Ground-based radar image and optical image matching method based on improved SIFT algorithm

Abstract

The invention relates to a ground-based radar image and optical image matching method based on an improved SIFT algorithm, and the method comprises the steps: carrying out the first processing of an optical image, and the first processing comprises the sky removal processing; extracting edge information of the optical image after the first processing, and taking the edge information as a part with obvious features of the optical image; performing second processing on the radar image, wherein the second processing comprises coherence windowing and coordinate conversion; extracting edge information of the radar image after the second processing, and taking the edge information as a part with obvious radar image features; and carrying out image registration by using an SIFT algorithm. The embodiments of the invention at least can overcome the problem of large image texture difference, and improve the accuracy of feature point extraction.

Description

Technical Field

[0001] The present disclosure relates to the technical fields of ground synthetic aperture radar images and optical image processing, and particularly to a method for matching ground radar images and optical images based on an improved SIFT algorithm. Background Art

[0002] In the prior art regarding the registration of radar images and optical images, there are problems such as poor extraction effect of feature information, unstable and incorrect matching, relatively high computational complexity, high requirements for computing resources, and limited generalization ability. Summary of the Invention

[0003] The present disclosure intends to provide a method for matching ground radar images and optical images based on an improved SIFT algorithm, which can at least overcome the problem of large differences in image textures and improve the accuracy of feature point extraction.

[0004] According to one aspect of the present disclosure, there is provided a method for matching ground radar images and optical images based on an improved SIFT algorithm, including:

[0005] Performing a first processing on the optical image, where the first processing includes sky removal processing;

[0006] Extracting the edge information of the optical image after the first processing as the part with obvious features of the optical image;

[0007] Performing a second processing on the radar image, where the second processing includes coherence windowing and coordinate transformation;

[0008] Extracting the edge information of the radar image after the second processing as the part with obvious features of the radar image;

[0009] Performing image registration using the SIFT algorithm.

[0010] In some embodiments, the sky removal processing includes:

[0011] Selecting the sky part in the optical image and converting each pixel point in the sky part to the YCbCr color space, where Y, Cb, and Cr respectively represent the brightness intensity, blue intensity, and red intensity of the image;

[0012] Selecting the pixels that simultaneously meet the conditions of Y, Cb, and Cr for sky removal processing.

[0013] In some embodiments, extracting the edge information of the optical image after the first processing includes:

[0014] Performing Gaussian filtering on the optical image after the first processing;

[0015] Calculating the gradient of the image after Gaussian filtering;

[0016] Compare the gradient amplitude of the current pixel along the gradient direction, retain the local maximum value, and suppress non-maximum values;

[0017] Set high and low thresholds for double-threshold detection and edge connection. Those with amplitudes greater than the high threshold are strong edges, and those with amplitudes between the high and low thresholds are weak edges. If there is a strong edge in the neighborhood of a weak edge, it is retained as a valid edge.

[0018] In some embodiments, wherein, coherence windowing includes:

[0019] Based on setting the mask data window, obtain the windowed echo data;

[0020] Perform visualization processing.

[0021] In some embodiments, wherein, coordinate transformation includes:

[0022] Reconstruct the windowed echo data in complex form to form a two-dimensional data matrix;

[0023] Calculate the amplitude of the echo data;

[0024] Convert from polar coordinates to rectangular coordinates;

[0025] Visualize the radar data to obtain the radar image in the rectangular coordinate system.

[0026] In some embodiments, wherein, extracting the edge information of the second processed radar image includes:

[0027] Extract the edge information of the radar image through local operations of dilation and erosion, and utilize the interaction between the structural element and the image.

