An automated method for checking the geometric accuracy of remote sensing satellite images

Abstract

The present invention provides a method for automatically checking the geometric accuracy of remote sensing satellite images, the method comprising: inputting an image to be checked, automatically acquiring a corresponding reference image according to the image to be checked; obtaining a set of same-name point pairs according to the image to be checked and the reference image by a method based on phase consistency image matching, wherein the set of same-name point pairs includes at least twenty same-name point pairs, which are key points evenly distributed in the image to be checked and matching points corresponding to the key points in the reference image; calculating the positioning error of the same-name point pairs according to the positions of the key points and matching points in the set of same-name point pairs and drawing an error direction map, thereby realizing automatic inspection of the geometric accuracy of remote sensing images. This method of obtaining a set of same-name point pairs by means of phase consistency matching reduces time and material consumption compared to traditional manual point puncturing; improves matching accuracy compared to the classic SIFT matching method, and effectively improves the demand for geometric accuracy inspection of large-scale remote sensing satellite image products.

Description

Technical Field

[0001] The present invention relates to the technical field of remote sensing image processing, and in particular to a method for automatically checking the geometric accuracy of remote sensing satellite images. The method first establishes reference image data using the downloaded original high-precision reference image as a test standard, and automatically obtains the corresponding reference image through remote sensing satellite image information. Then, the same-name point pairs between the remote sensing satellite image and the reference image are obtained through automatic matching, the same-name point pair errors are calculated, and the error direction map is drawn, thereby realizing the automatic inspection of the geometric accuracy of remote sensing satellite products. This method of obtaining a set of same-name points by means of phase consistency matching reduces the time and material consumption compared to traditional manual point puncturing; it improves the matching accuracy compared to the classic SIFT matching method, and meets the geometric accuracy inspection requirements of large-scale remote sensing satellite image products. Background Art

[0002] The ground positioning accuracy of remote sensing image products is an important indicator for measuring image data products. Quantitative evaluation of product positioning accuracy is of great significance for improving the geometric quality of remote sensing images and remote sensing applications.

[0003] With the advent of the global observation era, multi-source, multi-temporal and multi-resolution remote sensing data has grown rapidly, bringing about the diversification and mass of remote sensing data. The traditional method of geometric accuracy inspection uses field GPS measurement, which has a low degree of automation and is time-consuming, and is difficult to meet the accuracy quality inspection needs of remote sensing products with large amounts of data; and the traditional SIFT matching method to obtain homonymous points has the problem of mismatching or uneven distribution of homonymous points due to image radiation distortion and geometric distortion.

[0004] Therefore, the need for automated accuracy inspection of remote sensing satellite image products with large data volumes remains a problem that needs to be solved urgently. Summary of the invention

[0005] The present invention provides a method for automatically checking the geometric accuracy of remote sensing satellite images, which is used to solve the defects of the prior art, such as low automation, long time consumption, and difficulty in meeting the requirements for precision quality inspection of remote sensing products with large amounts of data, thereby improving the inspection efficiency and meeting the requirements for automated inspection of the geometric accuracy of remote sensing images with large amounts of satellite products.

[0006] The present invention provides a method for automatically checking the geometric accuracy of remote sensing satellite images, comprising:

[0007] Input an image to be inspected, and automatically obtain a corresponding reference image according to the image to be inspected;

[0008] Obtaining a set of same-name point pairs according to the image to be tested and the reference image by a method based on phase congruency image matching, wherein the set of same-name point pairs includes at least twenty same-name point pairs, and the same-name point pairs are key points evenly distributed in the image to be tested and matching points corresponding to the key points in the reference image;

[0009] The positioning error of the same-name point pairs is calculated according to the key points and the matching points in the same-name point pair set, and an error direction map is drawn.

[0010] According to a method for automatically checking the geometric accuracy of remote sensing satellite images provided by the present invention, the step of automatically acquiring a corresponding reference image based on the image to be checked comprises:

[0011] Acquiring original reference image data according to the attribute information of the image to be inspected and a preset mapping relationship;

[0012] According to the image to be inspected, automatically matching one or more associated reference images covered by the geographical location of the image to be inspected in the original reference image data;

[0013] Preprocessing the one or more associated reference images to obtain the reference image.

