Building complex combination straight line pattern recognition method and device combining feature classification and node matching

Abstract

The application belongs to the technical field of building group pattern recognition, and discloses a building group complex combination straight line pattern recognition method and device combining feature classification and node matching, which comprises the following steps: step 1, classifying building groups according to area and main direction; step 2, respectively constructing building group space proximity relations and identifying simple straight line patterns contained in each type of building group; step 3, preliminarily identifying building group complex combination straight line patterns by identifying simple straight line pattern groups based on the identified simple straight line patterns; and step 4, using a node matching strategy to identify combination structures and realizing accurate identification of building group complex combination straight line patterns. The application can identify complex combination straight line patterns contained in building groups with complex distribution and different spatial granularities.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of building group pattern recognition, and particularly relates to a building group complex combination straight line pattern recognition method and device combining feature classification and node matching. BACKGROUND

[0002] As an important geographical element, the spatial distribution characteristics of building groups have always been a hotspot and difficulty in the field of map generalization. On small-scale maps, building groups consider spatial and semantic characteristics, and focus on the division of natural attributes and network elements (roads, water systems, etc.); for the distribution of building groups at medium and large scales, current researches mainly focus on two aspects: ① clustering based on spatial proximity; ② pattern recognition of spatial distribution characteristics based on visual perception theory with multi-parameter constraints. Existing researches are mostly aimed at the spatial distribution patterns of building groups with uniform size and regular distribution, while the researches on the spatial structure characteristics of building groups with complex distribution and irregularity are limited to spatial distribution patterns with specific organizational structures, such as recognizing typical letter-type distribution patterns of building groups by constructing a library of letter-type templates and using a matching strategy.

[0003] However, in medium and large scale maps with low generalization degree, in addition to building groups with specific structural distribution, there are also building groups with regular distribution but without specific structure, which are organized in a straight line pattern as a whole. Straight line pattern recognition can effectively explore the distribution characteristics of building groups and provide support for subsequent map generalization. For building groups with complex distribution and different size at medium and large scales, due to the obvious local heterogeneity and complex spatial relationships, existing straight line pattern recognition methods cannot meet the complex straight line pattern detection requirements. SUMMARY

[0004] The application proposes a building group complex combination straight line pattern recognition method and device combining feature classification and node matching to solve the problem that existing straight line pattern recognition methods cannot meet the complex straight line pattern detection requirements.

[0005] To achieve the above purpose, the application adopts the following technical solutions:

[0006] The application proposes a building group complex combination straight line pattern recognition method combining feature classification and node matching, which comprises the following steps:

[0007] Step 1: classifying building groups according to area and main direction;

[0008] Step 2: respectively constructing spatial proximity relationships of building groups and recognizing simple straight line patterns contained in each type of building groups;

[0009] Step 3, based on the identified simple straight line patterns, the complex combined straight line patterns of the building group are preliminarily identified by identifying simple straight line pattern groups;

[0010] Step 4, the combined structure is identified by using the node matching strategy, and the accurate identification of the complex combined straight line patterns of the building group is realized.

[0011] Further, the step 1 comprises:

[0012] The k-means++ clustering algorithm is used to classify the building group, and the area and main direction features of the building are mapped to the Euclidean distance space, so as to measure the similarity between the buildings by the Euclidean distance.

[0013] Further, the step 2 comprises:

[0014] The weak adjacent search strategy is used to construct the spatial adjacent relationship of the building group, and the search range is set to the buildings within the second-order adjacent;

[0015] The structured parameters constructed are combined to establish the extraction rules as shown in formula 4.2:

[0016]

[0017] In the formula, R lenth , R width , S building , LD orient , OD orient respectively represent the adjacent long axis length ratio, the adjacent short axis length ratio, the interval distance ratio between the building groups in the pattern, the local direction difference and the overall direction difference, δ lenth , δ width , δ space , δ orient are the threshold values of the long axis length ratio, the short axis length ratio, the interval distance ratio and the pattern direction difference.

[0018] Further, the step 3 comprises:

[0019] Step 3.1, the simple straight line patterns identified from the building groups of each type are merged into a set PatternSet; the chain table set in which the simple straight line patterns form the complex combined straight line patterns is ListSet, and step 3.2 is executed.

[0020] Step 3.2, any two patterns P A and P B are taken from the set PatternSet, if P A and P B do not contain the same elements and satisfy the condition of formula 4.4, P A and P B can constitute a complex combined straight line pattern, and PA With P B Add to the linked list array and execute step 3.3;

[0021]

[0022] Where, PO orient 、S pattern , Dis, and Deg are mode direction difference, mode interval distance ratio, mode distance, and mode alignment, respectively. orient , δ space , δ dis and δ deg are the thresholds for pattern direction difference, separation distance ratio, pattern distance, and pattern alignment, respectively;

[0023] Step 3.3, traverse any linked list List in the collection ListSet i , if List i If the intersection with array is not empty, insert the non-intersecting elements in array into List i In; if List i If the intersection with array is empty, add array to ListSet; return to step 3.2 until all pattern combinations are traversed.

