Construction method based on rural settlement spatial form model

Abstract

The invention discloses a construction method based on a rural settlement spatial form model, and relates to the field of spatial form modeling. The construction method based on the rural settlement spatial form model comprises the following steps: obtaining visual angle sequence three-dimensional point cloud data and visual angle sequence image data of a plurality of areas of a to-be-constructed rural; based on the visual angle sequence three-dimensional point cloud data of the corresponding area, analyzing the area three-dimensional point cloud data of the corresponding area; performing comprehensive analysis on regional three-dimensional point cloud data and visual angle sequence image data of a corresponding region based on a pre-trained spatial structure recognition model to obtain a spatial form feature set of the to-be-constructed country, including a spatial structure discrete index and a spatial organization coupling index; according to the method, the rural settlement spatial form model is constructed through the spatial form feature set of the to-be-constructed rural, so that spatial structure features in the rural settlement are captured, the spatial features of the rural settlement can be expressed more clearly and more accurately, and spatial form modeling can be better carried out.

Description

Technical Field

[0001] The present invention relates to the field of spatial form modeling, and specifically to a construction method based on a rural settlement spatial form model. Background Art

[0002] Traditional rural planning methods mainly rely on experience and manual judgment, lacking scientific and quantitative spatial analysis tools, resulting in certain subjectivity and limitations in the planning results; Existing rural settlement spatial form analysis methods mainly focus on two-dimensional plane maps and simple three-dimensional modeling, and it is difficult to fully consider the three-dimensional spatial structure of rural settlements and the complexity of their functional areas. For example, many methods focus on simple land use map analysis and fail to deeply explore the spatial distribution characteristics, topological relationships between regions, and the coupling degree between spatial units within the region. Therefore, there is a lack of in-depth multi-dimensional and global analysis of the rural settlement spatial form, making it difficult to provide accurate basis for the refined planning and spatial optimization of rural areas.

[0003] The limitations of the existing technology at least include the following problems: It is difficult for the existing technology to accurately integrate multi-source heterogeneous data, such as image data and three-dimensional point cloud data. Among them, although image data has the advantage of wide coverage, it lacks spatial depth information and is difficult to accurately reflect complex terrain and building structures; while three-dimensional point cloud data contains rich spatial information, but usually has limited perspectives and is affected by problems such as occlusion and sparse point clouds, making it difficult to completely depict the diverse spatial forms of rural areas. Due to the lack of an effective multi-source data fusion mechanism, it is further difficult for the existing technology to achieve the deep fusion and complementarity of perspective sequence images and three-dimensional point cloud data, and then it is easy to lead to one-sided and inaccurate spatial structure recognition. Summary of the Invention

[0004] Aiming at the deficiencies of the existing technology, the present invention provides a construction method based on a rural settlement spatial form model, which solves the problem that it is difficult for the existing technology to accurately identify the spatial structure characteristics of rural settlements.

[0005] To achieve the above objectives, the present invention is realized through the following technical solutions: A construction method based on a rural settlement spatial form model includes the following steps: obtaining three-dimensional point cloud data and perspective sequence image data of several regions of the rural area to be constructed; based on the three-dimensional point cloud data of the perspective sequence of each region of the rural area to be constructed, analyzing the regional three-dimensional point cloud data of its corresponding region, including the regional voxel value and regional three-dimensional coordinates of each voxel point; comprehensively analyzing the regional three-dimensional point cloud data and perspective sequence image data of each region of the rural area to be constructed based on a pre-trained spatial structure recognition model to obtain a spatial form feature set of the rural area to be constructed, including a spatial structure discrete index and a spatial organization coupling index; and constructing a rural settlement spatial form model based on the spatial form feature set of the rural area to be constructed.

[0006] Further, the three-dimensional point cloud data of the perspective sequence includes the three-dimensional point cloud data of each perspective, specifically the voxel value and three-dimensional coordinates of each voxel point. The specific steps for analyzing the regional three-dimensional point cloud data of each region of the rural area to be constructed are as follows: performing spatial registration processing on the three-dimensional point cloud data of the perspective sequence of each region of the rural area to be constructed to obtain the registered voxel value and registered three-dimensional coordinates of each voxel point in its corresponding region, and performing neighborhood point matching processing to obtain several groups of matching voxel point sets in its corresponding region, including several matching voxel points; comprehensively analyzing the three-dimensional point cloud data of the perspective sequence of each region of the rural area to be constructed to obtain the weight coefficient of each matching voxel point in each group of matching voxel point sets in each region of the rural area to be constructed, and performing weighted processing with the registered voxel value and registered three-dimensional coordinates of its corresponding matching voxel point to obtain the regional three-dimensional point cloud data of its corresponding region.

[0007] Further, the specific steps for obtaining several groups of matching voxel point sets in its corresponding region are as follows: analyzing the spatial distance value between the registered three-dimensional coordinates of each voxel point in each region of the rural area to be constructed and the registered three-dimensional coordinates of each voxel point within its preset neighborhood radius; analyzing several groups of matching voxel point sets in its corresponding region based on the spatial distance value between each voxel point in each region of the rural area to be constructed and each voxel point within its preset neighborhood radius.

[0008] Further, the specific steps for obtaining the weight coefficient of each matching voxel point in each group of matching voxel point sets in each region of the rural area to be constructed are as follows: reading the three-dimensional point cloud data of each perspective of each region of the rural area to be constructed, analyzing the local shape similarity index, local surface normal similarity index, and visibility value of each voxel point in its corresponding perspective, and performing comprehensive analysis to obtain the structure information coverage factor of its corresponding voxel point; performing correlation processing on the structure information coverage factor of each voxel point in each perspective of each region of the rural area to be constructed to obtain the weight coefficient of each matching voxel point in each group of matching voxel point sets in each region of the rural area to be constructed.

[0009] Further, the specific formula for calculating the structure information coverage factor of a voxel point from a certain perspective in a certain area of the rural area to be constructed is as follows: ; Wherein, is the structure information coverage factor of a voxel point from a certain perspective in a certain area of the rural area to be constructed, is the local morphological similarity index of a voxel point from a certain perspective in a certain area of the rural area to be constructed, is the geometric coefficient stored in the database, is the local surface normal similarity index of a voxel point from a certain perspective in a certain area of the rural area to be constructed, is the normal coefficient stored in the database, is the visibility value of a voxel point from a certain perspective in a certain area of the rural area to be constructed, is the visibility coefficient stored in the database.