[0014] According to a method for automatically checking the geometric accuracy of remote sensing satellite images provided by the present invention, the step of obtaining a set of point pairs with the same name based on the image to be checked and the reference image by a method based on phase consistency image matching comprises:

[0015] Detecting the image to be tested by a non-maximum suppression detection algorithm, and determining the largest maximum point in a preset area of ​​the image to be tested as the key point;

[0016] Dividing the image to be inspected and the reference image into uniform grids in sequence to obtain one-to-one corresponding sub-image blocks to be inspected and reference sub-image blocks;

[0017] Extracting edge feature image 1 of the sub-image block to be inspected and edge feature image 2 of the reference sub-image block using a phase consistency model; calculating phase correlation pulses of the edge feature image 1 and the edge feature image 2; and obtaining a relative offset between the sub-image block to be inspected and the corresponding reference sub-image block according to the phase correlation pulses;

[0018] Preliminarily determining the predicted matching point position of the key point in the reference image according to the relative offset between the key point and the sub-image block to be checked;

[0019] Taking the key point and the predicted matching point as the center respectively, extracting the directional phase consistency feature image of the discrete direction set in the neighborhood based on the directional phase consistency model, and obtaining the key point feature descriptor and the predicted point feature descriptor;

[0020] Transforming the key point feature descriptor and the predicted matching point feature descriptor into the frequency domain, calculating the similarity between the key point feature descriptor and the predicted matching point feature descriptor; and determining an initial matching point according to the similarity;

[0021] The residual model parameters are calculated using the least squares method according to the key points and the initial matching points; the residuals of the key points are calculated according to the residual model; and a set of same-name point pairs is obtained using the key points whose residuals are less than a preset threshold and the corresponding initial matching points as the key points in a set of same-name point pairs and the matching points corresponding to the key points.

[0022] According to a remote sensing satellite image geometric accuracy inspection method provided by the present invention, the step of calculating the positioning error of the same-name point pair according to the key points and the matching points in the same-name point pair set and drawing the error direction map comprises:

[0023] Calculate the positioning error of the same-name point pair according to the key points and the matching points in the same-name point pair set;

[0024] Calculate the mean error based on the positioning error of the same-name point pair;

[0025] An error direction map is generated according to the mean error, the positioning error of the same-name point pair, the key point, the matching point and the image to be checked.

[0026] The automated inspection method for geometric accuracy of remote sensing satellite images provided by the present invention inputs an image to be inspected, and automatically obtains a corresponding reference image according to the image to be inspected; obtains a set of same-name point pairs according to the image to be inspected and the reference image through a method based on phase consistency image matching, wherein the set of same-name point pairs includes at least twenty same-name point pairs, and the same-name points are key points evenly distributed in the image to be inspected and matching points corresponding to the key points in the reference image; calculates the positioning error of the same-name point pairs according to the key points and the matching points in the same-name point pair set and draws an error direction map. Compared with traditional geometric accuracy inspection, this method improves inspection efficiency and meets the accuracy inspection requirements of remote sensing satellite image products with large amounts of data. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0028] Figure 1 This is one of the flow charts of the method for automatically checking the geometric accuracy of remote sensing satellite images provided by the present invention;

[0029] Figure 2 This is the second flow chart of the automated inspection method for geometric accuracy of remote sensing satellite images provided by the present invention;

[0030] Figure 3 This is the third flow chart of the automated inspection method for geometric accuracy of remote sensing satellite images provided by the present invention;

[0031] Figure 4 is a schematic diagram of a flow chart for calculating a relative offset provided by the present invention;

[0032] Figure 5 It is a schematic diagram of the process of calculating matching points provided by the present invention;

[0033] Figure 6 This is an example diagram of the error direction diagram provided by the present invention. DETAILED DESCRIPTION

[0034] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0035] The ground positioning accuracy of remote sensing image products is an important indicator for measuring data products. Quantitative evaluation of product positioning accuracy is of great significance for improving the geometric quality of remote sensing images and remote sensing applications.

[0036] With the advent of the global observation era, multi-source, multi-temporal and multi-resolution remote sensing data has grown rapidly, bringing about the diversification and mass of remote sensing data. The traditional geometric accuracy inspection method uses field GPS measurement, which has a low degree of automation and is time-consuming, making it difficult to meet the accuracy quality inspection needs of remote sensing products with large amounts of data.

