analyzing the geolocation accuracy of medium-resolution satellite imagery, the method comprising:

[0073] Step 11: Obtain the control point images corresponding to the ground truth images and the geographic location information of the ground truth images;

[0074] Step 12: Obtain satellite imagery from the reference channel and satellite imagery location information;

[0075] Step 13: Use the Otsu thresholding method to segment the satellite image to obtain images of cloud areas and non-cloud areas;

[0076] Step 14: Use Gaussian filtering to remap the ground truth image to obtain the remapped image;

[0077] Step 15: Extract feature points from the cloud region and non-cloud region images, and from the remapped image;

[0078] Step 16: Use the SIFT feature point matching algorithm to perform feature matching to obtain matching point pairs;

[0079] Step 17: Based on the matching point pairs, use the RANSAC algorithm to remove outlier matching point pairs to obtain accurate matching point pairs;

[0080] Step 18: Based on the geographic location information of the ground truth image and the geographic location information of the satellite image, extract the location information of the exact matching point pairs and convert it into positioning error;

[0081] Step 19: Perform a positioning accuracy check based on the positioning error to obtain the positioning accuracy check result.

[0082] In this embodiment of the invention, positioning accuracy testing is used to evaluate the accuracy and precision of satellite images. The positioning accuracy test yields the positioning error, i.e., the difference between the satellite image and the ground truth image, thus evaluating the accuracy of the positioning algorithm. The positioning accuracy test result reflects the accuracy of satellite image positioning, thereby assessing the usability and applicability of the satellite image. If the positioning accuracy test result is large, indicating a significant difference between the position of the satellite image and the position of the ground truth image, further improvement and optimization of the positioning algorithm are needed. Positioning accuracy testing evaluates the merits of the positioning algorithm and provides a basis for improving it. By analyzing and comparing positioning errors, problems and shortcomings in the positioning algorithm can be identified, allowing for targeted improvements and optimizations to enhance the positioning accuracy of satellite images. The results of the positioning accuracy test can also be used to compare different positioning algorithms, identify the optimal positioning algorithm, and provide more accurate and reliable satellite image positioning.

[0083] like Figure 1 As shown, step 11 involves obtaining the control point image corresponding to the ground truth image and the geographic location information of the ground truth image from the high-precision control point image database, including:

[0084] Step 111: Obtain the ground truth image from the high-precision control point image library:

[0085] Step 112: Based on the high-precision control point image library, retrieve the corresponding control point image of the ground truth image to obtain the control point image corresponding to the ground truth image.

[0086] Step 113: Extract geographic location information from the ground truth image and the corresponding control point image to obtain geographic location information of the ground truth image.

[0087] In this embodiment of the invention, obtaining the control point image corresponding to the ground truth image and the geographic location information of the ground truth image from the high-precision control point image library provides a reference and basis for subsequent positioning accuracy verification. By obtaining the control point image corresponding to the ground truth image, a reference image of the ground truth image in the high-precision control point image library can be obtained, which can be used as a standard for comparison with the ground truth image. By comparing the ground truth image and the corresponding control point image, the positioning accuracy and precision of the satellite image can be evaluated. Simultaneously, geographic location information can be extracted from the ground truth image and the corresponding control point image to obtain the geographic location coordinates of the ground truth image. The geographic location coordinates can be used in subsequent positioning accuracy verification, compared with the geographic location of the satellite image, and the positioning error and positioning accuracy can be calculated. By comparing the geographic location information of the ground truth image and the satellite image, the positioning accuracy and positioning error of the satellite image can be evaluated. Therefore, the effect of obtaining the control point image corresponding to the ground truth image and the geographic location information of the ground truth image from the high-precision control point image library is to provide accurate ground truth image and geographic location information for subsequent positioning accuracy verification, thereby evaluating the positioning accuracy and precision of the satellite image.

[0088] like Figure 1 As shown, step 12 involves obtaining satellite imagery from the payload database to be evaluated, including reference channel imagery and geographic location information of the satellite images.

[0089] Step 121: Obtain satellite images from the payload database to be evaluated;

[0090] Step 122: Based on the payload database to be evaluated, retrieve the reference channel image of the satellite image to obtain the reference channel image of the satellite image.

