Linear distribution pattern recognition method and system for complex building group pattern spots

Abstract

The invention discloses a linear distribution pattern recognition method and system for complex building group pattern spots, and the method comprises the steps: firstly, extracting building group pattern spot data in a target region, carrying out the pattern spot editing, data storage and quality inspection optimization, and forming qualified basic pattern spot data; secondly, performing semantic information escape on the building group pattern spots, and constructing a three-level semantic hierarchical structure; thirdly, defining a pattern spot linear distribution mode feature system and a measurement parameter set thereof in combination with features such as pattern spot areas, geometric shapes, directions and densities; secondly, designing a pattern spot constraint type clustering method suitable for different spatial scales; in addition, based on the semantic, morphological and density characteristics of the pattern spots, a recognition range delimiting algorithm considering semantic and spatial scales is constructed; and finally, based on a clustering result, adopting a multi-constraint strategy, designing a target mode automatic identification algorithm taking building group characteristics into consideration, and establishing a verification method in combination with expert knowledge, so as to quantitatively calculate the identification accuracy and recall ratio and carry out comprehensive analysis.

Description

Technical Field

[0001] The present invention belongs to the fields of architectural mapping, national land space governance, spatial planning and space management, and specifically relates to a linear distribution pattern recognition method and system for complex building group patterns. Background Art

[0002] A building cluster is a collection of buildings within a specific area, formed through functional, spatial, or morphological connections. It encompasses both the natural evolution of traditional villages and the products of planning and design, such as science and technology parks and commercial complexes. As a microcosm of urban and rural development, building clusters hold core social, economic, and ecological value. Extracting and analyzing the spatial distribution patterns of building clusters helps understand and optimize architectural spatial structure and spatial resource allocation, and holds important theoretical and practical significance.

[0003] The linear distribution pattern is the basic spatial distribution pattern of building clusters. At present, the recognition of linear distribution patterns is mainly aimed at single buildings in urban areas with no difference in semantic type, relatively close area, relatively regular shape and relatively uniform distribution density. It is often combined with the definition of geometric features of the pattern, and adopts pattern parameter measurement, deep learning or knowledge graph methods. However, these methods still have certain limitations when facing the linear distribution pattern recognition task of large-scale complex building clusters. The main manifestations are: (1) Due to the weak or even lack of semantic refinement expression ability of the target pattern recognition results, the reference value for applications in fields such as urban planning and management is limited; (2) Due to the limitations of the application of methods such as deep learning, the corresponding target pattern recognition results are difficult to obtain complete, and the calculation and parameter adjustment process is relatively difficult; (3) Complex building clusters often have the characteristics of different areas, complex shapes and different distribution densities, making it more difficult to automatically and accurately recognize the target pattern. Therefore, how to simply and effectively realize the automatic and accurate recognition of the linear distribution pattern of large-scale complex building clusters has become a technical problem that needs to be solved in this field. Summary of the Invention

[0004] In order to solve the technical problems existing in the background technology, the present invention aims to provide a linear distribution pattern recognition method and system for complex building clusters, including: determining and extracting building cluster data in the target area, carrying out pre-processing work such as cluster editing, data storage, quality inspection and optimization, and forming qualified basic cluster data; referring to the semantic classification of data such as surface coverage, interpreting the semantic information of building clusters, and constructing a three-level semantic hierarchy; combining the cluster area, geometric shape, direction difference and density difference, etc., to define the cluster linear distribution pattern feature system and its measurement parameter set; based on the target pattern feature system and Based on the semantic hierarchy of patches, a constrained clustering and merging method for patches of different spatial scales is designed; combining the semantic, morphological and density characteristics of the original patches with the constrained clustering and merging method, a reasonable range delineation algorithm for patch recognition that takes into account semantics and spatial scale is constructed; based on the above clustering results, combined with the target pattern characteristics and measurement parameters, a combined multiple constraint strategy is adopted to design an automatic target pattern recognition algorithm that takes into account the complex characteristics of building groups; combining expert knowledge and experience with spatial cognitive ability, a verification method for the automatic target pattern recognition results is constructed, and quantitative calculations are carried out based on the recognition accuracy and recall rate, and a comprehensive qualitative analysis is conducted.

[0005] In order to solve the technical problem, the technical solution of the present invention is:

[0006] A linear distribution pattern recognition method for complex building clusters, comprising:

[0007] S1: Based on the semantic classification structure of the surface cover data, the CC codes of the original qualified building cluster patches are converted into semantic classification information, and a three-level semantic hierarchy is constructed to obtain building cluster patches with three-level semantic hierarchy.

[0008] S2: Using the patch data output in step S1, a new feature system of the patch linear distribution pattern is determined, including combining area and geometric shape indicators to construct geometric morphological features; combining the differences in the directions of the patches themselves and between them to construct directional features; combining the differences in spatial distribution density to construct position and spacing features, and combining to form a linear distribution feature system and measurement parameters;

[0009] S3: Based on the patch data output from step S1 and the output results from step S2, the required level of classification is used to express the patch semantics, and a progressive clustering method for patches of different spatial scales is designed to obtain the patch data of the building group after preliminary clustering and merging.

[0010] S4: Based on the semantic, morphological, and density characteristics of complex building clusters, combined with the preliminary clustering data output by S3, the three-level semantic hierarchy and spatial scale information are integrated to design a reasonable range delineation algorithm for patch identification that takes into account semantic consistency and spatial proximity. This algorithm constructs a set of spacing limits, iteratively determines the mergibility between patches, and dynamically controls the area change and boundary fidelity during the merging process, thereby maintaining the geometric accuracy and semantic relevance of the original patches, and ultimately forming a new set of building cluster patches after cluster optimization;

[0011] S5: Based on the clustering results output from step S4 and the feature system output from step S2, a combined multi-constraint strategy is used to further design an automatic target pattern recognition algorithm that takes into account the complex characteristics of the building cluster pattern, and obtain the automatically recognized linear distribution pattern result of the building cluster;

[0012] S6: Using the recognition results of step S5 and the manual recognition results, the automatic recognition accuracy rate (ARR) and the recall rate (DR) are calculated by using methods such as intersection and difference sets, and a comprehensive evaluation is performed. Finally, a verification report and recognition performance indicators are output.