[0037] To solve the above problems, the present invention provides a method for automatically checking the geometric accuracy of remote sensing satellite images. Figure 1As shown, the following steps are included:

[0038] S11, input an image to be inspected, and automatically obtain a corresponding reference image according to the image to be inspected.

[0039] S12. Obtain a set of same-name point pairs according to the image to be tested and the reference image by a phase-consistency image matching method, wherein the set of same-name point pairs includes at least twenty same-name point pairs, and the same-name point pairs are key points evenly distributed in the image to be tested and matching points corresponding to the key points in the reference image.

[0040] S13, calculating the positioning error of the same-name point pairs according to the key points and the matching points in the same-name point pair set and drawing an error direction map.

[0041] Among them, this method can quantitatively evaluate the geometric positioning accuracy of remote sensing image products based on the positioning errors of same-name point pairs.

[0042] In an embodiment of the present invention, by inputting an image to be tested, a corresponding reference image is automatically obtained according to the image to be tested; a set of same-name point pairs is obtained according to the image to be tested and the reference image through a method based on phase consistency image matching, wherein the set of same-name point pairs includes at least twenty same-name point pairs, and the same-name point pairs are key points evenly distributed in the image to be tested and matching points corresponding to the key points in the reference image; the positioning error of the same-name point pairs is calculated according to the key points and the matching points in the same-name point pair set, and an error direction map is drawn. Furthermore, the geometric positioning accuracy of remote sensing image products can be quantitatively evaluated based on the positioning error of the same-name point pairs. Compared with traditional geometric accuracy inspection, this method improves the inspection efficiency and meets the accuracy inspection requirements of remote sensing satellite image products with large amounts of data.

[0043] According to the method for automatically checking the geometric accuracy of remote sensing satellite images provided by the present invention, the step of automatically acquiring a corresponding reference image based on the image to be checked;

[0044] like Figure 2 As shown, specifically including:

[0045] S21. Acquire original reference image data according to the attribute information of the image to be inspected and a preset mapping relationship.

[0046] Specifically, the image to be inspected has attribute information, and the attribute information may include satellite name, payload name, number of bands, resolution, column width, and width.

[0047] This method can search and obtain the corresponding map image locally as the original reference image data according to the attribute information of the image to be inspected and the preset mapping relationship. If the corresponding map image is not obtained locally, the corresponding map image can be searched and obtained from the server through the Internet. When obtaining the map image, the coordinates can be selected in the WGS84 geographic coordinate system for block download and hierarchical storage.

[0048] In one example, the preset mapping relationship of the image to be inspected is shown in Table 1. The map image is a label of an image corresponding to the satellite, payload, number of bands, resolution, column width, and width, and the corresponding map image can be obtained as the original reference image data according to the map image label.

[0049] Table 1 Preset mapping relationship table

[0050]

[0051]

[0052] S22. According to the image to be inspected, automatically matching one or more associated reference images covered by the geographical location of the image to be inspected in the original reference image data.

[0053] Specifically, the method can search for one or more corresponding associated reference images from the original reference image data according to the geographical location information of the image to be inspected.

[0054] S23: Process the one or more associated reference images to obtain the reference image.

[0055] Specifically, the method can process one or more associated reference images, including but not limited to one or more of cropping, splicing, and reprojection, to finally obtain a reference image.

[0056] In an example, the input image coordinate system is a projection coordinate system (UTM 44N) with a resolution of 2.6m. The corresponding map image is determined by referring to Table 1.<level 16> , get<level 16> The corresponding map image is used as the original reference data. Four associated reference images covering the geographic location of the image to be inspected are matched in the original reference data, and the four associated reference images are cropped and spliced, and reprojected to the projection coordinate system consistent with the image to be inspected (UTM44N) to obtain the reference image.

[0057] In an embodiment of the present invention, original reference image data is obtained through attribute information of the image to be tested, and then one or more associated reference images in the original reference image data are matched according to the geographical location of the image to be tested, and the associated images are processed to obtain the final reference image, thereby easily obtaining an accurate reference image.

