As-built drawing and remote sensing image geographical registration method, device and equipment and storage medium

Abstract

The application provides a method, device and equipment for geographic registration of a completed map and a remote sensing image, and a storage medium. The application relates to the field of geographic information. The method comprises the following steps: acquiring a completed map to be registered and a remote sensing satellite image containing geographic information; respectively pre-processing the completed map and the remote sensing satellite image; performing multi-modal image matching on the pre-processed completed map and the pre-processed remote sensing satellite image to obtain a matched feature point pair; calculating geographic transformation parameters of the completed map to the remote sensing satellite image according to the matched feature point pair; transforming the completed map into a coordinate system of the remote sensing satellite image by using polynomial transformation to obtain a transformed completed map; embedding a reference coordinate system and the geographic transformation parameters into the transformed completed map; and saving the transformed completed map with embedded metadata as a registration image in a set format. The application can realize accurate registration and provide image data with unified geographic information for scenes such as high-standard farmland construction projects.

Description

Technical Field

[0001] The present application relates to the field of geographic information technology, and in particular to a method, device, equipment and storage medium for geographic registration of as-built drawings and remote sensing images. Background Art

[0002] In the current management of high-standard farmland construction projects, accurate geo-registration is crucial for the fusion analysis of project completion drawings and remote sensing images. Traditional registration methods mainly rely on the manual selection of feature points, and determine the geometric transformation relationship between the two images through manual measurement and calculation. For example, in some early farmland construction projects, workers would mark obvious land features such as road intersections and river bends on the completion drawings and remote sensing images, and then use simple mathematical models to calculate translation, rotation, and scaling parameters to achieve preliminary image registration. Although this method is simple in principle, it is extremely inefficient, and the registration accuracy depends largely on the operator's experience and skill level, which is prone to human error.

[0003] With the development of computer technology and image processing algorithms, automated registration methods based on feature matching have gradually been applied. Among them, the scale-invariant feature transform (SIFT) algorithm is a relatively common one. The SIFT algorithm detects scale-invariant feature points in the image, calculates their feature descriptors, and then uses the similarity between feature descriptors to match feature points, thereby determining the transformation relationship between images. In farmland image registration, this algorithm can, to a certain extent, overcome the influence of factors such as image scale changes, rotation, and illumination changes, thereby improving the accuracy of registration. In addition, the accelerated robust feature SURF algorithm has also been used in some application scenarios with high real-time requirements due to its fast calculation speed and certain robustness to noise.

[0004] In recent years, deep learning-based registration methods have also begun to gain prominence. Deep learning models, such as convolutional neural networks (CNNs), can automatically learn high-level image features, enabling more accurate image matching and registration. Some studies have attempted to use CNNs to extract features from as-built drawings and remote sensing imagery and train models to predict the transformation parameters between the two. This approach has demonstrated good performance when processing complex scene images and is adaptable to different types of farmland landscapes and image features.

[0005] Although existing technologies have made some progress in the geo-registration of completion drawings of high-standard farmland construction projects with remote sensing images, there are still many shortcomings.

[0006] Traditional manual registration methods and early automated methods based on feature matching have significant limitations in accuracy. Due to the complex and diverse features of farmland objects, and the fact that some features appear differently in different images, the accuracy of feature point matching is difficult to guarantee. For example, in some mountainous farmlands with complex terrain, the extraction and matching of object features are prone to errors due to the undulating terrain and obstruction by vegetation. This results in significant deviations in key information such as farmland boundaries and water conservancy facilities in the registered images, making them unable to meet the requirements of refined management of high-standard farmland construction projects.

[0007] When processing large-scale, high-standard farmland construction project data, existing methods are often inefficient. Traditional feature matching algorithms consume a lot of time and computing resources in the calculation of feature points and the matching process. Especially when the image data volume is large or the resolution is high, the processing speed drops significantly. For some application scenarios that require real-time registration results, such as real-time monitoring of farmland construction progress, this inefficient registration method cannot meet actual needs. Although deep learning-based methods have certain advantages in accuracy, model training requires a large amount of labeled data and powerful computing equipment. The training process is time-consuming, and the deployment and operation of the model also have high hardware requirements, which limits its widespread application in actual projects.

[0008] Existing registration methods also have problems with applicability. As-built drawings and remote sensing images from different sources and qualities vary significantly in imaging conditions, resolution, and color modes, making it difficult for existing registration algorithms to effectively adapt to these complex and changing image data. For example, some older as-built drawings may exhibit image blur and distortion, while remote sensing images acquired at different times may be affected by factors such as weather and season, resulting in changes in image features. In such cases, existing registration methods often fail to accurately identify the correspondence between the two images, significantly reducing the registration effectiveness. Summary of the Invention

[0009] This application provides a method, device, equipment and storage medium for geo-registration of as-built drawings and remote sensing images to solve the problems of insufficient accuracy, low efficiency and poor adaptability of existing registration methods when processing geo-registration of as-built drawings and remote sensing images of high-standard farmland construction projects.

[0010] In a first aspect, the present application provides a method for geo-registration of as-built drawings and remote sensing images, comprising:

[0011] Obtaining an as-built drawing to be registered and a remote sensing satellite image containing geographic information; wherein the as-built drawing is a raster image that does not contain geographic information, and the remote sensing satellite image is a reference image that contains geographic information;

[0012] Preprocessing the as-built drawings and remote sensing satellite images respectively, wherein the preprocessing includes grayscale processing, noise reduction processing, and image enhancement processing;

[0013] The pre-processed as-built drawings and pre-processed remote sensing satellite images are input into the MINIMA multimodal image matching unified framework, and the RoMa image matching model is used to perform multimodal image matching. The feature points of the pre-processed as-built drawings and pre-processed remote sensing satellite images are extracted and matched to obtain matched feature point pairs.

[0014] Calculating geographic transformation parameters from the as-built drawing to the remote sensing satellite image based on the matched feature point pairs, and transforming the as-built drawing into the coordinate system of the remote sensing satellite image using a polynomial transformation to obtain a transformed as-built drawing;

[0015] embedding a reference coordinate system and geographic transformation parameters into the transformed as-built drawing;

[0016] Save the transformed as-built drawing with embedded metadata as a registered image in a specified format.

[0017] In one possible design, the grayscale processing includes:

[0018] Obtaining values ​​of the red channel, green channel, and blue channel of the as-built drawing and the remote sensing satellite image;

[0019] The grayscale value of the as-built drawing or remote sensing satellite image is calculated using the following formula:

[0020] ,

[0021] in, Gray Represents the grayscale value after conversion, R Represents the red component of the pixel, G represents the green component, B Represents the blue component, with weights of 0.299, 0.587, and 0.114 reflecting the sensitivity ratios of the human eye to red, green, and blue, respectively.

[0022] In one possible design, the noise reduction processing includes:

[0023] The grayscale image obtained based on the grayscale processing is filtered using the following formula:

[0024] ,

[0025] in, Represents the grayscale value of the center pixel after filtering, Represents each pixel in the neighborhood ( i , j )’s original grayscale value, Indicates the pixel point (x , y ) is the neighborhood window centered on .

[0026] In one possible design, the image enhancement processing includes:

[0027] Based on the denoised image obtained by the denoising process, performing statistics on the grayscale values ​​of all pixels in the denoised image, recording the frequency of occurrence of each grayscale level, and obtaining a grayscale histogram;

[0028] Normalizing the histogram to obtain a probability density function, wherein the probability density function reflects the distribution probability of each gray value in the image;

[0029] According to the probability density function, the cumulative distribution function value is calculated by the following formula:

[0030] ,

[0031] in, T ( k ) represents the grayscale k The cumulative distribution function value of n i Indicates grayscale i The number of pixels that appear, N Indicates the total number of pixels in the image;

[0032] The gray value of each pixel in the original denoised image r Mapped to a new grayscale value, the mapping function is:

[0033] ,

[0034] in, s Represents the new grayscale value after mapping, L Represents the total number of gray levels, T ( r ) is the grayscale value r The cumulative distribution function value of .

