High-precision vectorization method for grid electronic map

Abstract

The invention relates to the technical field of image data processing, in particular to a high-precision vectorization method of a grid electronic map, and solves the technical problem of poor vectorization effect of the grid electronic map in the prior art. The method comprises the following steps: acquiring a grid electronic map of a to-be-processed region; carrying out region segmentation on the grid electronic map based on a graph segmentation algorithm; wherein the shortest path of the graph cut algorithm is calculated by adopting an adaptive ant colony algorithm; and carrying out vectorization processing on the segmented grid electronic map.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image data processing, and particularly relates to a high-precision vectorization method of a raster electronic map. BACKGROUND

[0002] With the rapid development of geographic information technology and remote sensing technology, raster electronic maps have become an important data basis in the fields of land resource management, urban and rural planning, and ecological protection. Based on the vectorization of raster maps, the source-sink function area identification, land resource pattern index calculation, and spatial optimization analysis can provide key decision-making basis for the scientific allocation and sustainable use of land resources.

[0003] However, in actual application, due to the data characteristics of aerial remote sensing raster electronic maps, such as resolution limitation, noise interference, and feature boundary ambiguity, the traditional raster electronic map vectorization method is difficult to effectively distinguish different functional areas with similar appearance characteristics in the process of feature extraction and identification classification, resulting in insufficient regional segmentation accuracy. This problem directly affects the accuracy of the subsequent vectorization results, and further reduces the reliability of the land resource pattern index calculation, and finally has an adverse effect on the scientificity of land resource management decision-making. SUMMARY

[0004] In order to solve the technical problem of poor raster electronic map vectorization effect in the prior art, the purpose of the present application is to provide a high-precision vectorization method of a raster electronic map, and the technical solution adopted is as follows: Obtain a raster electronic map of a region to be processed; Perform regional segmentation on the raster electronic map based on a graph cut algorithm; wherein the shortest path of the graph cut algorithm is calculated using an adaptive ant colony algorithm; Perform vectorization processing on the segmented raster electronic map.

[0005] In one possible implementation, the method comprises: Constructing an undirected graph structure of the raster electronic map; the undirected graph structure takes the pixel points in the raster electronic map as nodes and the connecting lines between adjacent pixel points as edges; Calculating the edge weight values between the corresponding nodes based on the feature similarity between the pixel points; the edge weight values are used to represent the similarity between the pixel points; Using the adaptive ant colony algorithm, searching for the shortest path in the undirected graph structure based on the edge weight values as heuristic information, and performing regional segmentation based on the shortest path as the regional segmentation boundary; the heuristic information is used to adjust the selection probability of the path in the process of searching for the shortest path.

[0006] In one possible implementation, the method comprises: For each pixel point in the grid electronic map, a color feature vector sequence and a texture feature vector sequence of the pixel point are obtained. For any two adjacent pixel points in the grid electronic map, an edge weight value between nodes corresponding to the two pixel points is determined according to the color feature vector sequences and the texture feature vector sequences of the two pixel points.

[0007] In a possible implementation, the color feature vector sequence includes pixel value components corresponding to a plurality of color channels respectively; and the texture feature vector sequence includes at least one of contrast, correlation, energy and entropy.

[0008] In a possible implementation, the texture feature vector sequence of the pixel point is determined in the following manner: The grid electronic map is subjected to grayscale processing to obtain a grayscale image; the grayscale image includes grayscale values of each pixel point in the grid electronic map; For each pixel point in the grid electronic map, the texture feature vector sequence of the pixel point is determined by a gray level co-occurrence matrix based on the grayscale values of each pixel point in a pixel region centered on the pixel point.

[0009] In a possible implementation, the method includes: For any two adjacent pixel points in the grid electronic map, a difference degree of the color feature vector sequences is determined according to the color feature vector sequences of the two pixel points; A difference degree of the texture feature vector sequences is determined according to the texture feature vector sequences of the two pixel points; A weight coefficient is determined according to the difference degree of the texture feature vector sequences; the weight coefficient is positively correlated with the difference degree of the texture feature vector sequences; The difference degrees of the color feature vector sequences and the texture feature vector sequences are weighted and fused according to the weight coefficient, to determine an edge weight value between nodes corresponding to the two pixel points; the edge weight value is negatively correlated with the difference degrees of the color feature vector sequences of the two pixel points and the difference degrees of the texture feature vector sequences of the two pixel points.

[0010] In a possible implementation, the method includes: An ant colony is initialized to set a number of ants in the ant colony, a starting node and a target node of each ant, and a maximum number of iterations of the adaptive ant colony algorithm; In each iteration, for each ant in the ant colony, a selection probability corresponding to each adjacent node is calculated according to an information concentration and heuristic information, and the next node is gradually selected based on the selection probability, starting from the starting node; the information concentration is used to indicate a historical preference degree of a path between two nodes being selected. After each iteration, the pheromone concentration on the path is updated according to the path selection of each ant in the iteration; After reaching the maximum number of iterations, the shortest path is determined according to the path selection of each ant, and regional segmentation is performed with the shortest path as the regional segmentation boundary.

[0011] In a possible implementation, the method comprises: After each iteration, for each path selected by each ant in the iteration, the pheromone evaporation coefficient used for updating the pheromone concentration is adjusted according to the difference degree of the texture feature vector sequences of the two nodes corresponding to the path; The pheromone concentration on the path is updated based on the adjusted pheromone evaporation coefficient.