[0058] According to the remote sensing satellite image geometric accuracy automatic inspection method provided by the present invention, Figure 3 As shown, step S12 specifically includes the following steps:

[0059] S31, detecting the image to be tested by using a non-maximum suppression detection algorithm, and determining the largest maximum point in a preset area of ​​the image to be tested as the key point.

[0060] Specifically, the method can calculate the maximum maximum point in multiple preset areas in the image to be tested through a non-maximum suppression detection algorithm, use the maximum maximum point as a key point, and obtain multiple key points evenly distributed in the image to be tested.

[0061] S32, dividing the image to be inspected and the reference image into uniform grids in sequence to obtain one-to-one corresponding sub-image blocks to be inspected and reference sub-image blocks.

[0062] Specifically, the method divides the image to be inspected into uniform grids according to a preset size to obtain sub-image blocks to be inspected; according to the image coordinates of the geographical location of the upper left corner of the sub-image block to be inspected in the reference image as the starting point for dividing the reference sub-image blocks, the reference image is divided according to the preset size to obtain corresponding reference sub-image blocks;

[0063] Specifically, in one example, the image to be tested is gridded, and the size of each sub-image block to be tested is 500×500 pixels; the reference image is gridded according to the sub-image blocks to be tested to obtain reference sub-image blocks with a size of 500×500 pixels that correspond one-to-one to the sub-image blocks to be tested.

[0064] S33, using a phase consistency model to extract edge feature image 1 of the sub-image block to be tested and edge feature image 2 of the reference sub-image block; calculating phase correlation pulses of the edge feature image 1 and the edge feature image 2; and obtaining a relative offset between the sub-image block to be tested and the corresponding reference sub-image block based on the phase correlation pulses.

[0065] Specifically, the method can use the phase consistency model to extract a matrix for representing the edge feature image 1 of the sub-image block to be checked.

[0066] The phase consistency model can be calculated using the Log Gabor filter. The method for calculating the phase consistency model is shown in the following formula (1):

[0067]

[0068] Where PC(x,y) is the phase consistency strength at (x,y); n is the scale multiple of the filter; θ represents the filtering direction, which increases in fixed steps between [0,π]; A s,θ (x, y) represents the amplitude value of the harmonic component after decomposition using a Log Gabor filter with a scale of s and along the θ direction; W θ (x,y) represents the weight coefficient of frequency expansion in the θ direction; is an operation that only takes positive values ​​and sets negative values ​​to 0; T is the noise threshold; ε is a constant used to avoid zero division; Calculated from the phase angle of the decomposed harmonic components.

[0069] By integrating the phase consistency strength in each direction, the maximum moment M of the edge feature image used to represent the edge feature of the image is obtained. The method for calculating the maximum moment M of the edge feature image is shown in the following formula (2) and formula (3):

[0070]

[0071] in,

[0072]

[0073] The meaning of each element is consistent with the description in formula (1) and will not be repeated here.

[0074] This method can also use the above formula (1) to formula (3) to extract the matrix used to represent the edge feature image 2 of the reference sub-image block.

[0075] Specifically, the method can convert the matrix of edge feature image 1 and the matrix of edge feature image 2 into the frequency domain, and use the following formula (4) to calculate the cross power spectrum H of the sub-image block to be tested and the corresponding reference sub-image block:

[0076]

[0077] Where F is the Fourier transform, F * represents the conjugate complex number of F, F1 and F2 are the Fourier transforms of the sub-image block to be tested and the reference sub-image block respectively, and (u, v) is the Fourier transform coordinate representation.

[0078] Specifically, the method can determine the phase correlation pulses of the edge feature image 1 and the edge feature image 2 according to the cross power spectrum.

[0079] According to the Fourier shift law: the displacement of an object in the spatial domain is equivalent to the phase shift in the frequency domain. That is, the mutual power spectrum H of the sub-image block to be tested and the reference sub-image block is inversely Fourier transformed to obtain the phase-correlated pulse images of edge feature image 1 and edge feature image 2. The phase-correlated pulse image is a peak value at the relative offset, and the other position values ​​are close to zero. The x-axis coordinate and y-axis coordinate at the peak of the phase-correlated pulse image are the relative offset (dx, dy) between the sub-image block to be tested and the corresponding reference sub-image block.