[0028] In some embodiments, wherein, using the SIFT algorithm for image registration includes:

[0029] Use the obtained image edge information as feature information for image registration, calculate the gradient amplitude of the feature points and the pixel gradient directions around each feature point, and statistically calculate the gradient direction histogram within the window;

[0030] Calculate the rotation invariant descriptor of the window;

[0031] The feature points that meet the conditions are used as matching points, calculate the transformation relationship between the two images and describe the transformation relationship using a homography matrix;

[0032] Utilize the obtained homography matrix to map one image to the coordinate system of the other image through this transformation matrix, thereby achieving image registration.

[0033] In some embodiments, wherein, find the feature points with similar local image structures in different images as matching points according to the descriptors.

[0034] In some embodiments, if the minimum distance of a feature point is within a predetermined multiple of the second smallest distance, it is regarded as a matching point.

[0035] In some embodiments, a set of matching points is randomly selected, the corresponding homography matrix is calculated, and it is verified whether most points conform to the transformation. If the number of conforming points is large, the transformation is considered valid.

[0036] The ground-based radar image and optical image matching method based on the improved SIFT algorithm according to various embodiments of the present disclosure performs at least a first processing on the optical image. The first processing includes sky removal processing; extracting the edge information of the first-processed optical image as the part with obvious features of the optical image; performing a second processing on the radar image. The second processing includes coherence windowing and coordinate transformation; extracting the edge information of the second-processed radar image as the part with obvious features of the radar image; using the SIFT algorithm for image registration, thereby being able to overcome the problem of large differences in image textures and improving the accuracy of feature point extraction. Moreover, the matching of the ground-based radar image and the optical image performed on the basis of edge detection has a relatively fast operation speed and saves time.

[0037] It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure as claimed. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In the drawings, which are not necessarily drawn to scale, like reference numerals in different views may represent like components. Like reference numerals with alphabetic suffixes or like reference numerals with different alphabetic suffixes may represent different instances of like components. The drawings generally illustrate various embodiments by way of example and not limitation, and are used in conjunction with the specification and the claims to explain the disclosed embodiments.

[0039] Figure 1 Shows a flowchart of the ground-based radar image and optical image matching method based on the improved SIFT algorithm according to an embodiment of the present disclosure;

[0040] Figure 2 Shows the result after edge detection in an embodiment of the present disclosure;

[0041] Figure 3 Shows the visualization result after windowing in an embodiment of the present disclosure;

[0042] Figure 4 Shows a radar image in a rectangular coordinate system obtained in an embodiment of the present disclosure;

[0043] Figure 5 Shows the edge information extracted in an embodiment of the present disclosure;

[0044] Figure 6 Shows a registration result in an embodiment of the present disclosure. Detailed implementation manners

[0045] In order to make the objectives, technical solutions and advantages of the embodiments of the present disclosure clearer, the technical solutions of the embodiments of the present disclosure will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present disclosure. Obviously, the described embodiments are some but not all of the embodiments of the present disclosure. All other embodiments obtained by those of ordinary skill in the art based on the described embodiments of the present disclosure without creative efforts shall fall within the scope of protection of the present disclosure.

[0046] Compared with sensors such as optics, Synthetic Aperture Radar (SAR) is not restricted by illumination and weather conditions and can achieve active earth observation all day and all weather. SAR image registration is one of the basic technologies in image processing. The goal of registration is to find the corresponding relationships between two or more SAR images, align them in the same coordinate system, and make them have the same geometric spatial information, so as to realize the comparison and analysis of SAR images at different times or under different conditions, and thus realize subsequent applications such as change detection, image fusion, and target detection. Therefore, the research on SAR image registration has strong practical significance.

[0047] There are great differences in the imaging mechanisms, spectral and radiation characteristics, as well as image noise and texture between SAR images and optical images. SAR images are easily affected by speckle noise and present speckled textures, while optical images have relatively smooth visual characteristics, which will also interfere with the accurate detection of feature points, resulting in the appearance of false feature points and seriously affecting the subsequent feature point matching work. Therefore, how to extract stable, effective, and highly repeatable feature points to achieve high-precision SAR image registration is an urgent problem to be solved in the current research on SAR image registration and is also a research hotspot of concern.

