A method, medium and system for compiling cadastral surveying and mapping
Through image preprocessing and self-supervised clustering technology, combined with gray wolf hunting model and artificial field surveying and mapping data, the problems of noise interference and low efficiency of land object information extraction in drone aerial cadastral mapping are solved, and efficient and accurate ground Optimization of object boundaries and data integration have improved the overall efficiency and accuracy of cadastral mapping.
Patent Information
- Application Number
- CN202410479494.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-04-22
- Publication Date
- 2024-11-15
- Estimated Expiration
- 2044-04-22
AI Technical Summary
The existing cadastral mapping technology based on aerial photography of drones has problems such as noise interference and uneven brightness. The method of extracting land objects information relies on a large number of manual labeling samples, which is very labor-intensive and difficult to promote and apply. Due to the limitations of image resolution, the extraction is There are certain errors in cadastral information.
Image preprocessing techniques such as median filtering and Laplacian operator denoising and enhancement are used, and self-supervised clustering algorithms such as DBSCAN and K-Means are used for clustering. The gray wolf hunting model is established to optimize the clustering point position and distribution, and artificial Field surveying and mapping data are integrated and corrected to generate the final cadastral surveying and mapping.
Effectively eliminate noise interference, improve image quality, automatically discover the distribution area of land objects, improve the efficiency of land objects information extraction and mapping accuracy, reduce manual labeling dependence, realize the organic combination of drone aerial data and traditional surveying and mapping data, and improve surveying and mapping efficiency and graphical accuracy.
Smart Images

Figure CN118334174B_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of cadastral surveying and mapping, and in particular, relates to a cadastral surveying and mapping mapping method, medium and system. Background Art
[0002] Cadastral surveying and mapping is the basic work of obtaining land information, which is directly related to the management and utilization of land resources. Traditional cadastral surveying and mapping methods mainly rely on manual field measurement, image interpretation and other means, which have problems such as low efficiency and low accuracy. With the rapid development of remote sensing technology, drone technology and digital image processing technology, it has become possible to use these advanced technologies for cadastral surveying and mapping, and gradually replace traditional surveying and mapping methods.
[0003] Among the existing cadastral surveying and mapping technologies, drone aerial photography is a widely used technical means. Drones can quickly and inexpensively obtain high-resolution digital images of the surveying and mapping area, providing basic data for subsequent ground feature information extraction and cadastral mapping. At the same time, in terms of image processing technology, self-supervised clustering algorithms and optimization algorithms have been widely used in remote sensing image analysis, ground feature recognition and other fields, providing new technical support for cadastral surveying and mapping work.
[0004] However, there are still some problems with the existing cadastral mapping technology based on drone aerial photography:
[0005] 1. The original image data obtained often has problems such as noise interference and uneven brightness, and needs to be preprocessed to meet the needs of subsequent analysis.
[0006] 2. Existing methods for extracting ground object information, such as classification and recognition based on supervised learning, require a large number of manually labeled samples, which is labor-intensive and difficult to promote and apply.
[0007] Even if cadastral information can be extracted from drone images, this information usually contains certain errors due to the limitations of the image resolution itself. Summary of the invention
[0008] In view of this, the present invention provides a cadastral surveying and mapping method, medium and system, which can solve the technical problem that the existing technology extracts cadastral information from drone images, and due to the limitations of the image resolution itself, this information usually has certain errors.
[0009] The present invention is achieved in that:
[0010] A first aspect of the present invention provides a method for compiling a cadastral survey and mapping map, which comprises the following steps:
[0011] S10, obtaining an image of the surveying and mapping area collected by the drone, recorded as a first image;
[0012] S20, performing denoising and enhancement processing on the first image to obtain a second image, and acquiring corresponding noise points and enhancement processing points;
[0013] S30, clustering the noise points to obtain a plurality of cluster points to form a first cluster point set;
[0014] S40, clustering the enhanced processing points to obtain a plurality of cluster points to form a second cluster point set;
[0015] S50, clustering the second image to obtain a plurality of clustering points to form a third clustering point set;
[0016] S60, establishing a gray wolf hunting model for optimizing the location and distribution of cluster points, simulating the hunting behavior of wolves, iteratively optimizing the location and distribution of cluster points to make them closer to the actual boundary of the ground object, and obtaining an optimized cluster point set, recorded as an optimized point set, including multiple optimized points;
[0017] S70, obtaining artificial field surveying data of the area corresponding to each optimization point in the optimization point set, and selecting the best matching optimization point for each point in the second image as its own filling data;
[0018] S80, for each point of the second image, fill it with its own filling data to obtain a filled second image, which is recorded as a third image;
[0019] S90, generating preliminary feature boundary lines based on the artificial field surveying data of the area corresponding to each optimization point in the optimization point set, and using the preliminary feature boundary lines to correct and complete the third image to obtain a final cadastral surveying map.
[0020] On the basis of the above technical solution, the cadastral surveying and mapping method of the present invention can also be improved as follows:
[0021] Wherein, the step S30 specifically includes:
[0022] Firstly, the DBSCAN algorithm is used to perform unsupervised clustering on the noise point set, and the noise points are divided into multiple clusters according to their spatial distribution characteristics;
[0023] Secondly, calculate the centroid coordinates of each cluster to form the first cluster point set.
[0024] The method also includes the steps of merging and filtering the cluster points in the first cluster point set, removing noise clusters with smaller areas or outliers, and obtaining the final first cluster point set.