[0012] In a possible implementation, the method comprises: Obtain remote sensing image data and geographic information system data of a region to be processed; Perform a correction operation on the remote sensing image data; the correction operation comprises radiation correction and geometric correction; Perform a preprocessing operation on the geographic information system data; the preprocessing operation comprises coordinate system unification, attribute field normalization processing, and topology checking; Generate a raster electronic map based on the corrected remote sensing image data and the preprocessed geographic information system data.

[0013] In a possible implementation, the method further comprises: Identify different types of source-sink functional areas according to the vectorized raster electronic map; the source-sink functional areas comprise source functional areas and sink functional areas; the source functional areas are areas outputting resources; the sink functional areas are areas absorbing or dissipating energy; Calculate a land resource regional pattern index based on the identified source-sink functional areas; the land resource regional pattern index is used to quantify the spatial configuration characteristics of land resources; Use the calculated land resource regional pattern index as a decision criterion, and use a multi-criteria decision method to sort the source-sink functional areas to determine the priority of land optimization configuration.

[0014] The present application has the following advantages: Based on the technical scheme, the grid electronic map of the to-be-processed region is acquired, and a graph cut algorithm combined with an adaptive ant colony algorithm is used for region segmentation, so that the segmentation accuracy in a complex ground feature scene is improved, and the local optimal problem of the traditional algorithm is avoided. Therefore, the grid electronic map after segmentation can be subjected to vectorization processing. The above scheme not only realizes high-precision vectorization of the grid electronic map and improves the vectorization effect of the grid electronic map, but also provides a data-driven decision basis for scientific planning and management of land resources, and significantly improves the efficiency and reliability of land resource management. BRIEF DESCRIPTION OF DRAWINGS

[0015] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present application or the prior art, a brief introduction will be given below to the drawings needed to be used in the embodiments or the prior art description. Obviously, the drawings in the following description are only some embodiments of the present application, and those skilled in the art can obtain other drawings according to these drawings without any creative effort.

[0016] Figure 1 A flowchart of a high-precision vectorization method of a grid electronic map provided by an embodiment of the present application; Figure 2 A flowchart of another high-precision vectorization method of a grid electronic map provided by an embodiment of the present application; Figure 3 A flowchart of another high-precision vectorization method of a grid electronic map provided by an embodiment of the present application; Figure 4 A flowchart of another high-precision vectorization method of a grid electronic map provided by an embodiment of the present application; Figure 5 A flowchart of another high-precision vectorization method of a grid electronic map provided by an embodiment of the present application. DETAILED DESCRIPTION

[0017] In order to further illustrate the technical means and effects adopted by the present application to achieve the predetermined invention purpose, the following describes the specific implementation, structure, features and effects of a high-precision vectorization method of a grid electronic map according to the present application in combination with the drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. In addition, the specific features, structures or characteristics in one or more embodiments can be combined in any suitable form.

[0018] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the present application belongs.

[0019] In view of the technical problem of poor vectorization effect of the prior art grid electronic map, the present application provides a high-precision vectorization method of a grid electronic map, which obtains a grid electronic map of a region to be processed, and performs regional segmentation by using a graph cut algorithm combined with an adaptive ant colony algorithm, thereby improving the segmentation accuracy in a complex ground feature scene and avoiding the local optimal problem of the traditional algorithm. Thus, the present application can perform vectorization processing on the segmented grid electronic map. The above scheme not only realizes high-precision vectorization of the grid electronic map and improves the vectorization effect of the grid electronic map, but also provides data-driven decision basis for scientific planning and management of land resources, and significantly improves the efficiency and reliability of land resource management.

[0020] The specific scheme of the high-precision vectorization method of a grid electronic map provided by the present application will be specifically described below in combination with the accompanying drawings.

[0021] Please refer to Figure 1 which shows a method flowchart of a high-precision vectorization method of a grid electronic map provided by an embodiment of the present application. The method includes the following steps: Step 101, obtaining a grid electronic map of a region to be processed.

[0022] In one possible implementation manner, the present application can obtain remote sensing image data and geographic information system data of the region to be processed, perform correction operation on the remote sensing image data, and perform preprocessing operation on the geographic information system (GIS) data. Then, based on the corrected remote sensing image data and the preprocessed geographic information system data, a grid electronic map is generated.

[0023] The correction operation includes radiation correction and geometric correction, and the preprocessing operation includes coordinate system unification, attribute field normalization processing, and topology checking.

[0024] For example, the remote sensing image data can be a red, green, and blue (RGB) remote sensing image taken by a remote sensing device carried by a UAV, which can clearly record the appearance information of the ground features in the region (such as green coverage of farmland, gray texture of buildings, etc.). The geographic information system data can be geographic vector data of the region to be processed, which contains geographic spatial position information such as roads and administrative boundary, and is used to assist in constructing the grid electronic map.

[0025] For the correction operation, radiometric correction is used to eliminate the radiation distortion caused by uneven illumination intensity, sensor sensitivity difference and other factors, such as correcting the brightness difference of images of the same plot taken at different light periods; geometric correction is used to correct the image deformation and position offset caused by the flight attitude fluctuation of the unmanned aerial vehicle, the terrain undulation or the lens distortion, so as to ensure that the remote sensing image can be accurately matched with the real geographical coordinates.