[0091] Step 123: Extract geographic location information from the satellite image reference channel image to obtain satellite image geographic location information.

[0092] In this embodiment of the invention, satellite imagery reference channel images and satellite imagery geographic location information are obtained from the payload database to be evaluated, providing a reference and basis for subsequent positioning accuracy verification. By obtaining the satellite imagery reference channel images, a reference image of the satellite imagery in the payload database to be evaluated can be obtained, which can be used as a benchmark for comparison. By comparing the satellite imagery and the reference channel images, the positioning accuracy and precision of the satellite imagery can be evaluated. Simultaneously, geographic location information is extracted from the satellite imagery reference channel images to obtain the geographic location coordinates of the satellite imagery. The geographic location coordinates can be used in subsequent positioning accuracy verification, compared with the geographic location of the ground truth image, to calculate the positioning error and positioning accuracy. By comparing the geographic location information of the satellite imagery and the ground truth image, the positioning accuracy and positioning error of the satellite imagery can be evaluated.

[0093] like Figure 1 As shown, step 13 involves segmenting the satellite image using the Otsu thresholding method to obtain images of cloud and non-cloud regions, including:

[0094] Step 131: Acquire satellite imagery reference channel imagery;

[0095] Step 132: Convert the satellite image reference channel image into a grayscale image;

[0096] Step 133: Calculate the histogram of the gray level of each pixel in the grayscale image to obtain the number of pixels at each gray level in the image.

[0097] Step 134: Divide the number of pixels at each gray level by the total number of pixels to obtain the pixel probability distribution at each gray level.

[0098] Step 135: Using Otsu's thresholding method, a threshold is set to segment the grayscale image into two categories. The variance between the two categories after segmentation is calculated to obtain the inter-class variance.

[0099] Between-class variance = w0 × w1 × (μ0 - μ1) 2

[0100] Where w0 and w1 are the sum of pixel probabilities of category 0 and category 1, respectively, and μ0 and μ1 are the average gray values ​​of category 0 and category 1, respectively;

[0101] Step 136: Based on the inter-class variance, select the threshold that maximizes the inter-class variance as the optimal threshold.

[0102] Step 137: Binarize the grayscale image according to the optimal threshold. Pixels with grayscale levels greater than the threshold are set as cloud regions, and pixels with grayscale levels less than or equal to the threshold are set as non-cloud regions.

[0103] In this embodiment of the invention, the Otsu thresholding method is used to segment satellite images into cloud regions and non-cloud regions, extracting information from the cloud regions to provide a basis for subsequent cloud detection and positioning accuracy verification. The Otsu thresholding method is an adaptive threshold selection method that determines the optimal threshold by calculating the inter-class variance. It can automatically select the optimal threshold based on the image's grayscale distribution, maximizing the variance between the two categories after segmentation, thereby achieving effective image segmentation. After converting the satellite image reference channel image into a grayscale image, the number of pixels at each grayscale level can be obtained by calculating the histogram of the grayscale levels of each pixel in the grayscale image. Then, based on the number of pixels and the total image... The number of pixels is calculated, and the pixel probability distribution at each gray level is determined. Using the Otsu thresholding method, the inter-class variance is calculated by iterating through the thresholds for different gray levels. The threshold that maximizes the inter-class variance is selected as the optimal threshold. Based on the optimal threshold, the grayscale image is binarized, with pixels having a gray level greater than the threshold designated as cloud regions and pixels having a gray level less than or equal to the threshold designated as non-cloud regions. Image segmentation using the Otsu thresholding method effectively separates cloud and non-cloud regions in satellite images. The resulting cloud and non-cloud region images can be used for subsequent cloud detection and positioning accuracy verification, further analyzing and evaluating the cloud conditions and positioning accuracy of satellite images.

[0104] like Figure 1 As shown, step 14 involves remapping the ground truth image using Gaussian filtering to obtain the remapped image, including:

[0105] Step 141: Obtain the control point image corresponding to the ground truth image;

[0106] Step 142: Based on the control point image corresponding to the true image, read the latitude and longitude range of the true image and find the corresponding area in the latitude and longitude coordinate map of the corresponding load.