[0010] Further, the perspective sequence image data is specifically the image data of each perspective, including the pixel value, two-dimensional coordinate and corresponding depth information value of each pixel point. The specific steps for obtaining the spatial morphological feature set of the rural area to be constructed are as follows: input the regional three-dimensional point cloud data and perspective sequence image data of each area of the rural area to be constructed into a pre-trained spatial structure recognition model for parsing and processing to obtain the spatial perception set of the rural area to be constructed, including the spatial edge complexity index, spatial hierarchical organization index, spatial heterogeneity index, and spatial connectivity coordination index; based on the spatial perception set of the rural area to be constructed, analyze the spatial morphological feature set of the rural area to be constructed, including the spatial structure discreteness index and spatial organization coupling index.

[0011] Further, the spatial structure recognition model is specifically a spatial perception network, including an input layer, a feature extraction layer, a topological analysis layer, and an output layer. The feature extraction layer includes a point cloud feature extraction layer, an image feature extraction layer, and a feature fusion layer. The topological analysis layer includes a graph construction layer and a topological feature extraction layer.

[0012] Further, the specific steps to obtain the spatial perception set of the to-be-constructed village are as follows: In the input layer of the spatial perception network, receive the regional three-dimensional point cloud data and perspective sequence image data of each region of the to-be-constructed village, and perform preprocessing; in the feature extraction layer of the spatial perception network, perform feature encoding processing on the preprocessed regional three-dimensional point cloud data and perspective sequence image data of each region of the to-be-constructed village to obtain the fused feature vector of each region of the to-be-constructed village; in the topological analysis layer of the spatial perception network, perform topological construction processing on the feature vectors of each target of the to-be-constructed village to obtain the spatial topology vector of the to-be-constructed village; in the output layer of the spatial perception network, perform mapping processing on the spatial topology vector of the to-be-constructed village to obtain the spatial perception set of the to-be-constructed village.

[0013] Further, the specific steps to analyze the spatial form feature set of the to-be-constructed village are as follows: Perform weighted processing on the spatial edge complexity index and spatial hierarchical organization index of the to-be-constructed village to obtain the spatial structure discreteness index of the to-be-constructed village; perform weighted processing on the spatial heterogeneity index and spatial connectivity coordination index of the to-be-constructed village to obtain the spatial organization coupling index of the to-be-constructed village.

[0014] Further, the rural settlement spatial form model is specifically as follows: ; Wherein, is the proportionality coefficient stored in the database, is the spatial structure discreteness index of the to-be-constructed village, is the discreteness coefficient stored in the database, is the spatial organization coupling index of the to-be-constructed village, is the coupling coefficient stored in the database, is the interaction coefficient stored in the database.

[0015] The present invention has the following beneficial effects: (1) The construction method based on the rural settlement spatial form model obtains the perspective sequence three-dimensional point cloud data and perspective sequence image data from multiple perspectives, and through spatial registration and neighborhood matching, constructs a regional three-dimensional point cloud data with clear structure and high precision, and inputs it together with the perspective sequence image data into the spatial recognition model, so as to simultaneously extract three-dimensional structural features and information such as texture and boundary in the image. A graph neural network is also introduced inside the model to perform graph analysis on the spatial relationships between different regions, extract the organizational connection and coupling relationships between regions, generate the overall spatial topology features, capture the spatial structure features in the rural settlement, and thus can express the spatial characteristics of the rural settlement more clearly and accurately, and then can better perform the modeling of the spatial form.

[0016] (2) The construction method of the rural settlement spatial form model unifies the multi-view three-dimensional point cloud coordinate systems through the iterative closest point algorithm, constructs a unified coordinate system under the reference view, eliminates the spatial deviation and overlapping error between different views, and ensures that all point cloud data can be analyzed within a unified coordinate system. Then, it determines the matching voxel points through neighborhood matching, constructs the structural information coverage factor, generates the weight coefficients of the matching voxel point set, and performs weighted fusion based on the weight coefficients, thus effectively fusing the three-dimensional point cloud data of multiple views to ensure that the structure represented by the regional three-dimensional point cloud data is more stable and the details are more accurate.

[0017] (3) The construction method of the rural settlement spatial form model constructs a rural settlement spatial form model that can reflect the overall spatial pattern by extracting the spatial perception set of the rural area to be constructed and generating the spatial form feature set, integrating factors such as the connection between regions and the density of spatial distribution, forming a clear and complete spatial form expression. The finally obtained model can well reflect the distribution characteristics of the village space and the coordination between regions, thus providing effective support for rural space planning and optimization.

[0018] Of course, it is not necessary for any product implementing the present invention to achieve all the above-mentioned advantages simultaneously. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 It is a flowchart of the construction method of a rural settlement spatial form model according to the present invention.

[0020] Figure 2 It is a flowchart of the specific steps for analyzing the regional three-dimensional point cloud data of each region of the rural area to be constructed in the construction method of a rural settlement spatial form model according to the present invention.

[0021] Figure 3 It is a schematic diagram of the local shape similarity index voxel sequence of a certain perspective of a certain region of the rural area to be constructed in the construction method of a rural settlement spatial form model according to the present invention.

[0022] Figure 4 It is a schematic diagram of the local surface normal similarity index voxel sequence of a certain perspective of a certain region of the rural area to be constructed in the construction method of a rural settlement spatial form model according to the present invention.

[0023] Figure 5 It is a schematic diagram of the visibility value voxel sequence of a certain perspective of a certain region of the rural area to be constructed in the construction method of a rural settlement spatial form model according to the present invention.

[0024] Figure 6This is a specific step flowchart for obtaining the spatial form feature set of the to-be-constructed rural area in a construction method of a rural settlement spatial form model according to the present invention. Detailed implementation manners

[0025] Please refer to Figure 1 , an embodiment of the present invention provides a technical solution: a construction method of a rural settlement spatial form model, including the following steps: obtaining perspective sequence three-dimensional point cloud data and perspective sequence image data of several regions of the to-be-constructed rural area; based on the perspective sequence three-dimensional point cloud data of each region of the to-be-constructed rural area, analyzing the regional three-dimensional point cloud data of its corresponding region, including the regional voxel value and regional three-dimensional coordinates of each regional voxel point; comprehensively analyzing the regional three-dimensional point cloud data and perspective sequence image data of each region of the to-be-constructed rural area based on a pre-trained spatial structure recognition model to obtain the spatial form feature set of the to-be-constructed rural area, including a spatial structure dispersion index and a spatial organization coupling index; and constructing a rural settlement spatial form model based on the spatial form feature set of the to-be-constructed rural area.