[0024] Furthermore, the step 4 includes:

[0025] Step 4.1, take a simple straight line pattern group from the pattern list set to create a list of objects to be matched i {P0,P1,…,P n}, select the matching object P with the least number of nodes i As the starting reference object, P i The progressive matching chain of adjacent objects to be matched is constructed as the starting point {P i ,P i-1 ,…,P0} and {P i ,P i+1 ,…,P n};

[0026] Step 4.2: Perform progressive node matching according to the established matching chain. The node matching algorithm steps for the reference object and the object to be matched are as follows:

[0027] Step 4.2.1, taking the node representing one end of the non-suspended building in the reference object as a starting point, if there is no node representing the non-suspended building, taking the first end point in the reference object as a starting point, recording the node in the to-be-matched object closest to the starting point as a matching pair, and sequentially recording the nodes in the reference object and the to-be-matched object directly adjacent to each other and directly adjacent to the corresponding matching pairs in the last group as matching pairs, and the matching ends when the node representing the suspended building in the reference object is taken for the first time, and if there is no node representing the suspended building, all nodes are taken, and all matching pairs are added to Set1; Step 4.2.2 is performed;

[0028] Step 4.2.2, taking the node representing one end of the non-suspended building in the reference object as a starting point, if there is no node representing the non-suspended building, taking the first end point in the reference object as a starting point, recording the node in the to-be-matched object closest to the starting point as a matching pair, and sequentially recording the nodes in the reference object and the to-be-matched object directly adjacent to each other and directly adjacent to the corresponding matching pairs in the last group as matching pairs, and the matching ends when the node representing the suspended building in the reference object is taken for the first time, and if there is no node representing the suspended building, all nodes are taken, and all matching pairs are added to Set1; Step 4.2.2 is performed;

[0029] Step 4.2.3, comparing the matching degrees of the matching results of Step 4.2.1 and Step 4.2.2, taking the result with a higher matching degree as the final matching result, and if the matching degrees are equal, taking Set1 as the final matching result;

[0030] Step 4.3, the matching pair set is combined in order, and then the combined structure accurately recognized as a pattern unit is obtained, and the building not matched is removed from the simple straight line pattern, and then the simple straight line pattern group constituting the complex combined straight line pattern is obtained; if the influence domain of the complex combined straight line pattern intersects with other buildings, the pattern is deleted.

[0031] Another aspect of the present application provides a building complex combined straight line pattern recognition device combining feature classification and node matching, comprising:

[0032] The classification module is used for classifying the building complex according to the area and the main direction;

[0033] The first recognition module is used for constructing the spatial adjacent relationship of the building complex and recognizing the simple straight line pattern contained in each type of building complex;

[0034] The second recognition module is used for preliminarily recognizing the building complex combined straight line pattern by recognizing the simple straight line pattern group based on the recognized simple straight line pattern;

[0035] The third recognition module is used for recognizing the combined structure by using the node matching strategy, and realizing the accurate recognition of the building complex combined straight line pattern.

[0036] Further, the classification module is specifically used for:

[0037] The k-means++ clustering algorithm is used to classify the building groups, the area and main direction features of the buildings are mapped to the Euclidean distance space, and the Euclidean distance is used to measure the similarity between the buildings.

[0038] Further, the first identification module is specifically used for:

[0039] A weak proximity search strategy is used to construct the spatial proximity relationship of the building groups, and the search range is set to the buildings within the second-order proximity;

[0040] Combined with the constructed structured parameters, the extraction rules such as formula 4.2 are established:

[0041]

[0042] In the formula, R lenth , R width , S building , LD orient , OD orient respectively represent the adjacent long axis length ratio, the adjacent short axis length ratio, the interval distance ratio between the building groups in the mode, the local direction difference and the overall direction difference, δ lenth , δ width , δ space , δ orient are the threshold values of the long axis length ratio, the short axis length ratio, the interval distance ratio and the mode direction difference.

[0043] Further, the second identification module is specifically used for:

[0044] Step 3.1, combine the simple straight line modes identified from the building groups of each category into a set PatternSet; set the linked list set ListSet formed by the simple straight line modes to form a complex combination straight line mode, and execute step 3.2;

[0045] Step 3.2, take any two modes P A and P B from the set PatternSet, if P A and P B do not contain the same elements and satisfy the condition of formula 4.4, P A and P B can constitute a complex combination straight line mode, P A and P B are added to the linked list array array, and step 3.3 is executed;

[0046]

[0047] In the formula, PO orient , S pattern, Dis, and Deg are mode direction difference, mode interval distance ratio, mode distance, and mode alignment, respectively. orient , δ space , δ dis and δ deg are the thresholds for pattern direction difference, separation distance ratio, pattern distance, and pattern alignment, respectively;

[0048] Step 3.3, traverse any linked list List in the collection ListSet i , if List i If the intersection with array is not empty, insert the non-intersecting elements in array into List i In; if List i If the intersection with array is empty, add array to ListSet; return to step 3.2 until all pattern combinations are traversed.

[0049] Furthermore, the third identification module is specifically configured to:

[0050] Step 4.1, take a simple straight line pattern group from the pattern list set to create a list of objects to be matched i {P0,P1,…,P n}, select the matching object P with the least number of nodes i As the starting reference object, P i The progressive matching chain of adjacent objects to be matched is constructed as the starting point {P i ,P i-1 ,…,P0} and {P i ,P i+1 ,…,P n};

[0051] Step 4.2: Perform progressive node matching according to the established matching chain. The node matching algorithm steps for the reference object and the object to be matched are as follows:

[0052] Step 4.2.1: Take the node representing one end of the non-hanging building in the reference object as the starting point. If there is no node representing a non-hanging building, take the first endpoint of the reference object as the starting point and record it and the node closest to the object to be matched as a matching pair. Then, sequentially take the nodes in the reference object that are directly adjacent to the object to be matched and directly adjacent to the previous set of corresponding matching pairs as matching pairs. The matching ends when the node representing the hanging building in the reference object is first taken. If there is no node representing a hanging building, take all nodes and add all matching pairs to Set1; then execute step 4.2.2;

[0053] Step 4.2.2, taking the node representing one end of the non-suspended building in the reference object as the starting point, if there is no node representing the non-suspended building, taking the first end point in the reference object as the starting point, recording the distance of the second closest node in the to-be-matched object as a matching pair, as above, obtaining other matching pairs, and adding them to the set Set2; executing step 4.2.3;

[0054] Step 4.2.3, comparing the matching degrees of the matching results of step 4.2.1 and step 4.2.2, taking the result with a higher matching degree as the final matching result, and if the matching degrees are equal, taking Set1 as the final matching result;

[0055] Step 4.3, merging the matching pair set in order, then obtaining the combined structure accurately recognized as a pattern unit, and eliminating the un-matched building from the simple straight line pattern, then obtaining the simple straight line pattern group which can constitute a complex combined straight line pattern; if the influence domain of the complex combined straight line pattern intersects with other buildings, the pattern is deleted.