[0035] In one possible design, the preprocessed as-built drawings and preprocessed remote sensing satellite images are input into the MINIMA multimodal image matching unified framework, and the RoMa image matching model is used for multimodal image matching. Feature points of the preprocessed as-built drawings and preprocessed remote sensing satellite images are extracted and matched to obtain matched feature point pairs, including:

[0036] performing multi-scale Gaussian blur processing on the pre-processed as-built drawing and remote sensing satellite image to generate a Gaussian pyramid, constructing a Gaussian difference pyramid by taking differences between adjacent Gaussian blurred images, and detecting extreme points in the scale space based on the Gaussian difference pyramid;

[0037] Positioning the extreme points, removing edge unstable points and low contrast points by fitting the gradient information of pixels around the extreme points, and retaining the key points;

[0038] Construct a direction histogram based on the gradient direction distribution of pixels in the neighborhood of each key point, and select the direction with the largest amplitude in the direction histogram as the main direction of the key point;

[0039] Taking the key point after the main direction is assigned as the center, sub-regions are divided in the neighborhood of the key point, and the gradient amplitude and direction information in each sub-region are counted to generate a feature vector composed of multiple direction histograms. Each key point corresponds to one feature vector;

[0040] Calculating the Euclidean distance between the feature vectors of the feature points in the as-built image and the feature vectors of the feature points in the remote sensing satellite image; selecting, according to a set distance threshold, feature point pairs whose Euclidean distances are less than the distance threshold as the feature point pairs for preliminary matching;

[0041] A part of the feature point pairs from the preliminary matched feature point pairs are randomly selected as candidate inlier points sets, and an initial geometric transformation model is calculated based on the candidate inlier points sets; the positional relationship of other feature point pairs is predicted using the initial geometric transformation model, and the number of feature point pairs that meet the geometric constraints of the initial geometric transformation model is counted as the number of inliers; the random sampling and model evaluation process is repeated, and the feature point pair corresponding to the geometric transformation model with the largest number of inliers is taken as the final matched feature point pair.

[0042] In one possible design, based on the matched feature point pairs, geographic transformation parameters from the as-built drawing to the remote sensing satellite image are calculated, and polynomial transformation is used to transform the as-built drawing into the coordinate system of the remote sensing satellite image to obtain a transformed as-built drawing, including:

[0043] A quadratic polynomial transformation model is established; wherein the quadratic polynomial transformation model is expressed as:

[0044] ,

[0045] ,

[0046] in, x and y are the horizontal and vertical coordinates in the as-built drawing, x 'and y ' is the horizontal and vertical coordinates in the remote sensing satellite image, a 0. a 1. a 2. a 3. a 4. a 5. b0. b 1. b 2. b 3. b 4 and b 5 are all polynomial coefficients;

[0047] The error of each pair of matched feature points is determined by the following formula e i :

[0048] ,

[0049] in, x i and y i The first i feature point coordinates, and for x i and y i Corresponding coordinates in remote sensing satellite images;

[0050] Based on the error of each pair of matched feature points, determine the total error objective function E for:

[0051] ,

[0052] in, n is the number of characteristic points in the as-built drawing;

[0053] Substitute all matched feature point pairs into the quadratic polynomial transformation model to construct a linear equation system, introduce matrix expression, express unknown parameters in vector form, and use matrix operation method to solve the total error objective function. E The minimum value of determines the polynomial coefficients and obtains the geographic transformation parameters;

[0054] The as-built drawing is subjected to coordinate transformation processing based on the geographic transformation parameters, so as to project all feature points in the as-built drawing into the coordinate system of the remote sensing image according to a quadratic polynomial transformation model, thereby obtaining a transformed as-built drawing.

[0055] In one possible design, the transformed as-built drawing with the metadata embedded therein is saved as a registered image in TIF format. When saving in TIF format, a quality check is performed on the generated registered image in TIF format, and the registered image that passes the quality check is used as the final registered image; wherein the quality check includes a geometric accuracy check and a metadata integrity check;

[0056] The geometric accuracy check includes: comparing the coordinates of known objects in the registered image with the actual geographic coordinates, calculating the error, and determining that the geometric accuracy check has passed if the error is within a set threshold range;

[0057] The metadata integrity check includes: reading metadata of the registered image, and verifying whether the read metadata is consistent with metadata generated during the registration process. If they are consistent, it is determined that the metadata integrity check has passed.

[0058] In a second aspect, the present application provides a device for geo-registration of as-built drawings and remote sensing images, the device comprising:

[0059] a data acquisition module configured to acquire an as-built drawing to be registered and a remote sensing satellite image containing geographic information; wherein the as-built drawing is a raster image that does not contain geographic information, and the remote sensing satellite image is a reference image that contains geographic information;

[0060] a data preprocessing module configured to preprocess the as-built drawings and remote sensing satellite images respectively, wherein the preprocessing includes grayscale processing, noise reduction processing, and image enhancement processing;

[0061] The feature matching module is configured to input the pre-processed as-built drawing and the pre-processed remote sensing satellite image into the MINIMA multimodal image matching unified framework, perform multimodal image matching using the RoMa image matching model, extract feature points of the pre-processed as-built drawing and the pre-processed remote sensing satellite image, and match them to obtain matched feature point pairs;

[0062] a projection transformation module configured to calculate geographic transformation parameters from the as-built drawing to the remote sensing satellite image based on the matched feature point pairs, and transform the as-built drawing into the coordinate system of the remote sensing satellite image using a polynomial transformation to obtain a transformed as-built drawing;

[0063] a data embedding module configured to embed a reference coordinate system and geographic transformation parameters into the transformed as-built drawing;

[0064] The image saving module is configured to save the transformed as-built drawing after the metadata is embedded as a registered image in a set format.

[0065] In a third aspect, an embodiment of the present application provides an electronic device comprising: at least one processor and a memory; the memory stores computer-executable instructions; the at least one processor executes the computer-executable instructions stored in the memory, so that the at least one processor executes the method for geographic registration of completion drawings and remote sensing images as described in the first aspect and various possible designs of the first aspect.

[0066] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, in which computer execution instructions are stored. When a processor executes the computer execution instructions, the method for geographic registration of completion drawings and remote sensing images as described in the first aspect and various possible designs of the first aspect is implemented.

[0067] In a fifth aspect, an embodiment of the present application provides a computer program product, including a computer program. When the computer program is executed by a processor, it implements the method for geo-registration of completion drawings and remote sensing images as described in the first aspect and various possible designs of the first aspect.

[0068] The method, apparatus, device, and storage medium for georeferencing as-built drawings and remote sensing images provided in this application have at least the following beneficial effects:

[0069] Improved accuracy: This application significantly improves the accuracy of registration by adopting advanced multimodal image matching technology, especially using the RoMa image matching model combined with the SIFT algorithm for feature extraction and matching, and using the RANSAC algorithm to remove mismatched point pairs. Experimental results show that compared with traditional registration methods, this application can control the registration error within a very small range. For example, when registering the completion map of a high-standard farmland construction project in a certain area with remote sensing images, the average registration error of the traditional method is about 10 pixels, while the average registration error of this application can be reduced to within 2 pixels. It can more accurately achieve the geographic spatial alignment of the completion map and the remote sensing image, providing a reliable data foundation for high-precision analysis such as farmland boundary demarcation and water conservancy facility positioning.