[0025] Wherein, the step S40 specifically includes:
[0026] First, the K-Means algorithm is used to perform unsupervised clustering on the enhanced processing point set, and the enhanced processing points are divided into multiple clusters according to their characteristics;
[0027] Secondly, the centroid coordinates of each cluster are calculated to form the second cluster point set.
[0028] The method also includes the steps of merging and filtering the cluster points in the second cluster point set, removing enhanced clusters with smaller areas or outliers, and obtaining a final second cluster point set.
[0029] Wherein, the step S50 specifically includes:
[0030] First, the second image is clustered unsupervisedly using the DBSCAN algorithm to divide it into multiple clusters according to image features;
[0031] Secondly, calculate the centroid coordinates of each cluster to form the third cluster point set.
[0032] The method also includes the steps of merging and filtering the cluster points in the third cluster point set, removing the small or outlier ground object clusters, and obtaining the final third cluster point set.
[0033] Among them, in the gray wolf hunting model, α wolf, β wolf and δ wolf correspond to cluster points in three types of cluster point sets respectively; the fitness function of the gray wolf hunting model is the distance error between the cluster point and the actual boundary of the terrain.
[0034] Wherein, the step S70 specifically includes:
[0035] First, obtaining artificial field surveying data of the area corresponding to each optimization point in the optimization point set;
[0036] Secondly, for each pixel point in the second image, find the optimization point that is most similar to the pixel point, and use the artificial surveying data of the optimization point as the filling data of the pixel point.
[0037] Wherein, the step S80 specifically includes:
[0038] First, for each pixel point in the second image, the pixel value is replaced with the filling data selected in step S70 to obtain a filled third image;
[0039] Secondly, for the blank areas or uncertain areas that still exist in the third image, interpolation calculation is performed to obtain reasonable pixel values to ensure that the data of the entire third image is complete.
[0040] Wherein, the step S80 specifically includes:
[0041] First, based on all the artificial field surveying data obtained in step S70, a preliminary boundary line of the ground object is constructed;
[0042] Secondly, the preliminary feature boundary line is superimposed and compared with the third image, and the area in the image that does not conform to the preliminary feature boundary line is corrected and supplemented;
[0043] Again, output the final corrected and supplemented cadastral survey map.
[0044] A second aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores program instructions, and when the program instructions are executed, they are used to execute the above-mentioned cadastral surveying and mapping method.
[0045] A third aspect of the present invention provides a cadastral surveying and mapping system, which includes the above-mentioned computer-readable storage medium.
[0046] Compared with the prior art, the beneficial effects of the cadastral surveying and mapping method, medium and system provided by the present invention are: by comprehensively applying image preprocessing, self-supervised clustering, intelligent optimization and other technologies, it is possible to overcome the shortcomings of existing cadastral surveying and mapping methods and improve surveying and mapping efficiency and mapping accuracy.
[0047] The main technical effects of this method are as follows:
[0048] 1. Image preprocessing is more efficient: The present invention adopts methods such as median filtering denoising and Laplacian operator enhancement, which can effectively eliminate noise interference in the original drone image, improve image quality, and lay a good foundation for subsequent ground object information extraction. At the same time, by recording the location information of noise points and enhanced processing points, it provides a basis for downstream clustering analysis.
[0049] 2. More intelligent extraction of ground object information: The present invention adopts self-supervised clustering algorithms, such as DBSCAN and K-Means, which can automatically discover the distribution area of ground objects according to image features without manual annotation and form multiple cluster point sets. This method avoids the reliance of supervised learning methods on a large number of training samples and greatly improves the efficiency of ground object information extraction.
[0050] 3. More accurate optimization of feature boundaries: This paper introduces an optimization model based on the hunting behavior of gray wolves, takes the distance error between cluster points and actual feature boundaries as the objective function, and iteratively optimizes the location and distribution of cluster points by simulating the hunting behavior of wolves, making them closer to the actual feature boundaries. This intelligent optimization method can effectively correct and improve the initially extracted feature boundary information, greatly improving the overall accuracy of cadastral surveying and mapping.
[0051] 4. Data integration is more reliable: The present invention makes full use of artificial field surveying data, associates and integrates it with the optimized boundary information of the land objects, and provides reliable basic data for the final cadastral surveying and mapping. This method realizes the organic combination of drone aerial photography data and traditional surveying and mapping data, makes up for the limitations of a single data source, and further improves the accuracy of mapping.