[0013] Furthermore, the target area to be identified is determined, the map data of the building group is collected, and the map editing, data storage, quality inspection and optimization preprocessing are carried out to form the original map data of the qualified building group.

[0014] Furthermore, the step S2 includes:

[0015] S201, construct morphological features:

[0016] In order to take into account the different area values ​​of the building group, the maximum ratio of the original area mapped by the target pattern line is set to areas ratio for:

[0017]

[0018] Among them, maxArea and minArea represent the maximum and minimum area values ​​of the three original patches respectively.

[0019] At the same time, taking into account the large differences in the boundary construction methods and geometric dimensions of building clusters under the semantic hierarchy, no rigid restrictions or changes are made to the geometric structure and details of the patches. In other words, the shape characteristics of the original patches are expressed truthfully and no special processing is made to their shapes as much as possible, so as to truly and fully mine richer target recognition line results.

[0020] S202, build direction features:

[0021] In order to simply and effectively constrain the differences between the directions of the spots themselves, this paper only designs the threshold direction of the difference in the angle of the long side of the minimum circumscribed rectangle of the original spot that intersects the three-dimensional line. diff. At the same time, in order to effectively control the arrangement angles between the spots and the connection form of the center points, the angle range of the angle θ between the two sides of the ternary line is set, and it is automatically determined by calculating the pairwise connection vectors based on the three center points. θ is defined as:

[0022]

[0023] in, and They are respectively represented by the vectors of the edges connected by two center points, and the two have a common connection point (i.e. and common starting point).

[0024] S203, build position and spacing features:

[0025] To avoid the large difference in spacing between the identified patches that is contrary to human cognition, the length ratio of the two sides of the triplet is set to sideLength ratio :

[0026]

[0027] in, and Represents vectors and The modulus length, where Greater than

[0028] At the same time, taking into account the differences in the density, boundary direction and topological relationship of the patches, in order to limit the unreasonable "insertion" of the original patches that were originally distributed separately by the algorithm, this paper adds the intersectLength, which is the ratio of the length of the three-dimensional line that is intersected (covered) by three patches to its own length. ratio :

[0029]

[0030] li represents the length of each line segment covered by the pattern. It represents the sum of the lengths of multiple covered line segments, and selfLength represents the length of the triplet itself.

[0031] Furthermore, the step S3 includes:

[0032] S301. Design the minimum upper image size of the cluster at the corresponding scale of the target spatial scale (patchSize) min, shape detail and the minimum distance between patches min , and set reasonable thresholds to improve the detail of data expression.

[0033] S302: When generalizing the boundary, the original corner points and edges are retained as much as possible, and the nearest node of the two edges to be connected is searched at the local merging position that meets the conditions, so as to constrain the merging of the two patches by enclosing the minimum merging area.

[0034] Furthermore, the step S4 includes:

[0035] S401. At the third semantic level and target spatial scale, calculate the corresponding spacing limit between patches, and construct and form a spacing value set step by step according to a certain small span between 0 and the limit.

[0036] S402 , start iteration, traverse the level-by-level spacing value set, obtain the patch set corresponding to each value in the set, traverse the set and count its semantic CC code and area value.

[0037] S403. Determine whether the code values ​​are the same. If they are the same, merge the corresponding two image patches in ascending order of area value, that is, execute the "line drawing and surface construction" solution in Section S3.

[0038] S404: According to the current spacing value, determine and find the position of the two spots to be merged, and then search for the four closest nodes on both sides of the two spots.

[0039] S405 , then draw lines on the node set and construct faces on the minimum area formed by the enclosed area.

[0040] S406. Finally, the facet small area is merged with the two adjacent original spots. After one round of iteration, the two old spots are eliminated and a new spot is generated.

[0041] S407, the new patch participates in a new round of calculation again, and the iteration continues until all values ​​of the step-by-step spacing value set gradually participate and complete the calculation, then the iteration ends and new patch clustering data is obtained.

[0042] S408: Check and modify the data's topology, attribute values, and other quality factors to obtain a qualified patch clustering result.

[0043] S409. Numerical calculation is performed on the number and area changes of the patches before and after clustering to verify the effectiveness of clustering. The calculation formula includes:

[0044]

[0045] QSCV(x)= aft Q ratio (x)- befQ ratio (x)

[0046]

[0047] ASCV(x)= aft A ratio (x)- bef A ratio (x)

[0048] Among them, x represents a certain type of CC code pattern, Q bef. (x), Q aft. (x) represents the number of CC code patches before and after clustering, Represents the total number of patches before and after clustering, bQ ratio (x), aQ ratio (x) represents the total ratio of the number of CC code patches before and after clustering; QC ratio (x) represents the rate of change of the number of CC code patches before and after clustering, QSCV(x) represents the change in the number of CC code patches of a certain type in the total ratio before and after clustering; A bef. (x), A aft. (x) represents the area of ​​a certain type of CC code pattern before and after clustering, Represents the total area of ​​the patch before and after clustering, bA ratio (x), aA ratio (x) represents the total area ratio of a certain type of CC code pattern before and after clustering; AC ratio (x) represents the area change rate of a certain CC code patch before and after clustering, and ASCV(x) represents the change value of the area ratio of a certain type of CC code patch before and after clustering.

[0049] Furthermore, the step S5 includes:

[0050] S501. Based on the patch clustering results obtained by the aforementioned algorithm and combined with semantic classification, a one-to-many mapping relationship is established with the original patches to clearly identify the patch set A involved in the aforementioned clustering behavior.

[0051] S502. Secondly, in the semantic classification system, the patch set A is decomposed into corresponding multiple patch sets B by combining spatial position differences and topological relationships.

[0052] S503. For each patch set B, extract the corresponding patch center point set C and use the combinations function in the Python itertools library to calculate the number of combinations, generating a random combination D with no duplication or omissions, with each group of three points, to cover all recognition possibilities. The combinations function is defined as:

[0053] C(n,k)=n! / (k!*(nk)!)

[0054] Where n represents the total number of elements in the set, k represents the number of elements in the combination, and "!" represents the factorial operation.