[0107] Step 143: In the region corresponding to the satellite payload, find the corresponding coordinate position on the ground truth image pixel by pixel;

[0108] Step 144: Based on the corresponding coordinates on the ground truth image, Gaussian filtering is used to extract the pixel mean, and the pixel mean is projected onto the coordinates of the corresponding position on the satellite payload to obtain the remapped image.

[0109] In this embodiment of the invention, Gaussian filtering is used to remap the ground truth image to project the pixel values ​​of the ground truth image from their positions on the ground truth image to their positions on the corresponding satellite payload image, so as to obtain the remapped image. This allows for the acquisition of pixel values ​​corresponding to the ground truth image from the satellite payload image, which are then used for subsequent feature extraction and matching. Before remapping the ground truth image, it is necessary to first acquire the control point image corresponding to the ground truth image. By reading the latitude and longitude range of the ground truth image and finding the corresponding region in the latitude and longitude coordinate map of the corresponding payload, the region in the satellite payload image corresponding to the ground truth image can be determined. Then, pixel by pixel, the corresponding coordinate position in the ground truth image is found in the region corresponding to the satellite payload. For each corresponding coordinate in the ground truth image, a Gaussian filter is used to extract the mean of its surrounding pixels, and this mean is projected onto the coordinates of the corresponding position on the satellite payload to obtain the remapped image. This allows for the acquisition of a satellite payload image corresponding to the ground truth image, ensuring consistency in pixel values ​​between the two. Through the remapped image, the pixel values ​​corresponding to the ground truth image can be acquired from the satellite payload image, thereby achieving comparison and matching between the two. This is crucial for subsequent feature extraction and matching algorithms, improving the accuracy and precision of matching and further evaluating the positioning accuracy of the satellite image.

[0110] like Figure 1 As shown, step 15 is characterized by extracting feature points from the cloud region and non-cloud region images, and the remapped image, including:

[0111] Step 151: Obtain images of the cloud region and the non-cloud region, and the remapped image;

[0112] Step 152: Based on the cloud region and non-cloud region images and the remapped image, detect the scale- and rotation-invariant interest points in the image at different scales using the difference of Gaussian function to obtain the extreme points;

[0113] Step 153: Based on the extreme points, determine the location and scale of the key points by fitting a fine-grained model;

[0114] Step 154: Assign a principal direction to each keypoint based on the local gradient direction of the image;

[0115] Step 155: Measure the local gradient of the image at a selected scale within the neighborhood of the key point, and generate the descriptor of the key point by calculating the local gradient of the image to obtain the feature points of the cloud region and non-cloud region images, and the remapped image.

[0116] In this embodiment of the invention, feature points are extracted from cloud and non-cloud region images and the remapped image to provide keypoint locations and descriptors for subsequent feature matching, used for cloud detection and positioning accuracy verification. First, cloud and non-cloud region images and the remapped image are acquired. The image contains information about the cloud and non-cloud regions, as well as the image after remapping the ground truth image to the satellite image. Then, by using the difference of Gaussian function at different scales to detect interest points in the image, extreme points can be obtained. Extreme points have scale and rotation invariance and can be used as keypoints. The location and scale of the keypoints are determined by fitting a refined model, which is achieved by calculating the pixel gradient and Hessian matrix of the region surrounding the keypoint. To accurately determine the location and scale of key points; to ensure the subsequent descriptor generation process is invariant to rotational changes, a principal direction is assigned to each key point based on the gradient direction histogram of the region surrounding the key point; finally, the local gradient of the image in the neighborhood around the key point is measured at the selected scale, and the image local gradient is calculated to generate the descriptor of the key point; the descriptor contains feature information of the region surrounding the key point, used to represent the features of the key point; by extracting feature points from cloud and non-cloud images and the remapped image, these feature points can be used for matching and comparison in subsequent feature matching; this helps in cloud detection and positioning accuracy verification, providing the location and descriptor of the key points for evaluating cloud conditions and positioning accuracy in satellite imagery.