[0026] For the preprocessing operation, the coordinate system is used to unify the GIS data to the preset coordinate system, so as to ensure the spatial consistency with the remote sensing image; the attribute field normalization processing includes unifying the data unit, completing the missing ground object type information and removing the duplicate vector records; the topological check is used to repair the overlapping polygons, the unclosed boundaries and other problems in the vector data, so as to ensure the spatial logic correctness of the GIS data.

[0027] In this way, based on the corrected remote sensing image data and the preprocessed GIS data, the related GIS software can be used for fusion, classification and rasterization processing, and finally the raster electronic map which can be used for subsequent analysis can be generated.

[0028] Step 102, regional segmentation is performed on the raster electronic map based on a graph cut algorithm.

[0029] In the graph cut algorithm, the shortest path is calculated by using an adaptive ant colony algorithm.

[0030] The raster electronic map is important basic data in the field of land resource management, and the identification and classification of source and sink functional areas are of great significance for ecological planning of the region. For example, building groups and farmland belong to source functional areas, which provide resources and output energy; forests, green lands and water areas belong to sink functional areas, which can absorb energy, such as forests absorbing carbon emissions of cities to maintain ecological balance. Accurate identification of source and sink functional areas is of great significance for land resource management and planning.

[0031] However, in some cases, the source and sink functional areas may exhibit similar image features, resulting in poor identification and classification effect. For example, in the image of the raster electronic map, some areas of water features exhibit similar color features to trees and vegetation due to the influence of suspended solids and plants, and the mountain bodies covered by plants and trees also exhibit similar features to farmland areas.

[0032] The above cases make it difficult for the traditional vectorization method of the raster electronic map to balance the accuracy and robustness of processing complex scenes, especially when facing high-resolution remote sensing images, resulting in the phenomenon of blurred boundaries of different areas and ineffective differentiation in the vectorization result, which directly affects the accuracy of subsequent source and sink functional area identification and spatial pattern analysis.

[0033] It should be noted that the existing graph cut algorithm referred to in the embodiments of the present application is a graph cut algorithm based on a minimum path (or geodesic) model. The algorithm is an important method in the field of image segmentation, and its core is to model the image as a weighted undirected graph, and to form a segmentation boundary by solving the shortest path (i.e. geodesic) between the pre-specified seed points (or regions) in the graph. Unlike another type of existing graph cut algorithm based on maximum flow / minimum cut, the geodesic model graph cut algorithm directly relies on accurate shortest path search to determine the region contour, so it has a higher requirement for the efficiency and global optimization ability of the path search strategy. However, the existing geodesic model graph cut algorithm usually uses a deterministic greedy strategy such as Dijkstra algorithm when searching for the shortest path. Such a strategy is prone to local optimal solution in complex feature boundary, weak edge or texture gradient region, resulting in blurred, broken or inaccurate segmentation boundary, which in turn affects the subsequent vectorization accuracy.

[0034] In view of this, in order to improve the vectorization accuracy of the raster electronic map, the present application can optimize the shortest path search process through an adaptive ant colony algorithm, avoid the problem that the traditional graph cut algorithm is prone to local optimal solution in complex feature scenarios, and thus realize accurate segmentation of different feature regions (such as farmland, forest, water area, building area) in the raster map, providing a clear regional boundary basis for subsequent vectorization.

[0035] Step 103, performing vectorization processing on the segmented raster electronic map.

[0036] In some embodiments, after the vectorization processing, the present application can also perform land resource analysis based on the vectorized raster electronic map.

[0037] In one possible implementation, the present application can identify different types of source-sink functional areas according to the vectorized raster electronic map, calculate land resource regional pattern indexes based on the identified source-sink functional areas, and then use the calculated land resource regional pattern indexes as decision criteria to sort the source-sink functional areas using a multi-criteria decision method, so as to determine the priority of land optimization allocation.

[0038] Among them, the source-sink functional area includes a source functional area and a sink functional area. The source functional area is an area that outputs resources, such as a building group or farmland, which can provide living space and agricultural products. The sink functional area is an area that absorbs or absorbs energy, such as forest green land or water area, which can absorb carbon emissions and purify water bodies. The land resource regional pattern index is used to quantify the spatial allocation characteristics of land resources, such as patch density (reflecting the degree of regional fragmentation) and connectivity (reflecting the spatial continuity of the same type of region).

[0039] For example, the above-mentioned vectorization processing can convert the continuous grid area obtained by segmentation into discrete vector graphics (such as polygons, lines, and points), for example, converting the segmented farmland area into a closed polygon vector and converting the river area into a line vector with a width attribute.

[0040] The multi-criteria decision-making method can be a Technique for Order Preference by Similarity to Ideal Solution (TOPSIS) for ranking each source-sink functional area and determining the priority of land optimization allocation, for example, prioritizing the protection and restoration of sink functional areas with low connectivity and high ecological value. In this way, the application can output land optimization suggestions to provide data-driven scientific basis for land resource management planning.

[0041] Based on the above technical solutions, the application obtains a grid electronic map of the region to be processed, and uses a graph cut algorithm combined with an adaptive ant colony algorithm for region segmentation, thereby improving the segmentation accuracy in complex ground feature scenarios and avoiding the local optimal problem of traditional algorithms. In this way, the application can perform vectorization processing on the segmented grid electronic map, and the above-mentioned scheme not only realizes high-precision vectorization of the grid electronic map and improves the vectorization effect of the grid electronic map, but also provides data-driven decision-making basis for scientific planning and management of land resources, thereby significantly improving the efficiency and reliability of land resource management.