[0055] S504. Perform a series of calculations and analyses on each triple of combination D, including the shorter two line segments, the angle between the line segments, the length ratio, the number of intersecting patches, the angle difference in the direction of the long side, the area ratio, and the proportion of the intersecting line length to the triple, to improve the recognition accuracy and save the matching line elements.

[0056] S505. Repeat the above steps until all the three-dimensional lines that meet the conditions are found, and form a result line file, that is, the quantified original pattern straight line distribution pattern automatic recognition result.

[0057] Furthermore, the step S6 includes:

[0058] S601. Establish a verification mechanism for target pattern recognition results. Combining expert knowledge, experience, and spatial cognition, the accuracy of the automatic recognition results is verified by comparing them with manual recognition results by senior engineers in the field of cartography. The calculation method is defined as:

[0059] ML=ARQ∩MQR

[0060] MML=AQR-MQR

[0061] MSL=MRQ-ARQ

[0062] ARR=ML / (ML+MML)

[0063] DR=ML / (ML+MSL)

[0064] Where ARQ is the number of automatic recognition results, MRQ is the number of manual recognition results, ML is the number of automatic recognition results that are the same as manual recognition results, MML is the number of automatic recognition results that are different from manual recognition results, MSL is the number of manual recognition results that are missed by the automatic method, ARR is the automatic recognition accuracy, and DR is the automatic recognition recall.

[0065] S602. Combined with the numerical calculation results, comprehensively analyze the simplicity and effectiveness of the proposed method, and further apply the method to refine the knowledge of the linear distribution pattern of building groups.

[0066] A linear distribution pattern recognition system for complex building clusters, the system being applied to any of the above methods, comprising:

[0067] Spot data preprocessing module: determines the target area to be identified, collects building cluster spot data, performs spot editing, data storage, quality inspection and optimization preprocessing, and forms qualified building cluster original spot data;

[0068] Image patch semantic classification processing module: Based on the semantic classification structure of the surface cover data, the CC code of the original image patch of the qualified building group obtained above is converted into semantic classification information, and a three-level semantic hierarchy structure is constructed to obtain a building group image patch with a three-level semantic hierarchy;

[0069] The patch linear distribution pattern feature construction module uses the patch data output above to determine a new patch linear distribution pattern feature system, including combining area and geometric shape indicators to construct geometric morphological features; combining the directional differences between patches themselves and between them to construct directional features; combining spatial distribution density differences to construct position and spacing features, and thus obtaining a linear distribution feature system and measurement parameters;

[0070] Patch Constrained Merging Module: Based on the above-mentioned patch data and output results, the required level of classification is used to express patch semantics, and a patch-constrained clustering and merging method for different spatial scales is designed to obtain preliminary clustered and merged building group patch data;

[0071] The patch recognition scoping module: Based on the semantic, morphological, and density characteristics of complex building cluster patches, combined with the preliminary clustering data output above, the algorithm integrates the three-level semantic hierarchy and spatial scale information to design a reasonable patch recognition scoping algorithm that takes into account semantic consistency and spatial proximity. This algorithm constructs a set of spacing limits, iteratively determines the mergibility of patches, and dynamically controls the area change and boundary fidelity during the merging process, thereby maintaining the geometric accuracy and semantic relevance of the original patches, and ultimately forming a new patch set of building clusters after cluster optimization.

[0072] Image patch linear distribution pattern recognition module: Based on the above-mentioned clustering results and output feature system, a combined multi-constraint strategy is used to further design a target pattern automatic recognition algorithm that takes into account the complex characteristics of building clusters, and automatically recognize the linear distribution pattern of building clusters;

[0073] Image patch linear distribution pattern recognition result verification module: Using the above recognition results and manual recognition results, the automatic recognition accuracy rate (ARR) and recall rate (DR) are calculated through methods such as intersection and difference, and a comprehensive evaluation is performed to finally output a verification report and recognition performance indicators.

[0074] A computer device includes a memory, a processor, and a computer program stored in the memory and executable by the processor. When executing the program, the processor implements any of the above-mentioned methods for recognizing linear distribution patterns of complex building clusters.

[0075] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements any of the above-mentioned methods for recognizing linear distribution patterns of complex building clusters.

[0076] Compared with the prior art, the advantages of the present invention are:

[0077] A more realistic target pattern recognition perspective: By taking into account the needs of urban space management and governance, the semantic expression of the patches is refined to take into account the hierarchical differences and detailed information of the semantics of the patches of actual building groups, more reasonably reflect the semantic content of the patches, and show differentiated distribution characteristics.

[0078] A more superior complex patch clustering method: Taking into account the semantic level and spatial scale factors of building cluster patches, combined with multi-scale analysis ideas, a progressive constrained clustering strategy is designed to achieve the maintenance advantages of the patches in both numerical and distribution dimensions before and after clustering, and to characterize the overall and local distribution characteristics.

[0079] More reasonable pattern recognition integration algorithm: Based on constructing a set of target pattern feature measurement parameters and defining a reasonable range of pattern recognition, a reasonable target pattern automatic recognition strategy is designed, making the implementation of the integration algorithm simple and effective, and better adaptable to the complex distribution characteristics of the original pattern.

[0080] More reliable target pattern recognition results: Through the application of the overall solution, four typical recognition results can be obtained. The performance indicators of recall rate and accuracy are better overall. The result lines are rich and consistent with the original features. The mined information can be used to promote knowledge discovery and application services required by the field. BRIEF DESCRIPTION OF THE DRAWINGS

[0081] Figure 1 , the technical roadmap of the present invention's linear distribution pattern recognition method for complex building clusters;

[0082] Figure 2 , semantic hierarchical structure diagram of building clusters;

[0083] Figure 3 , logic diagram of the algorithm for demarcating reasonable range of pattern recognition;

[0084] Figure 4 , logic diagram of the algorithm for automatic recognition of the linear distribution pattern of patches. DETAILED DESCRIPTION

[0085] The specific implementation of the present invention is described below in conjunction with embodiments:

[0086] It should be noted that the structures, proportions, sizes, etc. shown in this specification are only used to match the contents disclosed in the specification for people familiar with this technology to understand and read, and are not used to limit the conditions under which the present invention can be implemented. Any structural modification, change in proportional relationship or adjustment of size should still fall within the scope of the technical content disclosed in the present invention without affecting the efficacy and purpose that can be achieved by the present invention.