[0117] like Figure 1 As shown, step 16 is characterized by using the SIFT feature point matching algorithm to perform feature matching to obtain matching point pairs, including:

[0118] Step 161: Obtain the location, scale, principal orientation, and descriptor of feature points in the cloud region and non-cloud region images, and the remapped image;

[0119] Step 162: Calculate the similarity between each feature point in the cloud region and non-cloud region images and all feature points in the remapped image;

[0120] Step 163: Based on the similarity threshold, feature points are filtered to obtain matching point pairs.

[0121] In this embodiment of the invention, feature matching using the SIFT feature point matching algorithm involves matching feature points from cloud and non-cloud region images with feature points from the remapped image to find corresponding matching point pairs for cloud detection and positioning accuracy verification. First, the position, scale, principal direction, and descriptors of feature points from the cloud and non-cloud region images and the remapped image are obtained; feature points contain local feature information of the image. Then, the similarity between each feature point in the cloud and non-cloud region images and all feature points in the remapped image is calculated; Euclidean distance is used to calculate the similarity. By calculating the similarity between two feature descriptors, the similarity of each feature point can be found. The optimal matching point in the remapped image; feature point filtering based on a similarity threshold to obtain matching point pairs; by setting a similarity threshold, matching points with low similarity can be filtered out, retaining only those with high similarity; this can improve the accuracy and reliability of matching; feature matching using the SIFT feature point matching algorithm can match feature points of cloud and non-cloud areas with feature points of the remapped image to find corresponding matching point pairs; matching point pairs can be used for cloud detection and positioning accuracy verification, evaluating the cloud conditions and positioning accuracy of satellite images; the quantity and quality of matching point pairs can serve as important indicators for evaluating positioning algorithms and image quality.

[0122] like Figure 1 As shown, in step 17, outlier matching pairs are removed using the RANSAC algorithm to obtain exact matching pairs, including:

[0123] Step 171: Obtain the matching point pairs and set them as the dataset;

[0124] Step 172: Randomly select the minimum required number of data points from the dataset as the initial model;

[0125] Step 173: Based on the initial model, estimate the parameters of the model using the selected data points to obtain the estimated model;

[0126] Step 174: Calculate the fitting error between the remaining matching point pairs in the dataset and the estimated model;

[0127] Step 175: According to the preset threshold, points with fitting errors to the estimated model less than the threshold are marked as inliers, and points with fitting errors greater than the threshold are marked as outliers.

[0128] Step 176: Compare the number of interior points of each estimated model and select the model with the most interior points as the best model.

[0129] Step 177: Repeat the above steps until the preset maximum number of iterations is reached or the number of interior points meets the preset conditions. Select the model with the most interior points as the final exact matching model.

[0130] Step 178: Using the final exact matching model, outliers are removed to obtain exact matching point pairs.

[0131] In this embodiment of the invention, the RANSAC algorithm is used to remove outlier matching pairs iteratively to obtain accurate matching pairs. This improves the accuracy and reliability of matching and reduces the interference of outliers on the matching results. First, the matching pairs are set as a dataset; in the RANSAC algorithm, the dataset contains all matching pairs. Then, the minimum required number of data points, such as two data points, are randomly selected from the dataset as the initial model. These data points are used to estimate the initial parameters of the model. Based on the initial model, the selected data points are used to estimate the parameters of the model to obtain the estimated model. For example, in the case of fitting a straight line, the slope and intercept of the line can be estimated using the selected data points. The fitting error between the remaining matching pairs in the dataset and the estimated model is calculated. A measurement method for the fitting error, such as the distance from a point to a line, can be used. Based on a preset threshold, the matching pairs are compared with the estimated model. Points with fitting errors less than a threshold are marked as inliers, and points with fitting errors greater than a threshold are marked as outliers. Then, the number of inliers in each estimated model is compared, and the model with the most inliers is selected as the best model, as models with more inliers have a higher probability of being correct. This process is repeated until a preset maximum number of iterations is reached or the number of inliers meets a preset condition. In each iteration, the model with the most inliers is selected as the current best model. Finally, the best model is used to remove outliers, resulting in accurate matching pairs. These matching pairs are the accurate matching results after outlier removal using the RANSAC algorithm. By using the RANSAC algorithm to remove outlier matching pairs, the interference of incorrect matches and outliers can be eliminated, resulting in more accurate and reliable matching pairs.