[0056] Compared with the prior art, the present application has the beneficial effects of:

[0057] (1) The straight line pattern recognition features and identification methods of the prior research are not applicable to the complex combined straight line pattern, and the present application can identify the complex combined straight line pattern contained in the building group with complex distribution and different spatial granularity.

[0058] (2) Invite one cartography expert to perform artificial pattern recognition on the experimental data of the present application, then compare the expert recognition result with the recognition result of the method, evaluate the recognition results by precision and recall, and calculate that the precision and recall are both 85%, and the result shows that the recognition result of the present application is reliable.

[0059] (3) For the building group synthesis with complex distribution and different granularity in a large-scale map, the complex combined straight line pattern can fully exploit the spatial distribution features of the atypical combined structure of the building group, provide an important basis for the selection of synthesis operators and algorithms in the building group synthesis process, and provide effective support for high-quality map synthesis. BRIEF DESCRIPTION OF DRAWINGS

[0060] Figure 1 is a straight line pattern of different types of building groups;

[0061] Figure 2 is a general flowchart of a building group complex combined straight line pattern recognition method combining feature classification and node matching according to an embodiment of the present application;

[0062] Figure 3 is an example diagram of the influence of adjacent relationship on straight line pattern recognition according to an embodiment of the present application;

[0063] Figure 4 Figure is a preliminary identification example of complex combined straight line pattern for the embodiment of the present application;

[0064] Figure 5 Figure is a precise identification example of complex combined straight line pattern for the embodiment of the present application;

[0065] Figure 6 Figure is an experimental result of complex combined straight line pattern identification for the embodiment of the present application;

[0066] Figure 7 Figure is an O-shaped letter distribution example for the embodiment of the present application;

[0067] Figure 8 Figure is an expert identification result for the embodiment of the present application;

[0068] Figure 9 Figure is a structural schematic diagram of a building complex combined straight line pattern identification device combining feature classification and node matching for the embodiment of the present application. DETAILED DESCRIPTION

[0069] The present application will be further explained in conjunction with the accompanying drawings and specific embodiments:

[0070] First, the complex combined straight line pattern of building complex is introduced:

[0071] Distribution pattern is one of the typical spatial distribution characteristics, which needs to be paid special attention and processed by cartographers, and this is also the original motivation of the research on map distribution pattern. The composition of pattern mainly depends on subjective factors such as cognitive psychology, for example, the tendency of perception to combine, such as proximity combination, similarity combination, and combination of good patterns (such as continuity, symmetry, and convergence, etc.). If a distribution repeatedly and continuously appears, it constitutes a certain pattern, which should be considered as a whole in map generalization, rather than being processed separately. How to make the computer "understand" the spatial cognitive law of human beings on building complex pattern is the key to automatic identification of building complex pattern.

[0072] Straight line pattern is a common and typical building complex distribution pattern. A large number of cognitive studies show that the topological perception relationship of straight line pattern has a hierarchy of "building (pattern unit) - straight line pattern", such as Figure 1 The straight line pattern of building complex in the dashed box D is formed by a plurality of buildings with similar characteristics (shape, direction, and size, etc.) arranged along a straight line. However, this cognitive feature is limited to the homogeneous building complex straight line pattern with single building as the pattern unit, and is not completely applicable to the case that the buildings in the dashed boxes A, B and C have obvious differences in size, direction and other characteristics, but the whole can still be identified as a straight line pattern. According to the Gestalt cognitive criteria and the theory of visual perception topology, the observer first pays attention to the overall spatial distribution of building complex when watching the map, such as Figure 1The building group in the dashed box A is regarded as a straight line pattern. Then the local structure and features are noticed, and it is found that the straight line pattern is formed by a number of building combination structures with similar features and structures, such as the combination of buildings 11, 13, 18 and 21. Further, the size and direction of the buildings in the building combination structure are obviously perceived.

[0073] For the straight line pattern formed by a number of building combination structures with similar features and structures, the present application introduces the concept of complex combination straight line pattern. The complex combination straight line pattern has the characteristics of multi-level cognition. At the pattern whole level, the building group is regularly distributed in a straight line pattern as a whole. At the pattern unit level, the pattern is composed of a group of building combination structures (i.e. pattern units) with similar structures and features, and the spatial distribution between the pattern units satisfies a specific arrangement - straight line arrangement. At the pattern element level, the pattern unit is composed of adjacent buildings with different features. For example, Figure 1 In the dashed box C, buildings 25, 30 and 32 form a combination structure, and a number of similar combination structures are arranged in a straight line to form a combination straight line distribution pattern.

[0074] From the perspective of the spatial correspondence between adjacent pattern units, the simple straight line pattern only contains 1:1 spatial correspondence, the extended combination straight line pattern contains 1:1, 1:n and m:n spatial correspondence combinations, and the complex combination straight line pattern only contains n:n type spatial correspondence. Therefore, the process of recognizing the complex combination straight line pattern can be regarded as the process of establishing the n:1 spatial correspondence between the buildings and the pattern units to which they belong and the 1:1 spatial correspondence between the pattern units.