[0042] It should be noted that the application divides the grid map into homogeneous regions by the image segmentation algorithm, and then performs vectorization on the segmentation result, which can significantly improve the boundary smoothness and region consistency. Among them, although the existing image segmentation algorithm can find the segmentation boundary by solving the shortest path between nodes in the graph, the shortest path optimization process of the traditional geodesic model graph cut algorithm is usually based on a greedy search strategy, which is easy to fall into a local optimal solution when finding the best segmentation boundary in a grid electronic map scenario with complex ground feature color characteristics and fuzzy ground feature boundaries, resulting in a fuzzy or broken segmentation boundary, especially in weak edge or gradual change regions. Therefore, the adaptive ant colony algorithm is used to optimize the shortest path search process of the graph cut algorithm, thereby further improving the accuracy of region segmentation.

[0043] As a possible embodiment of the application, the adaptive ant colony algorithm is combined with the graph cut algorithm to perform region segmentation on the grid electronic map. Figure 1 As shown in Figure 2 The above-mentioned step 102 can be implemented by the following steps: Step 201, constructing an undirected graph structure of the grid electronic map.

[0044] In this undirected graph structure, pixels in the raster electronic map are used as nodes, and lines connecting adjacent pixels are used as edges. For example, this undirected graph structure can be represented as follows: ,in, Indicates the first 1 pixel The corresponding node, Indicates the first 1 pixel The corresponding node and the first 1 pixel The edges between corresponding nodes. That is, each pixel in the raster electronic map is an independent node in an undirected graph structure. In this embodiment, the edges in the undirected graph structure are the edges between nodes corresponding to two adjacent pixels. For example, adjacent can refer to being adjacent in the four directions (up, down, left, right), or in the eight directions including the diagonal. Through this step, this application can transform intuitive image information into a mathematical structure that can be computed using graph theory algorithms, providing a computational framework for subsequent segmentation.

[0045] Step 202: Calculate the edge weights between corresponding nodes based on the feature similarity between pixels.

[0046] Among them, the edge weights are used to characterize the similarity between pixels. The edge weights can be determined by... This means that two adjacent pixels... and The similarity between pixels is considered. Higher similarity results in a larger edge weight, while lower similarity results in a smaller edge weight. This edge weight can be determined based on the feature information of each pixel. For example, adjacent pixels within the same farmland area have high similarity and large edge weights; adjacent pixels at the boundary between farmland and forest have low similarity and small edge weights.

[0047] Step 203: Adaptive ant colony algorithm is used to search for the shortest path in the undirected graph structure using edge weights as heuristic information, and the shortest path is used as the region segmentation boundary for region segmentation.

[0048] Heuristic information is used to adjust the probability of path selection during the search for the shortest path. For example, heuristic information can be represented by the reciprocal of the edge weight. That is, the smaller the edge weight (the lower the pixel similarity, the more likely it is to be a region boundary), the larger the heuristic information, and the higher the probability that the ants will choose that edge, thereby guiding the ant colony to search towards potential region boundaries.

[0049] The adaptive ant colony algorithm refers to finding the shortest path (with the minimum total weight of the path) from the starting node to the target node in an undirected graph by simulating the collective search behavior of ants. This shortest path is the dividing boundary between different regions in the raster map.

[0050] It should be noted that the traditional graph cut algorithm usually adopts the greedy search strategy of Dijkstra algorithm in the shortest path search, however, the algorithm cannot backtrack optimization, and selects a path in the shortest path search, which will lock the local area of the path. This search strategy is easy to cause the search to fall into local optimization in the raster electronic map scene with complex color features and fuzzy feature boundaries, and is easy to ignore the real boundary of different regions when exploring the boundary of the region with similar color features but different texture features, resulting in error in boundary segmentation.

[0051] Therefore, the application can replace the greedy search strategy of Dijkstra algorithm with the ant colony algorithm based on adaptive region texture features to realize the search of the shortest path, complete the optimal segmentation of the image boundary, and improve the segmentation accuracy.

[0052] Based on the above technical solution, the application converts the region segmentation problem of the raster electronic map into the shortest path search problem in graph theory by constructing an undirected graph structure, reduces the calculation complexity of complex image segmentation, quantifies the edge weight based on the pixel similarity, can accurately reflect the region affiliation correlation of adjacent pixels, provides reliable judgment basis for path search, and the adaptive ant colony algorithm taking the edge weight as heuristic information can effectively guide the search direction, avoid the local optimization problem of the traditional shortest path method in the complex scene, thereby effectively inhibiting the boundary blur and over-segmentation phenomenon, realizing high-precision region segmentation, and laying a clear region foundation for subsequent vectorization processing.

[0053] In addition, the application can also quantify the edge weight by combining the color features and texture features of the pixel points, solve the problem that similar color regions are difficult to distinguish, and improve the distinguishing ability of the edge weight.

[0054] As a possible embodiment of the application, in combination with Figure 2 As shown in Figure 3 , the above step 202 can be realized by the following steps: Step 301, for each pixel point in the raster electronic map, acquiring a color feature vector sequence and a texture feature vector sequence of the pixel point.

[0055] In some embodiments, the color feature vector sequence includes a plurality of color channel corresponding pixel value components; and the texture feature vector sequence includes at least one of contrast, correlation, energy and entropy.