[0132] like Figure 1 As shown, step 18 involves extracting the location information of precisely matched point pairs based on the geographic location information of the ground truth image and the geographic location information of the satellite image, and converting this information into positioning error, including:

[0133] Step 181: Obtain the precise matching points between the ground truth image and the satellite image, and obtain the latitude and longitude location information of the matching points;

[0134] Step 182: Based on the latitude and longitude location information of the matching point, find four grid points in the neighborhood of the location of the matching point, and obtain the latitude and longitude coordinates of these four grid points;

[0135] Step 183: Calculate the position of the matching point in sub-pixel coordinates based on the latitude and longitude location information of the matching point and the latitude and longitude grid of the load image;

[0136] Step 184: Fit the elevation information of the matching point in the sub-pixel coordinates based on the sub-pixel coordinates of the matching point.

[0137] Step 185: Calculate the latitude and longitude coordinates of the matching point in the sub-pixel coordinates based on the sub-pixel coordinates of the matching point.

[0138] Step 186: Subtract the grid coordinates of the matching point from its position in the ground truth image to obtain the positioning error along the track and along the track direction.

[0139] In this embodiment of the invention, based on the geographic location information of the ground truth image and the geographic location information of the satellite image, the location information of the precise matching point pairs is extracted and converted into positioning error to evaluate the positioning accuracy of the satellite image and determine the positioning difference between the satellite image and the ground truth image. First, the precise matching points of the ground truth image and the satellite image are obtained, and the latitude and longitude location information of the matching points is obtained. The matching points are the precise matching results obtained through the previous processing steps. Then, based on the latitude and longitude location information of the matching points, four grid points in the neighborhood of the location of the matching point are found, and the latitude and longitude coordinates of these four grid points are obtained. The grid points can be used as reference points to calculate the position of the matching point in subpixel coordinates. Then, based on the latitude and longitude location information of the matching points and the latitude and longitude grid of the payload image, the position of the matching point in subpixel coordinates can be calculated. Location; subpixel coordinates of the matching point can be obtained through interpolation or other methods; based on the subpixel coordinates of the matching point, the elevation information of the matching point is fitted in the subpixel coordinates, and the fitted model is used to estimate the elevation value of the matching point; finally, based on the subpixel coordinates of the matching point, the latitude and longitude coordinates of the matching point in the subpixel coordinates can be calculated; the subpixel coordinates can be converted into latitude and longitude coordinates through inverse projection or other methods; by subtracting the grid coordinate position of the matching point from its position in the ground truth image, the positioning error of the matching point in the direction of intersection with and along the track can be obtained; the positioning error can be used to evaluate the positioning accuracy of satellite images and determine the positioning difference between satellite images and ground truth images; the magnitude and direction of the positioning error can provide information about satellite image positioning, which can be used to evaluate and improve the accuracy and reliability of positioning algorithms.

[0140] Specifically, let the latitude and longitude of the matching point be (lat0, log0), and its four neighboring grid points (x, y, y) be... i y i The latitude and longitude coordinates corresponding to i = 1, 2, 3, 4 are respectively (lat i log i If i = 1, 2, 3, 4, and the sub-pixel coordinates of its grid point are (x0, y0), then we have

[0141]

[0142] Among them, latx =lat2-lat1,lat y =lat3-lat1, log x =log2-log1, log y =log3-log1

[0143] Therefore, the grid coordinates corresponding to (lat0, log0) are (x0+x1, y0+y1). After obtaining the grid coordinates of the corresponding point pair, the difference between their coordinate positions is used to obtain the intersection and track error.

[0144] like Figure 1 As shown, step 19 involves performing a positioning accuracy check based on the positioning error to obtain the positioning accuracy check result, including:

[0145] Step 191: Obtain the positioning error in the direction of intersection and along the track, and calculate the statistics of the positioning error;

[0146] Step 192: Based on the statistics of the positioning error, obtain the positioning accuracy test results.