[0075] From the cognitive features of the complex combination straight line pattern, for any building in any pattern unit of the complex combination straight line pattern, there must be a building with similar features and arranged in a straight line in other pattern units, such as Figure 1 In the dashed box C, buildings 32, 41, 50 and 63 can form a simple straight line pattern. Therefore, the complex combination straight line pattern can also be regarded as a group of simple straight line patterns that are spatially adjacent and have similar distribution features (direction, interval) arranged in the vertical direction of the pattern. The adjacent buildings between different simple straight line patterns are sequentially combined to form a pattern unit, and a number of pattern units form a complex combination straight line pattern.

[0076] Spatial proximity and geometric similarity are two important factors for human cognition of the spatial distribution pattern of building groups. Existing pattern recognition methods are mostly based on the construction of structured parameters of the pattern and the design of pattern recognition strategies. The mainstream idea is to first cluster the building groups based on spatial proximity, and then use a "top-down" repeated pruning connection edge strategy or a "bottom-up" step-by-step proximity search strategy. Since the adjacent buildings forming complex combination straight line patterns usually have heterogeneity, the edges connecting these buildings in the spatial proximity expression model all belong to the heterogeneous edges that need to be deleted, or the heterogeneity between buildings will block the proximity search. In addition, there is a strategy of matching the typical letter type distribution pattern of building groups using a template library, while the building distribution in the complex combination straight line pattern proposed by the present invention is complex and does not have specific structural features. In summary, the existing pattern recognition strategies are not completely applicable to complex combination straight line patterns. Therefore, in combination with the cognitive characteristics of complex combination straight line patterns, the present invention proposes a building group complex combination straight line pattern recognition method combining feature classification and node matching, the overall process of which is shown in Figure 2 as follows:

[0077] Step 1: Classify the building groups according to area and main direction;

[0078] Step 2: Respectively construct the spatial proximity relationship of the building groups and identify the simple straight line patterns contained in each type of building group;

[0079] Step 3: Based on the identified simple straight line patterns, preliminarily identify the complex combination straight line patterns of building groups by identifying simple straight line pattern groups;

[0080] Step 4: Use the node matching strategy to identify the combination structure and achieve accurate recognition of the complex combination straight line patterns of building groups.

[0081] Further, the step 1 includes:

[0082] Considering that the adjacent buildings constituting the complex combination straight line pattern have obvious heterogeneity, the strategy of classification is used to group the buildings with high homogeneity of area and main direction, which are more likely to form simple straight line patterns, so as to respectively identify and extract the simple straight line patterns that can constitute complex combination straight line patterns. The k-means++ clustering algorithm, as a classic and effective clustering method based on partition, is widely used, so the present invention uses the k-means++ clustering algorithm to classify the building groups. The similarity measurement index used by the clustering algorithm is an important basis for clustering. The present invention maps the area and main direction features of the buildings to the Euclidean distance space to measure the similarity between the buildings. Since the different dimensions of attribute feature variables have a great influence on distance measurement, it is necessary to standardize the attribute feature data to unify the dimensions. Here, the standard deviation standardization (Z-score standardization) processing method is used, and the i-th measurement value of the attribute feature f isfi , the normalized metric value Z fi is calculated as follows:

[0083]

[0084] where μ f and σ f are the mean and standard deviation of the attribute feature f metric value set, respectively.

[0085] Further, the step 2 comprises:

[0086] Humans tend to combine spatially adjacent, continuous and similar buildings together, and mainly consider the adjacency, similarity and continuity when identifying the simple linear pattern contained in the building group. Considering the linearity, compactness and similarity of the simple linear pattern, combined with the Gestalt cognitive criteria of human vision, the structured parameters in the cognitive sense are established from the size, direction and distance of the building as the constraint conditions for the extraction of the simple linear pattern. Here, the length ratio, interval distance ratio, local direction difference and overall direction difference are selected.

[0087] On the basis of the building group classification, the "bottom-up" adjacent search strategy is used to gradually identify the linear pattern implied in each type of building group. The construction of the spatial adjacent relationship of the building group is an important step for pattern recognition, and the constrained Delaunay triangulation is used to construct the spatial adjacent relationship of each type of building group. Considering that the strategy of searching by using direct adjacent relationship cannot effectively identify the linear pattern with similar and staggered arrangement of features in the building group, such as Figure 3 , only the simple linear pattern connected by the dashed line can be identified, therefore, the weak adjacent search strategy is adopted in the present application, and the search range is set to the buildings within the second-order adjacent. The rules followed by the pattern recognition are the key to the identification, and the extraction rules as shown in formula 4.2 are established combined with the structured parameters constructed. Among them, the length ratio of the long axis and the short axis is used to ensure the similarity of the size of the building, the interval distance ratio is used to ensure the equal interval distribution characteristics of the building, the local direction difference is used to ensure the similarity of the main direction of the building, and the overall direction difference is used to ensure the linear distribution characteristics of the building.

[0088]

[0089] where R lenth , R width , S building , LD orient , OD orient represent the adjacent long axis length ratio, adjacent short axis length ratio, interval distance ratio between the building groups in the pattern, local direction difference and overall direction difference, respectively, δ lenth , δ width , δ space , δ orientThresholds of length ratio of long axis, length ratio of short axis, interval distance ratio and difference of pattern direction, respectively.

[0090] Further, the key to recognizing the complex combined straight line pattern is to determine all the combined structures contained in the pattern. According to the cognitive features of the complex combined straight line pattern, the complex combined straight line pattern contains multiple combined structures or pattern groups formed by two or more simple straight line patterns, and the combined structures and the simple straight line pattern groups are interrelated and influenced. The present application first preliminarily recognizes the complex combined straight line pattern by recognizing the simple straight line pattern group, and then extracts the combined structure by using the node matching strategy, and finally realizes the accurate recognition of the complex combined straight line pattern.