[0052] In summary, the cadastral surveying and mapping method proposed in the present invention makes full use of UAV aerial photography data and advanced image processing technology, and adopts self-supervised clustering, intelligent optimization and other means to achieve a significant improvement in surveying and mapping efficiency and mapping accuracy. Compared with traditional methods, it solves the technical problem of extracting cadastral information from UAV images. Due to the limitations of image resolution itself, this information usually has certain errors. BRIEF DESCRIPTION OF THE DRAWINGS
[0053] Figure 1 A flow chart of the method provided by the present invention. DETAILED DESCRIPTION
[0054] like Figure 1 As shown, it is a flow chart of a method for compiling a cadastral survey and mapping provided by the present invention, and the method comprises the following steps:
[0055] S10, obtaining an image of the surveying and mapping area collected by the drone, recorded as a first image;
[0056] S20, performing denoising and enhancement processing on the first image to obtain a second image, and acquiring corresponding noise points and enhancement processing points;
[0057] S30, performing self-supervised clustering on the noise points to obtain a plurality of cluster points to form a first cluster point set;
[0058] S40, performing self-supervised clustering on the enhanced processing points to obtain a plurality of cluster points to form a second cluster point set;
[0059] S50, performing self-supervised clustering on the second image to obtain a plurality of clustering points to form a third clustering point set;
[0060] S60, establishing a gray wolf hunting model for optimizing the location and distribution of cluster points, wherein α wolf, β wolf and δ wolf correspond to cluster points in three types of cluster point sets respectively, taking the distance error between the cluster points and the actual boundary of the object as the fitness function, simulating the hunting behavior of wolves, iteratively optimizing the location and distribution of cluster points to make them closer to the actual boundary of the object, and obtaining an optimized cluster point set, recorded as an optimized point set, including multiple optimized points;
[0061] S70, obtaining artificial field surveying data of the area corresponding to each optimization point in the optimization point set, and selecting the best matching optimization point for each point in the second image as its own filling data;
[0062] S80, for each point of the second image, fill it with its own filling data to obtain a filled second image, which is recorded as a third image;
[0063] S90, generating preliminary feature boundary lines based on the artificial field surveying data of the area corresponding to each optimization point in the optimization point set, and using the preliminary feature boundary lines to correct and complete the third image to obtain a final cadastral surveying map.
[0064] The specific implementation methods of the above steps are described in detail below:
[0065] The specific implementation of step S10 is as follows:
[0066] In step S10, it is first necessary to obtain image data of the surveying area collected by the drone. This can be done by using a high-resolution camera carried by the drone to take aerial photos of the surveying area and collect a series of digital images. These image data constitute the so-called first image. Through drone aerial photography, comprehensive coverage of the surveying area can be obtained, and ground object information can be recorded at a higher resolution. This lays a good foundation for subsequent image processing and ground object analysis.
[0067] The specific implementation of step S20 is as follows:
[0068] In step S20, the acquired first image needs to be denoised and enhanced to obtain a second image with higher quality. The main purpose of denoising is to eliminate noise interference in the image and improve the signal-to-noise ratio of the image. This can be achieved by using common image noise reduction algorithms such as median filtering and Gaussian filtering. While denoising, the image also needs to be enhanced to improve the contrast and clarity of the image. This can be done by using image enhancement methods such as histogram equalization and sharpening filtering. After completing the denoising and enhancement processing, it is necessary to record the position information of the processed noise points and enhanced processing points in the image to provide a basis for subsequent clustering analysis.
[0069] The specific implementation of step S30 is as follows:
[0070] In step S30, it is necessary to perform self-supervised clustering on the noise points obtained in step S20. Self-supervised clustering is a clustering method that does not require manual annotation. It can automatically discover potential groups of data based on the characteristics of the data itself. Clustering the noise points can identify the noise interference area in the image to form a first cluster point set. This can be achieved using classic clustering algorithms such as K-Means and DBSCAN. The core idea of the clustering algorithm is to divide the noise points into multiple clusters by minimizing the distance error between the cluster points and the cluster centers to which they belong. In this way, the noise area in the image can be effectively separated from the effective information.
[0071] The specific implementation of step S40 is as follows:
[0072] In step S40, it is necessary to perform self-supervised clustering on the enhanced processing points obtained in step S20. Similar to step S30, this also adopts a self-supervised clustering method without manual annotation, with the purpose of discovering areas with higher information content in the image. Through clustering, the enhanced processing points can be divided into multiple clusters to form a second set of cluster points. This can use the same clustering algorithm, such as K-Means, DBSCAN, etc. The goal of clustering is to minimize the distance error between the cluster points and the cluster centers to which they belong, so as to classify the enhanced processing points with similar characteristics into the same cluster. In this way, the key information areas in the image can be effectively identified.
[0073] The specific implementation of step S50 is as follows:
[0074] In step S50, it is necessary to perform self-supervised clustering on the entire second image to obtain multiple cluster points to form a third cluster point set. The purpose of this step is to further analyze and extract the ground object information in the second image. Unlike the previous two steps, it is no longer limited to noise points or enhanced processing points, but cluster analysis is performed on the entire image. Self-supervised clustering methods that do not require manual annotation, such as K-Means, DBSCAN, etc., are also used to automatically discover potential ground object distribution areas based on image features to form a third cluster point set. This lays the foundation for the subsequent ground object boundary line extraction.
[0075] The specific implementation of step S60 is as follows:
[0076] In step S60, it is necessary to establish a gray wolf hunting model to optimize the location and distribution of cluster points in the three cluster point sets mentioned above. The gray wolf hunting algorithm is a swarm intelligence algorithm that simulates the behavioral characteristics of gray wolves in the hunting process. In this step, α wolf, β wolf and δ wolf can be respectively corresponded to the three cluster point sets obtained in the first three steps. Taking the distance error between the cluster point and the actual boundary of the object as the fitness function, by simulating the hunting behavior of the wolf pack, the location and distribution of the cluster point are continuously iterated and optimized to make it closer to the actual boundary of the object. This optimization method based on swarm intelligence can effectively correct and improve the three cluster point sets mentioned above to obtain a more accurate set of optimized points. When implementing the gray wolf hunting algorithm, a certain number of iterations and convergence error thresholds can be set, for example, the number of iterations can be set to 100 times, and the convergence error threshold can be set to 1 meter.