[0070] Improved efficiency: During the data processing process, this application optimizes various links, including the use of efficient preprocessing algorithms, fast feature extraction and matching algorithms, and optimized parameter calculation methods, which greatly improves the processing speed of registration. Taking the processing of a 1000×1000 pixel as-built map and the corresponding remote sensing image as an example, the traditional method requires about 10 minutes to complete the registration. However, this application uses parallel computing and algorithm optimization to complete the registration task within 1 minute, greatly shortening the project cycle and improving work efficiency. It meets the real-time requirements of modern high-standard farmland construction projects. For example, in the real-time monitoring of farmland construction progress, the registered images can be quickly obtained to keep abreast of the project progress.

[0071] Enhanced universality: The method of this application has strong universality and can adapt to completion drawings and remote sensing images of different complexity and sources. Regardless of whether the image has quality problems such as noise, blur, deformation, or there are large differences in imaging conditions, resolution, color mode, etc., this application can achieve accurate registration through corresponding preprocessing and matching strategies. In actual applications, when processing various types of high-standard farmland construction project data obtained in different regions and at different times, this application has achieved good registration effects, expanding the scope of application of this method in different scenarios and providing general technical support for the management of high-standard farmland construction projects across the country. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0073] Figure 1 An overall flow chart of a method for geo-registration of as-built drawings and remote sensing images provided in an embodiment of the present application;

[0074] Figure 2 A specific flow chart of a method for geo-registration of as-built drawings and remote sensing images provided in an embodiment of the present application;

[0075] Figure 3 A multimodal image matching flow chart for a method for geo-registration of as-built drawings and remote sensing images provided in an embodiment of the present application;

[0076] Figure 4 This is a structural diagram of the device for geo-registering as-built drawings and remote sensing images provided in an embodiment of the present application.

[0077] The above drawings illustrate specific embodiments of the present application, which will be described in more detail below. These drawings and the textual description are not intended to limit the scope of the present application in any way, but rather to illustrate the concepts of the present application to those skilled in the art by reference to specific embodiments. DETAILED DESCRIPTION

[0078] Exemplary embodiments will be described in detail herein, with examples illustrated in the accompanying drawings. In the following description, when referring to the drawings, identical numerals in different figures represent identical or similar elements, unless otherwise indicated. The embodiments described in the following exemplary embodiments are not intended to represent all embodiments consistent with the present application. Rather, they are merely examples of apparatus and methods consistent with certain aspects of the present application, as detailed in the appended claims.

[0079] In the technical solution of this application, the collection, storage, use, processing, transmission, provision and disclosure of information such as financial data or user data involved comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0080] It should be noted that in the embodiments of the present application, certain software, components, models and other existing solutions in the industry may be mentioned. They should be regarded as exemplary. Their purpose is only to illustrate the feasibility of implementing the technical solution of the present application, but it does not mean that the applicant has or will necessarily use the solution.

[0081] The following specific embodiments describe in detail the technical solution of the present application and how the technical solution of the present application solves the above-mentioned technical problems. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of the present application will be described below in conjunction with the accompanying drawings.

[0082] Example 1:

[0083] The present application embodiment provides a method for geo-registration of as-built drawings and remote sensing images, such as Figure 1 The figure shows the overall flow chart of the geo-registration method of the as-built drawing and remote sensing image provided by the embodiment of the present application. The overall process of the geo-registration method of the as-built drawing and remote sensing image is as follows: from the beginning, the data preparation stage is entered, the as-built drawing and remote sensing satellite image are first obtained, and then the spatial coverage and resolution are checked to see if they meet the requirements. If not, the data needs to be re-prepared; if they meet the requirements, the pre-processing stage is entered, and then multi-modal image matching and geo-transformation metadata are embedded to generate the registered image, which is then saved in TIF format and quality checked. It is then checked whether the requirements are met. If not, the registration process is re-checked. If the requirements are met, the registration is completed and the process ends.

[0084] Specifically, if Figure 2 FIG. 1 is a flow chart of a method for geo-referencing a completed drawing with a remote sensing image according to an embodiment of the present invention. The method for geo-referencing a completed drawing with a remote sensing image comprises the following steps S100 to S600.

[0085] S100: Acquire an as-built drawing to be registered and a remote sensing satellite image containing geographic information; wherein the as-built drawing is a raster image without geographic information, and the remote sensing satellite image is a reference image containing geographic information.

[0086] In some embodiments, the input images are a completed map of a high-standard farmland construction project to be registered and a remote sensing satellite image containing geographic information. The completed map is an electronic map raster image in a format such as jpg or png that does not contain geographic information. This type of image is typically generated from surveying and mapping data during the project construction process to visually present the actual situation of farmland construction, but lacks precise geographic positioning information. The remote sensing satellite image is a TIF reference image containing geographic information. It is acquired through ground observation by satellite sensors and has precise geographic coordinate information, reflecting the actual geographic location and spatial distribution of the farmland.

[0087] When acquiring images, it is important to ensure that the spatial coverage of the as-built drawing and the remote sensing satellite imagery is roughly the same. This is because if the spatial coverage differs significantly, there will be insufficient common area for matching, making it difficult to accurately find correspondences. The image resolutions should also be similar. Large differences in resolution can lead to inconsistent image details, which can also affect the registration results. For example, if the resolution of the as-built drawing is high, while the resolution of the remote sensing image is low, some small features in the as-built drawing will not be clearly visible in the remote sensing image, making accurate matching impossible.

[0088] S200: Preprocessing the as-built drawings and remote sensing satellite images respectively, including grayscale processing, noise reduction processing and image enhancement processing.

[0089] In this embodiment, the as-built drawing and remote sensing satellite imagery are preprocessed separately, including grayscale processing, noise reduction processing, and image enhancement processing, to improve image quality and lay a good foundation for subsequent image matching and georegistration. The grayscale processing is performed on the input as-built drawing and remote sensing satellite imagery, outputting the corresponding grayscale image. The grayscale image is subjected to noise reduction processing to obtain a reduced-noise image. Finally, the reduced-noise image is subjected to image enhancement processing to obtain the preprocessed as-built drawing and remote sensing satellite imagery.

[0090] In some embodiments, during the grayscale conversion process, each pixel of the as-built image and the color remote sensing satellite image is first traversed to extract the values ​​of the red channel R, green channel G, and blue channel B. Considering the differences in the human eye's sensitivity to different colors, a weighted average method is used to convert the color image into a grayscale image. This method assigns different weights to the three channels based on color sensitivity, typically setting green to have the highest weight, followed by red, and blue to have the lowest weight. The grayscale value is calculated as follows:

[0091] ,

[0092] Here, Gray represents the converted grayscale value, R represents the red component of the pixel, G represents the green component, and B represents the blue component. The weights 0.299, 0.587, and 0.114 reflect the relative sensitivity of the human eye to red, green, and blue, respectively. This processing method compresses the image's color information into single-channel luminance information, effectively preserving the image's structure and texture features while reducing the redundancy of the image data, thereby improving the efficiency and accuracy of subsequent noise reduction, enhancement, and feature matching. This unified grayscale processing also eliminates interference caused by color differences, providing a more stable foundation for feature alignment between multimodal images.