[0048] In order to improve the registration accuracy of radar images and optical images, in some existing technologies, first is the feature-based image registration technology, which mainly relies on extracting feature points (such as corner points, edges, etc.) from the images, and then performing matching and transformation estimation. However, the extraction of feature information depends on the image content. For images with less texture and lack of obvious features, there are problems such as poor extraction effect of feature information, unstable matching, and false matching. The second type of method is the region-based image registration method, which performs matching and transformation estimation by comparing the gray level, texture, or statistical features of corresponding regions in the images. Different from the feature-based method, it does not rely on feature points but directly uses the local information of the images. However, it needs to calculate the similarity of the entire region (such as mutual information or correlation coefficient), rather than just processing a small number of feature points. Therefore, the computational complexity is relatively high, and many region-based registration methods are not suitable for large-scale rotation, scaling, and perspective changes. The third type is the deep learning-based image registration technology, which uses models such as convolutional neural network (CNN), self-supervised learning, generative adversarial network (GAN), or deformation network (such as U-Net, Siamese network, STN (Spatial Transformer Network)) to achieve image registration. Compared with traditional methods, deep learning can automatically learn features and optimize registration parameters in an end-to-end framework. It has advantages in feature learning, non-rigid deformation processing, and computational speed, but it relies on a large amount of training data, has high requirements for computing resources, limited generalization ability, is difficult to handle extreme registration situations, has high training costs, and high requirements for data annotation and synthetic data.

[0049] Combined with the content recorded in the foregoing background art section, the present disclosure exemplarily records corresponding solutions in the form of embodiments to solve the defects existing in the prior art, but does not serve as a limitation on the scope of patent rights claimed by the present disclosure.

[0050] As one of the solutions, as Figure 1 FIG. shows a schematic flowchart of the implementation process of the method for matching ground-based radar images and optical images based on an improved SIFT algorithm (Scale-invariant feature transform or SIFT) according to an embodiment of the present disclosure. The embodiment of the present disclosure provides a method for matching ground-based radar images and optical images based on an improved SIFT algorithm, including:

[0051] Perform a first processing on the optical image, and the first processing includes sky removal processing;

[0052] Extract the edge information of the optical image after the first processing as the part with obvious features of the optical image;

[0053] Perform a second processing on the radar image, and the second processing includes coherence windowing and coordinate transformation;

[0054] Extract the edge information of the second processed radar image as the part with obvious radar image features;

[0055] Use the SIFT algorithm for image registration.

[0056] Regarding the foregoing content, each embodiment of the present disclosure aims to propose a method for matching ground-based radar images and optical images based on an improved SIFT algorithm, which can overcome the problem of large differences in image textures and improve the accuracy of feature point extraction. Moreover, the matching of ground-based radar images and optical images performed on the basis of edge detection has a relatively fast operation speed and saves time.

[0057] The steps of the embodiments of the present disclosure are exemplarily described below by taking Steps 1 to 5 as examples to further illustrate the method for matching ground-based radar images and optical images based on the improved SIFT algorithm of the present disclosure.

[0058] In some specific implementation manners, the present disclosure may be sky removal processing, including:

[0059] Select the sky part in the optical image, and convert each pixel point in the sky part to the YCbCr color space, where Y, Cb, and Cr respectively represent the brightness intensity, blue intensity, and red intensity of the image;

[0060] Select the pixels that simultaneously meet the Y, Cb, and Cr conditions for sky removal processing.

[0061] Step 1: Sky removal processing of the optical image.

[0062] In order to reduce the influence of the sky and other factors that are not conducive to extracting image feature information, the embodiments of the present disclosure first perform sky removal processing on the optical image, decompose the image into the YCbCr color space, where Y, Cb, and Cr respectively represent the brightness intensity, blue intensity, and red intensity of the image. The specific process is as follows:

[0063] In the first step, manually select the sky part in the optical image, and convert each pixel point in the sky part from the RGB color space to the YCbCr color space. The conversion formula is as follows:

[0064] Y = 0.299R + 0.587G + 0.114B

[0065] Cb = 128 + (-0.168736R - 0.331264G + 0.5B)

[0066] Cr = 128 + (0.5R - 0.418688G - 0.081312B).