[0026] The rural settlement spatial form model is specifically as follows: ; Among them, is the proportionality coefficient stored in the database, and its value is 0.500 in this embodiment, is the spatial structure dispersion index of the to-be-constructed rural area, is the dispersion coefficient stored in the database, is the spatial organization coupling index of the to-be-constructed rural area, is the coupling coefficient stored in the database, is the interaction coefficient stored in the database.

[0027] It should be explained that the dispersion coefficient stored in the database is obtained as follows: reading the spatial position coordinates of each region of the to-be-constructed rural area in the graph construction layer, and respectively calculating the Euclidean distance values between adjacent regions of the to-be-constructed rural area based on the Euclidean distance formula, and respectively performing mean processing and standard deviation processing to obtain the Euclidean distance mean and Euclidean distance standard deviation value of the to-be-constructed rural area, and performing ratio processing, that is, Euclidean distance standard deviation value / Euclidean distance mean, and marking the obtained result as the dispersion coefficient ; The coupling coefficient stored in the database is obtained as follows: reading the feature similarity between adjacent nodes in the graph construction layer, and performing mean processing, and marking the obtained result as the coupling coefficient ; The interaction coefficient The acquisition steps are as follows: Obtain the historical perspective sequence three-dimensional point cloud data and historical perspective sequence image data of several regions in several time periods of the rural area to be constructed, and conduct comprehensive analysis to obtain the historical spatial structure discrete index (consistent with the acquisition logic of the spatial structure discrete index) and historical spatial organization coupling index (consistent with the acquisition logic of the spatial organization coupling index) for each historical time period, and perform normalization processing to map the values of the two to the interval [0, 1] to unify the dimension; Multiply the historical spatial structure discrete index and historical spatial organization coupling index for each historical time period after normalization processing to obtain the interaction index for each historical time period, and perform mean and standard deviation processing respectively to obtain the mean value of the interaction index and the standard deviation of the interaction index, and perform a ratio processing, that is, the standard deviation of the interaction index / the mean value of the interaction index, and mark the obtained result as the interaction coefficient 。

[0028] Specifically, as Figure 2 shown, the perspective sequence three-dimensional point cloud data includes the three-dimensional point cloud data for each perspective (including but not limited to the top-down perspective directly above, the horizontal perspective directly in front, the left perspective, the right perspective, etc.), and specifically, it is the voxel value and three-dimensional coordinates of each voxel point. The specific steps for analyzing the regional three-dimensional point cloud data of each region of the rural area to be constructed are as follows: Perform spatial registration processing on the perspective sequence three-dimensional point cloud data of each region of the rural area to be constructed to obtain the registered voxel value and registered three-dimensional coordinates of each voxel point in its corresponding region, and perform neighborhood point matching processing to obtain several sets of matching voxel point sets in its corresponding region, including several matching voxel points; Conduct comprehensive analysis on the perspective sequence three-dimensional point cloud data of each region of the rural area to be constructed to obtain the weight coefficient of each matching voxel point in each set of matching voxel point sets of each region of the rural area to be constructed, and perform weighted processing with the registered voxel value and registered three-dimensional coordinates of its corresponding matching voxel point to obtain the regional three-dimensional point cloud data of its corresponding region.

[0029] Among them, the specific steps of spatial registration processing are as follows: Set the horizontal front view of each area as the reference view, and set the corresponding three-dimensional point cloud data as the reference point cloud data. On this basis, perform fine registration processing on the three-dimensional point cloud data of each view except the reference view in this area using the Iterative Closest Point algorithm (ICP). Specifically: Take the minimization of the sum of the squares of the Euclidean distances between the spatial corresponding points of the three-dimensional point cloud data of each view except the reference view and the reference point cloud data of the reference view as the objective function, iteratively calculate the three-dimensional rigid body transformation parameters required to align the three-dimensional point cloud data of each view to the reference point cloud data, that is, the rotation matrix and the translation vector, and update the rotation matrix and the translation vector in each iteration until the average registration error between point pairs is lower than the set convergence threshold. Based on the finally determined rotation matrix and translation vector of each view and the reference view, perform coordinate transformation on the three-dimensional coordinates of each voxel point of each view to obtain the three-dimensional coordinates of each voxel point of each view in the reference coordinate system, and mark them as registered three-dimensional coordinates, and at the same time mark the corresponding voxel value as the registered voxel value.

[0030] The specific steps to obtain several sets of matching voxel point sets in its corresponding area are as follows: Based on the registered three-dimensional coordinates of each voxel point in each area of the rural area to be constructed and the registered three-dimensional coordinates of each voxel point within its preset neighborhood radius, analyze its spatial distance value from each voxel point within the preset neighborhood radius (that is, calculate the distance between the registered three-dimensional coordinates of each voxel point and the registered three-dimensional coordinates of each voxel point within its preset neighborhood radius based on the Euclidean distance formula, and mark the result as the spatial distance value). Based on the spatial distance value between each voxel point in each area of the rural area to be constructed and each voxel point within the preset neighborhood radius, analyze several sets of matching voxel point sets in its corresponding area. Specifically: Judge whether the spatial distance value of its corresponding voxel point is lower than the preset spatial distance threshold; if it is lower than the preset spatial distance threshold, mark it and the corresponding voxel point within the preset neighborhood radius as a matching voxel point pair, and perform statistical analysis (summarize all voxel point pairs with spatial distance values lower than the preset spatial distance threshold to form a matching voxel point set), obtain several sets of matching voxel point sets in its corresponding area, and mark each voxel point in the corresponding set of matching voxel point sets as each matching voxel point; if it is not lower than the preset spatial distance threshold, do not mark.