[0091] Specifically, the step 3 comprises:

[0092] According to the cognitive features of the complex combined straight line pattern, the complex combined straight line pattern can be regarded as a group of simple straight line patterns which are spatially adjacent and have similar distribution characteristics (direction, interval) and are arranged in the mode vertical direction. Therefore, the building group containing the complex combined straight line pattern can be preliminarily recognized by recognizing the simple straight line pattern group meeting the above features. In combination with the spatial features of the simple straight line pattern group, the present application takes the pattern direction difference, the pattern interval distance ratio, the pattern distance and the pattern alignment degree as the constraint conditions for extracting the simple straight line pattern group.

[0093] The pattern direction difference (PO orient ) describes the direction difference between the main directions of different patterns, and the main direction of the pattern is the direction of the geometric center point connecting line of the first and last ends of the pattern unit in the pattern.

[0094] The pattern interval distance ratio (S pattern ) describes the interval distance difference between the patterns, and the interval distance of the pattern is the average value of the interval distances between all adjacent buildings in the pattern.

[0095] The pattern distance (Dis) describes the degree of proximity between the patterns, and in order to reasonably measure the degree of proximity between the patterns, the concept of influence domain is introduced. Analysis shows that the influence domains of the simple straight line patterns forming the combined pattern will have the spatial relationships of intersection, inclusion or separation, and according to the spatial relationships between the influence domains, the pattern distance has different calculation methods. Assuming that the influence domains of the simple straight line patterns P a and P b are U a and U b , the distance between P a and P b is:

[0096]

[0097] In the formula, d is the distance between U a and Ub The shortest distance to the upper node.

[0098] Definition 1: Pattern influence domain: Construct a convex hull for all buildings within a simple straight line pattern, and the spatial area enclosed by this convex hull is the influence domain of the pattern.

[0099] The pattern alignment degree (Deg) describes the degree to which the simple straight line pattern is aligned in the vertical direction of the pattern. a 、P b The influence domain is U a 、U b , with U b As a reference, make a reference target U in its main direction (main direction of the mode) a The projection of U a To U b The opposite projection length ratio r is the ratio of the maximum overlapping length of the projection to the total projection length in that direction, then the mode P a 、P b The alignment degree Deg(P a ,P b )=r.

[0100] by Figure 1 Take the building complex in the dotted box A as an example (e.g. Figure 4 As shown in Figure 2, based on the identification of simple straight line patterns implied by various types of building complexes, the steps for preliminary identification of complex combined straight line patterns are as follows:

[0101] Step 3.1: Merge the simple straight line patterns identified from various building groups into the set PatternSet. Figure 4 As shown by the solid line in the middle; let the linked list set that forms a complex combined straight line pattern with a simple straight line pattern be ListSet, and execute step 3.2;

[0102] Step 3.2: Take any two patterns P from the set PatternSet A and P B , if P A and P B Does not contain the same elements and satisfies the conditions of formula 4.4, then P A With P B Can form complex combination straight line patterns ( Figure 4 Mode P1 and P2, P2 and P3, P3 and P4), P A With P B Add to the linked list array and execute step 3.3;

[0103] Step 3.3: Traverse any linked list in the ListSet i , if List iIf the intersection of array is not empty, then insert the non-intersection elements in array into List i ; if the intersection of List i and array is empty, then add array to ListSet; return to step 3.2 until all combinations of patterns are traversed. As shown in Figure 4 , the patterns P1, P2, P3 and P4 are initially identified as a group of simple linear patterns which can form a complex combination linear pattern.

[0104]

[0105] wherein PO orient , S pattern , Dis and Deg are the pattern direction difference, interval distance ratio, pattern distance and pattern alignment degree respectively, and δ orient , δ space , δ dis and δ deg are the threshold values of the pattern direction difference, interval distance ratio, pattern distance and pattern alignment degree respectively.

[0106] Specifically, the step 4 comprises:

[0107] According to the distribution characteristics of the complex combination linear pattern, the combination structure of the complex combination linear pattern is formed by the adjacent buildings between different simple linear patterns in the associated pattern group. Therefore, the combination structure can be identified by using the simple linear pattern group which forms the complex combination linear pattern, so as to accurately identify the complex combination linear pattern.

[0108] The building is abstracted as a point, and the simple linear pattern is expressed by a line connected by points. The combination structure is identified by sequentially matching the adjacent nodes on the adjacent lines. The node matching process is constrained by the cognitive characteristics of the complex combination linear pattern. It can be analyzed that the upper limit of the number of combination structures in the complex combination linear pattern depends on the minimum number of buildings contained in each simple linear pattern in the associated pattern group. Therefore, the simple linear pattern containing the least buildings must be used as the starting reference object for node matching. Since each building can only belong to one combination structure, when a node in the object to be matched is determined as a matching pair, the node cannot participate in the matching of other nodes. Generally, the closest buildings are more likely to be divided into a combination structure. When considering the perception of space scene by human as a subject, the buildings which are regularly and integrally organized in a pattern are more likely to be divided into a combination structure. Figure 5As shown in the figure, although buildings 1 and 20 are closest, buildings 1 and 4 should be classified as a composite structure. However, using the closest distance as the matching criterion only extracts two composite structures, so no pattern can be formed. Therefore, matching should prioritize the number of matching pairs while also considering the closest distance criterion.

[0109] Combined with the above analysis, and to facilitate the description of the recognition method, the following two concepts are defined:

[0110] Definition 2: Matching degree: If the reference object and the object to be matched have n matching pairs, then the matching degree between the reference object and the object to be matched is n.

[0111] Definition 3: Overhanging buildings: Buildings that are separated from the overlapping area of ​​influence of adjacent simple straight line patterns are called overhanging buildings.