[0067] In the second step, calculate the maximum value and minimum value of each component, and select the pixels that simultaneously meet the Y, Cb, and Cr conditions for sky removal processing. The formula is as follows:

[0068] threshold1 = (Y ∈ [Y min , Y max ) ∩ (Cb ∈ [Cb min , Cb max ) ∩ (Cr ∈ [Cr min , Cr max )

[0069] Among them, threshold1 is the sky pixel region that simultaneously satisfies Y, Cb, and Cr.

[0070] In some specific embodiments, the present disclosure may be to extract the edge information of the first processed optical image, including:

[0071] Perform Gaussian filtering on the first processed optical image;

[0072] Calculate the gradient of the image after Gaussian filtering;

[0073] Compare the gradient magnitude of the current pixel along the gradient direction, retain the local maximum value, and suppress non-maximum values;

[0074] Set high and low thresholds for double-threshold detection and edge connection. If the magnitude is greater than the high threshold, it is a strong edge. If the magnitude is between the high threshold and the low threshold, it is a weak edge. When there is a strong edge in the neighborhood of the weak edge, it is retained as a valid edge.

[0075] The embodiment of the present disclosure may further include step 2 on the basis of the above step 1.

[0076] Step 2: Use the canny operator to extract the edge information of the optical image.

[0077] In order to obtain a more obvious feature part, the canny algorithm is used to perform edge detection on the optical image obtained in step 1, and the edge part is extracted as the part with obvious features. The specific steps are as follows:

[0078] The first step is to perform Gaussian filtering on the optical image obtained in step 1. The purpose is to reduce the interference of noise on the subsequent gradient calculation. The formula is as follows:

[0079]

[0080] Among them, σ is the standard deviation, x and y are pixel coordinates, representing the spatial position of the image.

[0081] The second step is to use the Sobel operator to calculate the gradient of the image after Gaussian filtering. The Sobel operator is divided into the horizontal direction operator G X and the vertical direction operator G Y, convolving the operator with the image to obtain the gradients in the x-axis and y-axis directions of the image, the formula is as follows:

[0082] G x = I * G x

[0083] G y = I * G y

[0084]

[0085] where I is the original image, * represents convolution, grad represents the gradient magnitude, and angle represents the gradient direction.

[0086] In the third step, compare the gradient magnitudes of the current pixel along the gradient direction, retain the local maximum, and suppress non-maximum values. The gradient direction angle is discretized into four main directions: 0°, 45°, 90°, and 135°. If the gradient direction is 0°, compare the magnitudes of the left and right pixels. If the gradient direction is 45°, compare the magnitudes of the upper-left and lower-right pixels. If the gradient direction is 90°, compare the magnitudes of the upper and lower pixels. If the gradient direction is 135°, compare the magnitudes of the lower-left and upper-right pixels. The formula for obtaining the maximum value is as follows:

[0087]

[0088] In the fourth step, set a high threshold T high and a low threshold T low to perform double-threshold detection and edge connection. Magnitude > T high is a strong edge; T low < Magnitude < T high is a weak edge; Magnitude < T low is directly removed. Check whether there is a strong edge in the neighborhood of the weak edge. If so, retain it as a valid edge. The result after edge detection is as Figure 2 shown.

[0089] In some specific implementation schemes, the present disclosure may be coherence windowing, including:

[0090] Based on setting a mask data window, obtain the echo data after windowing;

[0091] Perform visualization processing.

[0092] The embodiment of the present disclosure may further include step 3 on the basis of the above step 2.