[0031] In this implementation scheme, by setting a unified reference perspective and performing spatial registration processing, the problem of inconsistent coordinates between multi-perspective three-dimensional point cloud data is effectively solved, and the voxel points collected from different angles can be expressed in the same coordinate system. Secondly, after the registration is completed, through neighborhood matching, voxel points with similar spatial positions are automatically identified and grouped into the same set of matching voxel points, thereby effectively removing duplicate data brought by multiple perspectives and reducing spatial deviations caused by different acquisition angles, thus improving the accuracy of the regional three-dimensional point cloud data.

[0032] Specifically, the specific steps for obtaining the weight coefficient of each matching voxel point in each set of matching voxel points in each region of the rural area to be constructed are as follows: Read the three-dimensional point cloud data of each perspective in each region of the rural area to be constructed, analyze the local morphological similarity index, local surface normal similarity index, and visibility value of each voxel point in its corresponding perspective, and conduct comprehensive analysis to obtain the structural information coverage factor of its corresponding voxel point; Perform correlation processing on the structural information coverage factors of each voxel point in each perspective in each region of the rural area to be constructed to obtain the weight coefficient of each matching voxel point in each set of matching voxel points in each region of the rural area to be constructed.

[0033] Among them, the specific steps for analyzing the local morphological similarity index, local surface normal similarity index, and visibility value of each voxel point in its corresponding perspective are as follows: Based on each voxel point in each perspective in each region of the rural area to be constructed, a preset neighborhood voxel window (such as a 3×3×3 voxel window) is set, and the difference analysis is performed on the three-dimensional coordinates of each voxel point in the preset neighborhood voxel window (that is, calculation processing is performed based on the Euclidean distance formula), and the distance difference between it and each voxel point in the preset neighborhood voxel window is obtained, and variance processing is performed to obtain the distance difference variance value, and a covariance matrix is constructed based on the three-dimensional coordinates of each voxel point in the preset neighborhood voxel window, and the three-axis eigenvalues (that is, the eigenvalues in the three coordinate axis directions) are solved, and the minimum eigenvalue is statistically obtained, and the ratio analysis is performed with the sum of the three-axis eigenvalues to obtain the curvature value, and weighted processing is performed with the distance difference variance value to obtain the local morphological similarity index of each voxel point in each perspective in each region of the rural area to be constructed; Based on each voxel point from each perspective in each area of the rural area to be constructed, a preset neighborhood voxel window (such as a 3×3×3 voxel window) is set, and a covariance matrix is constructed with the three-dimensional coordinates of each voxel point within the preset neighborhood voxel window. The three-axis eigenvalues (i.e., the eigenvalues in the three coordinate axis directions) and the corresponding eigenvectors are obtained by solving. The minimum eigenvalue and its corresponding eigenvector are statistically obtained and marked as the unit normal vector corresponding to this voxel point. For each voxel point within the preset neighborhood voxel window, its normal vector is obtained (the acquisition logic is the same as that of the unit normal vector corresponding to this voxel point), the cosine similarity (i.e., the dot product of the normal vectors) between the unit normal vector corresponding to this voxel point and it is calculated, and the mean value is processed to obtain the local surface normal similarity index of each voxel point from each perspective in each area of the rural area to be constructed; The three-dimensional coordinates of the center point of the acquisition device for each perspective in each area of the rural area to be constructed are obtained, and analyzed respectively with the three-dimensional coordinates of each voxel point in the corresponding perspective to obtain the Euclidean distance of each voxel point (from the center point of the acquisition device). A maximum visible distance threshold is set. If the Euclidean distance of this voxel point is less than or equal to the maximum visible distance threshold, the visibility value is calculated by linear normalization, that is, 1 - Euclidean distance / maximum visible distance threshold. If the Euclidean distance of this voxel point is higher than the maximum visible distance threshold, the visibility value of this voxel point is assigned 0.

[0034] The specific steps of the association process are as follows: Each voxel point from each perspective in each area of the rural area to be constructed is associated with each matching voxel point in each set of matching voxel points in the corresponding area, and the structure information coverage factor of each voxel point is marked as the contribution factor of the corresponding matching voxel point. The contribution factors of each matching voxel point in each set of matching voxel points are summed to obtain the contribution sum value, and the contribution factor of each matching voxel point is respectively analyzed by ratio with the contribution sum value, and the ratio analysis result is used as its corresponding weight coefficient.

[0035] The specific formula for calculating the structure information coverage factor of a certain voxel point from a certain perspective in a certain area of the rural area to be constructed is as follows: ; Among them, is the structure information coverage factor of a certain voxel point from a certain perspective in a certain area of the rural area to be constructed, is the local morphological similarity index of a certain voxel point from a certain perspective in a certain area of the rural area to be constructed, is the geometric coefficient stored in the database, is the local surface normal similarity index of a certain voxel point from a certain perspective in a certain area of the rural area to be constructed, is the normal coefficient stored in the database, is the visibility value of a voxel point from a certain perspective in a certain area of the rural area to be constructed, is the visibility coefficient stored in the database.

[0036] It should be noted that the geometric coefficient stored in the database is obtained as follows: For the voxel points from a certain perspective in a certain area of the rural area to be constructed, set a fixed voxel neighborhood window (such as 3×3×3). Based on the three-dimensional coordinates of all voxel points within the window and the three-dimensional coordinates of the central point (i.e., this voxel point), construct a three-dimensional coordinate covariance matrix and extract its eigenvalues; perform a ratio analysis on the minimum eigenvalue and the sum of the three-axis eigenvalues to obtain the geometric curvature value of this point. Sample multiple voxel points in different areas and different perspectives, repeat the above process, collect their geometric curvature values, and perform an average process. The obtained result is marked as the geometric coefficient ; The normal coefficient stored in the database is obtained as follows: Read the local surface normal similarity index of each voxel point from each perspective in each area of the rural area to be constructed, and perform standard deviation processing and average processing to obtain the standard deviation of the local surface normal similarity index, the average of the local surface normal similarity index, and perform a ratio analysis, that is, the standard deviation of the local surface normal similarity index / the average of the local surface normal similarity index. The obtained result is marked as the normal coefficient ; The visibility coefficient stored in the database is obtained as follows: Read the visibility value of each voxel point from each perspective in each area of the rural area to be constructed, and perform standard deviation processing and average processing to obtain the standard deviation of the visibility value, the average of the visibility value, and perform a ratio analysis, that is, the standard deviation of the visibility value / the average of the visibility value. The obtained result is marked as the visibility coefficient .