[0087] At the same time, the terms such as "upper", "lower", "left", "right", "middle" and "one" quoted in this specification are only for the convenience of description and are not used to limit the scope of implementation of the present invention. Changes or adjustments to their relative relationships should be regarded as the scope of implementation of the present invention without substantially changing the technical content.

[0088] Example 1:

[0089] like Figure 1 As shown, the present invention provides a linear distribution pattern recognition method for complex building clusters, comprising the following steps:

[0090] S1: Constructing qualified original image patches of building groups:

[0091] Combine online map resources or import map resources to determine the target area to be identified, import or collect building group map data, implement a series of preprocessing tasks such as map editing, data storage, quality inspection and optimization, and form qualified original map data, laying a good foundation for subsequent data reprocessing.

[0092] S2: Build a semantic hierarchy that fits the reality:

[0093] With reference to the semantic classification structure of surface cover data, the semantic classification information of the CC code of the original map of qualified building groups obtained by S1 is translated to construct a three-level semantic hierarchy, and the three-level semantic features of the complex building group maps are formed to meet the application needs in the field of urban space management and governance.

[0094] S3: Define the linear distribution pattern characteristics of the spots:

[0095] Combining the area, geometric shape and other indicators of the original spots, the geometric morphological characteristics and corresponding measurement parameters are constructed; combining the original spots themselves and the directional differences between them, the directional characteristics and corresponding measurement parameters are constructed; combining the spatial distribution density differences of the original spots, the position and spacing characteristics and corresponding measurement parameters are constructed.

[0096] S4: Implement constrained clustering merging of original patches:

[0097] Based on the linear distribution pattern feature system of patches defined in S3, the required level of classification is used to express the semantics of patches, and a limited merging method of patches at different spatial scales (i.e., the "line construction and surface drawing" scheme) is implemented to effectively control the area after clustering while retaining the original characteristics of the patch boundaries, thereby achieving a better clustering effect.

[0098] S5: Delineate the reasonable identification range of the original image patch:

[0099] Combining the differences in semantic, morphological and density characteristics of complex building clusters, and integrating the method proposed in S4, an algorithm for reasonable range delineation of patch recognition that takes into account semantics and spatial scale is implemented, which better maintains the geometric accuracy and semantic relationship of the original patches and supports the clear expression of the overall and local characteristics of the patches.

[0100] S6: Identify the linear distribution pattern of the spots and quantify the results:

[0101] Based on the original patch clustering results formed by the implementation of the S5 algorithm, combined with the target pattern feature system constructed by S3, a combined multiple constraint strategy is used to further implement the target pattern automatic recognition algorithm that takes into account the complex characteristics of the building group patches, and finally form a quantitative recognition result.

[0102] S7: Verify and analyze the automatic recognition results of the target pattern:

[0103] Based on the expert knowledge and experience as well as their spatial cognition ability of the target pattern, combined with the verification scheme for the above recognition results, the distribution of the target pattern recognition results is quantitatively calculated from the two aspects of recognition accuracy and recall rate, and the simplicity and effectiveness of the proposed method are comprehensively analyzed to grasp the implementation results.

[0104] In this embodiment, the morphological features, directional features, and position and spacing features of the original pattern are constructed in S3:

[0105] S31. Constructing morphological features:

[0106] In order to take into account the different area values ​​of the building group, the maximum ratio of the original area mapped by the target pattern line is set to areas ratio for:

[0107]

[0108] Among them, maxArea and minArea represent the maximum and minimum area values ​​of the three original patches respectively.

[0109] At the same time, taking into account the large differences in the boundary construction methods and geometric dimensions of building clusters under the semantic hierarchy, no rigid restrictions or changes are made to the geometric structure and details of the patches. In other words, the shape characteristics of the original patches are expressed truthfully and no special processing is made to their shapes as much as possible, so as to truly and fully mine richer target recognition line results.

[0110] S32. Construct directional features

[0111] In order to simply and effectively constrain the differences between the directions of the spots themselves, this paper only designs the threshold direction of the difference in the angle of the long side of the minimum circumscribed rectangle of the original spot that intersects the three-dimensional line. diff. At the same time, in order to effectively control the arrangement angles between the spots and the connection form of the center points, the angle range of the angle θ between the two sides of the ternary line is set, and it is automatically determined by calculating the pairwise connection vectors based on the three center points. θ is defined as:

[0112]

[0113] in, and They are respectively represented by the vectors of the edges connected by two center points, and the two have a common connection point (i.e. and common starting point).

[0114] S33. Constructing position and spacing features

[0115] To avoid the large difference in spacing between the identified patches that is contrary to human cognition, the length ratio of the two sides of the triplet is set to sideLength ratio :

[0116]

[0117] in, and Represents vectors and The modulus length, where Greater than

[0118] At the same time, taking into account the differences in the density, boundary direction and topological relationship of the patches, in order to limit the unreasonable "insertion" of the original patches that were originally distributed separately by the algorithm, this paper adds the intersectLength, which is the ratio of the length of the three-dimensional line that is intersected (covered) by three patches to its own length. ratio :

[0119]

[0120] li represents the length of each line segment covered by the pattern. It represents the sum of the lengths of multiple covered line segments, and selfLength represents the length of the triplet itself.

[0121] In this embodiment, the “line drawing” method between the original image spots is executed in S4:

[0122] S41. The minimum upper image size of the cluster at the scale corresponding to the design target spatial scale is patchSize min , shape detail and the minimum distance between patches min , and set reasonable thresholds to improve the detail of data expression.

[0123] S42. When generalizing the boundary, the original corner points and edges are retained as much as possible, and the nearest node of the two edges to be connected is searched at the local merging position that meets the conditions, so as to constrain the merging of the two patches by enclosing the minimum merging area.

[0124] In this embodiment, the reasonable recognition range of the original image spot is defined in S5:

[0125] S51. At the third semantic level and target spatial scale, calculate the corresponding spacing limit between the patches, and construct and form a spacing value set step by step according to a certain small span between 0 and the limit.