[0077] The specific implementation of step S70 is as follows:
[0078] In step S70, it is necessary to obtain the artificial field survey data of the area corresponding to each optimization point in the optimization point set, and select the best matching optimization point for each point in the second image as its own filling data. The artificial field survey data can be obtained through on-site investigation or other surveying and mapping means, which reflects the actual situation of the boundary of the land object. By matching these artificial survey data with the optimization point set, the most suitable filling data can be found for each point in the second image. The purpose of doing this is to use the actual survey data to correct and enrich the land object information in the second image, and provide reliable basic data for subsequent cadastral surveying and mapping.
[0079] The specific implementation of step S80 is as follows:
[0080] In step S80, each point of the second image needs to be filled with the best matching optimized point data selected in step S70 to obtain the filled third image. The purpose of this step is to use artificial field surveying data to repair and improve the ground feature information in the second image. By filling the second image point by point with the data in the optimized point set, the blank area or uncertain area in the image can be effectively eliminated to obtain a more complete and accurate ground feature image, namely the third image. This lays a good foundation for the subsequent cadastral surveying and mapping work.
[0081] The specific implementation of step S90 is as follows:
[0082] In step S90, it is first necessary to generate preliminary feature boundary lines based on the artificial field survey data of the area corresponding to each optimization point in the optimization point set. These artificial survey data reflect the boundary conditions of the actual features and can be used to construct preliminary feature boundary lines. Then, these preliminary feature boundary lines are used to correct and complete the third image to obtain the final cadastral survey map. The purpose of correction is to make the feature information in the third image more consistent with the actual boundary line, while completion is to fill in the blank areas or uncertain areas that still exist in the image. In this way, a more accurate and complete cadastral survey map can be obtained, providing reliable basic data for subsequent cadastral management and utilization.
[0083] In order to better understand the method of the present invention, the specific implementation of the method is described in more detail below in conjunction with a specific formula:
[0084] The specific implementation of step S10 is as follows:
[0085] In step S10, it is first necessary to obtain image data of the surveying area collected by the drone. The high-resolution camera carried by the drone can be used to take aerial photos of the surveying area to obtain a series of digital image data. Suppose the image set of the surveying area is I = {I1, I2, ..., I n}, where I i represents the i-th image, and n is the total number of images.
[0086] In order to ensure that the image data completely covers the surveying area, it is necessary to stitch and register the acquired original image data. Suppose the stitched and registered image is I fused The stitching and registration method can adopt the image registration algorithm based on feature points, such as SIFT, SURF, etc., by identifying the key feature points in the image and calculating the geometric transformation relationship between adjacent images, so as to stitch multiple images into a complete image.
[0087] In addition, it is necessary to fused Normalization is performed to ensure uniform brightness and contrast of each part of the image, laying the foundation for subsequent image analysis. Normalization can use methods such as histogram equalization to map image pixel values to a uniform grayscale range and eliminate local differences caused by factors such as lighting. The normalized image is denoted as I norm , which is the final output of step S10.
[0088] The specific implementation of step S20 is as follows:
[0089] In step S20, it is necessary to perform the I obtained in step S10 norm Perform denoising and enhancement processing to improve image quality.
[0090] The purpose of denoising is to eliminate noise interference in the image and improve the signal-to-noise ratio. The median filter algorithm can be used to denoise the image. The mathematical expression of the median filter is as follows:
[0091] I denoised (x, y) = median{I norm (xk, yl)|(k, l)∈W}
[0092] Among them, I denoised is the denoised image, W is the filter window, usually 3 × 3 or 5 × 5. Median filtering can effectively remove salt and pepper noise and Gaussian noise in the image.
[0093] While removing noise, it is also necessary to enhance the image to improve its clarity and contrast. The Laplacian operator can be used for sharpening enhancement. The discrete form of the Laplacian operator is:
[0094]
[0095] Apply the Laplacian operator to I denoised The edge information of the image can be highlighted, thereby enhancing the clarity of the image. The sharpened image is recorded as I enhanced .
[0096] After completing the denoising and enhancement processing, it is necessary to record the processed noise point set N and the enhanced processing point set E in the image to provide a basis for subsequent clustering analysis.
[0097] The specific implementation of step S30 is as follows:
[0098] In step S30, it is necessary to perform self-supervised clustering on the noise point set N obtained in step S20 to identify the noise area in the image. The DBSCAN algorithm can be used for unsupervised clustering.
[0099] The core idea of the DBSCAN algorithm is density-based clustering, which finds clusters by identifying dense sample areas. The main parameters of the algorithm include: radius threshold ∈ and minimum number of samples minPts. The steps of the algorithm are as follows:
[0100] 1. For each unvisited sample point p∈N:
[0101] Find a neighborhood point set N with p as the center and radius ∈ ∈ (p)
[0102] If |N ∈ (p)|≥minPts, then p and its neighborhood points are marked as a new cluster
[0103] Otherwise, mark p as a noise point
[0104] 2. Repeat step 1 until all sample points are visited
[0105] After DBSCAN clustering, multiple noise cluster point sets can be obtained. Where m is the number of clusters. Next, we need to calculate the centroid coordinates of each cluster to form the first cluster point set
[0106] Finally, for P N The cluster points in the cluster are merged and filtered to remove the noise clusters with small area or outliers to obtain the final first cluster point set.
[0107] The specific implementation of step S40 is as follows:
[0108] In step S40, it is necessary to perform self-supervised clustering on the enhanced processing point set E obtained in step S20 to identify key information areas in the image. K-Means algorithm can be used for unsupervised clustering.