[0093] Step 3: Perform coherence windowing and coordinate transformation on the radar image

[0094] When performing target monitoring tasks, the synthetic aperture radar scans the target object in a rectangular shape, which will cause the generated echo data to include data of other objects in addition to the monitored target. This will pose a great difficulty for subsequent image registration. In order to retain only the target information in the radar image, we perform coherence windowing on the echo data, set a mask data window, where the target information to be retained is represented inside the window with a value of 1, and the interference information to be removed is represented outside the window with a value of 0. The windowing formula is as follows:

[0095] masked data = original data × mask

[0096] Among them, original data represents the original echo data of the radar, and masked data represents the echo data after windowing. The visualization result after windowing is as shown in Figure 3 shown.

[0097] In some specific implementation manners, the present disclosure may be coordinate transformation, including:

[0098] Reconstruct the windowed echo data in complex form to form a two-dimensional data matrix;

[0099] Calculate the amplitude of the echo data;

[0100] Use polar coordinates to find rectangular coordinates;

[0101] Visualize the radar data to obtain a radar image in the rectangular coordinate system.

[0102] In order to facilitate subsequent image registration, it is necessary to convert the radar image after windowing from the polar coordinate system to the rectangular coordinate system. The specific process is as follows:

[0103] The first step is to reconstruct the windowed echo data in complex form to form a two-dimensional data matrix

[0104] defo = real part + j × imaginary part.

[0105] The second step is to calculate the amplitude of the echo data

[0106]

[0107] Among them, r represents the azimuth vector of the radar, and q represents the radar angle vector.

[0108] The third step is to use polar coordinates to find rectangular coordinates

[0109] x = rcos(θ), y = rsin(θ).

[0110] The fourth step is to visualize the radar data to obtain a radar image in the rectangular coordinate system as shown in Figure 4 shown.

[0111] In some specific embodiments, the present disclosure may be to extract the edge information of the second processed radar image, including:

[0112] Extract the edge information of the radar image by local operations of dilation and erosion, using the interaction between the structuring element and the image.

[0113] The embodiment of the present disclosure may further include step S4 on the basis of the above step S3.

[0114] Step 4: Use the morphological method to extract the edge information of the radar image.

[0115] The morphological method extracts the edge information of the radar image by local operations of dilation and erosion, using the interaction between the structuring element and the image. The extracted edge information is used as the part with obvious features in the radar image.

[0116] In the first step, performing a dilation operation on the image obtained in step 3 will expand the bright regions in the image, emphasize the boundaries of the objects in the image, and highlight the outside of the edges. The formula is as follows:

[0117]

[0118] Where, represents the dilation operation, and S is the structuring element.

[0119] In the second step, performing an erosion operation on the image obtained in step 3 will shrink the bright regions in the image, remove the small regions in the image, and highlight the inside of the edges. The formula is as follows:

[0120]

[0121] In the third step, perform a difference operation on the dilated and eroded images to extract the edges by the difference between the dilated image and the eroded image. The specific formula is as follows:

[0122]

[0123] Where, and represent the dilation and erosion operations respectively, S is the structuring element, I is the original image, and E is the extracted edge.

[0124] The embodiment of the present disclosure can enhance the edge information of the radar image through the morphological method. The extracted edge information is as Figure 5 shown.

[0125] In some specific embodiments, the present disclosure may be to perform image registration using the SIFT algorithm, including:

[0126] Use the obtained image edge information as feature information for image registration, calculate the gradient magnitude of the feature points and the pixel gradient directions around each feature point, and statistically calculate the gradient direction histogram within the window;

[0127] Calculate the rotation invariant descriptor of the window;

[0128] The feature points that meet the conditions are used as matching points, calculate the transformation relationship between the two images and use the homography matrix to describe the transformation relationship;

[0129] Utilize the obtained homography matrix to map one image to the coordinate system of another image through this transformation matrix, thereby achieving image registration.