[0126] S52 , start iteration, traverse the level-by-level spacing value set, obtain the patch set corresponding to each value in the set, traverse the set and count its semantic CC code and area value.

[0127] S53. Determine whether the code values ​​are the same. If they are the same, merge the corresponding two image patches in ascending order of area value, that is, execute the "line drawing and surface construction" solution in Section 3.

[0128] S54: According to the current spacing value, determine and find the position of the two spots to be merged, and then search for the four nearest nodes on both sides of the two spots.

[0129] S55. Then, a line is drawn on the node set, and a surface is constructed on the minimum area formed by the enclosed area.

[0130] S56. Finally, the small facet area is merged with the two adjacent original spots. After one round of iteration, the two old spots are eliminated and a new spot is generated.

[0131] S57, the new patch participates in a new round of calculation again, and the iteration continues until all values ​​of the step-by-step spacing value set gradually participate and complete the calculation, then the iteration ends and new patch clustering data is obtained.

[0132] S58. Check and modify the data's topology, attribute values, and other quality factors to obtain qualified patch clustering results.

[0133] S59. Numerical calculation of the number and area changes of the patches before and after clustering is performed to verify the effectiveness of clustering. The calculation formula includes:

[0134]

[0135] QSCV(x)= aft Q ratio (x)- bef Q ratio (x)

[0136]

[0137] ASCV(x)= aft A ratio (x)- bef A ratio (x)

[0138] Among them, x represents a certain type of CC code pattern, Q bef. (x), Q aft. (x) represents the number of CC code patches before and after clustering, Represents the total number of patches before and after clustering, bQ ratio (x), aQ ratio (x) represents the total ratio of the number of CC code patches before and after clustering; QC ratio (x) represents the rate of change of the number of CC code patches before and after clustering, QSCV(x) represents the change in the number of CC code patches of a certain type in the total ratio before and after clustering; A bef. (x), A aft. (x) represents the area of ​​a certain type of CC code pattern before and after clustering, Represents the total area of ​​the patch before and after clustering, bA ratio (x), aA ratio (x) represents the total area ratio of a certain type of CC code pattern before and after clustering; AC ratio (x) represents the area change rate of a certain CC code patch before and after clustering, and ASCV(x) represents the change value of the area ratio of a certain type of CC code patch before and after clustering.

[0139] In this embodiment, an automatic recognition algorithm for the original spot straight line distribution pattern based on multiple strategies is executed in S6:

[0140] S61. Based on the patch clustering results obtained by the aforementioned algorithm and combined with semantic classification, a one-to-many mapping relationship is established with the original patches to clearly identify the patch set A involved in the aforementioned clustering behavior.

[0141] S62. Secondly, under the semantic classification system, combining spatial position differences and topological relationships, the patch set A is decomposed into corresponding multiple patch sets B.

[0142] S63. For each patch set B, extract the corresponding patch center point set C and use the combinations function in the Python itertools library to calculate the number of combinations, generating a random combination D with no duplication or omissions, with each group of three points, to cover all recognition possibilities. The combinations function is defined as:

[0143] C(n,k)=n! / (k!*(nk)!)

[0144] Where n represents the total number of elements in the set, k represents the number of elements in the combination, and "!" represents the factorial operation.

[0145] S64. Perform a series of calculations and analyses on each triple of combination D, including the shorter two line segments, the angle between line segments, the length ratio, the number of intersecting patches, the angle difference in the direction of the long side, the area ratio, and the proportion of the intersecting line length to itself, to improve the recognition accuracy and save the matching line features.

[0146] S65. Repeat the above steps until all the three-dimensional lines that meet the conditions are found, and form a result line file, that is, the automatic recognition result of the quantified original pattern straight line distribution pattern.

[0147] In this embodiment, a verification mechanism is used in S7 to test and analyze the above identification results:

[0148] S71. Establish a verification mechanism for target pattern recognition results. Combining expert knowledge, experience, and spatial cognition, the accuracy of the automatic recognition results is verified by quantitative calculations and a comprehensive qualitative analysis is performed by comparing the results with those of manual recognition by senior engineers in the field of cartography. The calculation method is defined as:

[0149] ML=ARQ∩MRQ

[0150] MML=ARQ-MRQ

[0151] MSL=MRQ-ARQ

[0152] ARR=ML / (ML+MML)

[0153] DR=ML / (ML+MSL)

[0154] Where ARQ is the number of automatic recognition results, MRQ is the number of manual recognition results, ML is the number of automatic recognition results that are the same as manual recognition results, MML is the number of automatic recognition results that are different from manual recognition results, MSL is the number of manual recognition results that are missed by the automatic method, ARR is the automatic recognition accuracy, and DR is the automatic recognition recall.

[0155] S72. Combined with the numerical calculation results, comprehensively analyze the simplicity and effectiveness of the proposed method, and further apply the method to refine the knowledge of the linear distribution pattern of building groups.

[0156] Example 2:

[0157] In one embodiment, the present invention proposes a linear distribution pattern recognition system for complex building clusters. The system can be used to implement the aforementioned linear distribution pattern recognition method. Specifically, it includes the following modules:

[0158] The patch data preprocessing module is used to link online map resources or import map resources, determine the target area to be identified, import or collect building group patch data, implement a series of preprocessing tasks such as patch editing, data warehousing, quality inspection and optimization, and form qualified original patch data.

[0159] The patch semantic classification processing module is used to provide a patch semantic classification processing template, refer to the semantic classification structure of surface cover and other data, interpret the semantic classification information of the original patches of qualified building groups obtained using the patch data preprocessing module, construct a multi-level semantic hierarchy structure of the patches and form new semantic information.

[0160] The module for constructing the features of the linear distribution pattern of patches is used to construct the measurement parameters of geometric features by combining indicators such as area and geometric shape; to construct the measurement parameters of directional features by combining the directional differences between patches themselves and between them; and to construct the measurement parameters of position and spacing features by combining the differences in spatial distribution density.