[0056] Illustratively, for the pixel point in the raster electronic map , the color feature can be acquired by the pixel value components of the RGB three channels, to form a set of color feature vector sequences, which can be expressed as , wherein, pixel points Pixel value components on R, G, B channels.

[0057] It should be noted that in the grid electronic map, it is difficult to distinguish the regions with similar color features but different source-sink functions, such as farmland, forest vegetation and some water body regions, only by relying on the color feature information (RGB color feature) of the image. In the growing season, crops in farmland and forest vegetation both show green color features in remote sensing images, and due to the influence of suspended solids and aquatic plants, inland lakes, rivers and other water bodies also show similar green color features, which affects the correct identification and classification of source-sink function regions.

[0058] However, the texture features of these different regions can reflect the spatial structure differences of ground objects, and therefore can be used as additional supplementary information of color features to improve the accuracy of identification and classification of source-sink function regions and avoid the situation of fuzzy boundary segmentation of different regions.

[0059] As for the performance of different regions in texture features, the texture features of farmland regions show periodicity and obvious directionality due to artificial management and ridge and furrow planting, and there is no large fluctuation in crop texture features in the same farmland. The forest vegetation region shows chaotic texture features due to the growth of natural vegetation without fixed direction, and the trend of mountain ridge line and the terrain undulation further cause the texture features of the mountain forest region to show high intensity variation level. In the water body region, there is no significant directionality in large-area continuous water body, and the water surface is smooth, showing smooth and weak texture performance in texture features.

[0060] In some embodiments, the texture feature vector sequence of a pixel point can be determined by: performing gray-scale processing on the grid electronic map to obtain a gray-scale image. For each pixel point in the grid electronic map, based on the gray-scale values of each pixel point in the pixel region centered on the pixel point, the texture feature vector sequence of the pixel point is determined by a gray-scale co-occurrence matrix.

[0061] The gray-scale image includes the gray-scale values of each pixel point in the grid electronic map.

[0062] For example, for the acquisition of texture features of different regions, first, the image is gray-scaled using the average value method, and then the gray-scale co-occurrence matrix method is used to calculate the texture features of the pixel points, the window size is set to 3*3, and the texture features of the pixel points in the window are assigned to the center pixel point. For example, the texture feature vector sequence can be represented as wherein, respectively represent the contrast, correlation, energy and entropy values of the pixel point .

[0063] Step 302, for any two adjacent pixel points in the grid electronic map, determining an edge weight value between nodes corresponding to the two pixel points according to color feature vector sequences and texture feature vector sequences of the two pixel points.

[0064] For example, the application can evaluate the difference between the two pixel points based on the color feature vector sequences and the texture feature vector sequences of the two pixel points, and determine the edge weight value between the nodes corresponding to the two pixel points based on the difference between the two pixel points.

[0065] Based on the above technical solution, the application introduces color features and texture features to avoid the problem of difficulty in distinguishing similar color regions caused by relying only on color features, such as effectively distinguishing green farmland from green mountains. The texture feature can accurately extract the spatial structure difference of the ground object, providing additional basis for regional division. Therefore, the application can significantly improve the quantization accuracy of the edge weight value by using color features and texture features, providing a reliable numerical basis for subsequent shortest path search, and further ensuring the accuracy of regional division.

[0066] As a possible embodiment of the application, in combination with Figure 3 As shown in Figure 4 , the above step 302 can be implemented by the following steps: Step 401, for any two adjacent pixel points in the grid electronic map, determining a difference degree of color feature vector sequences according to color feature vector sequences of the two pixel points.

[0067] For example, the difference degree of the color feature vector sequences of the adjacent pixel points can be calculated using the Euclidean distance. The larger the distance value, the more significant the color difference between the two pixel points (such as the color difference degree of a red building and a green farmland). In this way, the overall difference between multi-dimensional color feature vector sequences (such as RGB three channels) can be intuitively and accurately quantified. For example, the difference degree of the color feature vector sequences can be represented as wherein, is the color feature vector sequence of the i-th pixel point, is the color feature vector sequence of the j-th pixel point.

[0068] In some embodiments, the application can also control the sensitivity of the influence of color features on node edge weight values through a color feature scale parameter. For example, the difference degree of the color feature vector sequence adjusted based on the color feature scale parameter can be represented as wherein, ​​​​The color feature scale parameter is a positive number, which can be set according to actual conditions, for example, it can be set to 0.3.

[0069] Step 402, determining the difference degree of the texture feature vector sequence according to the texture feature vector sequences of the two pixel points.

[0070] Similarly, the difference degree of the texture feature vector sequences of adjacent pixel points can also be calculated by using the Euclidean distance. The greater the distance value, the more significant the texture difference between the two pixel points (for example, the difference degree of the regular texture of farmland and the chaotic texture of mountains and forests is large), so that the difference of each dimension (such as contrast, correlation, energy, and entropy) in the texture feature can be comprehensively considered, and the spatial structure difference can be fully reflected. For example, the difference degree of the texture feature vector sequence can be represented as wherein, is the texture feature vector sequence of the i-th pixel point, is the texture feature vector sequence of the j-th pixel point.

[0071] In some embodiments, the present application can also control the sensitivity of the texture feature to the node edge weight by using a texture feature scale parameter. For example, the difference degree of the texture feature vector sequence adjusted based on the texture feature scale parameter can be represented as wherein, is the texture feature scale parameter, which is a positive number, which can be set according to actual conditions, for example, it can be set to 0.2.

