Sar target positioning method fusing geometric positioning and heterogeneous matching

Abstract

The present application relates to remote sensing image processing and target positioning technical field, in view of the existing SAR target positioning strong dependence on radar parameter, iterative solution is easy to diverge, the problem of difficult heterogenous matching, provide a kind of fusion geometry positioning and SAR target positioning method of heterogenous matching, comprising: obtaining airborne radar space position, straight down point position and motion velocity vector;Based on the space triangle formed by airborne radar, straight down point and target point, the rough coordinate of target is directly solved using geometric relationship, and is not dependent on Doppler frequency and radar wavelength;According to rough coordinate, determine SAR real-time map target template area and optical reference map search area, extract local gradient direction feature map, make local gray inversion area present consistent direction expression by nonlinear correspondence;Adopt similarity measurement matching to local gray inversion tolerance, obtain target accurate coordinate.The present application does not need iteration, reduces parameter dependence, improves robustness and precision.

Description

Technical Field

[0001] This invention belongs to the field of remote sensing image processing and target localization technology, specifically relating to a SAR target localization method that integrates geometric localization and heterogeneous matching. Background Technology

[0002] Synthetic Aperture Radar (SAR) is an active microwave imaging sensor with all-weather, all-day operation capabilities. SAR target localization refers to determining the three-dimensional position of a target point in an absolute coordinate system using SAR image data, target image point coordinates, and related imaging parameters, combined with a SAR imaging model. Compared to optical image localization, SAR image localization has stronger environmental adaptability, functioning normally even in adverse weather conditions such as cloudy, rainy, and foggy conditions. Therefore, it has broad application prospects in fields such as civilian surveying and mapping, disaster monitoring, and resource exploration.

[0003] SAR image localization methods mainly include stereo SAR image localization, single-frame SAR image localization, and interferometric SAR image localization. Among them, single-frame SAR image localization has received widespread attention due to its convenient data acquisition and simple processing procedures. Currently, single-frame SAR target localization technology mainly revolves around the range-Doppler model and its derived algorithms. This model establishes a rigorous geometric relationship between the target's geographic coordinates and the range and azimuth axes of the SAR image by constructing range equations and Doppler equations. It has a clear theoretical basis and has been successfully applied to various spaceborne and airborne systems.

[0004] However, existing methods have the following shortcomings: First, there is a high degree of dependence on radar system parameters and imaging geometry. Positioning accuracy fundamentally depends on the accuracy of input parameters, including platform orbit data, attitude data, radar timing delay, slant range history, Doppler center frequency, and radar wavelength. Any measurement error or estimation deviation of any parameter will be transmitted and amplified in the positioning result through complex geometric relationships. In the absence of high-precision calibration data or when using lightweight, low-cost SAR systems, this dependence becomes a major bottleneck in improving absolute positioning accuracy.

[0005] Second, the inherent instability of the iterative solution process. Range-Doppler positioning is essentially a process of solving a system of nonlinear equations, which usually requires numerical methods such as Newton's iteration method. This process is very sensitive to the choice of initial values. When the initial values ​​deviate far from the true solution or the target is in complex terrain such as overlapping or shadows, the iteration may converge slowly or even diverge, leading to positioning failure or gross errors.

[0006] Third, matching images from different sources is difficult. To obtain higher positioning accuracy, it is often necessary to match real-time SAR images with high-resolution optical reference images. However, the imaging mechanisms of SAR and optical images are very different, and there are significant differences between them in grayscale characteristics and texture structure. The speckle noise, nonlinear grayscale changes, and non-uniform geometric distortion in SAR images make accurate matching with optical images very difficult.

[0007] In summary, existing SAR target localization methods still have significant shortcomings in terms of parameter dependence, iterative stability, and robustness of heterogeneous image matching. To address this, this invention proposes a SAR target localization method that integrates geometric localization and heterogeneous image matching. This method employs a non-iterative geometric direct localization technique to eliminate dependence on Doppler parameters and radar wavelength. Simultaneously, by constructing a heterogeneous image consistency feature representation based on gradient phase field and designing a similarity metric tolerant to local gray-level inversion, highly robust heterogeneous image matching is achieved. This enables high-precision and high-reliability target localization results on a low-cost airborne SAR platform. Summary of the Invention

[0008] To address the aforementioned technical problems, this invention proposes a SAR target localization method that integrates geometric localization and heterogeneous image matching. The method employs a joint framework of non-iterative geometric direct localization and heterogeneous image matching: it directly calculates the target's coarse coordinates using the spatial triangle cosine theorem and vector cross product constraints, without relying on Doppler center frequency and radar wavelength parameters; it doubles the gradient direction angle of the pixel neighborhood to construct a local gradient feature map; it extracts common structural information from heterogeneous images using covariance principal component analysis driven by complex square relationships; and it uses similarity metrics with angle inversion and threshold truncation for feature matching, enhancing tolerance to grayscale inversion and improving matching robustness. This mitigates the parameter sensitivity of traditional RD models and the risk of mismatches caused by grayscale inversion in heterogeneous matching, thereby improving the autonomy and reliability of target localization on low-cost airborne SAR platforms.