[0080] In an example, Figure 4 As shown, image a is the sub-image block to be tested of the image to be tested, and image b is the reference sub-image block in the reference image corresponding to the sub-image block to be tested. Edge feature a is edge feature image 1 calculated based on the sub-image block to be tested, and edge feature b is edge feature image 2 calculated based on the reference sub-image block. The cross-power spectrum of the sub-image block to be tested and the reference sub-image block is calculated by edge feature image 1 corresponding to edge feature a and edge feature image 2 corresponding to edge feature b. The cross-power spectrum is inversely Fourier transformed to obtain a phase-correlated pulse diagram. The x-axis coordinate and y-axis coordinate corresponding to the peak value in the phase-correlated pulse diagram are the relative offset (dx, dy) between the sub-image block to be tested and the corresponding reference sub-image block.

[0081] S34: Preliminarily determine the predicted matching point position of the key point in the reference image according to the relative offset between the key point and the sub-image block to be checked.

[0082] Specifically, the method can determine the predicted matching point position according to the key point and the relative offset between the to-be-checked sub-image block where the key point is located and the corresponding reference sub-image block.

[0083] When the latitude, longitude and resolution of the upper left corner of the sub-image block to be tested and the reference sub-image block are completely consistent, the image coordinate relationship between the key point coordinate (x, y) and its corresponding initial matching point position (x′, y′) is shown in formula (5):

[0084]

[0085] Wherein, dx is the relative offset in the x direction between the sub-image block to be checked where the key point is located and the corresponding reference sub-image block, and dy is the relative offset in the y direction between the sub-image block to be checked where the key point is located and the corresponding reference sub-image block.

[0086] S35. Taking the key point and the predicted matching point as the center respectively, extract the directional phase consistency feature image of the discrete direction set in the neighborhood based on the directional phase consistency model to obtain the key point feature descriptor and the predicted matching point feature descriptor.

[0087] Specifically, this method can extract the phase consistency feature intensity of the specified direction by using the Log Gabor wavelet at the specified angle with the key point as the center, and arrange and combine the phase consistency feature intensity of the discrete direction set according to the direction into a three-dimensional directional phase consistency feature map as the key point feature descriptor. The calculation formula of the directional phase consistency model at the specified angle is shown in formula (6):

[0088]

[0089] Among them, PC θ (x, y) is the phase consistency strength along the θ filtering direction at (x, y), and the meanings of the remaining elements are consistent with the description in formula (1) and will not be repeated here.

[0090] Similarly, the directional phase consistency model obtained by the above formula (6) can be used to apply the above method to obtain the feature descriptor of the predicted matching point.

[0091] In an example, with the key point and the predicted matching point as the center, the neighborhood radius is 50 pixels, the search radius is 10 pixels, and the Log Gabor wavelet in 6 directions (0°, 30°, 60°, 90°, 120°, 150°) is constructed. The directional phase consistency features (PC) in 6 directions are extracted based on the phase consistency model. θi , i=1,2,3,4,5,6) as the feature descriptors of key points and predicted matching points, and the dimension of the feature descriptor is 120×120×6.

[0092] S36, transforming the key point feature descriptor and the predicted matching point feature descriptor into the frequency domain, calculating the similarity between the key point feature descriptor and the predicted matching point feature descriptor; and determining the matching point according to the similarity.

[0093] Specifically, the method can transform the key point feature descriptor and the predicted matching point feature descriptor into the frequency domain, and calculate the similarity between the key point feature descriptor and the initial predicted matching point feature descriptor.

[0094] The similarity between descriptors can be measured using the Sum of Squared Differences (SSD) algorithm. The calculation formula is shown in formula (7):

[0095] SSD(x,y)=∑ i ∑ j [S(i,j)-T(i,j)] 2 (7)

[0096] Among them, SSD(x,y) represents the sum of squared errors at the point (x,y), T represents the feature descriptor of the key point, S represents the feature descriptor of the corresponding initial predicted matching point, and (i,j) is the point coordinate of the feature descriptor.