[0072] Step 403, determining the weight coefficient according to the difference degree of the texture feature vector sequence.

[0073] wherein, the weight coefficient is positively correlated with the difference degree of the texture feature vector sequence.

[0074] It should be noted that when the texture difference degree of the two pixel points is small (such as adjacent pixels in the same farmland), the value of the weight coefficient is small, at this time, the edge weight calculation can enhance the influence degree of the color feature, and avoid over-segmentation (i.e., mistakenly dividing the same region into multiple regions) in subsequent segmentation; when the texture difference degree of the two pixels is large (such as the boundary pixels of farmland and mountains), the value of the weight coefficient is large, at this time, the influence degree of the texture feature can be enhanced, the difference representation at the boundary can be strengthened, and the region boundary can be accurately captured in subsequent segmentation.

[0075] For example, the weight coefficient satisfies the following formula: wherein, ​​​​is a weight coefficient, which takes a value in the range of [0, 1], is an adjustment factor of the texture feature similarity between nodes, used to limit the influence degree of the texture feature on the edge weight, for example, it can be set to 0.7. is the texture feature vector sequence of the th pixel point, is the texture feature vector sequence of the th pixel point, is the texture feature vector sequence of the th pixel point.

[0076] Step 404, according to the weight coefficient, the difference degree of the color feature vector sequence and the difference degree of the texture feature vector sequence are weighted and fused to determine the edge weight between the nodes corresponding to the two pixel points.

[0077] Wherein, the edge weight is negatively correlated with the difference degree of the color feature vector sequence of the two pixel points, and is negatively correlated with the difference degree of the texture feature vector sequence of the two pixel points.

[0078] Exemplarily, the edge weight between the nodes corresponding to the two pixel points satisfies the following formula: Wherein, is the edge weight between the nodes corresponding to the th pixel point and the th pixel point, is a weight coefficient, is the texture feature vector sequence of the th pixel point, is the texture feature vector sequence of the th pixel point. is a color feature scale parameter. is the texture feature vector sequence of the th pixel point, is the texture feature vector sequence of the th pixel point. is a texture feature scale parameter. is the texture feature vector sequence of the th pixel point, is the texture feature vector sequence of the th pixel point. is the texture feature vector sequence of the th pixel point.

[0079] Based on the above technical scheme, the difference degrees of the two types of features of color and texture are determined, and the dynamic weight coefficient is set, so that the edge weight calculation can adaptively adjust the contribution degrees of the two features according to the pixel texture difference, which avoids over-segmentation of the same region and ensures accurate differentiation of the boundary region. Finally, the edge weight obtained can accurately reflect the similarity and difference of the pixels, providing a reliable weight basis for the subsequent adaptive ant colony algorithm to search for the shortest path (segmentation boundary), and fundamentally solving the problem of insufficient segmentation accuracy in the similar color feature scene.

[0080] As a possible embodiment of the present application, in combination with Figure 2 As shown in Figure 5 , the above step 203 can be realized by the following steps: Step 501, initialize the ant colony to set the number of ants in the ant colony, the starting node of each ant, the target node and the maximum iteration number of the adaptive ant colony algorithm.

[0081] For example, the number of ants can be set to 100 (which can be adjusted according to the number of map pixels) to ensure the parallelism and diversity of group search; the starting node and the target node can be specified by the user (such as manually labeling seed points on the edge of the region to be segmented) or automatically determined by the adaptive seed point selection algorithm (such as selecting two different region centers with high color and texture feature uniformity as the starting node and the target node, respectively); the maximum iteration number T can be set to 100 (which can be adjusted according to the convergence speed, and when the shortest path length is unchanged for 5 consecutive times in iteration, the iteration can be terminated in advance) to avoid waste of computing resources caused by excessive iteration. In addition, other related parameters involved in the algorithm can also be set, such as the initial amount of pheromone set to 0.1, and the preset pheromone evaporation coefficient set to 0.5.

[0082] Step 502, in each iteration, for each ant in the ant colony, starting from the starting node, calculate the selection probability corresponding to each adjacent node according to the pheromone concentration and heuristic information, and select the next node based on the selection probability.

[0083] The pheromone concentration is used to indicate the historical preference degree of the path between the two nodes.

[0084] For example, the selection probability corresponding to the adjacent node satisfies the following formula: wherein, is the selection probability of the th ant from node to node , and From node To the node The upper path is at the The pheromone concentration during each iteration. Represents a node To the node Heuristic information about the paths between them This is the pheromone index parameter, which controls the non-linearity of the pheromone's influence; it is set to 0.3. This is the importance parameter for heuristic information, used to adjust the influence of pheromones and heuristic information on path selection, and is set to 0.2. This indicates that the ant starts from the node. To the node All accessible nodes (i.e., nodes) The set of adjacent nodes.

[0085] Step 503: After each iteration, update the pheromone concentration on the path based on the path selection of each ant during the iteration.

[0086] For example, the updated pheromone concentration satisfies the following formula: in, For the updated node To the node The upper path is at the The pheromone concentration during each iteration. For the node To the node The pheromone evaporation coefficient of the upper path, For nodes To the node The upper path is at the The pheromone concentration during each iteration. This indicates the number of ants in the ant colony. Indicates the first Only ants are at the node To the node The pheromone increment left along the path.