[0093] In some embodiments, during the noise reduction process, to effectively remove salt and pepper noise from as-built drawings and remote sensing images, a median filtering algorithm is used to smooth the grayscaled image. This method is a typical nonlinear filtering approach, the core concept of which is to replace the original pixel value with local statistics to enhance the image's noise resistance. Specifically, for each pixel in the image, a fixed-size window is constructed with that pixel as the center. Common window sizes are 3×3 or 5×5. Within this window, the grayscale values ​​of all pixels are extracted and sorted from smallest to largest. After sorting, the middle value in the sorted sequence is selected as the new grayscale value for the center pixel. For example, when using a 3×3 window, the 5th value of the nine grayscale values ​​within the window is sorted, and this value replaces the original value of the center pixel in the window. The essence of median filtering is to replace the original value with the local statistical median, avoiding the edge blurring that can occur with mean filtering. It is particularly effective in suppressing sharp salt and pepper noise. If there is a noise point with an abnormally high or low grayscale value in the original image, the point will not be in the middle position in the median sorting, and will be effectively replaced by a value closer to the grayscale of the surrounding pixels. The formula is expressed as:

[0094] ,

[0095] in Represents the grayscale value of the center pixel after filtering, Represents the original grayscale value of each pixel in the neighborhood, represents a neighborhood window centered at (x,y). This processing step not only effectively removes salt-and-pepper noise introduced by factors such as sensor errors and signal interference, but also maintains the clarity of image edges and the integrity of details, providing clean and continuous input data for subsequent image enhancement and feature extraction.

[0096] In some embodiments, during image enhancement, to enhance the contrast of local details in as-built drawings and remote sensing images and make image features more prominent in multimodal matching, a histogram equalization method is used to enhance the denoised image (which is a grayscale image). First, the grayscale values ​​of all pixels in the denoised image are counted, and the frequency of occurrence of each grayscale level is recorded to obtain a grayscale histogram. This histogram is then normalized to obtain a probability density function, which reflects the distribution probability of each grayscale value in the image. The cumulative distribution function is calculated based on the probability density function, and its formula is:

[0097] ,

[0098] in T ( k ) represents the grayscale k The cumulative distribution function value of n i Indicates grayscale i The number of pixels that appear, N Represents the total number of pixels in the image. Then the grayscale value of each pixel in the original denoised image r Map to new grayscale value s , the mapping function is:

[0099] ,

[0100] in s Represents the new grayscale value after mapping, L Indicates the total number of gray levels, usually 256. T ( r ) is the grayscale value r The cumulative distribution function value of is . Through this nonlinear transformation, the grayscale of pixels originally concentrated in a narrow grayscale range is redistributed to a wider grayscale interval, thereby improving the overall contrast of the image and making edges, textures, and details more prominent. Histogram equalization enhances the brightness levels of low-contrast images while maintaining the continuity of the image structure. This makes the subsequent feature extraction algorithm more stable and accurate when processing different image types, effectively improving the robustness of the multimodal image matching process.

[0101] S300: Input the preprocessed as-built drawing and the preprocessed remote sensing satellite image into the MINIMA multimodal image matching unified framework, use the RoMa image matching model to perform multimodal image matching, extract feature points of the preprocessed as-built drawing and the preprocessed remote sensing satellite image, and match them to obtain matched feature point pairs.

[0102] In this embodiment, the preprocessed as-built drawing and the preprocessed remote sensing satellite image are input into the MINIMA (Modality Invariant Image Matching) unified framework, and the RoMa (Robust Dense Feature Matching) image matching model is used to perform multimodal image matching, extract feature points from the two, and match them to obtain matched feature point pairs.

[0103] In some embodiments, as Figure 3 As shown in the figure, in the multimodal image matching step, the RoMa image matching model is used to extract features from the preprocessed as-built drawings and remote sensing satellite images. The core algorithm uses the SIFT algorithm to extract scale-invariant features in the image and obtain stable and repeatable feature vectors. The first step in SIFT feature extraction is to construct a scale space. By performing multi-scale Gaussian blurring on the image, a Gaussian pyramid is generated. Then, using the differences between adjacent Gaussian blurred images, a Gaussian difference pyramid (DoG) is constructed to detect extreme points in the scale space. These extreme points are robust to scale, rotation, and illumination changes. The construction of the scale space is based on the following function:

[0104] ,

[0105] in L Represents images at different scales, G The scale is σ Gaussian kernel function, I is the original image, and * represents the convolution operation.

[0106] After extreme point detection, each candidate keypoint is precisely located. Unstable edge points and low-contrast points are removed by fitting the gradient information of surrounding pixels, retaining keypoints with significant features. A main direction is then assigned to each keypoint, and a directional histogram is constructed based on the gradient direction distribution of pixels in its neighborhood. The direction with the largest amplitude is selected as the main direction, thus making the keypoint rotationally invariant. Finally, with the keypoint as the center, the neighborhood is divided into subregions, and the gradient amplitude and direction information within each region are counted. A feature descriptor is generated by concatenating multiple directional histograms, with each keypoint corresponding to a 128-dimensional vector. This feature vector accurately describes the local structural information of the image and is highly robust to changes in illumination, rotation, and scale. It facilitates high-precision cross-modal feature point matching in the RoMa model, ensuring accurate spatial correspondence between the as-built drawings and remote sensing images.

[0107] In multimodal image matching, to achieve high-precision feature alignment between preprocessed as-built drawings and remote sensing satellite images, a feature point matching method based on Euclidean distance is used, combined with the RANSAC algorithm to eliminate false matches. First, for each key point extracted by the SIFT algorithm, a corresponding 128-dimensional feature vector is constructed. All feature vectors in the as-built drawings and remote sensing images are compared pairwise, and their Euclidean distance in feature space is calculated to determine their similarity. The formula for calculating Euclidean distance is:

[0108] ,

[0109] in A and B Represent the feature vectors of the as-built drawing and remote sensing image respectively, Ai and Bi is the first i components, and 128 is the dimension of the SIFT feature vector. The smaller the distance, the higher the similarity between the two feature points in the image structure. Feature point pairs with high similarity are screened out according to the set distance threshold as the preliminary matching results. In order to further improve the matching accuracy, the RANSAC algorithm is used to optimize the preliminary matching point pairs. The core idea of ​​RANSAC is to randomly select a small number of samples from all matching point pairs as the candidate inlier set, use this set to calculate the initial geometric transformation model, and then use this model to predict the positional relationship of other point pairs. The number of all point pairs that meet the geometric constraints of the model is counted and regarded as inliers. Random sampling and model evaluation are repeated, and the model with the largest number of inliers is recorded as the final matching result, and the inconsistent outliers are eliminated. The feature point pairs that are finally retained are highly consistent in spatial geometry and local image structure, which can provide a reliable basis for the accurate estimation of subsequent geographic transformation parameters, ensuring that the as-built drawings have high accuracy and stability when aligned to the remote sensing image coordinate system.

[0110] S400: Calculating geographic transformation parameters from the as-built drawing to the remote sensing satellite image based on the matched feature point pairs, and transforming the as-built drawing into the coordinate system of the remote sensing satellite image using a polynomial transformation to obtain a transformed as-built drawing.

[0111] In some embodiments, geographic transformation parameters from the as-built drawing to the remote sensing satellite image are calculated based on the matched feature point pairs, and a polynomial transformation method is used to transform the as-built drawing into the coordinate system of the remote sensing satellite image to obtain the transformed as-built drawing. Specifically, a quadratic polynomial transformation model is used to establish a coordinate transformation relationship between the as-built drawing and the remote sensing satellite image. The polynomial coefficients are solved using the least squares method to obtain the geographic transformation parameters, including translation parameters, rotation parameters, scaling parameters, and polynomial coefficients.

[0112] Establishing transformation model: The general form of quadratic polynomial transformation model is:

[0113] ,

[0114] ,

[0115] in, x and y are the horizontal and vertical coordinates in the as-built drawing, x 'and y ' is the horizontal and vertical coordinates in the remote sensing satellite image, a 0. a 1. a 2. a 3. a 4. a 5. b 0. b 1. b 2. b 3. b 4 and b 5 are all polynomial coefficients. This model can better describe the complex nonlinear geometric transformation relationship between images and adapt to the image registration requirements under different terrains and imaging conditions.