[0009] This invention provides a SAR target localization method that integrates geometric localization and heterogeneous matching, comprising: Step S110: Obtain the spatial position of the airborne radar, the spatial position of the downward-facing point, and the velocity vector of the airborne radar; the airborne radar is a synthetic aperture radar. Step S120: Based on the spatial triangle formed by the airborne radar, the downward viewpoint, and the target point to be located, calculate the first distance between the target point and the downward viewpoint using trigonometric relationships; determine the direction vector of the target point relative to the downward viewpoint according to spatial vector constraints, and directly calculate the coarse coordinates of the target point in the spatial coordinate system to obtain the coarse positioning result; The spatial vector constraint relationship includes: the line vector connecting the target point and the downward viewpoint is perpendicular to both the motion velocity vector and the vector from the Earth's center to the downward viewpoint; The direct calculation process does not depend on the Doppler center frequency and radar wavelength parameters; Step S130: Determine the target template region in the real-time synthetic aperture radar image and the search region in the pre-prepared optical reference image based on the coarse positioning result; extract feature maps based on local gradient directions from the real-time synthetic aperture radar image and the optical reference image respectively; The feature value of each pixel in the feature map has a non-linear correspondence with the gradient direction angle in the neighborhood of the pixel, which is used to ensure that the region where local grayscale inversion occurs presents a consistent directional expression on the feature map. Step S140: Using a similarity metric that is tolerant of local grayscale inversion, the feature map of the real-time synthetic aperture radar image and the feature map of the optical reference image are matched within the search area to determine the matching position and obtain the precise coordinates of the target point.

[0010] Preferably, the nonlinear correspondence includes: calculating the sum of the squares of the gradient vectors in the neighborhood of the pixel, and obtaining the feature value of the pixel based on the phase of the summation result, such that the feature value is numerically equal to the average gradient direction angle in the neighborhood; the average gradient direction angle is the statistical average of the gradient direction angles of all pixels in the neighborhood.

[0011] Preferably, in step S130, the method for extracting feature maps based on local gradient directions further includes: For each pixel in the neighborhood of the pixel, the gradient magnitude is calculated using the horizontal and vertical gradients of the pixel. Calculate the average gradient magnitude of all pixels in the neighborhood to obtain the average gradient intensity in the neighborhood; When the average gradient intensity is lower than a preset threshold, the neighborhood of the pixel is determined to be a textureless region, and the feature value output of the pixel is suppressed.

[0012] Compared with the prior art, the beneficial effects of the present invention include: (1) The geometric direct positioning method without iteration is adopted, which fundamentally avoids the risk of iterative divergence. The positioning process is stable and reliable, and the computational efficiency is high.

[0013] (2) The geometric direct positioning process does not depend on the Doppler center frequency and radar wavelength parameters, thus eliminating the absolute dependence on high-precision radar parameters and making it suitable for low-cost, low-precision airborne SAR equipment.

[0014] (3) A gradient phase field-based heterogeneous image feature extraction method is adopted. By doubling the angle, the features of the local gray-scale inversion region are expressed in a consistent direction, which effectively suppresses nonlinear gray-scale distortion and local gray-scale inversion between multimodal images.

[0015] (4) A similarity metric that is tolerant of local grayscale reversal was designed. Combined with angle inversion and threshold truncation strategies, the robustness of matching was enhanced.

[0016] (5) A fusion strategy combining coarse and fine positioning is adopted. The coarse position provided by geometric direct positioning is used to significantly reduce the search area for heterogeneous matching, thereby improving the matching success rate and positioning accuracy. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the structures shown in these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating the steps of a SAR target localization method that integrates geometric localization and heterogeneous source matching in one embodiment of the present invention. Figure 2 This is a schematic diagram illustrating the principle of the geometric direct positioning method in one embodiment of the present invention; Figure 3 This is a schematic diagram illustrating the radiation invariance of the gradient phase field in one embodiment of the present invention, wherein, Figure 3 (a) is a visible light and infrared image with random grayscale transformation in four quadrants. Figure 3 (b) is the corresponding principal component angle image; Figure 4 This is a schematic diagram of a principal component orientation angle similarity measurement method in one embodiment of the present invention, wherein, Figure 4 (a) is a schematic diagram illustrating the calculation principle of principal component orientation angles greater than 90°. Figure 4 (b) is a schematic diagram illustrating the calculation principle of principal component orientation angles greater than 45° and less than 90°. Figure 4 (c) is a schematic diagram illustrating the calculation principle of principal component orientation angles less than 45°; Figure 5 This is an example diagram of SAR and visible light image matching results based on gradient phase field in one embodiment of the present invention, wherein, Figure 5 (a) is a SAR image of the Gobi Desert, with the points to be matched marked with crosshairs; Figure 5 (b) is a visible light satellite image of the Gobi Desert, with the matching results marked with crosshairs; Figure 5(c) is a SAR image of a sparsely textured mountainous area, with the points to be matched marked with crosshairs; Figure 5 (d) is a visible light image of a sparsely textured mountainous area, with the matching results marked with crosshairs; Figure 6 This is a complete flowchart of a SAR target localization method that integrates geometric localization and heterogeneous matching in one embodiment of the present invention; Figure 7 This is a schematic diagram of the scene matching results of SAR images with multiple feature points in the experiment of this invention, wherein, Figure 7 (a) is a SAR image of rural scene 1 with a large tilt angle, and the points to be matched are marked by crosshairs; Figure 7 (b) is a visible light image of rural scene 1 at a large tilt angle, with the matching results marked by crosshairs; Figure 7 (c) is a SAR image of rural scene 2 with a large tilt angle, and the points to be matched are marked by crosshairs; Figure 7 (d) is a visible light image of rural scene 2 with a large tilt angle, and the matching results are marked by crosshairs; Figure 8 This is a schematic diagram of the scene matching results of multiple feature points of optical satellite imagery in the experiment of this invention, wherein, Figure 8 (a) is a SAR image of rural scene 3 with a large tilt angle, and the points to be matched are marked by crosshairs; Figure 8 (b) is a visible light image of rural scene 3 at a large tilt angle, with the matching results marked by crosshairs; Figure 8 (c) is a SAR image of rural scene 4 with a large tilt angle, and the points to be matched are marked by crosshairs; Figure 8 (d) is a visible light image of rural scene 4 with a large tilt angle, and the matching results are marked by crosshairs. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] In one embodiment, such as Figure 1 As shown, this invention provides a SAR target localization method that integrates geometric localization and heterogeneous source matching, comprising: Step S110: Obtain the spatial position of the airborne radar, the spatial position of the downward-facing point, and the velocity vector of the airborne radar; the airborne radar is a synthetic aperture radar. Step S120: Based on the spatial triangle formed by the airborne radar, the downward viewpoint, and the target point to be located, calculate the first distance between the target point and the downward viewpoint using trigonometric relationships; determine the direction vector of the target point relative to the downward viewpoint according to spatial vector constraints, and directly calculate the coarse coordinates of the target point in the spatial coordinate system to obtain the coarse positioning result; The spatial vector constraint relationship includes: the line vector connecting the target point and the downward viewpoint is perpendicular to both the motion velocity vector and the vector from the Earth's center to the downward viewpoint; The direct calculation process does not depend on the Doppler center frequency and radar wavelength parameters; Step S130: Determine the target template region in the real-time synthetic aperture radar image and the search region in the pre-prepared optical reference image based on the coarse positioning result; extract feature maps based on local gradient directions from the real-time synthetic aperture radar image and the optical reference image respectively; The feature value of each pixel in the feature map has a non-linear correspondence with the gradient direction angle in the neighborhood of the pixel, which is used to ensure that the region where local grayscale inversion occurs presents a consistent directional expression on the feature map. Step S140: Using a similarity metric that is tolerant of local grayscale inversion, the feature map of the real-time synthetic aperture radar image and the feature map of the optical reference image are matched within the search area to determine the matching position and obtain the precise coordinates of the target point.