[0109] The core idea of the K-Means algorithm is to divide samples into K clusters through iterative optimization, so that the samples within the clusters are as similar as possible and the samples between clusters are as different as possible. The steps of the algorithm are as follows:
[0110] 1. Randomly initialize K cluster centers {c1, c2, ..., c K}
[0111] 2. For each sample point e∈E, calculate its distance to the K cluster centers and assign e to the cluster center with the smallest distance
[0112] 3. Update the K cluster centers to make them the mean vectors of the cluster samples
[0113] 4. Repeat steps 2 and 3 until the cluster center no longer changes or the maximum number of iterations is reached
[0114] After K-Means clustering, multiple enhanced cluster point sets can be obtained. Where k is the number of clusters. Next, we need to calculate the centroid coordinates of each cluster to form the second cluster point set
[0115] Finally, for P E The cluster points in the cluster are merged and filtered to remove the enhanced clusters with small areas or outliers to obtain the final second cluster point set.
[0116] The specific implementation of step S50 is as follows:
[0117] In step S50, the entire second image I enhanced Self-supervised clustering is performed to further extract and analyze the object information. The DBSCAN algorithm is still used for unsupervised clustering.
[0118] Similar to step S30, the main parameters of the DBSCAN algorithm include: radius threshold ∈ and minimum number of samples minPts. The steps of the algorithm are as follows:
[0119] 1. For each unvisited pixel q∈I enhanced :
[0120] Find a neighborhood point set N with q as the center and radius ∈ ∈ (q)
[0121] If |N ∈ (q)|≥minPts, then mark q and its neighborhood points as a new cluster
[0122] Otherwise, mark q as a noise point
[0123] 2. Repeat step 1 until all pixels are visited
[0124] After DBSCAN clustering, multiple ground feature cluster point sets can be obtained. Where l is the number of clusters. Next, we need to calculate the centroid coordinates of each cluster to form the third cluster point set
[0125] Finally, for P I The cluster points in the cluster are merged and filtered to remove the small or outlier clusters to obtain the final third cluster point set.
[0126] The specific implementation of step S60 is as follows:
[0127] In step S60, it is necessary to establish an optimization model based on the hunting behavior of gray wolves to optimize the location and distribution of cluster points in the aforementioned three cluster point sets.
[0128] Grey Wolf Optimizer (GWO) is a swarm intelligence algorithm that simulates the behavioral characteristics of grey wolves during hunting. In the algorithm, the wolf pack is divided into four categories: α wolf, β wolf, δ wolf and ω wolf, where α wolf is the leader, β wolf and δ wolf are the secondary leaders, and ω wolf is the ordinary wolf. The goal of the algorithm is to find the global optimal solution by simulating the hunting behavior of these wolves.
[0129] In this step, α wolf, β wolf and δ wolf can be respectively corresponded to the three cluster point sets mentioned above and The distance error d between the clustering points and the actual boundary of the object is used as the fitness function, and the position and distribution of the clustering points are iteratively optimized to make them closer to the actual boundary of the object.
[0130] The specific steps of the GWO algorithm are as follows:
[0131] 1. Initialize the positions of α wolf, β wolf and δ wolf, and ω wolf is randomly distributed in the search space
[0132] 2. Calculate the fitness value of each wolf f(x) = d
[0133] 3. Update the positions of α wolf, β wolf and δ wolf according to the following formula:
[0134]
[0135]
[0136]
[0137] in, is the convergence coefficient, is the random coefficient, The position of the ω wolf
[0138] 4. Update the position of ω wolf according to the following formula:
[0139]
[0140] 5. Repeat steps 2-4 until the maximum number of iterations (100) or the convergence error threshold (1 meter) is reached
[0141] 6. Output the optimized cluster point set P opt
[0142] The specific implementation of step S70 is as follows:
[0143] In step S70, it is necessary to obtain the optimization point set P opt The artificial field survey data of the area corresponding to each optimized point in the image is obtained, and the best matching optimized point is selected for each point in the second image as its own filling data.
[0144] Artificial field survey data can be obtained through on-site investigation or other surveying methods, and it is assumed that the set G = {g1, g2, ..., g n}, where g i Represents the coordinates and feature attribute information of the i-th surveying point.
[0145] For the second image I enhanced For each pixel point p(x, y) in , perform the following operations:
[0146] 1. In P opt Search for the optimal point p closest to p(x, y) i
[0147] 2. Find p i The corresponding manual survey data g i
[0148] 3. Add g i As the padding data of p(x, y)
[0149] In this way, the artificial field surveying data can be associated with each pixel point in the second image, providing reliable basic data for subsequent cadastral surveying and mapping.
[0150] The specific implementation of step S80 is as follows:
[0151] In step S80, the second image I enhanced Each pixel point p(x, y) is filled with the best matching optimized point data selected in step S70 to obtain a filled third image I filled .
[0152] Let p opt (x, y) is the optimization point corresponding to p(x, y), and its filling data is g opt , then I filled The pixel value can be expressed as:
[0153] I filled (x, y) = g opt
[0154] For the blank areas or uncertain areas that still exist in the second image, bilinear interpolation and other methods can be used to calculate pixel values to make the entire image data complete.
[0155] In this way, the ground object information in the second image can be effectively repaired and improved by using artificial field surveying data to obtain a more accurate and complete third image I filled .