[0116] In the georegistration process between as-built drawings and remote sensing satellite images, matching feature point pairs provides observational data of spatial correspondence. By constructing a geographic transformation model and optimizing the model parameters using the least squares method, the optimal transformation coefficients can be obtained to achieve high-precision coordinate alignment. The core idea of ​​the least squares method is to minimize the sum of squared errors between each feature point in the transformed as-built drawing and its corresponding point in the remote sensing image. For each pair of matching points, the error is composed of the difference between the actual remote sensing coordinates and the model prediction value, which can be expressed as:

[0117] ,

[0118] in x i and y i The first i feature point coordinates, and for x i and y i The corresponding coordinates in the remote sensing satellite image, a0 to a5, b0 to b5 are the coefficients to be determined of the quadratic polynomial model. This error function reflects the spatial deviation of each pair of feature points under the model transformation. The sum of squared errors for all n matching points is calculated to obtain the total error objective function E, which is expressed as:

[0119] ,

[0120] The model parameters are determined by solving the minimum value of the total error objective function. In the actual solution process, all point pairs are substituted into the model expression to construct a linear equation system, introduce matrix expression, express the unknown parameters in vector form, and solve them using matrix operation methods to finally obtain the optimal parameter set. These parameters accurately describe the spatial mapping relationship between the as-built drawing and the remote sensing image, including translation, rotation, scaling, and nonlinear deformation information. Based on the obtained transformation parameters, the coordinate transformation of the as-built drawing is performed, that is, all pixel points in the image are projected into the coordinate system of the remote sensing image according to the polynomial model, completing the geometric alignment process and providing an accurate spatial basis for the subsequent embedding of geographic metadata and generation of registered images.

[0121] S500: Embed the reference coordinate system and geographic transformation parameters into the transformed as-built drawing.

[0122] Metadata such as the reference coordinate system and geographic transformation parameters are embedded in the transformed as-built drawing. The reference coordinate system includes both geographic and projected coordinate systems. The geographic coordinate system uses the WGS84 coordinate system, a widely adopted international system that provides a globally unified geographic positioning benchmark. The projected coordinate system uses the Gauss-Krüger or UTM projections. The Gauss-Krüger projection is a conformal elliptical cylindrical projection that maintains constant angles and is suitable for mapping low- and mid-latitude areas. The UTM projection, also a conformal elliptical cylindrical projection, divides the Earth into 60 projection zones. The central meridian of each projection zone is projected as a straight line with a constant length, making it suitable for global map production and geographic information analysis. Geographic transformation parameters include the transformation model, transformation coefficients, and error parameters. These parameters record the coordinate transformation relationship and accuracy information between the as-built drawing and the remote sensing image.

[0123] When embedding metadata, follow the metadata storage format for TIF images. The TIF image format supports the storage of rich metadata information. Through specific tags and structures, information such as the reference coordinate system and geographic transformation parameters can be accurately embedded in the transformed as-built drawing. This allows GIS software to read this metadata and correctly understand the image's geospatial information and transformation process when subsequently using the registered image, enabling accurate image display and analysis.

[0124] S600: Saving the transformed as-built drawing after embedding the metadata as a registered image in a set format.

[0125] In some embodiments, the transformed as-built drawing with embedded metadata is saved as a registered image in TIF format, which can be correctly imported into geographic information system software such as QGIS and ArcGIS for display. When saving in TIF format, the generated TIF format registered image is subjected to quality checks, including geometric accuracy checks and metadata integrity checks.

[0126] Geometric Accuracy Check: By comparing the coordinates of known features in the registered image with their actual geographic coordinates, the error is calculated to assess whether the geometric accuracy of the registered image meets the requirements. For example, select several feature points that are easily identifiable in both the as-built drawings and remote sensing images, such as road intersections and building corners. Use GIS software to measure the coordinates of these points in the registered image and compare them with their actual geographic coordinates. If the error is within the allowable range, the geometric accuracy of the registered image meets the requirements. If the error is too large, the registration process needs to be re-examined to identify the problem and make corrections.

[0127] Metadata integrity check: Check the integrity of the metadata embedded in the registered image, including the accuracy of information such as the reference coordinate system and geographic transformation parameters. Use GIS software to read the metadata and verify its consistency with the metadata generated during the registration process. If metadata is missing or incorrect, it is supplemented and corrected promptly to ensure that the registered image can be correctly imported into the GIS software for display and application, providing reliable data support for the management and analysis of the high-standard farmland construction project.

[0128] Example 2:

[0129] An embodiment of the present application provides a method for geo-registration of as-built drawings and remote sensing images. The method is applied to the geo-registration of as-built drawings and remote sensing images of a high-standard farmland construction project in a plain area, and includes the following steps S1 to S7.

[0130] S1. Data Preparation: Obtain a completed map of a high-standard farmland construction project in a plain area. The map is in JPG format and sized at 2000×2000 pixels. This map was drawn by project surveyors based on field survey data. It primarily displays information such as the distribution of farmland plots, irrigation channels, and road layout, but lacks geographic coordinates. Also obtain remote sensing satellite imagery covering the area in TIF format with a resolution of 0.5 meters. This image contains precise geographic information and was taken close to the time the completed map was drawn to ensure minimal changes in ground features.

[0131] S2. Preprocessing, including the following steps S201-S203.

[0132] S201. Grayscale Processing: Grayscale the as-built drawings and remote sensing satellite images using the formula: Grayscale value = 0.299 × red component + 0.587 × green component + 0.114 × blue component. This processing reduces the image data size, speeds up subsequent processing, and eliminates color interference in image matching. For example, feature confusion caused by color differences in original color images is resolved, facilitating subsequent feature extraction.

[0133] S202, Noise Reduction: The grayscale image is denoised using a median filter algorithm with a 3×3 neighborhood window. For each pixel, the grayscale values ​​of the nine pixels in its neighborhood are sorted, and the median value replaces the original grayscale value. This noise reduction effectively removes salt-and-pepper noise caused by sensor noise, resulting in a smoother image. For example, scattered noise points disappear, and features such as field boundaries become clearer, providing a more accurate data foundation for subsequent feature extraction and matching.

[0134] S203. Image Enhancement: The denoised image is contrast-enhanced using histogram equalization. By counting the frequency of each grayscale level in the image, a grayscale histogram is generated. The cumulative distribution function is then calculated based on the histogram, and the grayscale values ​​of the original image are mapped according to the cumulative distribution function. After this process, the image contrast is significantly enhanced, and previously blurred field details and road edges become clearer. For example, small field boundaries and irrigation channel details, which were difficult to distinguish in low-contrast images, are now clearly visible in the enhanced image, facilitating subsequent feature extraction and matching operations.

[0135] S3. Multimodal image matching, including the following steps S301-S303.

[0136] S301, Feature Extraction: The pre-processed as-built drawings and remote sensing satellite images are input into the Minima multimodal image matching unified framework, and feature extraction is performed using the RoMa image matching model combined with the SIFT algorithm. When constructing the Gaussian difference pyramid DoG, a series of images of different scales are generated by convolving the image with Gaussian kernels of different scales. In these images, scale space extreme points are detected, and a total of approximately 5,000 key points are detected. These key points are then precisely located, and unstable edge points and low-contrast points are removed, ultimately obtaining approximately 3,000 stable key points. Next, based on the gradient direction distribution within the neighborhood of the key point, a main direction is assigned to each key point to make it rotation invariant. Finally, with the key point as the center, the gradient direction histogram is calculated within its neighborhood to generate a 128-dimensional feature vector to complete feature extraction.