[0021] Specifically, in step S110, the airborne radar parameters are obtained, including at least the spatial position of the airborne radar, the spatial position of the downward-facing point, and the motion velocity vector of the airborne radar; the airborne radar is a synthetic aperture radar.

[0022] In this embodiment, the airborne radar is mounted on the aircraft. The spatial position coordinates and velocity vector of the aircraft are obtained through the positioning and navigation system on the aircraft. The geographical coordinates (longitude, latitude, altitude) of the aircraft's position are converted into coordinates in a spatial rectangular coordinate system, and the velocity vector of the aircraft in the North-Sky-East coordinate system is also converted into a spatial rectangular coordinate system.

[0023] It should be noted that, in this embodiment, the direct downward viewpoint refers to the vertical projection point of the airborne radar onto the Earth's surface (or reference ellipsoid), that is, the intersection of the airborne radar position towards the Earth's center (or the local vertical direction) and the Earth's surface. The spatial location of the direct downward viewpoint can be calculated from the spatial location of the airborne radar and the Earth model.

[0024] Figure 2 This is a schematic diagram illustrating the principle of the geometric direct positioning method in this invention. Figure 2As shown, the geometric direct positioning method of this invention uses the Earth's spatial rectangular coordinate system: with the Earth's center... The origin of the coordinate system is... The axis points to the Earth's North Pole. The axis points to the intersection of the Greenwich Meridian plane and the Earth's equator. Axis perpendicular to The plane forms a right-handed coordinate system. Indicates the center position of the aircraft's radar. Indicates the downward viewpoint of the aircraft's radar. Indicates the target point. If... Indicates from point to The vector, then for and The intersection, It is a right angle.

[0025] Obtain the spatial position of the airborne radar in the spatial rectangular coordinate system. Spatial position of the downward viewpoint and the velocity vector of the airborne radar , This represents the transpose of a vector / matrix. These parameters can be obtained from an airborne navigation system (such as a GPS / INS integrated navigation system). Let the coordinates of the target point in a spatial rectangular coordinate system be denoted as... .

[0026] Furthermore, in step S120, based on the spatial triangle formed by the airborne radar, the downward viewpoint, and the target point to be located, the first distance between the target point and the downward viewpoint is calculated using trigonometric relationships; the direction vector of the target point relative to the downward viewpoint is determined according to spatial vector constraints, and the coarse coordinates of the target point in the spatial coordinate system are directly calculated to obtain a coarse positioning result.

[0027] The spatial vector constraint relationship includes: the line vector connecting the target point and the downward viewpoint is perpendicular to both the motion velocity vector and the vector from the Earth's center to the downward viewpoint; the direct solution process does not depend on the Doppler center frequency and radar wavelength parameters.

[0028] The step S120 employs a non-iterative geometric direct positioning method, fundamentally avoiding the risk of iterative divergence. This ensures a stable and reliable positioning process with high computational efficiency. Furthermore, the geometric direct positioning process does not rely on the Doppler center frequency and radar wavelength parameters, eliminating the absolute dependence on high-precision radar parameters and making it suitable for low-cost, low-precision airborne SAR equipment.

[0029] Specifically, the process of calculating the first distance between the target point and the downward-facing viewpoint using trigonometric relationships further includes: Combined with the Earth's equatorial radius and airborne radar slant range Based on the Earth's spherical model, the geocentric angle of the triangle formed by the Earth's center, the airborne radar, and the target point is calculated using the law of cosines. The geocentric angle It is the line connecting the Earth's center to the airborne radar. The line connecting the Earth's center to the target point The slant range of the airborne radar is calculated by the time difference between the radar's transmitted pulse and the received target echo, or it can be directly read from the SAR system's auxiliary data.

[0030] Central angle The formula for calculation is: ; in, The distance from the Earth's center to the airborne radar. The distance is Euclidean. The distance from the Earth's center to the target point is the Earth's equatorial radius, assumed under the Earth's spherical model. Approximate; Airborne radar slant range It is the slant distance from the phase center of the radar antenna to the target point.