[0112] by Figure 1 Take the building complex in the dotted box A as an example (e.g. Figure 5 As shown), the algorithm flow and steps for accurate recognition of complex combined straight line patterns are as follows:

[0113] Step 4.1: Build a matching chain. Take a simple straight line pattern group from the pattern list set to build a list of objects to be matched i {P0,P1,…,P n}, select the matching object P with the least number of nodes i As the starting reference object, P i The progressive matching chain of adjacent objects to be matched is constructed as the starting point {P i ,P i-1 ,…,P0} and {P i ,P i+1 ,…,P n}; If all objects to be matched have the same number of nodes, P0 is taken as the starting reference object, and the constructed progressive matching chain is {P0, P1, ..., P n}.like Figure 5 In (a), P1 is used as the starting reference object, and the constructed matching chain is {P1, P2, P3, P4}.

[0114] Step 4.2: Progressive node matching. Perform progressive node matching according to the constructed matching chain to match the chain {P0, P1, ..., P n For example, we use P0 and P1 as reference objects and objects to be matched, respectively. Then, we use the matched node set in P1 as the reference object and P2 as the object to be matched. We continue matching in this way until the last object in the matching chain is matched. The node matching algorithm steps for reference objects and objects to be matched are as follows:

[0115] Step 4.2.1: Take the node representing one end of the non-suspended building in the reference object as the starting point, if there is no node representing the non-suspended building, take the first end point in the reference object as the starting point, record the nearest node in the to-be-matched object as the matching pair, and take the directly adjacent nodes in the reference object and the to-be-matched object as the matching pairs in turn, and record them as the matching pairs, and take the node representing the suspended building in the reference object as the starting point, if there is no node representing the suspended building, take all the nodes, and add all the matching pairs to the set Setl; execute step 4.2.2; if Figure 5 In (a), the matching pairs of P1 and P2 are (1, 20), (23, 19) and (22, 18), and the matching degree is 3;

[0116] Step 4.2.2: Take the node representing one end of the non-suspended building in the reference object as the starting point, if there is no node representing the non-suspended building, take the first end point in the reference object as the starting point, record the second nearest node in the to-be-matched object as the matching pair, and obtain other matching pairs as described above, and add them to the set Set2; execute step 4.2.3; if Figure 5 In (b), the matching pairs of P1 and P2 are (1, 4), (23, 20), (22, 19) and (21, 18), and the matching degree is 4;

[0117] Step 4.2.3: Compare the matching degrees of the matching results of step 4.2.1 and step 4.2.2, and take the result with the higher matching degree as the final matching result, if the matching degrees are equal, take Setl as the final matching result. If Figure 5 In (b), the nodes connected by the dashed line are the final matching result.

[0118] Step 4.3: Pattern extraction and post-processing. Combine the matching pair set in order, and then obtain the combined structure as the pattern unit for accurate recognition, such as Figure 5 In (c), the building connected by the dashed line is the recognized combined structure, and the recognized complex combined straight line pattern can be represented as CP = {(1, 4, 2, 3), (23, 20, 15, 16), (22, 19, 12, 14), (21, 18, 11, 13)}; remove the unmatched buildings from the simple straight line pattern, and then obtain the simple straight line pattern group that can constitute the complex combined straight line pattern. Considering that other buildings may conflict with the complex combined straight line pattern, if the influence domain of the complex combined straight line pattern intersects with other buildings, the pattern is deleted.

[0119] To verify the effect of the present invention, the above method was implemented using C# and ArcGIS Engine programming. The experimental data is a part of the building complex in Rotterdam, the Netherlands at a scale of 1:10,000. The experimental area contains a total of 1,892 buildings, such as Figure 6 (a) As shown. The sizes and shapes of the buildings in the selected experimental area are somewhat different. Most buildings have similar orientations. The building clusters are locally densely distributed, and there is obvious local spatial heterogeneity. In visual perception, they present a clear complex combination of straight line patterns. The complex combination of straight line pattern extraction process involves multiple parameters. Referring to previous experience and after repeated tests, in the building classification stage, the number of categories (Num c )Threshold δ n =5; in the stage of recognizing simple straight line patterns, set the length ratio threshold δ len =0.85,δ wid =0.75, the interval distance is greater than the threshold δ spa =0.85, direction difference threshold δ ori =10°; in the stage of identifying complex combined straight line patterns, set the pattern interval distance ratio threshold δ spa =0.85, mode direction difference threshold δ ori =10°, mode distance threshold δ dis =10m, mode alignment threshold δ deg =0.7. The building group classification results are as follows Figure 6 As shown in (b), the recognition results of complex combined linear patterns expressed in combined structures are as follows Figure 6 (c) shown.

[0120] The statistical and comparative analysis of the recognition results of the method of the present invention and the comparative experiment showed that the method of the present invention finally extracted 13 complex combination straight line patterns, and the comparative experiment identified a total of 95 simple straight line patterns. The analysis showed that the method proposed by the present invention can identify complex combination straight line patterns contained in building complexes with complex distributions and different spatial granularities. The more buildings each combination structure as a pattern unit in the complex combination straight line pattern contains, the more complex the structure of the combination structure is. The complex combination straight line pattern combination structure identified in this experiment contains a maximum of 5 buildings. In addition, the results recognized by the method of the present invention also contain complex combination straight line patterns with typical character-type combination structures, for example, Figure 6 (c) The combined structure of the combined pattern in the solid line frame is Figure 7 If it is organized in the form of connecting lines, it will show obvious structural characteristics of the O-shaped character.

[0121] Invite a mapping expert to perform manual pattern recognition on the experimental data of the present invention (the recognition results are as follows Figure 8The expert identification result is compared with the identification result of the method, the identification result is evaluated in terms of the precision rate and the recall rate, it is calculated that the precision rate and the recall rate are both 85%, and the result shows that the identification result of the application is reliable.

[0122] In conclusion, the application proposes a new linear mode, complex combination linear mode, aiming at the phenomenon that there is no specific structure but regularity and the whole is in linear mode organization and distribution in large-scale building groups.