[0037] The specific implementation example of calculating the structural information coverage factor of a voxel point from a certain perspective in a certain area of the rural area to be constructed is as follows. There is the following data, including the local morphological similarity index, local surface normal similarity index, and visibility value of 5 voxel points (randomly selected) from a certain perspective in a certain area of the rural area to be constructed, as shown in Table 1 and Figures 3 - 5 shown as follows: Table 1 Example of voxel point sequence data from a certain perspective in a certain area of the rural area to be constructed Local morphological similarity index Local surface normal similarity index Visibility value Voxel point 1 0.887 0.859 0.921 Voxel point 2 0.726 0.816 0.793 Voxel point 3 0.824 0.836 0.821 Voxel point 4 0.568 0.426 0.764 Voxel point 5 0.637 0.549 0.718 The geometric coefficient stored in the database is approximately: 0.627; The normal coefficient stored in the database is approximately: 0.142; The visibility coefficient stored in the database Approximately: 0.134; Substitute the data in Table 1 and the above coefficients into the specific formula for the structural information coverage factor of a voxel point from a certain perspective in a certain area of the village to be constructed, and obtain: The structural information coverage factor of the first voxel point from a certain perspective in a certain area of the village to be constructed = ln(((1 + 0.627×0.887 + √(0.142×0.859) + 0.921^0.134)^(1 / (1 + 0.627 + 0.142 + 0.134)))) ≈ 0.558; The structural information coverage factor of the second voxel point from a certain perspective in a certain area of the village to be constructed = ln(((1 + 0.627×0.726 + √(0.142×0.816) + 0.793^0.134)^(1 / (1 + 0.627 + 0.142 + 0.134)))) ≈ 0.533; The structural information coverage factor of the third voxel point from a certain perspective in a certain area of the village to be constructed = ln(((1 + 0.627×0.824 + √(0.142×0.836) + 0.821^0.134)^(1 / (1 + 0.627 + 0.142 + 0.134)))) ≈ 0.546; The structural information coverage factor of the fourth voxel point from a certain perspective in a certain area of the village to be constructed = ln(((1 + 0.627×0.568 + √(0.142×0.426) + 0.764^0.134)^(1 / (1 + 0.627 + 0.142 + 0.134)))) ≈ 0.489; The structural information coverage factor of the fifth voxel point from a certain perspective in a certain area of the village to be constructed = ln(((1 + 0.627×0.637 + √(0.142×0.549) + 0.718^0.134)^(1 / (1 + 0.627 + 0.142 + 0.134)))) ≈ 0.508.

[0038] In this implementation plan, by analyzing the characteristics of voxel points from different perspectives to determine which points are more reliable and representative, and by setting a local neighborhood window, comprehensively measuring the geometric changes, surface direction consistency, and visibility of each point, calculating its structural information coverage factor, so as to more accurately distinguish which points are key and need to be retained, and which points may be redundant or unstable. Finally, when fusing multi-angle data, information can be selectively integrated according to these weights to reduce errors, improve the accuracy and coherence of the regional 3D point cloud data, and facilitate subsequent spatial structure analysis and settlement modeling analysis.

[0039] Specifically, such as Figure 6As shown, the perspective sequence image data is specifically the image data of each perspective, including the pixel value, two-dimensional coordinates, and corresponding depth information value of each pixel point. The specific steps to obtain the spatial morphological feature set of the rural area to be constructed are as follows: Input the regional three-dimensional point cloud data and perspective sequence image data of each region of the rural area to be constructed into a pre-trained spatial structure recognition model for parsing and processing to obtain the spatial perception set of the rural area to be constructed, including the spatial edge complexity index, spatial hierarchical organization index, spatial heterogeneity index, and spatial connectivity coordination index; Based on the spatial perception set of the rural area to be constructed, analyze the spatial morphological feature set of the rural area to be constructed, including the spatial structure dispersion index (measuring the distribution dispersion degree of each region in the rural space) and the spatial organization coupling index (measuring the organizational association and coupling coordination degree between regions in the rural space).

[0040] The spatial structure recognition model is specifically a spatial perception network (constructed based on PointNet++ and multi-view convolutional neural networks, etc.), including an input layer, a feature extraction layer, a topological analysis layer, and an output layer. The feature extraction layer includes a point cloud feature extraction layer, an image feature extraction layer, and a feature fusion layer, and the topological analysis layer includes a graph construction layer and a topological feature extraction layer.

[0041] The specific steps to obtain the spatial perception set of the rural area to be constructed are as follows: In the input layer of the spatial perception network, receive the regional three-dimensional point cloud data and perspective sequence image data of each region of the rural area to be constructed (i.e., the regional voxel value, regional three-dimensional coordinates of each regional voxel point, and the pixel value, two-dimensional coordinates, and corresponding depth information value of each pixel point of each perspective), and perform preprocessing; In the feature extraction layer of the spatial perception network, perform feature encoding processing on the preprocessed regional three-dimensional point cloud data and perspective sequence image data of each region of the rural area to be constructed to obtain the fusion feature vector of each region of the rural area to be constructed; In the topological analysis layer of the spatial perception network, perform topological construction processing on the feature vector of each target of the rural area to be constructed to obtain the spatial topological vector of the rural area to be constructed; In the output layer of the spatial perception network, perform mapping processing on the spatial topological vector of the rural area to be constructed to obtain the spatial perception set of the rural area to be constructed.

[0042] The specific steps to analyze the spatial morphological feature set of the rural area to be constructed are as follows: Perform weighted processing on the spatial edge complexity index and spatial hierarchical organization index of the rural area to be constructed to obtain the spatial structure dispersion index of the rural area to be constructed; Perform weighted processing on the spatial heterogeneity index and spatial connectivity coordination index of the rural area to be constructed to obtain the spatial organization coupling index of the rural area to be constructed.