[0031] According to the geocentric angle Relative to the Earth's equatorial radius Calculate the first distance: Firstly, based on geometric relationships, the first distance... Theoretically, it can be expressed as: ; And because Substituting into the above formula, we can obtain the first distance: .

[0032] Since this step only requires coarse positioning of the target point, the error introduced by the assumption that the Earth's surface is a plane is completely within the tolerable range.

[0033] Determine the first distance Later, due to The coordinates of the point are known; we only need to determine... The direction can determine the target point The coordinates. Based on the aforementioned spatial vector constraints, the vector... Simultaneously perpendicular to the aircraft's flight velocity vector and the vector from the geocentric point to the positive downward viewpoint ,Right now The direction vector is The unit vector. Therefore, the direction vector of the target point relative to the downward viewpoint is given by the following formula: ; in, This represents the vector from the Earth's center to the directly below viewpoint. Represents the velocity vector of the motion. For the first distance, The coordinates of the target point in a spatial rectangular coordinate system.

[0034] The direct solution process described above does not rely on the Doppler center frequency and radar wavelength parameters, requires no iteration, and fundamentally avoids the divergence risk of traditional RD models. The positioning process is stable, reliable, and computationally efficient. Furthermore, it eliminates the absolute dependence on high-precision radar parameters, making it suitable for low-cost, low-precision airborne SAR equipment.

[0035] After calculating the target point's coordinates in the Earth's spatial rectangular coordinate system using the above formula, in order to use them in the heterogeneous image matching module, it is necessary to transform the spatial rectangular coordinate system to the SAR real-time image coordinate system and the satellite reference image (i.e., the optical reference image) coordinate system to obtain the target point's coordinates in the SAR real-time image coordinate system. Coordinates in the satellite image reference coordinate system The template image and search image for heterogeneous image matching can be obtained through the following methods: Centered on, with A template image is obtained by extracting the width and height from the SAR image; Centered on, respectively Extract the search image from the satellite reference image for width and height. For the search length, It should be greater than the coarse positioning error (unit: pixels) to ensure that the search image can completely contain the template image.

[0036] Next, in step S130, the search area is determined and the feature map is extracted, including determining the target template area in the real-time synthetic aperture radar image and the search area in the pre-prepared optical reference image based on the coarse localization result; and extracting the feature maps of the real-time synthetic aperture radar image and the optical reference image based on the local gradient direction.

[0037] The feature value of each pixel in the feature map has a non-linear correspondence with the gradient direction angle in the neighborhood of the pixel, which is used to ensure that the region where local grayscale inversion occurs presents a consistent directional expression on the feature map.

[0038] The aforementioned geometric direct positioning method can control the SAR target positioning error within 100 meters. However, due to the limitations of low-cost equipment accuracy, it is difficult to improve accuracy through geometric calculations. Therefore, this invention utilizes image matching to directly locate the target position on the satellite reference image, further improving target positioning accuracy. Specifically, it employs a heterogeneous image feature extraction method based on gradient phase fields. By doubling the angle, the features of local gray-level inversion regions are presented in a consistent direction, effectively suppressing nonlinear gray-level distortion and local gray-level inversion between multimodal images.

[0039] Due to the significant differences in imaging mechanisms, SAR images and visible light reference images do not have a strict correspondence in the grayscale domain. Grayscale values ​​of the same pattern may be completely or partially reversed, and the original grayscale distribution patterns of the scene may be completely disrupted, making it difficult to accurately describe scene content using grayscale information. Therefore, this invention employs a gradient phase field to describe scene content, enhances the saliency of the scene's core structure through gradient intensity, and finally calculates the similarity of the matching windows using a similarity metric.

[0040] Unlike grayscale representation, the gradient phase field of this invention calculates the principal component orientation angles of the gradient vectors in the neighborhood of each pixel and combines them to form a consistent representation of heterogeneous images. The feature value of each pixel in the feature map has a non-linear correspondence with the gradient orientation angles in the neighborhood of that pixel. Figure 3 This is a schematic diagram illustrating the radiation invariance of the gradient phase field in one embodiment of the present invention, wherein, Figure 3 (a) is a visible light and infrared image with random grayscale transformation in four quadrants. Figure 3 (b) is the corresponding principal component angle image. For example... Figure 3 As shown, the nonlinear correspondence specifically includes: calculating the sum of the squares of the gradient vectors in the neighborhood of the pixel, and obtaining the feature value of the pixel based on the phase of the summation result, such that the feature value is numerically equal to the average gradient direction angle in the neighborhood; the average gradient direction angle is the statistical average of the gradient direction angles of all pixels in the neighborhood.

[0041] The feature map extraction method specifically includes the following steps: (1) Calculate the horizontal gradient of each pixel in the neighborhood of the pixel. and vertical gradient The gradient vectors of all pixels in the neighborhood are obtained. The gradient direction angle within the neighborhood of the pixel is the direction of the gradient vector. For a given pixel in image I... Within its range neighborhood Within, its vertical and horizontal gradients can form a mass of... The local gradient field matrix.

[0042] (2) Construct the covariance matrix of the gradient vector. : ; in, , The components on the main diagonal of the covariance matrix, The components on the second diagonal of the covariance matrix, This represents the summation of all pixels in the neighborhood of a given pixel.

[0043] (3) Solve for the principal component orientation angles of the covariance matrix. The principal component orientation angle is used as the feature value of the pixel. The principal component orientation angle is obtained by summing the squares of the gradient vectors in the neighborhood and taking the half-phase angle, and its value is equal to the average gradient orientation angle in the neighborhood.