[0123] For the identification problem of the complex combination linear mode in building groups, a building group complex combination linear mode identification method combining feature classification and node matching is proposed. Firstly, the cognitive characteristics and definition of the combination linear mode are analyzed and studied; then the improved k-means algorithm (k-means++) is used to classify the building groups taking the area and main direction of the building as the index; the simple linear modes of various building groups are identified considering the linearity, compactness and similarity of the linear mode and other constraint conditions; the simple linear mode groups that can constitute the combination linear mode are preliminarily identified combining the cognitive characteristics of the combination linear mode; finally, the combination structure is identified using the node matching strategy, and the accurate identification of the combination linear mode is realized.

[0124] On the basis of the above-mentioned embodiments, as Figure 9 shown, the application also proposes a building group complex combination linear mode identification device combining feature classification and node matching, comprising:

[0125] The classification module is used for classifying the building groups according to the area and the main direction;

[0126] The first identification module is used for respectively constructing the spatial proximity relationship of the building groups and identifying the simple linear modes contained in various building groups;

[0127] The second identification module is used for preliminarily identifying the building group complex combination linear mode by identifying the simple linear mode groups based on the identified simple linear modes;

[0128] The third identification module is used for realizing the accurate identification of the building group complex combination linear mode by identifying the combination structure using the node matching strategy.

[0129] Further, the classification module is specifically used for:

[0130] The k-means++ clustering algorithm is used to classify the building groups, and the area and main direction characteristics of the building are mapped to the Euclidean distance space, so as to measure the similarity between the buildings in terms of the Euclidean distance.

[0131] Further, the first identification module is specifically used for:

[0132] The weak adjacent search strategy is adopted to construct the spatial adjacent relationship of the building group, and the search range is set as the buildings within the second-order adjacent;

[0133] The extraction rule as shown in formula 4.2 is established in combination with the constructed structured parameters:

[0134]

[0135] In the formula, R lenth , R width , S building , LD orient , OD orient respectively represent the adjacent long axis length ratio, the adjacent short axis length ratio, the interval distance ratio between the building groups in the mode, the local direction difference and the overall direction difference, δ lenth , δ width , δ space , δ orient are the threshold values of the long axis length ratio, the short axis length ratio, the interval distance ratio and the mode direction difference.

[0136] Further, the second identification module is specifically configured to:

[0137] Step 3.1, combine the simple straight line modes identified from various building groups into a set PatternSet; set the chain table set of the complex combination straight line modes formed by the simple straight line modes as ListSet, and execute step 3.2;

[0138] Step 3.2, take any two modes P A and P B from the set PatternSet, if P A and P B do not contain the same elements and satisfy the condition of formula 4.4, P A and P B can constitute a complex combination straight line mode, P A and P B are added to the chain table array array, and step 3.3 is executed;

[0139]

[0140] In the formula, PO orient , S pattern , Dis, Deg respectively represent the mode direction difference, the interval distance ratio, the mode distance, the mode alignment degree, δ orient , δ space , δ dis and δ deg are the threshold values of the mode direction difference, the interval distance ratio, the mode distance and the mode alignment degree;

[0141] Step 3.3, traverse any chain List in the set ListSet i , if the intersection of List i and array is not empty, insert the non-intersection elements in array into List i ; if the intersection of List i and array is empty, add array to ListSet; return to step 3.2 until all combinations of patterns are traversed.

[0142] Further, the third identification module is specifically configured to:

[0143] Step 4.1, take a simple straight pattern group from the set of pattern chains to establish a set of to-be-matched objects List i {P0, P1, …, P n}, select the to-be-matched object P i containing the least number of nodes as the starting reference object, and construct the progressive matching chain of the adjacent to-be-matched objects as {P i , P i-1 , …, P0} and {P i , P i+1 , …, P n} with P i as the starting point.

[0144] Step 4.2, perform progressive node matching according to the constructed matching chain, and the node matching algorithm steps of the reference object and the to-be-matched object are as follows:

[0145] Step 4.2.1, take the node representing one end of the non-suspended building in the reference object as the starting point, or take the first end point in the reference object as the starting point if there is no node representing the non-suspended building, record the node closest to the to-be-matched object as a matching pair, and record the nodes directly adjacent to the corresponding matching pair of the last group in the reference object and the to-be-matched object as matching pairs in turn, and the matching ends after the first time to the node representing the suspended building in the reference object, or all nodes are taken if there is no node representing the suspended building, and all matching pairs are added to the set Set1; execute step 4.2.2.

[0146] Step 4.2.2, take the node representing one end of the non-suspended building in the reference object as the starting point, or take the first end point in the reference object as the starting point if there is no node representing the non-suspended building, record the node next closest to the to-be-matched object as a matching pair, and obtain other matching pairs in the same way and add them to the set Set2; execute step 4.2.3.

[0147] Step 4.2.3, comparing the matching degrees of the matching results of step 4.2.1 and step 4.2.2, taking the result with higher matching degree as the final matching result, and if the matching degrees are equal, taking Set 1 as the final matching result;

[0148] Step 4.3, merging the matching pair sets in order, then obtaining the accurate recognition of the combined structure as the mode unit, and eliminating the unmatched buildings from the simple straight line mode, then obtaining the simple straight line mode group which can constitute the complex combined straight line mode; if the influence domain of the complex combined straight line mode intersects with other buildings, the mode is deleted.

[0149] In summary,

[0150] (1) The straight line mode cognitive characteristics and recognition method of the existing research are not suitable for the complex combined straight line mode, and the present application can recognize the complex combined straight line mode contained in the building group with complex distribution and different spatial granularity.