[0097] Expanding and simplifying the above formula (7), we can get formula (8):

[0098] SSD=S*S-2S*T+T*T (8)

[0099] Among them, SSD represents the sum of squared errors, T represents the key point feature descriptor, S represents the feature descriptor of the corresponding predicted matching point, and * is the convolution operator.

[0100] S*S and T*T are fixed values, so we can continue to simplify and get the similarity expression formula (9):

[0101] Sim=S*T=F S ×F T (9)

[0102] Among them, F S and F T are the Fourier transform results of S and T respectively. It can be seen from the calculation formula (7) that SSD is the point-by-point subtraction operation of S and T. Therefore, the more similar the input key point feature descriptor and the predicted matching point feature descriptor are, the smaller the SSD value is, and the larger Sim is. That is, the position of the maximum value point in the image of the similarity Sim is the initial matching point position corresponding to the key point.

[0103] Specifically, the position of the maximum value point of the similarity Sim in the image may be determined as the position of the initial matching point corresponding to the key point.

[0104] In an example, Figure 5 As shown, the predicted matching point B' is determined in the reference image block according to the key point A in the sub-image block to be tested, and the phase consistency features in 6 directions are extracted as their respective feature descriptors with A and B' as the center. The feature descriptors are converted to the frequency domain to calculate the similarity, and the position of B at the maximum similarity is the actual initial matching point position.

[0105] In the embodiment of the present invention, the predicted matching point position is determined by the relative offset between the key point and the sub-image block to be checked where the key point is located and the corresponding reference sub-image block, the key point feature descriptor and the predicted matching point feature descriptor based on the preset discrete direction set with the key point and the predicted matching point position as the center are calculated respectively, the similarity between the key point feature descriptor and the predicted matching point feature descriptor is calculated, and the initial matching point is determined according to the similarity, so as to realize the accurate calculation of the initial matching point corresponding to the key point.

[0106] S37. Calculate the residual model parameters using the least squares method according to the key points and the initial matching points; calculate the residuals of the key points according to the residual model; and obtain a set of same-name point pairs using the key points whose residuals are less than a preset threshold and the corresponding initial matching points as the key points in the set of same-name point pairs and the matching points corresponding to the key points.

[0107] Specifically, after determining the initial matching points corresponding to all key points according to the above steps, the method can calculate the residual model parameters using the least squares method according to the key points and the corresponding initial matching points, and calculate the residual of the key points according to the residual model.

[0108] The residual model determined in this method is shown in the following formula (10):

[0109]

[0110] Among them, (x, y) is the coordinate of the key point on the image to be tested, (Δx, Δy) is the geometric position deviation of the initial matching point corresponding to the key point, that is, the relative offset, (a0, a1, a2) and (b0, b1, b2) are model parameters.

[0111] The method can use the least square method to solve the residual model parameters. The method can use the key points whose residuals are less than a preset threshold and the corresponding initial matching points as the key points in the same-name point pair set and the matching points corresponding to the key points to obtain the same-name point pair set, the same-name point pair set includes at least 20 same-name point pairs, and one same-name point pair includes one key point and one matching point corresponding to the key point.

[0112] In the embodiment of the present invention, the residual model parameters are calculated using the least square method based on the key points and the initial matching points, the residual model is used to calculate the residual of the key points, and the key points with residuals less than a preset threshold and the corresponding initial matching points are used as the key points in the same-name point pair set and the matching points corresponding to the key points to generate a same-name point pair set. The key points with large residuals and the corresponding initial matching points are eliminated, so that the same-name point pairs in the generated same-name point pair set have higher accuracy.

[0113] According to the automatic inspection method for geometric accuracy of remote sensing satellite images provided by the present invention, step S13 includes:

[0114] S131, calculating the positioning error of the same-name point pairs according to the key points and the matching points in the same-name point pair set.

[0115] S132, calculating the mean error based on the positioning errors of the plurality of same-name point pairs.

[0116] S133, generating an error direction map according to the mean error, the positioning error of the same-name point pair, the key point, the matching point and the image to be tested.

[0117] Specifically, the calculation formula for the positioning error of the same-name point pair is:

[0118]

[0119] In the formula, Δx=x i -x b , Δy=y i -y b , x i and i represents the ground positioning coordinates of the key point in the image to be tested, x b and b Represents the ground coordinates of the matching point corresponding to the key point in the reference image.