[0161] The patch-constrained merging module is used to design and verify the "line construction and surface drawing" scheme for multiple patches at different spatial scales based on the set of feature measurement parameters obtained by the patch straight line distribution pattern feature construction module, combined with the defined patch semantic classification information, and provide cluster area experimental comparison and patch boundary analysis report.

[0162] The patch recognition range delineation module is used to combine the semantic, morphological and density differences of the patches of complex building groups in the target area. Based on the results of the patch constraint merging module, it simultaneously delineates a reasonable patch recognition range from both semantic and spatial scale dimensions, and clearly expresses the overall and local characteristics of the patches.

[0163] The patch linear distribution pattern recognition module is used to implement the target pattern quantitative recognition results for complex building cluster patches in the target area based on the results of the patch recognition range demarcation module, combined with the measurement parameter set of the patch linear distribution pattern feature construction module, and using a combined multiple constraint strategy.

[0164] The patch straight line distribution pattern recognition result verification module is used to combine expert knowledge and experience with the spatial cognition ability of the target pattern. Through the constructed recognition result verification scheme, it focuses on the recognition accuracy and recall rate, quantitatively calculates the distribution of the target pattern recognition results, and conducts visual comparative analysis.

[0165] Example 3:

[0166] In one embodiment of the present invention, a terminal device is provided, which includes a processor and a memory, wherein the memory is used to store a computer program, the computer program includes program instructions, and the processor is used to execute the program instructions stored in the computer storage medium. The processor may be a central processing unit (CPU), or may be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal, which is suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions to implement the corresponding method flow or corresponding function; the processor described in the embodiment of the present invention can be used for the operation of the linear distribution pattern recognition method for complex building group patterns, including the following steps:

[0167] S1. Determine the target area to be identified, collect building group patch data, carry out a series of pre-processing work such as patch editing, data storage, quality inspection and optimization, and form qualified original patch data, laying a good foundation for subsequent data reprocessing.

[0168] S2. Referring to the semantic classification structure of the surface cover data, the CC code of the original qualified building group obtained in S1 is converted into semantic classification information, a three-level semantic hierarchy is constructed, and the three-level semantic features inherent in the complex building group are formed.

[0169] S3. Define a new feature system for the linear distribution pattern of patches, which mainly includes: combining indicators such as area and geometric shape to construct geometric morphological features; combining the directional differences between patches themselves and between them to construct directional features; combining the differences in spatial distribution density to construct position and spacing features.

[0170] S4. Based on the linear distribution pattern feature system of patches defined in S3, the required level of classification is used to express the semantics of patches, and a limited merging method for patches of different spatial scales (i.e., the "line construction and surface drawing" scheme) is designed to effectively control the area after clustering while retaining the original characteristics of the patch boundaries.

[0171] S5. Combining the differences in semantic, morphological, and density characteristics of complex building clusters, and integrating the method proposed in S4, a reasonable range delineation algorithm for patch recognition is designed that takes into account semantics and spatial scale, so as to better maintain the geometric accuracy and semantic relationship of the original patches and support the clear expression of the overall and local characteristics of the patches.

[0172] S6. Based on the original pattern clustering results formed by the S5 algorithm, combined with the target pattern feature system constructed by S3, through a combined multiple constraint strategy, further design a target pattern automatic recognition algorithm that takes into account the complex characteristics of the building group patterns to form a quantitative recognition result.

[0173] S7. Based on the expert knowledge and experience as well as the spatial cognition ability of the target pattern, a verification scheme for the above recognition results is constructed, focusing on quantitatively calculating the distribution of the target pattern recognition results from the two aspects of recognition accuracy and recall rate, and comprehensively analyzing the simplicity and effectiveness of the proposed method.

[0174] Example 4:

[0175] In one embodiment of the present invention, the present invention further provides a storage medium, specifically a computer-readable storage medium (Memory), which is a memory device in a terminal device for storing programs and data. It can be understood that the computer-readable storage medium here can include both the built-in storage medium in the terminal device and, of course, the extended storage medium supported by the terminal device. The computer-readable storage medium provides a storage space, which stores the operating system of the terminal. In addition, one or more instructions suitable for being loaded and executed by the processor are also stored in the storage space. These instructions can be one or more computer programs (including program codes). It should be noted that the computer-readable storage medium here can be a high-speed RAM memory or a non-volatile memory, such as at least one disk memory.

[0176] The processor may load and execute one or more instructions stored in a computer-readable storage medium to implement the corresponding steps of the method for recognizing a linear distribution pattern for a complex building cluster in the above embodiment. The processor may load and execute the following steps:

[0177] S1. Determine the target area to be identified, collect building group patch data, carry out a series of pre-processing work such as patch editing, data storage, quality inspection and optimization, and form qualified original patch data, laying a good foundation for subsequent data reprocessing.

[0178] S2. Referring to the semantic classification structure of the surface cover data, the CC code of the original qualified building group obtained in S1 is converted into semantic classification information, a three-level semantic hierarchy is constructed, and the three-level semantic features inherent in the complex building group are formed.

[0179] S3. Define a new feature system for the linear distribution pattern of patches, which mainly includes: combining indicators such as area and geometric shape to construct geometric morphological features; combining the directional differences between patches themselves and between them to construct directional features; combining the differences in spatial distribution density to construct position and spacing features.

[0180] S4. Based on the linear distribution pattern feature system of patches defined in S3, the required level of classification is used to express the semantics of patches, and a limited merging method for patches of different spatial scales (i.e., the "line construction and surface drawing" scheme) is designed to effectively control the area after clustering while retaining the original characteristics of the patch boundaries.

[0181] S5. Combining the differences in semantic, morphological, and density characteristics of complex building clusters, and integrating the method proposed in S4, a reasonable range delineation algorithm for patch recognition is designed that takes into account semantics and spatial scale, so as to better maintain the geometric accuracy and semantic relationship of the original patches and support the clear expression of the overall and local characteristics of the patches.

[0182] S6. Based on the original pattern clustering results formed by the S5 algorithm, combined with the target pattern feature system constructed by S3, through a combined multiple constraint strategy, further design a target pattern automatic recognition algorithm that takes into account the complex characteristics of the building group patterns to form a quantitative recognition result.