[0137] S302, Feature Point Matching: Feature point matching is performed using the Euclidean distance metric to measure the similarity between feature vectors. A Euclidean distance threshold is set to 0.8, and feature point pairs with a distance less than this threshold are considered matching points. Approximately 1,000 matching point pairs are initially obtained. The RANSAC algorithm is then used to remove mismatched point pairs. After 50 iterations, four pairs of matching points are randomly selected each time to calculate the transformation model, and the remaining matching points are verified. Ultimately, approximately 200 pairs of mismatched points are removed, resulting in approximately 800 pairs of accurately matched feature point pairs, significantly improving matching accuracy.

[0138] S303, geographic transformation: Based on the 800 matched feature point pairs, a quadratic polynomial transformation model is used to calculate the geographic transformation parameters from the as-built drawing to the remote sensing satellite image. The following transformation model is established:

[0139] ,

[0140] ,

[0141] The least squares method is used to solve the polynomial coefficients. Multiple iterative calculations are performed to obtain the geographic transformation parameters, including translation parameters (a0, b0), rotation parameters, and scaling parameters. The rotation parameters are calculated using a1-a5 and b1-b5, and the scaling parameters are determined based on the polynomial coefficients a0-a5 and b0-b5 calculated from the model. Based on these parameters, the coordinates of the as-built drawing are transformed and mapped into the coordinate system of the remote sensing satellite imagery to obtain the transformed as-built drawing.

[0142] S4. Metadata Embedding: Metadata, including the reference coordinate system and geographic transformation parameters, are embedded into the transformed as-built drawing. The reference coordinate system uses the WGS84 geographic coordinate system and the Gauss-Krüger projection. Geographic transformation parameters include the transformation model (quadratic polynomial transformation), the transformation coefficients a0-a5 and b0-b5, and the matching error calculated using the error parameters. Accurately embedding this metadata into the transformed as-built drawing, in accordance with the metadata storage format of TIF images, ensures that subsequent GIS software can correctly read and understand the image's geospatial information.

[0143] S5. Generate registered image: Save the transformed as-built image with the embedded metadata as a registered image in TIF format. During the saving process, perform a quality check.

[0144] S6. Geometric Accuracy Check: We selected 50 easily identifiable feature points from the as-built drawings and remote sensing images, such as road intersections and building corners. We used GIS software to measure the coordinates of these points in the registered image and compared them with their actual geographic coordinates. The calculated average error was approximately 1.5 pixels, meeting the project's geometric accuracy requirement of less than 2 pixels.

[0145] S7. Metadata integrity check: Read the metadata through the geographic information system software and compare it with the metadata generated during the registration process to verify that the reference coordinate system, geographic transformation parameters and other information are accurate, ensuring that the registered image can be correctly imported into the QGIS geographic information system software for display and analysis.

[0146] Example 3:

[0147] An embodiment of the present application provides a method for geo-registration of as-built drawings and remote sensing images. The method is applied to the geo-registration of as-built drawings and remote sensing images of a high-standard farmland construction project in a hilly area, and includes the following steps S1 to S7.

[0148] S1. Data Preparation: Obtain a completed map of a high-standard farmland construction project in a hilly area. The image is in PNG format and sized at 1500×1500 pixels. Due to the complex terrain, the map features a rich variety of features but also exhibits some distortion. Also obtain the corresponding remote sensing satellite image in TIF format with a resolution of 1 meter. Due to the shadows cast by the terrain, some areas are dimmed.

[0149] S2. Preprocessing, including the following steps S201-S203.

[0150] S201. Grayscale processing: The weighted average method is also used for grayscale processing. The formula is grayscale value = 0.299 × red component + 0.587 × green component + 0.114 × blue component. This effectively reduces the amount of data and eliminates color interference, allowing subsequent processing to focus more on the brightness and texture characteristics of the image.

[0151] S202, Noise Reduction: Given the high noise content of this image, a median filter algorithm with a 5×5 neighborhood window is used for noise reduction. For each pixel, the grayscale values ​​of the 25 pixels within its 5×5 neighborhood are sorted and the median value is used to replace the original grayscale value. This successfully removes a large amount of noise, such as imaging noise caused by terrain undulations and noise generated during transmission. This makes the image clearer and provides a more reliable image foundation for subsequent feature extraction.

[0152] S203, Image Enhancement Processing: Use histogram equalization to enhance image contrast. By adjusting the image grayscale histogram, the grayscale value distribution is made more uniform. The details of the areas with low brightness originally affected by the terrain shadow are enhanced. For example, some terraced fields and mountain trails that were difficult to distinguish under low brightness become clearly visible in the enhanced image, facilitating subsequent feature extraction and matching operations.

[0153] S3. Multimodal image matching, including the following steps S301-S303.

[0154] S301, Feature Extraction: Feature extraction is performed on the preprocessed image using the RoMa image matching model combined with the SIFT algorithm. When constructing a Gaussian Difference Pyramid to detect key points, due to the complex terrain, approximately 6,000 key points were detected. After precise positioning and removal of unstable points, approximately 3,500 stable key points were obtained. These key points were assigned principal directions and 128-dimensional feature vectors were generated to complete feature extraction.

[0155] S302, Feature Point Matching: Feature point matching was performed using the Euclidean distance metric for feature vector similarity, with a threshold of 0.75. Approximately 1,200 pairs of matching points were initially obtained. The RANSAC algorithm was used to remove mismatched points. After 80 iterations, four pairs of matching points were randomly selected each time to calculate the transformation model and verify the remaining matching points. Ultimately, approximately 300 pairs of mismatched points were removed, resulting in approximately 900 pairs of accurately matched feature points, ensuring matching accuracy.

[0156] S303, Geographic Transformation: Based on the 900 matching feature point pairs, a quadratic polynomial transformation model is used to calculate the geographic transformation parameters. The polynomial coefficients are solved using the least squares method. After complex matrix operations and multiple iterations, the translation, rotation, and scaling parameters and polynomial coefficients required for the geographic transformation are obtained. These parameters are used to transform the coordinates of the as-built drawing, aligning it with the coordinate system of the remote sensing satellite imagery, resulting in the transformed as-built drawing.

[0157] S4. Metadata embedding: The WGS84 geographic coordinate system and the UTM projection coordinate system are used as reference coordinate systems, and are embedded into the transformed as-built drawing together with geographic transformation parameters including transformation model, coefficients and error parameters in accordance with the TIF image metadata storage format to ensure the integrity and readability of the image's geospatial information.

[0158] S5. Generate registration image: Save the as-built drawing with embedded metadata as a registration image in TIF format and perform quality check.

[0159] S6. Geometric accuracy check: 60 ground feature points were selected for coordinate comparison, and the average error was calculated to be approximately 1.8 pixels, which meets the geometric accuracy requirements of the high-standard farmland construction project in the hilly area, that is, the error is less than 2.5 pixels.

[0160] S7. Metadata integrity check: Verify the integrity and accuracy of metadata through geographic information system software to ensure that the registered images can be smoothly imported into ArcGIS geographic information system software for display and further analysis, providing accurate data support for the management and decision-making of high-standard farmland construction projects in the region.