[0147] In this embodiment of the invention, positioning accuracy testing based on positioning errors can evaluate the positioning accuracy of satellite images and obtain positioning accuracy test results, which are used to assess the positioning accuracy and reliability of satellite images. First, the positioning errors in the orbital and along-orbit directions are obtained. The positioning errors in the orbital and along-orbit directions can be obtained by subtracting the grid coordinates of the matching points from their positions in the ground truth image. Then, the statistics of the positioning errors are calculated. The average, standard deviation, maximum, and minimum values ​​of the positioning errors can be calculated, as well as other relevant statistical indicators as needed. These statistics can provide an overall picture and distribution characteristics of the positioning errors in the satellite images. Based on the statistics of the positioning errors, the positioning accuracy test results can be obtained. By comparing... By comparing the statistical measures of the positioning error with the predetermined accuracy requirements or standards, we can assess whether the positioning accuracy of the satellite image meets the requirements. If the positioning error is small and all statistical indicators meet the accuracy requirements, the positioning accuracy of the satellite image can be considered high. Conversely, if the positioning error is large, or some statistical indicators do not meet the requirements, further improvement and optimization of the positioning algorithm are needed to improve the positioning accuracy of the satellite image. The positioning accuracy test results can provide a quantitative assessment of the positioning accuracy of the satellite image, helping users understand the positioning accuracy of the satellite image and providing a reliable reference for the application of the satellite image. Through the positioning accuracy test results, we can determine whether the satellite image is suitable for a specific application scenario and whether further improvement and optimization of the positioning algorithm is needed.

[0148] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for analyzing the geolocation accuracy of medium-resolution satellite imagery, characterized in that, The method includes: Obtain the control point images corresponding to the ground truth images and the geographic location information of the ground truth images; Acquire satellite imagery reference channel images and satellite imagery geographic location information; The Otsu thresholding method was used to segment the satellite image to obtain images of cloud areas and non-cloud areas; The ground truth image is remapped using a Gaussian filter to obtain the remapped image. Extract feature points from cloud and non-cloud areas of the image, and from the remapped image; The SIFT feature point matching algorithm is used to perform feature matching to obtain matching point pairs; Based on the matching point pairs, the RANSAC algorithm is used to remove outlier matching point pairs in order to obtain accurate matching point pairs; Based on the geographic location information of the ground truth image and the geographic location information of the satellite image, the location information of the exact matching point pairs is extracted and converted into positioning error; The positioning accuracy is checked based on the positioning error to obtain the positioning accuracy check result; The process involves remapping the ground truth image using Gaussian filtering to obtain the remapped image, including: Obtain the control point image corresponding to the ground truth image; Based on the control point image corresponding to the true image, read the latitude and longitude range of the true image and find the corresponding region in the latitude and longitude coordinate map of the corresponding load; Within the region corresponding to the satellite payload, find the corresponding coordinate position on the ground truth image pixel by pixel; Based on the corresponding coordinates on the ground truth image, Gaussian filtering is used to extract the pixel mean, and the pixel mean is projected onto the coordinates of the corresponding position on the satellite payload to obtain the remapped image. Based on the geographic location information from the ground truth image and satellite image, the location information of the exact matching point pairs is extracted and converted into positioning error, including: Obtain the precise matching points between the ground truth image and the satellite image, and obtain the latitude and longitude location information of the matching points; Based on the latitude and longitude location information of the matching point, find four grid points in the neighborhood of the matching point and obtain the latitude and longitude coordinates of these four grid points; Based on the latitude and longitude location information of the matching point and the latitude and longitude grid of the payload image, the position of the matching point in sub-pixel coordinates is calculated; Based on the sub-pixel coordinates of the matching points, the elevation information of the matching points in the sub-pixel coordinates is fitted. Calculate the latitude and longitude coordinates of the matching point in the subpixel coordinates. The difference between the grid coordinates of the matching point and its position in the ground truth image is used to obtain the positioning error along the track and along the track direction.

2. The geolocation accuracy analysis method for medium-resolution satellite imagery according to claim 1, characterized in that, Obtain the ground truth image of the corresponding control point and its geographic location information from the high-precision control point image database, including: Obtain ground truth images from a high-precision control point image library: Based on the high-precision control point image database, the corresponding control point image of the ground truth image is retrieved to obtain the control point image corresponding to the ground truth image. Geographic location information is extracted from the ground truth image and the corresponding control point image to obtain the geographic location information of the ground truth image.