[0120] The calculation formula of mean error is:

[0121]

[0122] In the formula, n represents the number of pairs of points with the same name, D i is the positioning error of the i-th pair of points with the same name.

[0123] Specifically, this method can calculate the mean error based on the positioning errors of multiple points with the same name, and can sample key points. In the image to be tested, the sampled key point position is used as the starting point, and the error arrow points to the actual geographical location of the key point, wherein the actual geographical location of the key point is the geographical location of the matching point in the reference image, and the magnification ratio of the error arrow is determined according to the mean error.

[0124] In an example, the error direction map generated based on the mean error, the positioning error of the same-name point pair, the key points, the matching points and the image to be tested is as follows: Figure 6 As shown, Figure 6 The mean error calculated based on the positioning errors of multiple pairs of points with the same name is 5m.

[0125] In an embodiment of the present invention, after calculating the error of the same-name point pair, the mean error can be calculated based on the positioning error of the same-name point pair, and an error direction map can be generated based on the positioning error of the same-name point pair, key points, matching points and the image to be tested, so that the error can be visualized and the size and direction of the image product error can be intuitively displayed.

[0126] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for automatically checking the geometric accuracy of remote sensing satellite images, characterized in that: include: Input an image to be inspected, and automatically obtain a corresponding reference image according to the image to be inspected; Obtaining a set of same-name point pairs according to the image to be tested and the reference image by a method based on phase congruency image matching, wherein the set of same-name point pairs includes at least twenty same-name point pairs, and the same-name point pairs are key points evenly distributed in the image to be tested and matching points corresponding to the key points in the reference image; Calculate the positioning error of the same-name point pair according to the key points and the matching points in the same-name point pair set and draw an error direction map; Wherein, the step of automatically acquiring the corresponding reference image according to the image to be inspected comprises: Acquiring original reference image data according to the attribute information of the image to be inspected and a preset mapping relationship; According to the image to be inspected, automatically matching one or more associated reference images covered by the geographical location of the image to be inspected in the original reference image data; Preprocessing the one or more associated reference images to obtain the reference image; The step of obtaining a set of point pairs with the same name based on the image to be inspected and the reference image by a method based on phase consistency image matching comprises: Performing feature detection on the image to be tested by a non-maximum suppression detection algorithm, and determining the largest maximum point in a preset area of ​​the image to be tested as the key point; Dividing the image to be inspected and the reference image into uniform grids in sequence to obtain one-to-one corresponding sub-image blocks to be inspected and reference sub-image blocks; Extracting edge feature image 1 of the sub-image block to be inspected and edge feature image 2 of the reference sub-image block using a phase consistency model; calculating phase correlation pulses of the edge feature image 1 and the edge feature image 2; and obtaining a relative offset between the sub-image block to be inspected and the corresponding reference sub-image block according to the phase correlation pulses; Preliminarily determining the predicted matching point position of the key point in the reference image according to the relative offset between the key point and the sub-image block to be checked; Taking the key point and the predicted matching point as the center respectively, extracting the directional phase consistency feature image of the discrete direction set in the neighborhood based on the directional phase consistency model, and obtaining the key point feature descriptor and the predicted matching point feature descriptor; Transforming the key point feature descriptor and the predicted matching point feature descriptor into the frequency domain, calculating the similarity between the key point feature descriptor and the predicted matching point feature descriptor; and determining an initial matching point according to the similarity; The residual model parameters are calculated using the least squares method according to the key points and the initial matching points; the residuals of the key points are calculated according to the residual model; and a set of same-name point pairs is obtained using the key points whose residuals are less than a preset threshold and the corresponding initial matching points as the key points in a set of same-name point pairs and the matching points corresponding to the key points.

2. The method for automatically checking the geometric accuracy of remote sensing satellite images according to claim 1, characterized in that: The step of calculating the positioning error of the same-name point pair according to the key points and the matching points in the same-name point pair set and drawing an error direction map comprises: Calculate the positioning error of the same-name point pair according to the key points and the matching points in the same-name point pair set; Calculate the mean error based on the positioning error of the same-name point pair; An error direction map is generated according to the mean error, the positioning error of the same-name point pair, the key point, the matching point and the image to be checked.

Images