[0183] S7. Based on the expert knowledge and experience as well as the spatial cognition ability of the target pattern, a verification scheme for the above recognition results is constructed, focusing on quantitatively calculating the distribution of the target pattern recognition results from the two aspects of recognition accuracy and recall rate, and comprehensively analyzing the simplicity and effectiveness of the proposed method.

[0184] It will be understood by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0185] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0186] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0187] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 The steps for the function specified in one or more boxes.

[0188] The preferred embodiments of the present invention are described in detail above, but the present invention is not limited to the above embodiments. Various changes can be made within the knowledge of ordinary technicians in this field without departing from the scope of the present invention.

[0189] Many other changes and modifications can be made without departing from the spirit and scope of the present invention. It should be understood that the present invention is not limited to the specific embodiments, and the scope of the present invention is defined by the appended claims.

Claims

1. A linear distribution pattern recognition method for complex building clusters, characterized by: The method comprises: S1: Based on the semantic classification structure of the surface cover data, the CC codes of the original qualified building cluster patches are converted into semantic classification information, and a three-level semantic hierarchy is constructed to obtain building cluster patches with three-level semantic hierarchy. S2: Using the patch data output in step S1, a new feature system of the patch linear distribution pattern is determined, including combining area and geometric shape indicators to construct geometric morphological features; combining the differences in the directions of the patches themselves and between them to construct directional features; combining the differences in spatial distribution density to construct position and spacing features, and combining to form a linear distribution feature system and measurement parameters; S3: Based on the patch data output from step S1 and the output results from step S2, the required level of classification is used to express the patch semantics, and a progressive constrained clustering method for patches of different spatial scales is designed to obtain the patch data of the building group after preliminary clustering and merging. S4: Based on the semantic, morphological, and density characteristics of complex building clusters, combined with the preliminary clustering data output by S3, the three-level semantic hierarchy and spatial scale information are integrated to design a reasonable range delineation algorithm for patch identification that takes into account semantic consistency and spatial proximity. This algorithm constructs a set of spacing limits, iteratively determines the mergibility between patches, and dynamically controls the area change and boundary fidelity during the merging process, thereby maintaining the geometric accuracy and semantic relevance of the original patches, and ultimately forming a new set of building cluster patches after cluster optimization; S5: Based on the clustering results output from step S4 and the feature system output from step S2, a combined multi-constraint strategy is used to further design an automatic target pattern recognition algorithm that takes into account the complex characteristics of the building cluster pattern, and obtain the automatically recognized linear distribution pattern result of the building cluster; S6: Using the recognition results of step S5 and the manual recognition results, the automatic recognition accuracy rate (ARR) and the recall rate (DR) are calculated through intersection and difference sets, and a comprehensive evaluation is performed to finally output a verification report and recognition performance indicators.

2. The method for recognizing linear distribution patterns of complex building clusters according to claim 1, characterized in that: Determine the target area to be identified, collect the building group patch data, perform patch editing, data storage, quality inspection and optimization preprocessing, and form the qualified building group original patch data.

3. The linear distribution pattern recognition method for complex building clusters according to claim 1, characterized in that: The step S2 includes: S201, construct morphological features: Set the maximum ratio of the original image area mapped by the target mode line to areas ratio for: Among them, maxArea and minArea represent the maximum and minimum area values ​​of the three original patches respectively; Considering the large differences in the boundary construction methods and geometric dimensions of building clusters under the semantic hierarchy, no rigid restrictions or changes are made to the geometric structure and details of the patches. In other words, the original patch shape characteristics are faithfully expressed without any special processing of its shape, so as to truly and fully mine richer target recognition line results. S202, build direction features: Design the threshold direction of the angle difference between the minimum circumscribed rectangle of the original patch and the trigonometric line. diff. At the same time, the angle range of the angle θ between the two sides of the ternary line is set, and it is automatically determined by calculating the pairwise connecting vectors based on the three center points. θ is defined as: in, and are the vector representations of the edges connected by two center points, and they share a connection point, that is, and common starting point; S203, build position and spacing features: Set the length ratio of the two sides of the triplet to sideLength ratio : in, and Represents vectors and The modulus length, where Greater than Add intersectLength, the ratio of the length of the three-dimensional line intersected by three patches to its own length ratio : li represents the length of each line segment covered by the pattern. It represents the sum of the lengths of multiple covered line segments, and selfLength represents the length of the triplet itself.

4. The method for recognizing linear distribution patterns of complex building clusters according to claim 1, wherein: The step S3 comprises: S301. Design the minimum upper image size of the cluster at the corresponding scale of the target spatial scale (patchSize) min , shape detail and the minimum distance between patches min , and set reasonable thresholds to improve the detail of data expression; S302: When generalizing the boundary, the original corner points and edges are retained as much as possible, and the nearest node of the two edges to be connected is searched at the local merging position that meets the conditions, and the two patches are constrainedly merged by enclosing the minimum merging area.

5. The method for recognizing linear distribution patterns of complex building clusters according to claim 1, wherein: The step S4 comprises: S401, at the third semantic level and the target spatial scale, calculating the corresponding spacing limit between patches, and constructing and forming a spacing value set step by step according to a certain small span between 0 and the limit; S402, start iteration, traverse the level-by-level spacing value set, obtain the patch set corresponding to each value in the set, traverse the set and count its semantic CC code and area value; S403, determining whether the code values ​​are the same, if they are the same, merging the corresponding two patches in ascending order of area value, i.e. executing the line drawing and surface construction scheme of section S3; S404: Determine and find the position of the two spots to be merged based on the current spacing value, and then search for the four closest nodes on both sides of the two spots; S405: Draw lines on the node set and construct faces on the minimum area formed by the enclosed area; S406, finally merging the facet small area with the two adjacent original spots, and after one round of iteration, the two old spots are eliminated and a new spot is generated; S407, the new patch participates in a new round of calculation again, and the iteration continues until all values ​​of the step-by-step spacing value set gradually participate and complete the calculation, then the iteration ends and new patch clustering data is obtained; S408: Check and modify the topology and attribute value quality of the data to obtain a qualified patch clustering result; S409. Numerical calculation is performed on the changes in the number and area of ​​the patches before and after clustering to verify the effectiveness of clustering. The calculation formula includes: QSCV(x)=aftQ ratio (x)-befQ ratio (x) ASCV(x)=aftA ratio (x)-befA ratio (x) Among them, x represents a certain type of CC code pattern, Q bef. (x), Q aft. (x) represents the number of CC code patches before and after clustering, Represents the total number of patches before and after clustering, bQ ratio (x), aQ ratio (x) represents the total ratio of the number of CC code patches before and after clustering; QC ratio (x) represents the rate of change of the number of CC code patches before and after clustering, QSCV(x) represents the change in the number of CC code patches of a certain type in the total ratio before and after clustering; A bef. (x), A aft. (x) represents the area of ​​a certain type of CC code pattern before and after clustering, Represents the total area of ​​the patch before and after clustering, bA ratio (x), aA ratio (x) represents the total area ratio of a certain type of CC code pattern before and after clustering; AC ratio (x) represents the area change rate of a certain CC code patch before and after clustering, and ASCV(x) represents the change value of the area ratio of a certain type of CC code patch before and after clustering.