[0044] Principal component orientation angle The formula for calculation is: ; in The range of values ​​is . The definition of is: .

[0045] The principal component orientation angle and the gradient vector satisfy a complex square relationship: ; in, It is the imaginary unit.

[0046] This complex square relation proves that the principal component vector is equivalent to squaring the length of the gradient vector, doubling the angle, summing them, and then restoring the original length by taking half an angle. By squaring the gradient vector (doubling the angle), gradient vectors with opposite directions will point in the same direction, and after summing, they will reinforce each other. Perpendicular gradient vectors will cancel each other out, effectively suppressing nonlinear gray-level distortion and local gray-level inversion between multimodal images, ensuring that regions experiencing local gray-level inversion exhibit a consistent directional representation on the feature map. This aligns with the imaging characteristics of multimodal sensors.

[0047] Furthermore, the principal component angle of a pixel is only related to the angle between the average gradient vectors of its neighborhood, and is independent of the absolute value of the pixel's gray level. This effectively addresses local gray-level distortion or local gray-level inversion between multimodal images, enabling consistent description and representation of multimodal images.

[0048] For textured regions, higher feature intensity results in a higher image signal-to-noise ratio, and noise interference has minimal impact on the calculation of principal component angles. However, for textureless regions, feature intensity is very low, leading to a low image signal-to-noise ratio, and noise significantly affects the phase angle. Therefore, average gradient intensity needs to be introduced during the calculation to extract salient features. Specifically, this includes: For each pixel in the neighborhood of the pixel, the horizontal gradient of the pixel is used. with vertical gradient Calculate the gradient magnitude: ; Calculate the average gradient magnitude of all pixels in the neighborhood to obtain the average gradient intensity in the neighborhood; When the average gradient intensity is lower than a preset threshold, the neighborhood of the pixel is determined to be a textureless region, and the feature value output of the pixel is suppressed, making the feature value of the pixel invalid; otherwise, the calculation result is output normally.

[0049] In engineering practice, the calculation of principal component orientation angles can be accelerated using integral images, and the calculation of principal component angles on the reference image can be completed in advance without consuming online computing resources; the calculation requires converting the angle from radians to degrees and expanding the range of angle values ​​from... translate to By performing the above operations, the principal component phase angle is converted from floating-point type to unsigned integer type, reducing memory usage and improving calculation speed. Angle calculation involves many judgment operations, which are not easy to optimize the calculation speed. A lookup table can be used to speed up the calculation.

[0050] Finally, in step S140, heterogeneous image matching and fine localization are performed, including: using a similarity metric that is tolerant of local gray-level inversion, matching the feature map of the real-time synthetic aperture radar image with the feature map of the optical reference image within the search area, determining the matching position, and obtaining the precise coordinates of the target point. This process designs a similarity metric that is tolerant of local gray-level inversion, and combines angle inversion and threshold truncation strategies to enhance the robustness of the matching.

[0051] Traditional similarity measurement methods are suitable for intensity-based feature representations but are less effective for direction-based representations, and even less capable of handling the frequent local gray-level distortions and gray-level inversions in multimodal images. Therefore, this invention designs a specific similarity measurement method for principal component angles, which is tolerant of local gray-level inversions.

[0052] In one embodiment, a similarity metric that tolerates local grayscale inversion is defined as follows: For the target template region and the sub-windows within the search region that have the same size as the target template region, a similarity score is defined. for: ; in: ; ; in, For the target template area at pixels eigenvalues ​​at that location For the sub-window at pixel points Eigenvalues ​​at; 、 These are the width and height of the target template area, respectively.

[0053] Figure 4 This is a schematic diagram of a principal component orientation angle similarity measurement method in one embodiment of the present invention, wherein, Figure 4 (a) is a schematic diagram illustrating the calculation principle of principal component orientation angles greater than 90°. Figure 4 (b) is a schematic diagram illustrating the calculation principle of principal component orientation angles greater than 45° and less than 90°. Figure 4 (c) is a schematic diagram illustrating the calculation principle of principal component orientation angles less than 45°. For example... Figure 4 As shown, the above similarity measure combines angle inversion and threshold truncation strategies. The specific design logic is as follows: when the angle between two vectors is greater than 90°, to accommodate the tendency for grayscale inversion in multi-modal images, one vector is inverted, and the angle between the inverted vector and the other vector is then calculated as the current angle between the two vectors. Figure 4 (a) shows the result; otherwise, the angle remains unchanged. After completing the above operations, continue to judge. If the angle between the vectors is greater than 45°, the two vectors are considered to have failed to match, and the contribution is 0; otherwise, the contribution is the difference between 90 degrees and the angle between the vectors. After these two steps, the closer the vector directions are, the greater the contribution value; when the vectors coincide (0°), the contribution value is the maximum value. This design fully considers the characteristics of multi-modal images, such as the tendency for grayscale inversion and pattern matching failures, thus enhancing matching robustness.

[0054] During the scene matching process, the SAR image target template is slid in the search area of ​​the reference image according to a preset step size, the similarity measure of each matching window is calculated, the maximum value of the similarity measure is selected as the final scene matching result output, the matching position is determined, and thus the precise coordinates of the target point are obtained. Figure 5 This is an example image illustrating the SAR-visible light scene matching result based on gradient phase field in one embodiment of the present invention, showing the SAR-visible light scene matching result obtained using the above method. Wherein, Figure 5 (a) is a SAR image of the Gobi Desert, with the points to be matched marked with crosshairs; Figure 5(b) is a visible light satellite image of the Gobi Desert, with the matching results marked with crosshairs; Figure 5 (c) is a SAR image of a sparsely textured mountainous area, with the points to be matched marked with crosshairs; Figure 5 (d) is a visible light image of a sparsely textured mountainous area, with the matching results marked with crosshairs.