[0130] The embodiments of the present disclosure may further include step S5 on the basis of the above step S4.

[0131] Step 5: Use the SIFT algorithm for image registration.

[0132] Use the image edge information obtained in step 2 and step 4 as feature information for image registration, calculate the gradient magnitude of the feature points and the pixel gradient directions around each feature point, and statistically calculate the gradient direction histogram within a 16x16 pixel window, and then calculate the rotation invariant descriptor of the window.

[0133] In some specific implementation manners, the present disclosure may be that, according to the descriptor, find the feature points with similar local image structures in different images as matching points.

[0134] According to the descriptor, find the feature points with similar local image structures in different images as matching points, and calculate the Euclidean distance between the two descriptors for the similarity measurement formula as follows:

[0135]

[0136] Where, d1 (i) and d2 (i) are the values of the two descriptors in the i-th dimension respectively.

[0137] In some specific implementation manners, the present disclosure may be that if the minimum distance of the feature point is within a predetermined multiple of the second smallest distance, it is used as a matching point.

[0138] To reduce false matches, a ratio test can be used. For example, if the minimum distance of the matching point is within 0.8 times of the second smallest distance, it is considered a matching point:

[0139]

[0140] Where, D min is the descriptor distance of the nearest neighbor, D secondIt is the descriptor distance of the second nearest neighbor.

[0141] In some specific embodiments, the present disclosure may be that a set of matching points is randomly selected, the corresponding homography matrix is calculated, and it is verified whether most points conform to the transformation. If the number of conforming points is large, the transformation is considered valid.

[0142] The feature points that meet the conditions are used as matching points, the transformation relationship between the two images is calculated, and the homography matrix is used to describe this transformation. To remove incorrect matches, the RANSAC (Random Sample Consensus) algorithm is used to randomly select a set of matching points, calculate the corresponding homography matrix, and verify whether most points conform to the transformation. If the number of conforming points is large, the transformation is considered valid. Using the obtained homography matrix, one image is mapped to the coordinate system of another image through this transformation matrix, thereby completing image registration. The registration result is as Figure 6 shown.

[0143] The ground-based radar image and optical image matching method based on the improved SIFT algorithm in each embodiment of the present disclosure has at least the following beneficial effects compared with the prior art:

[0144] 1. The imaging mechanisms of optical images and radar images are very different, and radar images are vulnerable to speckle noise. Optical images have relatively smooth visual features, which will result in low accuracy in feature point extraction, incorrect matching points, and affect subsequent feature point matching work. Each embodiment of the present disclosure can overcome the problem of large differences in image textures and improve the accuracy of feature point extraction;

[0145] 2. Each embodiment of the present disclosure is based on edge detection, and its operation speed is relatively fast, saving time.

[0146] As one of the solutions, the embodiment of the present disclosure provides a ground-based radar image and optical image matching device based on the improved SIFT algorithm, including at least one processing module, which can be configured to implement the ground-based radar image and optical image matching method based on the improved SIFT algorithm described in the above steps 1 to 5.

[0147] Specifically, one of the inventive concepts of the present disclosure is a method for matching ground-based radar images and optical images based on an improved SIFT algorithm in various embodiments of the present disclosure. At least the optical image is subjected to a first process, and the first process includes sky removal processing; extracting edge information of the optical image after the first process as the part with obvious features of the optical image; performing a second process on the radar image, and the second process includes coherence windowing and coordinate transformation; extracting edge information of the radar image after the second process as the part with obvious features of the radar image; using the SIFT algorithm for image registration, thereby being able to overcome the problem of large differences in image textures and improving the accuracy of feature point extraction. Moreover, the matching of ground-based radar images and optical images performed on the basis of edge detection has a relatively fast operation speed and saves time.