[0161] Example 4:

[0162] The present application also provides a device for geo-registration of as-built drawings and remote sensing images. Figure 4 As shown, the as-built drawing and remote sensing image georeferencing device includes:

[0163] The data acquisition module 401 is configured to acquire an as-built drawing to be registered and a remote sensing satellite image containing geographic information; wherein the as-built drawing is a raster image that does not contain geographic information, and the remote sensing satellite image is a reference image that contains geographic information;

[0164] A data preprocessing module 402 is configured to preprocess the as-built drawings and remote sensing satellite images, respectively, wherein the preprocessing includes grayscale processing, noise reduction processing, and image enhancement processing;

[0165] The feature matching module 403 is configured to input the pre-processed as-built drawing and the pre-processed remote sensing satellite image into the MINIMA multimodal image matching unified framework, perform multimodal image matching using the RoMa image matching model, extract feature points from the pre-processed as-built drawing and the pre-processed remote sensing satellite image, and match them to obtain matched feature point pairs;

[0166] The projection transformation module 404 is configured to calculate geographic transformation parameters from the as-built drawing to the remote sensing satellite image based on the matched feature point pairs, and transform the as-built drawing into the coordinate system of the remote sensing satellite image using a polynomial transformation to obtain a transformed as-built drawing;

[0167] a data embedding module 405 configured to embed a reference coordinate system and geographic transformation parameters into the transformed as-built drawing;

[0168] The image saving module 406 is configured to save the transformed as-built drawing after the metadata is embedded as a registered image in a set format.

[0169] An embodiment of the present application provides an electronic device, which may include a processor and a memory, wherein the processor and the memory can communicate with each other; illustratively, the processor and the memory communicate with each other via a communication bus.

[0170] The processor executes the computer-executable instructions stored in the memory, so that the processor implements the solutions in the above embodiments. The processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0171] The communication bus can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus. System buses can be categorized as address buses, data buses, and control buses. Transceivers facilitate communication between the database access device and other computers (e.g., clients, read-write libraries, and read-only libraries). Memory may include random access memory (RAM) or non-volatile memory.

[0172] The electronic device provided in the embodiment of the present application may be the terminal device of the above embodiment.

[0173] An embodiment of the present application also provides a computer-readable storage medium, which stores computer instructions. When the computer instructions are executed on a computer, the computer executes the technical solution of the method for geo-registration of completion drawings and remote sensing images in the above embodiment.

[0174] An embodiment of the present application also provides a computer program product, which includes a computer program stored in a computer-readable storage medium. At least one processor can read the computer program from the computer-readable storage medium. When at least one processor executes the computer program, it can implement the technical solution of the method for geographic registration of completion drawings and remote sensing images in the above-mentioned embodiment.

[0175] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of modules is only a logical function division. In actual implementation, there may be other division methods, such as multiple modules can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interface, device or module, which can be electrical, mechanical or other forms.

[0176] Modules described as separate components may or may not be physically separate, and components shown as modules may or may not be physical units, that is, they may be located in one place or distributed across multiple network elements. Some or all of these modules may be selected to implement the solution of this embodiment based on actual needs.

[0177] In addition, the functional modules in the various embodiments of the present application may be integrated into a single processing unit, or each module may exist physically separately, or two or more modules may be integrated into a single unit. The above-mentioned modules may be implemented in the form of hardware or hardware plus software functional units.

[0178] The integrated modules implemented in the form of software function modules can be stored in a computer-readable storage medium. The software function modules stored in a storage medium include a number of instructions for causing a computer device (which can be a personal computer, server, or network device, etc.) or a processor to execute some of the steps of the methods of various embodiments of the present application.

[0179] It should be understood that the processor described above may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), etc. A general-purpose processor may be a microprocessor or any conventional processor. The steps of the method disclosed in the present invention may be directly executed by a hardware processor or by a combination of hardware and software modules within the processor.

[0180] The memory may include a high-speed RAM memory, and may also include non-volatile storage NVM, such as at least one disk memory, and may also be a USB flash drive, a mobile hard disk, a read-only memory, a magnetic disk or an optical disk.

[0181] A bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus. Buses can be divided into address buses, data buses, and control buses.

[0182] The aforementioned storage medium can be realized by any type of volatile or nonvolatile storage devices or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk or optical disk. The storage medium can be any available medium that can be accessed by a general or special purpose computer.

[0183] An exemplary storage medium is coupled to the processor so that the processor can read information from, and write information to, the storage medium. Of course, the storage medium can be a part of the processor. The processor and the storage medium can be located in an application specific integrated circuit (ASIC). Of course, the processor and the storage medium can be located in a remote terminal or a host computer.

[0184] Those skilled in the art can understand that all or part of the steps of the above-mentioned method embodiments can be completed by program instruction related hardware. The aforementioned program can be stored in a computer readable storage medium. When the program is executed, the steps of the above-mentioned method embodiments are executed; and the aforementioned storage medium includes ROM, RAM, magnetic disk or optical disk and various storage media that can store program codes.

[0185] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions recorded in the above embodiments, or make equivalent replacement for part or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for geo-registration of as-built drawings and remote sensing images, characterized in that: The method comprises: Obtaining an as-built drawing to be registered and a remote sensing satellite image containing geographic information; wherein the as-built drawing is a raster image that does not contain geographic information, and the remote sensing satellite image is a reference image that contains geographic information; Preprocessing the as-built drawings and remote sensing satellite images respectively, wherein the preprocessing includes grayscale processing, noise reduction processing, and image enhancement processing; The pre-processed as-built drawings and pre-processed remote sensing satellite images are input into the MINIMA multimodal image matching unified framework, and the RoMa image matching model is used to perform multimodal image matching. The feature points of the pre-processed as-built drawings and pre-processed remote sensing satellite images are extracted and matched to obtain matched feature point pairs. Calculating geographic transformation parameters from the as-built drawing to the remote sensing satellite image based on the matched feature point pairs, and transforming the as-built drawing into the coordinate system of the remote sensing satellite image using a polynomial transformation to obtain a transformed as-built drawing; embedding a reference coordinate system and geographic transformation parameters into the transformed as-built drawing; Save the transformed as-built drawing with embedded metadata as a registered image in a set format; The pre-processed as-built drawings and pre-processed remote sensing satellite images are input into the MINIMA multimodal image matching unified framework. The RoMa image matching model is used for multimodal image matching. The feature points of the pre-processed as-built drawings and pre-processed remote sensing satellite images are extracted and matched to obtain matching feature point pairs, including: performing multi-scale Gaussian blur processing on the pre-processed as-built drawing and remote sensing satellite image to generate a Gaussian pyramid, constructing a Gaussian difference pyramid by taking differences between adjacent Gaussian blurred images, and detecting extreme points in the scale space based on the Gaussian difference pyramid; Positioning the extreme points, removing edge unstable points and low contrast points by fitting the gradient information of pixels around the extreme points, and retaining the key points; Construct a direction histogram based on the gradient direction distribution of pixels in the neighborhood of each key point, and select the direction with the largest amplitude in the direction histogram as the main direction of the key point; Taking the key point after the main direction is assigned as the center, sub-regions are divided in the neighborhood of the key point, and the gradient amplitude and direction information in each sub-region are counted to generate a feature vector composed of multiple direction histograms. Each key point corresponds to one feature vector; Calculating the Euclidean distance between the feature vectors of the feature points in the as-built image and the feature vectors of the feature points in the remote sensing satellite image; selecting, according to a set distance threshold, feature point pairs whose Euclidean distances are less than the distance threshold as the feature point pairs for preliminary matching; A part of the feature point pairs from the preliminary matched feature point pairs are randomly selected as candidate inlier points sets, and an initial geometric transformation model is calculated based on the candidate inlier points sets; the positional relationship of other feature point pairs is predicted using the initial geometric transformation model, and the number of feature point pairs that meet the geometric constraints of the initial geometric transformation model is counted as the number of inliers; the random sampling and model evaluation process is repeated, and the feature point pair corresponding to the geometric transformation model with the largest number of inliers is taken as the final matched feature point pair.