[0055] This invention employs a fusion strategy that combines coarse and fine positioning. By utilizing the coarse position provided by direct geometric positioning, the search area for heterogeneous matching is significantly reduced, thereby improving the matching success rate and positioning accuracy.

[0056] In one embodiment, the pre-preparation process of the optical reference image is completed offline, including: calculating a feature map of the optical reference image based on local gradient directions; and calibrating the transformation relationship between the coordinates of pixels in the optical reference image and geodetic coordinates. The optical reference image is an optical satellite image, a digital orthophoto, or a high-precision digital map. The reference map preparation is performed offline, without requiring online computing resources.

[0057] In a preferred embodiment, after determining the matching location, the method further includes: performing a confidence analysis on the matching location; if the confidence level is greater than a preset confidence threshold... If the matching position is successful, the coordinates corresponding to the matched position are output as the positioning result; otherwise, the coarse positioning result is directly output as the final positioning result. This step ensures that the system can still output a usable positioning result even if scene matching fails, improving the robustness and practicality of the method.

[0058] In one embodiment, the confidence threshold The ratio of the maximum similarity to the second maximum similarity is greater than 1.2.

[0059] Figure 6 A complete flowchart of the SAR target localization method integrating geometric localization and heterogeneous source matching provided by this invention is given. According to the implementation process, the method provided by this invention can be subdivided into the following key processes: 1) Offline preparation of satellite imagery reference maps, specifically including: (1.1) Calculating the gradient phase field of the satellite imagery reference map (i.e., feature map based on local gradient direction), and extracting features as shown in the figure; (1.2) Geodetic coordinate calibration, i.e., the transformation relationship between the pixel coordinate system of the reference map and the geodetic coordinate system. The reference map preparation is carried out offline, without occupying online computing resources.

[0060] 2) Calculate the gradient phase field of the SAR image.

[0061] 3) Calculate the target's coordinates in the Earth's rectangular coordinate system using SAR imaging parameters based on the geometric direct positioning method, and then deduce the search area on the satellite reference image. That is, determine the search center based on the target coarse positioning result and determine the search radius based on the target coarse positioning error.

[0062] 4) Analyze the adaptability of the SAR image target template within the given search area of ​​the reference image. If the requirements are met, proceed to the next step.

[0063] 5) SAR image scene matching, the specific implementation process is as follows: (5.1) Slide the SAR image target template in the search area of ​​the reference map according to the preset step size; (5.2) Calculate the similarity measure of each matching window; (5.3) Select the maximum value of the similarity measure as the final scene matching result output.

[0064] 6) Perform confidence analysis on the target scene matching results. If the confidence is greater than the preset confidence threshold, output the coordinates corresponding to the matching position as the final positioning result; otherwise, directly output the coarse positioning result as the final positioning result.

[0065] Furthermore, to verify the positioning accuracy and reliability of the method of this invention, coarse and fine target positioning experiments were conducted using real airborne SAR images of a certain region. The SAR images were acquired during winter, and due to flight conditions, the weather was mostly sunny or cloudy, without rain, snow, or other extreme weather. The image coverage area is mostly plains, with some hills and valleys, and the ground elevation is between 350 and 400 meters. The main ground scenes include rivers, urban road networks, urban building complexes, valleys, farmland, and hills.

[0066] The SAR target coarse localization experiment is mainly based on the SAR imaging model and SAR imaging parameters, and has two important purposes: first, to determine the search area for matching SAR images with satellite imagery. The coarse localization accuracy directly determines the size of the search area for image matching, and the smaller the search area, the higher the reliability of image matching; second, to directly serve as the output of SAR target localization results when the image matching confidence does not meet the requirements.

[0067] To analyze the accuracy of airborne SAR imagery in locating ground targets, multiple sets of SAR imaging parameters were selected for target localization experiments. These parameters included: radar range (i.e., airborne radar slant range), radar Doppler center frequency, radar installation center latitude and longitude, and the three-dimensional velocity components of the aircraft in the north-sky-east coordinate system. Without loss of generality, the target point to be located was selected at the center of the SAR image in the experiment. For target points at other locations, the location could be achieved by modifying the function input. The coarse SAR target localization results are shown in Table 1.

[0068] Table 1. SAR target coarse localization results

[0069] As shown in Table 1, after coarse positioning using the method of this invention, the maximum positioning error of the target is 107.4926 meters, the minimum positioning error is 25.2017 meters, and the average positioning error is 62.7850 meters. Therefore, coarse SAR target positioning can pinpoint the target to a relatively small range, confirming the correctness and effectiveness of the coarse SAR target positioning method of this invention, and significantly reducing the search range for matching SAR images with satellite imagery. This also demonstrates the advantage of the fusion strategy combining coarse and fine positioning adopted in this invention: utilizing the coarse position provided by geometric direct positioning significantly reduces the search area for heterogeneous matching, improving the matching success rate and positioning accuracy.

[0070] Coarse SAR target localization results are significantly affected by SAR imaging parameters, ground elevation, and terrain undulations, often making it difficult to achieve satisfactory measurement accuracy. Therefore, scene matching is necessary to further improve target localization accuracy. This experiment uses scene matching based on steps S130 and S140. Before scene matching begins, the coarse SAR target localization results (latitude and longitude) need to be converted to the satellite image coordinate system (pixels). After scene matching, the fine target localization results (pixels) on the satellite image need to be converted to the geodetic coordinate system (latitude and longitude). Optical satellite images were downloaded in batches from Google Maps and underwent block-by-block fine calibration using Google Earth before use. The latitude and longitude accuracy corresponding to each pixel is within 10 meters, and the satellite image resolution is 1 meter / pixel.