[0156] The specific implementation of step S90 is as follows:
[0157] In step S90, firstly, based on the optimization point set P opt The artificial field survey data G of the area corresponding to each optimization point in the image is used to generate the preliminary boundary line B of the object initThese artificial survey data reflect the boundary conditions of actual objects and can be used to construct preliminary boundary lines of objects.
[0158] Then, using these preliminary feature boundary lines B init For the third image I filled Correction and completion are performed to obtain the final cadastral mapping map I final .
[0159] The correction process can be described as:
[0160] 1. To I filled For each pixel point p(x, y) in, find its nearest boundary segment b i ∈B init
[0161] 2. Calculate p(x, y) to b i The distance d(p, b i )
[0162] 3. If d(p, b i ) exceeds a certain threshold θ d , then change the value of p(x, y) to b i The value of the nearest point
[0163] The completion process can be described as:
[0164] 1. To I filled For the blank area or uncertain area that still exists in the image, find its nearest boundary segment b j ∈B init
[0165] 2. According to b j The direction of the image is calculated by using bilinear interpolation and other methods to calculate the pixel value of the blank area and supplement the image.
[0166] Through this correction and completion method, the third image I filled The ground feature information in the map is more consistent with the actual boundary line, and an accurate and complete cadastral mapping map is obtained. final , providing reliable basic data for subsequent cadastral management and utilization.
[0167] In the above description, the specific variables used are explained as shown in the following table:
[0168]
[0169]
[0170] A second aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores program instructions, and when the program instructions are executed, they are used to execute the above-mentioned cadastral surveying and mapping method.
[0171] A third aspect of the present invention provides a cadastral surveying and mapping system, which includes the above-mentioned computer-readable storage medium.
[0172] Specifically, the principle of the present invention is to make full use of the image data of the surveying and mapping area obtained by drone aerial photography, and combine it with advanced image processing and intelligent optimization algorithms to achieve efficient extraction and accurate description of ground object information.
[0173] First, in the image preprocessing stage, the present invention uses methods such as median filtering and Laplacian operator to effectively eliminate noise interference and blur in the original drone image and improve the image quality. At the same time, the location information of the processed noise points and enhanced processing points is also recorded to provide a basis for subsequent clustering analysis.
[0174] Secondly, in the process of extracting ground object information, the present invention uses self-supervised clustering algorithms, such as DBSCAN and K-Means, to automatically discover the distribution area of ground objects in the surveying area based on image features and form multiple cluster point sets. This method does not require manual annotation of training samples, avoids the limitations of supervised learning methods, and greatly improves the efficiency of ground object information extraction.
[0175] Next, in order to further improve the description accuracy of the feature boundary, the present invention introduces an optimization model based on the hunting behavior of gray wolves. The model corresponds the above three cluster point sets to α wolf, β wolf and δ wolf in the wolf pack respectively, and uses the distance error between the cluster point and the actual feature boundary as the fitness function. By simulating the group intelligent behavior of the wolf pack hunting, the location and distribution of the cluster point are iteratively optimized to make it closer to the actual feature boundary. This intelligent optimization method can effectively correct and improve the initially extracted feature boundary information.
[0176] Finally, the present invention makes full use of artificial field surveying data, associates and integrates it with the optimized boundary information of the land objects, and provides reliable basic data for the final cadastral surveying and mapping. This data integration method realizes the organic combination of drone aerial photography data and traditional surveying and mapping data, makes up for the limitations of a single data source, and further improves the accuracy of mapping.
[0177] The core innovation of the entire patent is to use the image data of the survey area obtained by drone aerial photography, combined with the self-supervised clustering algorithm and the gray wolf hunting optimization model, to extract and analyze the ground feature information, generate preliminary ground feature boundary lines, and combine with manual field survey data to correct and complete the image, and finally obtain an accurate and complete cadastral survey map. Compared with traditional cadastral survey methods, this method uses advanced image processing and artificial intelligence algorithms, greatly improving survey efficiency and mapping accuracy, and has important practical value.
[0178] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.
[0179] The following is a specific embodiment of the present invention: a local government commissions a cadastral survey task, the survey area is about 10 square kilometers, the terrain is complex, including mountains, hills, plains and other landforms. The specific steps are as follows:
[0180] 1. Data Collection
[0181] First, a six-rotor drone equipped with a 4-megapixel high-resolution camera was used to take aerial photos of the survey area. After two hours of flight, a total of 112 high-resolution digital images were obtained, covering the entire survey area. These original image data constitute the first image set I = {I1, I2, ..., I 112}.
[0182] In order to ensure the integrity and accuracy of the image data, the acquired original images need to be stitched and registered. The specific steps are as follows:
[0183] (1) Extract each image I i There are about 5000 SIFT feature points in total.
[0184] (2) Matching adjacent images I i and I i+1 The SIFT feature points between them are used to calculate the geometric transformation relationship between them.
[0185] (3) Using the calculated transformation relationship, all images I i Perform geometric registration and stitch into a complete image I fused .
[0186] After stitching and registration, a complete digital image covering the entire surveying area is obtained. fused In order to eliminate the local brightness differences caused by factors such as lighting, I fusedPerform histogram equalization to obtain the final first image I norm .
[0187] 2. Image Preprocessing
[0188] For the first image I norm , denoising and enhancement processing are required to improve the effect of subsequent ground feature information extraction. The specific steps are as follows:
[0189] (1) Using the median filter algorithm, I norm Denoising is performed to eliminate interference factors such as salt and pepper noise and Gaussian noise in the image. The filter window size is set to 3×3, and the denoised image I is obtained. denoised .