[0148] The present disclosure also provides a computer-readable storage medium, on which computer-executable instructions are stored. When the computer-executable instructions are executed by a processor, it mainly implements the method for matching ground-based radar images and optical images based on the improved SIFT algorithm as described above, and at least includes:

[0149] Performing a first process on the optical image, and the first process includes sky removal processing;

[0150] Extracting edge information of the optical image after the first process as the part with obvious features of the optical image;

[0151] Performing a second process on the radar image, and the second process includes coherence windowing and coordinate transformation;

[0152] Extracting edge information of the radar image after the second process as the part with obvious features of the radar image;

[0153] Using the SIFT algorithm for image registration.

[0154] The above embodiments are only exemplary embodiments of the present disclosure and are not used to limit the present disclosure. The protection scope of the present disclosure is defined by the claims. Those skilled in the art can make various modifications or equivalent replacements within the essence and protection scope of the present disclosure, and such modifications or equivalent replacements should also be regarded as falling within the protection scope of the present disclosure.

Claims

1. A method for matching ground-based radar images and optical images based on an improved SIFT algorithm, comprising: Performing a first processing on the optical image, the first processing including sky removal processing; Extracting the edge information of the optical image after the first processing as the part with obvious features of the optical image; Performing a second processing on the radar image, the second processing including coherence windowing and coordinate transformation; Extracting the edge information of the radar image after the second processing as the part with obvious features of the radar image; Performing image registration using the SIFT algorithm.

2. The method according to claim 1, wherein, The sky removal processing includes: Selecting the sky part in the optical image, and converting each pixel point in the sky part to the YCbCr color space, where Y, Cb, and Cr respectively represent the brightness intensity, blue intensity, and red intensity of the image; Selecting the pixels that satisfy the conditions of Y, Cb, and Cr simultaneously for sky removal processing.

3. The method according to claim 2, wherein, Extracting the edge information of the optical image after the first processing includes: Performing Gaussian filtering on the optical image after the first processing; Calculating the gradient of the image after Gaussian filtering; Comparing the gradient amplitude of the current pixel along the gradient direction, retaining the local maximum value, and suppressing non-maximum values; Setting a high threshold and a low threshold for double-threshold detection and edge connection. The amplitude greater than the high threshold is a strong edge, and the amplitude between the high threshold and the low threshold is a weak edge. When there is a strong edge in the neighborhood of the weak edge, it is retained as a valid edge.

4. The method according to claim 3, wherein, The coherence windowing includes: Based on setting a mask data window, obtaining the windowed echo data; Performing visualization processing.

5. The method according to claim 4, wherein The coordinate transformation includes: Reconstructing the windowed echo data in complex form to form a two-dimensional data matrix; Calculating the amplitude of the echo data; Calculating the rectangular coordinates using polar coordinates; Visualizing the radar data to obtain the radar image in the rectangular coordinate system.

6. The method according to claim 5, wherein, Extracting the edge information of the radar image after the second processing includes: Extracting the edge information of the radar image through local operations of dilation and erosion, and using the interaction between the structuring element and the image.

7. The method according to claim 6, wherein, Performing image registration using the SIFT algorithm includes: Using the obtained image edge information as feature information for image registration, calculating the gradient amplitude of the feature points and the pixel gradient direction around each feature point, and statistically calculating the gradient direction histogram within the window; Calculating the rotation-invariant descriptor of the window; The feature points that meet the conditions are used as matching points, calculating the transformation relationship between the two images and using the homography matrix to describe the transformation relationship; Using the obtained homography matrix to map one image to the coordinate system of the other image through the transformation matrix, thereby performing image registration.

8. The method according to claim 7, wherein, Searching for feature points with similar local image structures in different images as matching points according to the descriptor.

9. The method according to claim 8, wherein If the minimum distance of the feature points is within a predetermined multiple of the second minimum distance, it is used as a matching point.

10. The method according to claim 9, wherein, Randomly selecting a group of matching points, calculating the corresponding homography matrix, and verifying whether most points conform to the transformation. If the number of conforming points is large, the transformation is considered valid.

Images