[0071] Figure 7 This is a schematic diagram of the scene matching results of SAR images with multiple feature points in the experiment of this invention, wherein, Figure 7 (a) is a SAR image of rural scene 1 with a large tilt angle, and the points to be matched are marked by crosshairs; Figure 7 (b) is a visible light image of rural scene 1 at a large tilt angle, with the matching results marked by crosshairs; Figure 7 (c) is a SAR image of rural scene 2 with a large tilt angle, and the points to be matched are marked by crosshairs; Figure 7 (d) is a visible light image of rural scene 2 with a large tilt angle, and the matching results are marked by crosshairs. Figure 8 This is a schematic diagram of the scene matching results of multiple feature points of optical satellite imagery in the experiment of this invention, wherein, Figure 8 (a) is a SAR image of rural scene 3 with a large tilt angle, and the points to be matched are marked by crosshairs; Figure 8 (b) is a visible light image of rural scene 3 at a large tilt angle, with the matching results marked by crosshairs; Figure 8 (c) is a SAR image of rural scene 4 with a large tilt angle, and the points to be matched are marked by crosshairs; Figure 8 (d) is a visible light image of rural scene 4 at a large tilt angle, with matching results marked by crosshairs. For example... Figure 7 and Figure 8 As shown, 13 points are selected at equal intervals along the center line of the SAR real-time image as the target center, with the 7th point being the center point of the SAR real-time image. The search area of ​​the center point on the satellite image is given by coarse SAR target localization, and the search areas of other points on the satellite image are given by their relative geometric position relationship with the center point.

[0072] Scene matching and precise localization results are as follows Figure 7 and Figure 8 As shown, it is not difficult to find that the scene matching result is correct and reliable through visual qualitative observation. In addition, in order to quantitatively give the accuracy of the SAR target fine positioning result of the present invention, the SAR target fine positioning result of the present invention method is compared with the target GPS positioning result. The specific implementation means are: (1) Select a point with obvious features on the SAR image as the target center point; (2) Perform fine positioning of the target through scene matching and give the target positioning result of the present invention method; (3) Find the spatial position of the target point manually and use differential GPS to measure the GPS positioning result of the target; (4) Compare the positioning result of the present invention method with the GPS true value. A total of 20 target positioning experiments were carried out independently, and the absolute difference between the positioning result of the present invention method and the GPS true value was within 10 meters.

[0073] In this invention's experiments, Google Maps was used as the reference map. However, due to specific reasons, even after segmented fine-tuning, the latitude and longitude accuracy of Google Maps remained around 10 meters, serving as the main source of error and significantly impacting the positioning accuracy of the method presented in this invention. In future experiments, if domestic high-resolution satellite imagery is used as the reference map, the positioning accuracy is expected to improve substantially.

[0074] In one embodiment, the present invention also provides a SAR target localization device that integrates geometric localization and heterogeneous matching, comprising: The first module is used to acquire the spatial position of the airborne radar, the spatial position of the downward-facing point, and the velocity vector of the airborne radar; the airborne radar is a synthetic aperture radar. The second module is used to calculate the first distance between the target point and the downward viewpoint based on the spatial triangle formed by the airborne radar, the downward viewpoint, and the target point to be located, using trigonometric relationships; determine the direction vector of the target point relative to the downward viewpoint according to spatial vector constraints, and directly solve the coarse coordinates of the target point in the spatial coordinate system to obtain the coarse positioning result; The spatial vector constraint relationship includes: the line vector connecting the target point and the downward viewpoint is perpendicular to both the motion velocity vector and the vector from the Earth's center to the downward viewpoint; The direct calculation process does not depend on the Doppler center frequency and radar wavelength parameters; The third module is used to determine the target template region in the real-time synthetic aperture radar image and the search region in the pre-prepared optical reference image based on the coarse positioning result; and to extract feature maps based on the local gradient direction from the real-time synthetic aperture radar image and the optical reference image respectively. The feature value of each pixel in the feature map has a non-linear correspondence with the gradient direction angle in the neighborhood of the pixel, which is used to ensure that the region where local grayscale inversion occurs presents a consistent directional expression on the feature map. The fourth module is used to match the feature map of the real-time synthetic aperture radar image with the feature map of the optical reference image within the search area using a similarity metric that is tolerant of local grayscale inversion, to determine the matching position and obtain the precise coordinates of the target point.

[0075] On the other hand, in one embodiment, the present invention provides a computer device including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps of the SAR target localization method fusion geometric positioning and heterogeneous matching provided in any of the above embodiments. The computer device may be a server. The computer device includes a processor, a memory, a network interface, and a database connected via a system bus. The processor of the computer device provides computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The database of the computer device stores sample data. The network interface of the computer device is used for communication with external terminals via a network connection.

[0076] On the other hand, in one embodiment of the present invention, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps of the SAR target localization method that fuses geometric localization and heterogeneous matching provided in any of the above embodiments.

[0077] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0078] Matters not covered in this invention are common knowledge.

[0079] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0080] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these modifications and improvements all fall within the protection scope of this application.