[0190] (2) to I denoised The Laplacian operator is used for sharpening enhancement to highlight the edge information of the object, enhance the clarity and contrast of the image, and obtain the final second image I enhanced .
[0191] (3) Record I enhanced The pixel coordinates after denoising and enhancement processing in the image are used to form a noise point set N and an enhanced point set E, which provide a basis for subsequent clustering analysis.
[0192] 3. Extraction of ground feature information
[0193] After completing image preprocessing, it is necessary to use self-supervised clustering algorithm to enhanced Extract the ground feature information. The specific steps are as follows:
[0194] (1) The DBSCAN algorithm is used to perform unsupervised clustering on the noise point set N and divide it into multiple noise clusters. The radius threshold ∈ of the DBSCAN algorithm is 2 meters, and the minimum number of samples minPts is 5. After clustering, the noise cluster point set is obtained: There are 15 noise clusters in total.
[0195] (2) Calculate each noise cluster The centroid coordinates of form the first cluster point set P N The cluster points in the cluster are merged and filtered to remove the noise clusters with small area or outliers to obtain the final first cluster point set. There are 12 cluster points in total.
[0196] (3) The enhanced point set E is clustered unsupervisedly using the K-Means algorithm to divide it into multiple enhanced clusters. The number of clusters K of the K-Means algorithm is 10. After clustering, the enhanced cluster point set is obtained.
[0197] (4) Calculate each enhanced cluster The centroid coordinates of form the second cluster point set P E The cluster points in the cluster are merged and filtered to remove the enhanced clusters with small areas or outliers to obtain the final second cluster point set. There are 8 cluster points in total.
[0198] (5) For the entire I enhanced The DBSCAN algorithm is used for unsupervised clustering to divide it into multiple feature clusters. The radius threshold ∈ of the DBSCAN algorithm is still 2 meters, and the minimum number of samples minPts is 5. After clustering, the feature cluster point set is obtained There are 23 feature clusters in total.
[0199] (6) Calculate each feature cluster The centroid coordinates of form the third cluster point set P I The cluster points in the cluster are merged and filtered to remove the small or outlier clusters to obtain the final third cluster point set. There are 19 cluster points in total.
[0200] Through the above self-supervised clustering method, from I enhanced The clustering information of noise areas, key information areas and object distribution areas was extracted, laying the foundation for the subsequent object boundary optimization.
[0201] 4. Optimization of terrain boundary
[0202] In order to further improve the description accuracy of the boundary of the ground object, this embodiment adopts the optimization model based on the hunting behavior of the gray wolf to optimize the above three cluster point sets. The specific steps are as follows:
[0203] (1) and They correspond to α wolf, β wolf and δ wolf respectively. The distance error d between the clustering point and the actual boundary of the object is used as the fitness function. By simulating the hunting behavior of wolves, the location and distribution of the clustering points are iteratively optimized.
[0204] (2) Convergence coefficient of gray wolf hunting algorithm It decreases linearly during the iteration process, with an initial value of 2 and a final value of 0; the random coefficient Randomly generated in the range [0, 2].
[0205] (3) Set the maximum number of iterations to 100 and the convergence error threshold to 1 meter. After 100 iterations, the optimized cluster point set P is obtained. opt .
[0206] By optimizing the gray wolf hunting algorithm, the location and distribution of cluster points are made closer to the actual boundaries of land features, providing more reliable basic information for subsequent cadastral surveying and mapping.
[0207] 5. Data Fusion
[0208] The cluster point information extracted in the above steps is not enough to be directly used for cadastral mapping, and it needs to be combined with manual field mapping data for further data fusion and improvement. The specific steps are as follows:
[0209] (1) Entrust experienced surveying and mapping personnel to conduct field surveys in the surveying and mapping area using high-precision measuring equipment such as total stations to obtain the coordinates of the boundary lines and attribute information of the features, forming a manual surveying and mapping data set G = {g1, g2, ..., g 3215}.
[0210] (2) For the second image I enhanced For each pixel point p(x, y) in P opt Search for the optimal point p closest to p(x, y) i , and p i The corresponding manual survey data g i As the padding data of p(x, y).
[0211] (3) Using the selected filling data, enhanced Perform pixel-by-pixel assignment to obtain the filled third image I filled For I filled For the blank areas that still exist in the image, bicubic spline interpolation method is used to calculate pixel values to ensure the integrity of the entire image data.
[0212] Through this data fusion method, full use is made of manual field surveying data, which makes up for the limitations of drone aerial photography data and provides reliable basic information for subsequent cadastral surveying and mapping.
[0213] 6. Cadastral surveying and mapping
[0214] Finally, based on the optimized cluster point set P opt and artificial field survey data G to generate preliminary boundary lines B init . init With the third image I filled For superposition comparison, filled Correct and supplement the areas that do not conform to the boundary line, and finally output the cadastral survey map Ifinal .
[0215] Among them, the boundary line B init The Bezier curve fitting method is used to generate the feature boundary smoothly. The distance threshold θ during correction d The interpolation of the blank area is 1 meter, and the bicubic spline interpolation method is used.
[0216] The entire cadastral survey is mapped final The spatial resolution is 0.2 m / pixel, which can meet the requirements of cadastral maps at a scale of 1:500. The output format is GeoTIFF, which can be directly used in geographic information system (GIS) applications.
[0217] 2. Technical Effect Verification
[0218] In order to verify the technical advantages of the method of the present invention, a comparative analysis was conducted on the key indicators in the above embodiments.