2. The method for geo-registration of as-built drawings and remote sensing images according to claim 1, characterized in that: The grayscale processing includes: Obtaining values ​​of the red channel, green channel, and blue channel of the as-built drawing and the remote sensing satellite image; The grayscale value of the as-built drawing or remote sensing satellite image is calculated using the following formula: ， in, Gray Represents the grayscale value after conversion, R Represents the red component of the pixel, G represents the green component, B Represents the blue component, with weights of 0.299, 0.587, and 0.114 reflecting the sensitivity ratios of the human eye to red, green, and blue, respectively.

3. The method for geo-registration of as-built drawings and remote sensing images according to claim 1, characterized in that: The noise reduction process includes: The grayscale image obtained based on the grayscale processing is filtered using the following formula: ， in, Represents the grayscale value of the center pixel after filtering, Represents each pixel in the neighborhood ( i , j )’s original grayscale value, Indicates the pixel point ( x , y ) is the neighborhood window centered on .

4. The method for geo-registration of as-built drawings and remote sensing images according to claim 1, characterized in that: The image enhancement processing includes: Based on the denoised image obtained by the denoising process, performing statistics on the grayscale values ​​of all pixels in the denoised image, recording the frequency of occurrence of each grayscale level, and obtaining a grayscale histogram; Normalizing the histogram to obtain a probability density function, wherein the probability density function reflects the distribution probability of each gray value in the image; According to the probability density function, the cumulative distribution function value is calculated by the following formula: ， in, T ( k ) represents the grayscale k The cumulative distribution function value of n i Indicates grayscale i The number of pixels that appear, N Indicates the total number of pixels in the image; The gray value of each pixel in the original denoised image r Mapped to a new grayscale value, the mapping function is: ， in, s Represents the new grayscale value after mapping, L Represents the total number of gray levels, T ( r ) is the grayscale value r The cumulative distribution function value of .

5. The method for geo-registration of as-built drawings and remote sensing images according to claim 1, characterized in that: Calculating geographic transformation parameters from the as-built drawing to the remote sensing satellite image based on the matched feature point pairs, and transforming the as-built drawing into the coordinate system of the remote sensing satellite image using a polynomial transformation to obtain a transformed as-built drawing, including: A quadratic polynomial transformation model is established; wherein the quadratic polynomial transformation model is expressed as: ， ， in, x and y are the horizontal and vertical coordinates in the as-built drawing, x 'and y ' is the horizontal and vertical coordinates in the remote sensing satellite image, a 0. a 1. a 2. a 3. a 4. a 5. b 0. b 1. b 2. b 3. b 4 and b 5 are all polynomial coefficients; The error of each pair of matched feature points is determined by the following formula e i : ， in, x i and y i The first i feature point coordinates, and for x i and y i Corresponding coordinates in remote sensing satellite images; Based on the error of each pair of matched feature points, determine the total error objective function E for: ， in, n is the number of characteristic points in the as-built drawing; Substitute all matched feature point pairs into the quadratic polynomial transformation model to construct a linear equation system, introduce matrix expression, express unknown parameters in vector form, and use matrix operation method to solve the total error objective function. E The minimum value of determines the polynomial coefficients and obtains the geographic transformation parameters; The as-built drawing is subjected to coordinate transformation processing based on the geographic transformation parameters, so as to project all feature points in the as-built drawing into the coordinate system of the remote sensing image according to a quadratic polynomial transformation model, thereby obtaining a transformed as-built drawing.

6. The method for geo-registration of as-built drawings and remote sensing images according to claim 1, characterized in that: Saving the transformed as-built drawing after embedding metadata as a registered image in TIF format; performing a quality check on the generated TIF format registered image when saving it in TIF format, and using the registered image that passes the quality check as the final registered image; wherein the quality check includes a geometric accuracy check and a metadata integrity check; The geometric accuracy check includes: comparing the coordinates of known objects in the registered image with the actual geographic coordinates, calculating the error, and determining that the geometric accuracy check has passed if the error is within a set threshold range; The metadata integrity check includes: reading metadata of the registered image, and verifying whether the read metadata is consistent with metadata generated during the registration process. If they are consistent, it is determined that the metadata integrity check has passed.

7. A device for geo-registering as-built drawings and remote sensing images, characterized in that: The device comprises: a data acquisition module configured to acquire an as-built drawing to be registered and a remote sensing satellite image containing geographic information; wherein the as-built drawing is a raster image that does not contain geographic information, and the remote sensing satellite image is a reference image that contains geographic information; a data preprocessing module configured to preprocess the as-built drawings and remote sensing satellite images respectively, wherein the preprocessing includes grayscale processing, noise reduction processing, and image enhancement processing; The feature matching module is configured to input the pre-processed as-built drawing and the pre-processed remote sensing satellite image into the MINIMA multimodal image matching unified framework, perform multimodal image matching using the RoMa image matching model, extract feature points of the pre-processed as-built drawing and the pre-processed remote sensing satellite image, and match them to obtain matched feature point pairs; a projection transformation module configured to calculate geographic transformation parameters from the as-built drawing to the remote sensing satellite image based on the matched feature point pairs, and transform the as-built drawing into the coordinate system of the remote sensing satellite image using a polynomial transformation to obtain a transformed as-built drawing; a data embedding module configured to embed a reference coordinate system and geographic transformation parameters into the transformed as-built drawing; An image saving module is configured to save the transformed as-built drawing after embedding metadata as a registered image in a set format; The pre-processed as-built drawings and pre-processed remote sensing satellite images are input into the MINIMA multimodal image matching unified framework. The RoMa image matching model is used for multimodal image matching. The feature points of the pre-processed as-built drawings and pre-processed remote sensing satellite images are extracted and matched to obtain matching feature point pairs, including: performing multi-scale Gaussian blur processing on the pre-processed as-built drawing and remote sensing satellite image to generate a Gaussian pyramid, constructing a Gaussian difference pyramid by taking differences between adjacent Gaussian blurred images, and detecting extreme points in the scale space based on the Gaussian difference pyramid; Positioning the extreme points, removing edge unstable points and low contrast points by fitting the gradient information of pixels around the extreme points, and retaining the key points; Construct a direction histogram based on the gradient direction distribution of pixels in the neighborhood of each key point, and select the direction with the largest amplitude in the direction histogram as the main direction of the key point; Taking the key point after the main direction is assigned as the center, sub-regions are divided in the neighborhood of the key point, and the gradient amplitude and direction information in each sub-region are counted to generate a feature vector composed of multiple direction histograms. Each key point corresponds to one feature vector; Calculating the Euclidean distance between the feature vectors of the feature points in the as-built image and the feature vectors of the feature points in the remote sensing satellite image; selecting, according to a set distance threshold, feature point pairs whose Euclidean distances are less than the distance threshold as the feature point pairs for preliminary matching; A part of the feature point pairs from the preliminary matched feature point pairs are randomly selected as candidate inlier points sets, and an initial geometric transformation model is calculated based on the candidate inlier points sets; the positional relationship of other feature point pairs is predicted using the initial geometric transformation model, and the number of feature point pairs that meet the geometric constraints of the initial geometric transformation model is counted as the number of inliers; the random sampling and model evaluation process is repeated, and the feature point pair corresponding to the geometric transformation model with the largest number of inliers is taken as the final matched feature point pair.

8. An electronic device, characterized in that: include: a processor, and a memory communicatively connected to the processor; The memory stores computer-executable instructions; The processor executes the computer-executable instructions stored in the memory to implement the method for geo-registration of as-built drawings and remote sensing images according to any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement the method for geo-registration of as-built drawings and remote sensing images according to any one of claims 1 to 6.

Images