[0081] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A SAR target localization method integrating geometric positioning and heterogeneous source matching, characterized in that, include: Step S110: Obtain the spatial position of the airborne radar, the spatial position of the downward-facing point, and the motion velocity vector of the airborne radar; The airborne radar is a synthetic aperture radar; Step S120: Based on the spatial triangle formed by the airborne radar, the downward viewpoint, and the target point to be located, calculate the first distance between the target point and the downward viewpoint using trigonometric relationships; determine the direction vector of the target point relative to the downward viewpoint according to spatial vector constraints, and directly calculate the coarse coordinates of the target point in the spatial coordinate system to obtain the coarse positioning result; The spatial vector constraint relationship includes: the line vector connecting the target point and the downward viewpoint is perpendicular to both the motion velocity vector and the vector from the Earth's center to the downward viewpoint; The direct calculation process does not depend on the Doppler center frequency and radar wavelength parameters; Step S130: Determine the target template region in the real-time synthetic aperture radar image and the search region in the pre-prepared optical reference image based on the coarse positioning result. Feature maps based on local gradient directions are extracted from the real-time synthetic aperture radar image and the optical reference image, respectively. The feature value of each pixel in the feature map has a non-linear correspondence with the gradient direction angle in the neighborhood of the pixel, which is used to ensure that the region where local grayscale inversion occurs presents a consistent directional expression on the feature map. Step S140: Using a similarity metric that is tolerant of local grayscale inversion, the feature map of the real-time synthetic aperture radar image and the feature map of the optical reference image are matched within the search area to determine the matching position and obtain the precise coordinates of the target point.

2. The SAR target localization method integrating geometric positioning and heterogeneous source matching according to claim 1, characterized in that, In step S120, the first distance between the target point and the downward-facing viewpoint is calculated using trigonometric relationships, including: Combining the Earth's equatorial radius and the slant range of the airborne radar, and based on an Earth spherical model, the geocentric angle of the triangle formed by the Earth's center, the airborne radar, and the target point is calculated using the law of cosines. The geocentric angle is the angle between the line connecting the Earth's center to the airborne radar and the line connecting the Earth's center to the target point. The slant range of the airborne radar is calculated by the time difference between the radar's transmitted pulse and the received target echo, or it can be directly read from SAR system auxiliary data. The first distance is calculated based on the geocentric angle and the Earth's equatorial radius: the Earth's equatorial radius multiplied by the sine of the geocentric angle.

3. The SAR target localization method integrating geometric localization and heterogeneous source matching according to claim 1, characterized in that, In step S120, the direction vector of the target point relative to the downward viewpoint is given by the following formula: ； in, This represents the vector from the Earth's center to the directly below viewpoint. Represents the velocity vector of the motion. For the first distance, The coordinates of the target point in a spatial rectangular coordinate system; Represents Euclidean distance. Represents the transpose of a vector / matrix.

4. The SAR target localization method integrating geometric localization and heterogeneous source matching according to claim 1, characterized in that, The nonlinear correspondence includes: calculating the sum of the squares of the gradient vectors in the neighborhood of the pixel, and obtaining the feature value of the pixel based on the phase of the summation result, such that the feature value is numerically equal to the average gradient direction angle in the neighborhood; the average gradient direction angle is the statistical average of the gradient direction angles of all pixels in the neighborhood.

5. The SAR target localization method integrating geometric localization and heterogeneous source matching according to claim 4, characterized in that, In step S130, the method for extracting feature maps based on local gradient directions includes: Calculate the horizontal and vertical gradients of each pixel within the neighborhood of the pixel to obtain the gradient vector of all pixels within the neighborhood; the gradient direction angle within the neighborhood of the pixel is the direction of the gradient vector. Construct the covariance matrix of the gradient vector; Solve for the principal component orientation angles of the covariance matrix, and use the principal component orientation angles as the feature values ​​of the pixel. The principal component direction angle is obtained by summing the squares of the gradient vectors in the neighborhood and taking the half-phase angle. The value of the principal component direction angle is equal to the average gradient direction angle in the neighborhood.

6. The SAR target localization method integrating geometric localization and heterogeneous matching according to claim 5, characterized in that, The principal component orientation angle and the gradient vector satisfy a complex square relationship: ； in, , The components on the main diagonal of the covariance matrix, The components on the second diagonal of the covariance matrix, This represents the summation of all pixels in the neighborhood of a given pixel. , Let be the horizontal gradient and vertical gradient of each pixel within the neighborhood, respectively, and let the gradient vector be represented as... ; It is the imaginary unit.

7. The SAR target localization method integrating geometric localization and heterogeneous source matching according to claim 5 or 6, characterized in that, In step S130, the feature map extraction method based on local gradient direction further includes: For each pixel in the neighborhood of the pixel, the gradient magnitude is calculated using the horizontal and vertical gradients of the pixel. Calculate the average gradient magnitude of all pixels in the neighborhood to obtain the average gradient intensity in the neighborhood; When the average gradient intensity is lower than a preset threshold, the neighborhood of the pixel is determined to be a textureless region, and the feature value output of the pixel is suppressed.

8. The SAR target localization method integrating geometric localization and heterogeneous source matching according to claim 1, characterized in that, In step S140, the similarity measure that is tolerant of local grayscale inversion is defined in the following way: For the target template region and the sub-windows within the search region that have the same size as the target template region, a similarity score is defined: ； in: ； ； in, For the target template area at pixels eigenvalues ​​at that location For the sub-window at pixel points Eigenvalues ​​at; , These are the width and height of the target template area, respectively.

9. The SAR target localization method integrating geometric localization and heterogeneous source matching according to claim 1, characterized in that, The pre-preparation process of the optical reference image is completed offline, including: Calculate the feature map of the optical reference image based on the local gradient direction; The transformation relationship between the coordinates of pixels in the optical reference image and the geodetic coordinates is determined; The optical reference image is an optical satellite image, a digital orthophoto, or a high-precision digital map.

10. The SAR target localization method integrating geometric localization and heterogeneous matching according to claim 1, characterized in that, The method further includes: after determining the matching position, performing a confidence analysis on the matching position: if the confidence is greater than a preset confidence threshold, then outputting the coordinates corresponding to the matching position as the positioning result; otherwise, directly outputting the coarse positioning result as the final positioning result.

Images