[0087] It should be noted that the pheromone concentration update rule of the traditional ant colony algorithm is based on a fixed preset pheromone evaporation coefficient. This results in the pheromone concentration left by ants in different regions having the same decay rate. In each iteration, it is impossible to highlight the importance of nodes with similar color features but different texture features. Consequently, the boundary segmentation of different regions with similar features is unclear, making it difficult to find the optimal segmentation boundary.

[0088] Therefore, this application can adaptively adjust the decay rate of pheromones according to the texture features of the nodes, thereby preserving the pheromone concentration on the path with more boundary segmentation discrimination features, accelerating the removal of pheromones on nodes in the same region, and improving the accuracy of boundary segmentation of different regions.

[0089] In one possible implementation, after each iteration of this application, for each path selected by each ant in the iteration, the pheromone evaporation coefficient used to update the pheromone concentration is adjusted according to the difference between the texture feature vector sequences of the two nodes corresponding to the path, and then the pheromone concentration on the path is updated based on the adjusted pheromone evaporation coefficient.

[0090] For example, the pheromone evaporation coefficient used to adjust and update pheromone concentration satisfies the following formula: in, For the node To the node The pheromone evaporation coefficient of the upper path, This is the volatile baseline value, which can range from 0 to 1, for example, it can be set to 0.3. For nodes The corresponding pixel texture feature vector sequence, For nodes The corresponding pixel texture feature vector sequence, Represents a node and nodes Texture difference penalty term.

[0091] Thus, when the ant colony explores areas with significant differences in texture features, such as farmland and forest areas, where pixels have similar color features but large differences in texture features, the adjusted... The smaller the pheromone content, the lower the pheromone decay rate, retaining more pheromone and enhancing the ant colony's probability of exploring significant boundaries. This strengthens the ants' continuous exploration of ambiguous boundaries, ultimately enabling enhanced searching between regions with significant texture differences. The increased exploration space allows for shortest path searching within these regions, avoiding the omission of true boundaries due to rapid pheromone evaporation, a common problem in traditional methods. Similarly, in regions with smaller texture feature differences, such as large, continuous areas of farmland or water, the adjusted pheromone content... A larger size can accelerate the evaporation rate of pheromones and avoid over-segmentation caused by ant colonies exploring invalid path information in homogeneous areas.

[0092] The adaptive pheromone updating strategy based on the texture feature guidance can adjust the pheromone evaporation rate according to the texture feature information of different regions, and the adaptive mechanism takes into account the global search ability and local boundary sensitivity of the complex scene, and optimizes the exploration ability of the ant colony in different texture feature regions. In the region with large texture difference, the pheromone decay is slow, more pheromones are retained on the path, the continuous exploration of the ants to the fuzzy boundary is enhanced, the traditional method is avoided to miss the real boundary due to the rapid evaporation of pheromones, and the optimal shortest path cannot be obtained, so that the boundary segmentation of different regions is fuzzy. In the homogeneous region, the removal of invalid path pheromones is accelerated, the ants are guided to focus on the obvious edge, and the over-segmentation in the homogeneous region is avoided.

[0093] Step 504: After reaching the maximum number of iterations, the shortest path is determined according to the path selection of each ant, and the region is segmented with the shortest path as the region segmentation boundary.

[0094] When the number of iterations reaches the maximum number of iterations, the paths searched by all ants in the iterations are counted, and the path with the minimum total weight is selected from the complete paths from the starting node to the target node searched by all ants as the shortest path in the undirected graph. The shortest path is the segmentation boundary of different regions in the grid electronic map.

[0095] For example, the application can obtain a plurality of path segments of the segmentation boundary by performing the above scheme multiple times, and then integrate the plurality of path segments obtained into a segmentation boundary network. Each path segment is a segment of the segmentation boundary network, and the end point of the path segment is a node of the network. All nodes in the segmentation boundary network are traversed, the nearest adjacent node of the node is determined from any unvisited node, and the nodes are connected in turn until the starting node is returned, thereby forming a closed region. The process is repeated until all nodes are visited, a closed boundary set covering the entire grid electronic map is obtained, and the region segmentation of the grid electronic map is realized.

[0096] Based on the above technical scheme, the application ensures the parallelism and efficiency of the ant colony search by reasonable setting of the initialization parameters, avoids resource waste and search blindness, and calculates the selection probability of the pheromone and the heuristic information, so that the ants can utilize historical search experience and respond to current feature of the ground object. The ants can effectively guide the ant colony to search for the potential segmentation boundary, and the historical experience and the current path quality are taken into account. Finally, the region segmentation is realized through the shortest path, the boundary is clear, the region consistency is high, and a high-quality segmentation result is provided for subsequent vectorization processing.

[0097] It is to be noted that the progressive order of the above-mentioned embodiments of the present application is only for the purpose of description, and does not represent the advantages and disadvantages of the embodiments. The processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multi-task processing and parallel processing are also possible or can be advantageous.

[0098] Each of the embodiments in the specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other. Each embodiment focuses on the difference from other embodiments.

Claims

1. A high-precision vectorization method of a grid electronic map, characterized in that, The method comprises the following steps: acquiring a raster electronic map of a region to be processed; performing region segmentation on the raster electronic map based on a graph cut algorithm, wherein a shortest path of the graph cut algorithm is calculated by using an adaptive ant colony algorithm; performing vectorization processing on the segmented raster electronic map.