[0043] Among them, the specific steps of the feature encoding processing are as follows: Point cloud feature extraction layer, which extracts the point cloud feature vectors of each region based on the PointNet++ network structure. Specifically: for the three-dimensional point cloud data of each region of the rural area to be constructed, a spatial neighborhood radius parameter is set, such as 0.05 meters. For each voxel point in the region, a spatial indexing structure (such as a kd-tree) is used to retrieve the set of neighboring voxel points within the set spatial neighborhood radius, forming the local adjacent point set of the voxel points in this region. If the number of adjacent points is less than the preset threshold (such as 10), the radius is gradually increased until the requirement for the number of adjacent points is met. Then, the adjacent point set is sampled at multiple spatial neighborhood radii (such as 0.05 meters, 0.1 meters, and 0.2 meters) to construct a multi-scale local neighborhood. For the voxel points in each local neighborhood, the three-dimensional coordinate deviation relative to the center point of this neighborhood (that is, the mean of the three-dimensional coordinates of all voxel points in all local neighborhoods, and the result is used as the three-dimensional coordinates of the center point of this neighborhood) is calculated, and combined with the normal direction and curvature value obtained by principal component analysis (PCA) to form an input feature vector. This input feature vector is input into a multi-layer perceptron (MLP). The MLP contains 3 fully connected layers, with 64, 128, and 256 neurons respectively. The activation function uses ReLU to extract the geometric features of each neighborhood. For the feature vectors of different-scale neighborhoods, a max pooling operation is performed by dimension to aggregate them, obtaining the point cloud feature vector; Image feature extraction layer, which extracts the image feature vectors of each region based on the multi-view convolutional neural network structure. The multi-view convolutional neural network consists of multiple convolutional layers in sequence. Each convolutional layer uses a convolutional kernel of size 3×3, and the stride is set to 1. After the convolutional operation, a batch normalization layer (Batch Normalization) is immediately followed to normalize the activation values of the convolutional output. Subsequently, the result after batch normalization is activated through the non-linear activation function ReLU to achieve non-linear transformation of the features. After every two convolutional operations in the network, a max pooling layer is set to complete the downsampling of the feature map. Through the successive processing of this multi-layer convolution, batch normalization, activation, and pooling, the network can gradually extract and encode the multi-level visual information of the input image, including but not limited to texture details, edge structures, and depth change features, generating a three-dimensional feature map tensor for each perspective image to represent the multi-channel feature table of this perspective image. An element-wise max pooling operation is performed on the three-dimensional feature map tensors of all perspectives in the channel dimension. This operation is to take the largest activation value among all perspectives at the corresponding channel position, merge the multi-view information, and retain the most significant feature responses to generate a unified multi-view feature vector, that is, the image feature vector; Feature fusion layer: The point cloud feature vectors and image feature vectors of each region are respectively input into two independent linear mapping layers. Through matrix multiplication and addition operations, the linear mapping layers adjust the dimensions of the input feature vectors to the same preset dimension. After the mapping is completed, concatenation is performed based on the feature dimension direction (simple vector connection) to form a comprehensive feature vector before fusion. Then, based on a multi-layer fully connected neural network (including several fully connected layers), non-linear feature fusion and information interaction are completed. Each fully connected layer performs a linear transformation on the input vector through a weight matrix and a bias, and combines a non-linear activation function (such as ReLU) to realize the expression and refinement of complex features, and outputs a fused feature vector, which reflects the geometric attributes of the three-dimensional space structure in the point cloud data and the texture, edge, and depth information contained in the multi-view images.

[0044] The specific steps of the topology construction process are as follows: Graph construction layer: According to the spatial position coordinates of each region of the rural area to be constructed (i.e., the mean of the three-dimensional coordinates of all voxel points in the region), each region is used as a node in the graph, and each node corresponds to a region. Using the spatial proximity relationship between nodes, edges between nodes are constructed based on a preset adjacency distance threshold (that is, for any two nodes in the graph, calculate the Euclidean distance between their spatial position coordinates. If the distance is less than or equal to the adjacency distance threshold, an undirected edge is established between the two nodes), forming an undirected weighted graph. The weight of the edge is determined by calculating the feature similarity between adjacent nodes (calculated based on cosine similarity) to reflect the spatial association strength between nodes. The attribute carried by each node in the graph is the fused feature vector output by the feature extraction layer, and thus a spatial structure topology graph is obtained. Topological feature extraction layer, which extracts features from the spatial structure topology graph based on the graph neural network. Specifically: taking the fused feature vector of each node in the spatial topology graph as input, the graph neural network interacts the fused feature vectors of each node with its adjacent nodes through a pre-defined adjacency matrix. That is, for each node in the graph, it is achieved by weighted summation or concatenation of the fused feature vectors of its adjacent nodes. The weights are automatically learned through network training and are updated through multiple layers of iteration (i.e., message passing and node feature update operations are performed in each layer) to obtain the node features of each layer. Among them, the node features of each layer are activated by a non-linear activation function (such as ReLU) after passing through a trainable linear transformation (such as a fully connected layer). After all graph neural network layers are processed, through a global pooling operation (such as average pooling or max pooling), the feature vectors of all nodes are aggregated into a graph-level feature vector with a fixed dimension, that is, a spatial topology vector, which reflects the topological relationship of the rural spatial structure and the complex interaction information between regions, such as boundary perturbation geometric features, such as the number of adjacent nodes (the number of adjacent nodes for each node, traversing each node and counting the number of its adjacent nodes), the distribution of adjacent weights (by performing variance processing on the weights of adjacent nodes), the weighted average of adjacent node features (calculating the weighted average of the node feature vectors of adjacent nodes to each node), hierarchical features (degree centrality value, betweenness centrality value, clustering coefficient, etc. of nodes), spatial heterogeneity-related features, such as the main diagonal values of the covariance matrix of all spatial node features (by calculating the covariance matrix of node features and extracting its main diagonal values, that is, the variance of each feature), the root mean square cosine distance between node feature vectors (calculating the similarity between two node feature vectors, using the cosine similarity to measure the similarity of the two vectors, and then calculating the root mean square value), the normalized information entropy of the feature channels between nodes (measuring the distribution difference of the feature channels between nodes, used to represent the uncertainty or information content of the features, by calculating the probability distribution of node features and using the information entropy formula to calculate), the overall connectivity features of the graph, such as the average edge weight of the adjacency structure (calculating the average value of the weights of all edges in the graph), the average shortest path length of the graph (calculating the shortest paths between all node pairs in the graph and then taking the average), the graph diameter (the distance between the two farthest nodes in the graph) and the connectivity distribution density, that is, calculating the ratio of the number of edges between nodes in the subgraph to the maximum number of connected edges, and its maximum number of connected edges = [total number of nodes × (total number of nodes - 1)] / 2.