6. The method for recognizing linear distribution patterns of complex building clusters according to claim 1, wherein: The step S5 comprises: S501. Based on the obtained patch clustering results and in combination with semantic classification, a one-to-many mapping relationship is established with the original patches to clearly identify the patch set A involved in the aforementioned clustering behavior. S502, decomposing the patch set A into corresponding multiple patch sets B based on spatial position differences and topological relationships under a semantic classification system; S503. For each patch set B, extract the corresponding patch center point set C and use the combinations function in the Python itertools library to calculate the number of combinations to generate a random combination D with no duplication or omissions, with each group of three points, to cover all recognition possibilities. The combinations function is defined as: C(n,k)=n! / (k!*(nk)!) Where n represents the total number of elements in the set, k represents the number of elements in the combination, and "!" represents the factorial operation; S504. Perform a series of calculations and analyses on each triple of combination D, including the shorter two line segments, the angle between the line segments, the length ratio, the number of intersecting patches, the angle difference in the direction of the long side, the area ratio, and the proportion of the length of the intersecting line to the triple, to improve the recognition accuracy and save the matching line features. S505. Repeat the above steps until all the three-dimensional lines that meet the conditions are found, and form a result line file, that is, the quantified original pattern straight line distribution pattern automatic recognition result.

7. The method for recognizing linear distribution patterns of complex building clusters according to claim 1, characterized in that: The step S6 comprises: S601. Establish a verification mechanism for target pattern recognition results. Combining expert knowledge, experience, and spatial cognition, the accuracy of the automatic recognition results is verified by computational verification by comparing them with manual recognition results by senior engineers in the field of cartography. The computational method is defined as: ML=ARQ∩MRQ MML=ARQ-MRQ MSL=MQR-ARQ ARR=ML / (ML+MML) DR=ML / (ML+MSL) Where ARQ represents the number of automatic recognition results, MRQ represents the number of manual recognition results, ML represents the number of automatic recognition results that are the same as manual recognition results, MML represents the number of automatic recognition results that are different from manual recognition results, MSL represents the number of manual recognition results that are missed by the automatic method, ARR represents the automatic recognition accuracy, and DR represents the automatic recognition recall rate. S602. Combined with the numerical calculation results, comprehensively analyze the simplicity and effectiveness of the proposed method, and further apply the method to refine the knowledge of the linear distribution pattern of building groups.

8. A linear distribution pattern recognition system for complex building clusters, characterized by: The system is applied to the method according to any one of claims 1 to 7, and the system includes: Spot data preprocessing module: determines the target area to be identified, collects building cluster spot data, performs spot editing, data storage, quality inspection and optimization preprocessing, and forms qualified building cluster original spot data; Image patch semantic classification processing module: Based on the semantic classification structure of the surface cover data, the CC code of the original image patch of the qualified building group obtained above is converted into semantic classification information, and a three-level semantic hierarchy structure is constructed to obtain a building group image patch with a three-level semantic hierarchy; The patch linear distribution pattern feature construction module uses the patch data output above to determine a new patch linear distribution pattern feature system, including combining area and geometric shape indicators to construct geometric morphological features; combining the directional differences between patches themselves and between them to construct directional features; combining spatial distribution density differences to construct position and spacing features, and thus obtaining a linear distribution feature system and measurement parameters; Patch Constrained Merging Module: Based on the above-mentioned patch data and output results, the required level of classification is used to express patch semantics, and a patch-constrained clustering and merging method for different spatial scales is designed to obtain preliminary clustered and merged building group patch data; The patch recognition scoping module: Based on the semantic, morphological, and density characteristics of complex building cluster patches, combined with the preliminary clustering data output above, the algorithm integrates the three-level semantic hierarchy and spatial scale information to design a reasonable patch recognition scoping algorithm that takes into account semantic consistency and spatial proximity. This algorithm constructs a set of spacing limits, iteratively determines the mergibility of patches, and dynamically controls the area change and boundary fidelity during the merging process, thereby maintaining the geometric accuracy and semantic relevance of the original patches, and ultimately forming a new patch set of building clusters after cluster optimization. Image patch linear distribution pattern recognition module: Based on the above-mentioned clustering results and output feature system, a combined multi-constraint strategy is used to further design a target pattern automatic recognition algorithm that takes into account the complex characteristics of building clusters, and automatically recognize the linear distribution pattern of building clusters; Image patch linear distribution pattern recognition result verification module: Using the above recognition results and manual recognition results, the automatic recognition accuracy rate (ARR) and recall rate (DR) are calculated through the intersection and difference method, and a comprehensive evaluation is performed to finally output a verification report and recognition performance indicators.

9. A computer device, characterized in that: The invention comprises a memory, a processor and a computer program stored in the memory and executable by the processor. When the processor executes the program, the method for recognizing a linear distribution pattern of a complex building group as described in any one of claims 1 to 8 is implemented.

10. A computer-readable storage medium, characterized in that A computer program is stored therein, which, when executed by a processor, implements the linear distribution pattern recognition method for complex building clusters as described in any one of claims 1 to 8.

Images