3. The geolocation accuracy analysis method for medium-resolution satellite imagery according to claim 2, characterized in that, Obtain satellite imagery reference channel images and satellite imagery geographic location information from the payload database to be evaluated, including: Obtain satellite images from the database of loads to be evaluated; Based on the payload database to be evaluated, retrieve the reference channel imagery of the satellite imagery to obtain the reference channel imagery of the satellite imagery; Geographic location information is extracted from the satellite imagery reference channel to obtain the satellite imagery geographic location information.

4. The geolocation accuracy analysis method for medium-resolution satellite imagery according to claim 3, characterized in that, The Otsu thresholding method was used to segment the satellite imagery to obtain images of cloud and non-cloud regions, including: Acquire satellite imagery reference channel images; Convert the satellite imagery reference channel to a grayscale image; Calculate the histogram of gray levels for each pixel in a grayscale image to obtain the number of pixels at each gray level in the image; Divide the number of pixels at each gray level by the total number of pixels to obtain the pixel probability distribution at each gray level; Using Otsu's thresholding method, a threshold is set to segment the grayscale image into two categories. The variance between the two categories after segmentation is calculated to obtain the inter-class variance. ， in, and These are the sums of pixel probabilities for category 0 and category 1, respectively. and These are the average gray values ​​for category 0 and category 1, respectively. Based on the inter-class variance, select the threshold that maximizes the inter-class variance as the optimal threshold. Based on the optimal threshold, the grayscale image is binarized. Pixels with grayscale levels greater than the threshold are designated as cloud regions, and pixels with grayscale levels less than or equal to the threshold are designated as non-cloud regions.

5. The method for analyzing the geolocation accuracy of medium-resolution satellite imagery according to claim 1, characterized in that, Feature points are extracted from cloud and non-cloud areas of the image, as well as from the remapped image, including: Acquire images of cloud and non-cloud areas, and the remapped images; Based on cloud and non-cloud images and remapped images, interest points in the image that are invariant to scale and rotation are detected at different scales using the difference of Gaussian function to obtain extreme points; Based on the extreme points, the location and scale of key points are determined by fitting a fine-grained model; Based on the gradient direction of the local image, a principal direction is assigned to each key point; Within the neighborhood of the keypoint, the local gradient of the image is measured at a selected scale. The descriptor of the keypoint is generated by calculating the local gradient of the image to obtain the feature points of the cloud region and non-cloud region images, as well as the remapped image.

6. The method for analyzing the geolocation accuracy of medium-resolution satellite imagery according to claim 5, characterized in that, Feature matching is performed using the SIFT feature point matching algorithm to obtain matching point pairs, including: Acquire the location, scale, principal orientation, and descriptor of feature points in cloud and non-cloud images, and the remapped image; Calculate the similarity between each feature point in the cloud region and the non-cloud region images and all feature points in the remapped image; Feature points are filtered based on a similarity threshold to obtain matching point pairs.

7. The method for analyzing the geolocation accuracy of medium-resolution satellite imagery according to claim 6, characterized in that, Based on the matching point pairs, the RANSAC algorithm is used to remove outlier matching point pairs to obtain exact matching point pairs, including: Get the matching point pairs and set them as the dataset; Randomly select the minimum required number of data points from the dataset as the initial model; Based on the initial model, the parameters of the model are estimated using selected data points to obtain the estimated model; Calculate the fitting error between the remaining matching point pairs in the dataset and the estimated model; Based on a preset threshold, points with a fitting error less than the estimated model are marked as inliers, and points with a fitting error greater than the threshold are marked as outliers. Compare the number of interior points of each estimated model and select the model with the most interior points as the best model; Repeat the above steps until the preset maximum number of iterations is reached or the number of interior points meets the preset conditions. Select the model with the most interior points as the final exact matching model. Using the final exact matching model, outliers are removed to obtain exact matching point pairs.

8. The method for analyzing the geolocation accuracy of medium-resolution satellite imagery according to claim 1, characterized in that, The positioning accuracy is verified based on the positioning error to obtain the positioning accuracy verification results, including: Obtain the positioning error along the track and in the track direction, and calculate the statistics of the positioning error; The positioning accuracy test results are obtained based on the statistics of the positioning error.

Images