[0045] The specific steps of the mapping process are as follows: The spatial edge complexity index takes the geometric features of boundary perturbations in the spatial topology vector as the main input basis, including the number of adjacent nodes, the distribution of adjacent weights, and the weighted average of the features of adjacent nodes. These are input into the mapping branch of the spatial edge complexity index, i.e., the fully connected layer. This layer will transform these features into a new representation space through linear mapping. Through scale normalization, all features are adjusted to a similar scale range. The normalized features will be weighted and combined, and finally, the spatial edge complexity index is output, reflecting the clarity of the rural boundary structure to be constructed and the complexity of perturbations. The spatial hierarchical organization index takes the hierarchical-related features in the spatial topology vector as the input basis, including the degree centrality value of nodes (the number of direct connections between a node and other nodes), the betweenness centrality value (the frequency of appearance on the shortest paths between other nodes, the number of shortest paths passing through this node / the number of shortest paths between all node pairs in the graph), the clustering coefficient (measuring the degree to which a node forms a triangle with its neighbor nodes, i.e., forms a closed subgraph, the actual number of edges between the neighbors of this node / the maximum number of edges that can exist in theory), etc. These are input into the mapping path of the spatial hierarchical organization index. Through linear mapping, these features are transformed into a new representation space, and through scale normalization, all features are adjusted to a similar scale range. The normalized features will be weighted and combined, and the spatial hierarchical organization index is output, indicating the hierarchical distribution and organizational orderliness of the rural area to be constructed. The spatial heterogeneity index takes the spatial heterogeneity-related features in the spatial topology vector as the main input basis, including the main diagonal values of the covariance matrix of all spatial node features, the root mean square cosine distance between node feature vectors, and the normalized information entropy of the feature channels between nodes. These are input into the feature mapping branch of the spatial heterogeneity index. Through linear mapping, these features are transformed into a new representation space, and through scale normalization, all features are adjusted to a similar scale range. The normalized features will be weighted and combined, and the spatial heterogeneity index is output, reflecting the differential distribution of the rural spatial system. The spatial connectivity coordination index selects the overall connectivity features of the graph in the spatial topology vector as the input basis, including the average edge weight of the adjacency structure, the average shortest path length of the graph, the graph diameter, and the connectivity distribution density of local subgraphs. These are input into the mapping path of the spatial connectivity coordination index. Through linear mapping, these features are transformed into a new representation space, and through scale normalization, all features are adjusted to a similar scale range. The normalized features will be weighted and combined, generating a connectivity coordination index used to measure the tight relationship of the overall spatial structure and the overall coordination of spatial distribution.

[0046] And the pre-training steps of the spatial perception network are as follows: Obtain the dataset of labeled spatial structure features, which is constructed by urban and rural planning experts based on remote sensing images, topographic survey data, and building layout planning maps of the rural area to be constructed, and combined with the current regional land use situation, historical village organizational structure, and manual on-site investigation records, perform quadruple labeling of the spatial structure, form label fields for the spatial edge complexity index, spatial hierarchical organization index, spatial heterogeneity index, and spatial connectivity and coordination index, and divide them into spatial training subsets and spatial validation subsets.

[0047] Initialize the spatial perception network, such as initializing it in the Xavier way or the Kaiming way.

[0048] Train based on the spatial training subset, set the number of training epochs (such as 150 epochs), and each epoch includes: the forward propagation stage, input the training data of each epoch into the spatial perception network to generate the predicted values of the spatial edge complexity index, spatial hierarchical organization index, spatial heterogeneity index, and spatial connectivity and coordination index in sequence; the loss function construction stage, calculate the MSE loss for the predicted values of each type of spatial perception index and its label field respectively to form a multi-objective weighted loss function; the backpropagation and parameter optimization stage, use the AdamW optimizer to iteratively update the parameters of each layer, and introduce the DropEdge and random node masking mechanisms to prevent overfitting of the spatial graph structure features.

[0049] After each epoch of training, perform an evaluation process on the spatial validation subset, compare the prediction results and the true values of the four types of spatial perception indexes respectively, and the statistical indicators include MSE, MAE, spatial structure classification accuracy (used for predicting the hierarchical organization degree and heterogeneity), and node connection prediction consistency index (used for predicting the connectivity and coordination degree). At the same time, record the change of the training loss decline curve and the stability of the validation indicators; if the accuracy of the validation set does not improve significantly in multiple consecutive training cycles, trigger the Early Stopping mechanism to terminate the training process in advance.

[0050] After training is completed, export the model interface package in the standard deployable format to support the subsequent online parsing of rural spatial topology vectors and the rapid generation of spatial perception sets for the deployment of the rural spatial layout auxiliary decision-making system.

[0051] In this implementation, by fully integrating two types of data, namely regional three-dimensional point cloud data and multi-view images, not only the geometric information of the spatial structure is retained, but also visual features such as textures and edges in the images are extracted, so as to comprehensively describe the spatial appearance of each region in the countryside. At the same time, topological mapping and graph neural network analysis are introduced to enable the model to identify the spatial connections, structural hierarchies, and distribution differences between regions, thereby more accurately reflecting the spatial morphological characteristics within the settlement. Finally, through a series of index outputs, the performance of rural space in terms of boundary clarity, organizational hierarchy, difference, and connectivity is clearly quantified, thus enhancing the depth of understanding and evaluation accuracy of the rural spatial structure.

[0052] Although the preferred embodiments of the present invention have been described, those skilled in the art can make additional changes and modifications to these embodiments once they know the basic creative concept. Therefore, the appended claims are intended to be construed as including the preferred embodiments as well as all changes and modifications falling within the scope of the present invention.

[0053] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalent technologies, the present invention is also intended to include these modifications and variations.