2. The method of high-precision vectorization of a raster electronic map according to claim 1, characterized in that, The region segmentation on the raster electronic map based on the graph cut algorithm comprises: constructing an undirected graph structure of the raster electronic map, wherein the undirected graph structure takes pixel points in the raster electronic map as nodes and connecting lines between adjacent pixel points as edges; calculating edge weights between corresponding nodes based on feature similarity between pixel points, wherein the edge weights are used to represent the similarity between pixel points; searching for a shortest path in the undirected graph structure by using the adaptive ant colony algorithm and taking the edge weights as heuristic information, and performing region segmentation by taking the shortest path as a region segmentation boundary, wherein the heuristic information is used to adjust the selection probability of the path in the process of searching for the shortest path.

3. The method of high-precision vectorization of a raster electronic map according to claim 2, characterized in that, The calculation of the edge weights between corresponding nodes based on the feature similarity between pixel points comprises: for each pixel point in the raster electronic map, acquiring a color feature vector sequence and a texture feature vector sequence of the pixel point; for any two adjacent pixel points in the raster electronic map, determining the edge weights between the corresponding nodes of the two pixel points according to the color feature vector sequences and the texture feature vector sequences of the two pixel points.

4. The method of high-precision vectorization of a raster electronic map according to claim 3, characterized in that, The color feature vector sequence comprises pixel value components corresponding to multiple color channels respectively, and the texture feature vector sequence comprises at least one of contrast, correlation, energy and entropy.

5. The method of high-precision vectorization of a raster electronic map according to claim 4, characterized in that, The texture feature vector sequence of the pixel point is determined by the following way: performing grayscale processing on the raster electronic map to obtain a grayscale image, wherein the grayscale image comprises grayscale values of each pixel point in the raster electronic map; for each pixel point in the raster electronic map, determining the texture feature vector sequence of the pixel point by a gray level co-occurrence matrix based on the grayscale values of each pixel point in a pixel region with the pixel point as a center point.

6. The method of high-precision vectorization of a raster electronic map according to claim 3, characterized in that, The determination of the edge weights between the corresponding nodes of the two pixel points according to the color feature vector sequences and the texture feature vector sequences of the two pixel points comprises: for any two adjacent pixel points in the raster electronic map, determining a difference degree of the color feature vector sequences according to the color feature vector sequences of the two pixel points; determining a difference degree of the texture feature vector sequences according to the texture feature vector sequences of the two pixel points; determining a weight coefficient according to the difference degree of the texture feature vector sequences, wherein the weight coefficient is positively correlated with the difference degree of the texture feature vector sequences; performing region segmentation by taking the shortest path as a region segmentation boundary. According to the weight coefficient, the difference degrees of the color feature vector sequences and the texture feature vector sequences are weighted and fused to determine an edge weight value between nodes corresponding to the two pixel points; the edge weight value is negatively correlated with the difference degrees of the color feature vector sequences of the two pixel points and is negatively correlated with the difference degrees of the texture feature vector sequences of the two pixel points.

7. The method of high-precision vectorization of a raster electronic map according to claim 2, characterized in that, The adaptive ant colony algorithm is used to search for a shortest path in the undirected graph structure by taking the edge weight value as heuristic information, and region segmentation is performed by taking the shortest path as a region segmentation boundary, including: Ant colony is initialized to set the number of ants in the ant colony, the starting node and the target node of each ant, and the maximum iteration number of the adaptive ant colony algorithm; In each iteration, for each ant in the ant colony, starting from the starting node, the selection probability of each adjacent node is calculated according to the pheromone concentration and heuristic information, and the next node is selected based on the selection probability step by step until the target node is reached; the pheromone concentration is used to indicate the historical preference degree of the path between two nodes being selected; After each iteration, the pheromone concentration on the path is updated according to the path selection of each ant in the iteration; After the maximum iteration number is reached, the shortest path is determined according to the path selection of each ant, and region segmentation is performed by taking the shortest path as a region segmentation boundary.

8. The method of high-precision vectorization of a raster electronic map according to claim 7, characterized in that, After each iteration, the pheromone concentration on the path is updated according to the path selection of each ant in the iteration, including: After each iteration, for each path selected by each ant in the iteration, the pheromone evaporation coefficient used for updating the pheromone concentration is adjusted according to the difference degrees of the texture feature vector sequences of the two nodes corresponding to the path; The pheromone concentration on the path is updated based on the adjusted pheromone evaporation coefficient.

9. The method for high-precision vectorization of a raster electronic map according to any one of claims 1 to 8, characterized in that, The raster electronic map of the region to be processed is obtained, including: Remote sensing image data and geographic information system data of the region to be processed are obtained; A correction operation is performed on the remote sensing image data; the correction operation includes radiation correction and geometric correction; A preprocessing operation is performed on the geographic information system data, including coordinate system unification, attribute field normalization processing, and topology checking; Based on the corrected remote sensing image data and the preprocessed geographic information system data, the raster electronic map is generated.

10. The method for high-precision vectorization of a raster electronic map according to any one of claims 1 to 8, characterized in that, The method further includes: Different types of source-sink functional areas are identified according to the vectorized raster electronic map; the source-sink functional areas include source functional areas and sink functional areas; the source functional areas are areas that output resources; the sink functional areas are areas that absorb or consume energy; Based on the identified source-sink functional areas, a land resource regional pattern index is calculated; the land resource regional pattern index is used to quantify the spatial configuration characteristics of land resources; The calculated land resource regional pattern index is used as a decision criterion, and a multi-criteria decision method is used to sort the source-sink functional areas to determine the priority of land optimization configuration.

Images