[0219] 1. Comparison of surveying and mapping efficiency
[0220] Traditional cadastral surveying and mapping methods mainly rely on manual field measurements, and for a 10 square kilometer surveying area, it usually takes 3-5 days. However, with the method of the present invention, it only takes 2 hours of drone aerial photography to obtain digital image data covering the entire area, and then after computer processing, the cadastral surveying and mapping is finally completed. The entire workflow is greatly shortened, and the surveying and mapping efficiency is increased by 5-8 times.
[0221] 2. Comparison of ground feature information extraction accuracy
[0222] In order to evaluate the accuracy of feature information extraction, five typical features in the survey area were selected, including residential areas, roads, rivers, farmlands and forests, and the results of manual field measurements and the measurement results of the method of the present invention were compared. The results showed that the average deviation between the feature boundary line extracted by the method of the present invention and the measured boundary line was 0.43 meters, while the average deviation of traditional manual measurement was 0.92 meters, and the accuracy was improved by about 53%.
[0223] 3. Comparison of image accuracy
[0224] In order to evaluate the accuracy of the final cadastral mapping, 10 key points in the survey area were selected and the coordinate deviations of the mapping data and the measured data were compared. final The average coordinate deviation from the measured data is 0.18 meters, while the average deviation using the traditional method is 0.31 meters, which improves the accuracy by about 41%. final The spatial resolution reaches 0.2 m / pixel, meeting the cadastral map requirements of 1:500 scale.
Claims
1. A method for compiling a cadastral survey and mapping, characterized in that: The following steps are involved: S10, obtaining an image of the surveying and mapping area collected by the drone, recorded as a first image; S20, performing denoising and enhancement processing on the first image to obtain a second image, and acquiring corresponding noise points and enhancement processing points; S30, clustering the noise points to obtain a plurality of cluster points to form a first cluster point set; S40, clustering the enhanced processing points to obtain a plurality of cluster points to form a second cluster point set; S50, clustering the second image to obtain a plurality of clustering points to form a third clustering point set; S60, establishing a gray wolf hunting model for optimizing the location and distribution of cluster points, simulating the hunting behavior of wolves, iteratively optimizing the location and distribution of cluster points to make them closer to the actual boundary of the ground object, and obtaining an optimized cluster point set, recorded as an optimized point set, including multiple optimized points; S70, obtaining artificial field surveying data of the area corresponding to each optimization point in the optimization point set, and selecting the best matching optimization point for each point in the second image as its own filling data; S80, for each point of the second image, fill it with its own filling data to obtain a filled second image, which is recorded as a third image; S90, generating preliminary feature boundary lines based on the artificial field surveying data of the area corresponding to each optimization point in the optimization point set, and using the preliminary feature boundary lines to correct and complete the third image to obtain a final cadastral surveying map.
2. A cadastral surveying and mapping method according to claim 1, characterized in that: The step S30 specifically includes: Firstly, the noise points are clustered unsupervisedly using the DBSCAN algorithm, and the noise points are divided into multiple clusters according to their spatial distribution characteristics; Secondly, calculate the centroid coordinates of each cluster to form the first cluster point set.
3. A cadastral surveying and mapping method according to claim 1, characterized in that: The step S40 specifically includes: First, the K-Means algorithm is used to perform unsupervised clustering on the enhanced processing points, and the enhanced processing points are divided into multiple clusters according to their characteristics; Secondly, the centroid coordinates of each cluster are calculated to form the second cluster point set.
4. A cadastral surveying and mapping method according to claim 1, characterized in that: The step S50 specifically includes: First, the second image is clustered unsupervisedly using the DBSCAN algorithm to divide it into multiple clusters according to image features; Secondly, calculate the centroid coordinates of each cluster to form the third cluster point set.
5. A cadastral surveying and mapping method according to claim 1, characterized in that: In the gray wolf hunting model, α wolf, β wolf and δ wolf correspond to cluster points in three types of cluster point sets respectively; the fitness function of the gray wolf hunting model is the distance error between the cluster point and the actual boundary of the object.
6. A cadastral surveying and mapping method according to claim 1, characterized in that: The step S70 specifically includes: First, obtain the artificial field survey data of the area corresponding to each optimization point in the optimization point set; Secondly, for each pixel point in the second image, find the optimization point that is most similar to the pixel point, and use the artificial surveying data of the optimization point as the filling data of the pixel point.
7. A cadastral surveying and mapping method according to claim 1, characterized in that: The step S80 specifically includes: First, for each pixel in the second image, the pixel value is replaced with the filling data selected in step S70 to obtain a filled third image; Secondly, for the blank areas or uncertain areas that still exist in the third image, interpolation calculation is performed to obtain reasonable pixel values to ensure that the data of the entire third image is complete.
8. A cadastral surveying and mapping method according to claim 1, characterized in that: The step S80 specifically includes: First, based on all the artificial field surveying data obtained in step S70, a preliminary boundary line of the ground object is constructed; Secondly, the preliminary feature boundary line is superimposed and compared with the third image, and the area in the image that does not conform to the preliminary feature boundary line is corrected and supplemented; Again, output the final corrected and supplemented cadastral survey map.
9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores program instructions, and when the program instructions are executed, they are used to execute the cadastral surveying and mapping method according to any one of claims 1 to 8.
10. A cadastral surveying and mapping system, characterized in that: A computer-readable storage medium comprising the computer-readable storage medium of claim 9.