Claims

1. A construction method based on a rural settlement spatial form model, characterized in that The method includes the following steps: Obtain the three-dimensional point cloud data and the image data of the perspective sequence of several regions of the rural area to be constructed; Based on the three-dimensional point cloud data of the perspective sequence of each region of the rural area to be constructed, analyze the three-dimensional point cloud data of the corresponding region, including the regional voxel value and the three-dimensional coordinates of each voxel point in each region; Based on the pre-trained spatial structure recognition model, comprehensively analyze the three-dimensional point cloud data and the image data of the perspective sequence of each region of the rural area to be constructed, and obtain the spatial form feature set of the rural area to be constructed, including the spatial structure discrete index and the spatial organization coupling index; And construct a rural settlement spatial form model based on the spatial form feature set of the rural area to be constructed.

2. The construction method based on the rural settlement spatial form model according to claim 1, characterized in that The three-dimensional point cloud data of the perspective sequence includes the three-dimensional point cloud data of each perspective, specifically the voxel value and the three-dimensional coordinates of each voxel point. The specific steps for analyzing the three-dimensional point cloud data of each region of the rural area to be constructed are as follows: Perform spatial registration processing on the three-dimensional point cloud data of the perspective sequence of each region of the rural area to be constructed to obtain the registered voxel value and the registered three-dimensional coordinates of each voxel point in the corresponding region, and perform neighborhood point matching processing to obtain several sets of matching voxel point sets in the corresponding region, including several matching voxel points; Comprehensively analyze the three-dimensional point cloud data of the perspective sequence of each region of the rural area to be constructed, obtain the weight coefficient of each matching voxel point in each set of matching voxel point sets in each region of the rural area to be constructed, and perform weighted processing with the registered voxel value and the registered three-dimensional coordinates of the corresponding matching voxel point to obtain the three-dimensional point cloud data of the corresponding region.

3. The construction method based on the rural settlement spatial form model according to claim 2, characterized in that, The specific steps for obtaining several sets of matching voxel point sets in the corresponding region are as follows: Based on the registered three-dimensional coordinates of each voxel point in each region of the rural area to be constructed and the registered three-dimensional coordinates of each voxel point within the preset neighborhood radius, analyze the spatial distance value between each voxel point within the preset neighborhood radius; Based on the spatial distance value between each voxel point in each region of the rural area to be constructed and each voxel point within the preset neighborhood radius, analyze several sets of matching voxel point sets in the corresponding region.

4. The construction method based on the rural settlement spatial form model according to claim 2, wherein The specific steps for obtaining the weight coefficient of each matching voxel point in each set of matching voxel point sets in each region of the rural area to be constructed are as follows: Read the three-dimensional point cloud data of each perspective of each region of the rural area to be constructed, analyze the local shape similarity index, the local surface normal similarity index, and the visibility value of each voxel point in the corresponding perspective, and perform comprehensive analysis to obtain the structure information coverage factor of the corresponding voxel point; Perform correlation processing on the structure information coverage factors of each voxel point in each perspective of each region of the rural area to be constructed to obtain the weight coefficient of each matching voxel point in each set of matching voxel point sets in each region of the rural area to be constructed.

5. The construction method based on the rural settlement spatial form model according to claim 4, characterized in that, The specific formula for calculating the structure information coverage factor of a voxel point in a certain perspective of a certain region of the rural area to be constructed is as follows: ； Among them, , , , are, in sequence, the structure information coverage factor, the local morphological similarity index, the local surface normal similarity index, and the visibility value of a voxel point from a certain perspective of a certain area of the rural area to be constructed. , , are, in sequence, the geometric coefficient, the normal coefficient, and the visibility coefficient stored in the database.

6. The construction method based on the rural settlement spatial form model according to claim 1, characterized in that The image data of the perspective sequence is specifically the image data of each perspective, including the pixel value, the two-dimensional coordinates, and the corresponding depth information value of each pixel point. The specific steps for obtaining the spatial form feature set of the rural area to be constructed are as follows: Input the regional three-dimensional point cloud data and perspective sequence image data of each area of the rural area to be constructed into a pre-trained spatial structure recognition model for parsing and processing, and obtain the spatial perception set of the rural area to be constructed, including the spatial edge complexity index, the spatial hierarchical organization index, the spatial heterogeneity index, and the spatial connectivity coordination index; Based on the spatial perception set of the rural area to be constructed, analyze the spatial form feature set of the rural area to be constructed, including the spatial structure dispersion index and the spatial organization coupling index.

7. The construction method based on the rural settlement spatial form model according to claim 6, wherein, The spatial structure recognition model is specifically a spatial perception network, including an input layer, a feature extraction layer, a topological analysis layer, and an output layer. The feature extraction layer includes a point cloud feature extraction layer, an image feature extraction layer, and a feature fusion layer. The topological analysis layer includes a graph construction layer and a topological feature extraction layer.

8. The construction method based on the rural settlement spatial form model according to claim 7, characterized in that The specific steps to obtain the spatial perception set of the rural area to be constructed are as follows: In the input layer of the spatial perception network, receive the regional three-dimensional point cloud data and perspective sequence image data of each area of the rural area to be constructed, and perform preprocessing; In the feature extraction layer of the spatial perception network, perform feature encoding processing on the preprocessed regional three-dimensional point cloud data and perspective sequence image data of each area of the rural area to be constructed, and obtain the fusion feature vector of each area of the rural area to be constructed; In the topological analysis layer of the spatial perception network, perform topological construction processing on the feature vectors of each target in the rural area to be constructed, and obtain the spatial topology vector of the rural area to be constructed; In the output layer of the spatial perception network, perform mapping processing on the spatial topology vector of the rural area to be constructed, and obtain the spatial perception set of the rural area to be constructed.

9. The construction method based on the rural settlement spatial form model according to claim 6, characterized in that, The specific steps to analyze the spatial form feature set of the rural area to be constructed are as follows: Perform weighted processing on the spatial edge complexity index and the spatial hierarchical organization index of the rural area to be constructed to obtain the spatial structure dispersion index of the rural area to be constructed; Perform weighted processing on the spatial heterogeneity index and the spatial connectivity coordination index of the rural area to be constructed to obtain the spatial organization coupling index of the rural area to be constructed.

10. The construction method based on the rural settlement spatial form model according to claim 1, wherein, The specific rural settlement spatial form model is as follows: ； Among them, is the proportionality coefficient stored in the database, , are, in sequence, the spatial structure discrete index and the spatial organization coupling index of the rural area to be constructed, , , are, in sequence, the discrete coefficient, coupling coefficient, and interaction coefficient stored in the database.

Images