[0151] (2) One cartography expert is invited to perform artificial mode recognition on the experimental data of the present application (the recognition result is shown in Table 1), and then the recognition result of the expert is compared with the recognition result of the method, and the recognition result is evaluated by precision and recall, and it is calculated that the precision and recall are both 85%, and the result shows that the recognition result of the present application is reliable. Figure 8

[0152] (3) For the building group synthesis with complex distribution and different granularity in large scale maps, the complex combined straight line mode can fully exploit the spatial distribution characteristics of the atypical combined structure of the building group, and provide an important basis for the selection of synthesis operator and algorithm in the building group synthesis process, and provide effective support for high quality map synthesis.

[0153] The above only shows the preferred embodiments of the present application, and it should be noted that for ordinary skilled in the art, without departing from the principles of the present application, a number of improvements and refinements can be made, and these improvements and refinements should be considered as the protection scope of the present application.​

Claims

1. A method for recognizing complex linear patterns of building complexes by combining feature classification with node matching, characterized in that: include: Step 1: Classify the building complex according to area and main direction; Step 2: construct spatial proximity relationships of building clusters and identify simple linear patterns within each type of building cluster; Step 3, based on the identified simple straight line pattern, preliminarily identify the complex combination straight line pattern of the building complex by identifying the simple straight line pattern group; Step 4: Use node matching strategy to identify the combined structure and achieve accurate recognition of the complex combined straight line pattern of the building complex; The step 3 includes: In step 3.1, merge the simple straight line patterns identified from various building groups into the set PatternSet. Let ListSet be the linked list set of complex combined straight line patterns formed from simple straight line patterns, and proceed to step 3.

2. Step 3.2, select any two patterns P from the set PatternSet A and P B , if P A and P B Does not contain the same elements and satisfies the conditions of formula 4.4, then P A With P B Can form complex combination straight line pattern, A With P B Add to the linked list array and execute step 3.3; Where, PO orient 、S pattern , Dis, and Deg are mode direction difference, mode interval distance ratio, mode distance, and mode alignment, respectively. orient , δ space , δ dis and δ deg are the thresholds for pattern direction difference, separation distance ratio, pattern distance, and pattern alignment, respectively; Step 3.3, traverse any linked list List in the collection ListSet i , if List i If the intersection with array is not empty, insert the non-intersecting elements in array into List i In; if List i If the intersection with array is empty, add array to ListSet; return to step 3.2 until all pattern combinations are traversed; The step 4 comprises: Step 4.1, take a simple straight line pattern group from the pattern list set to create a list of objects to be matched i {P0,P1,…,P n }, select the matching object P with the least number of nodes i As the starting reference object, P i The progressive matching chain of adjacent objects to be matched is constructed as the starting point {P i ,P i-1 ,…,P0} and {P i ,P i+1 ,…,P n }; Step 4.2: Perform progressive node matching according to the established matching chain. The node matching algorithm steps for the reference object and the object to be matched are as follows: Step 4.2.1: Take the node representing one end of the non-hanging building in the reference object as the starting point. If there is no node representing a non-hanging building, take the first endpoint of the reference object as the starting point and record it and the node closest to the object to be matched as a matching pair. Then, sequentially take the nodes in the reference object that are directly adjacent to the object to be matched and directly adjacent to the previous set of corresponding matching pairs as matching pairs. The matching ends when the node representing the hanging building in the reference object is first taken. If there is no node representing a hanging building, take all nodes and add all matching pairs to Set1; then execute step 4.2.2; In step 4.2.2, take the node representing one end of the non-hanging building in the reference object as the starting point. If no node representing a non-hanging building exists, take the first endpoint in the reference object as the starting point. Record it and the next closest node in the object to be matched as a matching pair. Similarly, obtain other matching pairs and add them to Set2; then proceed to step 4.2.

3. Step 4.2.3: Compare the matching degrees of the matching results of step 4.2.1 and step 4.2.2, and take the result with the higher matching degree as the final matching result. If the matching degrees are equal, Set1 is taken as the final matching result. In step 4.3, the matching pair sets are merged in sequence to obtain a precisely identified combination structure as a pattern unit. Unmatched buildings are removed from the simple straight line pattern to obtain a group of simple straight line patterns that can form a complex combined straight line pattern. If the influence domain of the complex combined straight line pattern intersects with other buildings, the pattern is deleted.

2. The method for recognizing complex combined linear patterns of buildings by combining feature classification with node matching according to claim 1 is characterized in that: The step 1 comprises: The k-means++ clustering algorithm is used to classify the building complexes. The area and main direction features of the buildings are mapped to the Euclidean distance space, and the Euclidean distance is used to measure the similarity between buildings.

3. The method for recognizing complex combined linear patterns of buildings by combining feature classification with node matching according to claim 1 is characterized in that: The step 2 includes: A weak proximity search strategy is used to construct the spatial proximity relationship of building groups, and the search range is set to buildings within the second-order proximity; Combined with the constructed structural parameters, the extraction rules are established as shown in formula 4.2: Where R lenth 、R width 、S building LD orient ,OD orient They represent the ratio of adjacent major axis lengths, the ratio of adjacent minor axis lengths, the ratio of the spacing distances between building groups within the pattern, the local direction difference, and the overall direction difference, respectively. lenth , δ width , δ space , δ orient are the thresholds for major axis length ratio, minor axis length ratio, interval distance ratio and pattern direction difference, respectively.

4. A device for recognizing complex linear patterns of building complexes by combining feature classification with node matching, characterized in that: include: A classification module for classifying building complexes based on area and main orientation; The first recognition module is used to construct the spatial proximity relationship of building clusters and identify the simple straight line patterns contained in various building clusters; The second recognition module is used to preliminarily recognize the complex combination straight line pattern of the building complex by recognizing the simple straight line pattern group based on the recognized simple straight line pattern; The third recognition module is used to identify the combined structure using the node matching strategy to achieve accurate recognition of the complex combined straight line pattern of the building complex; The second identification module is specifically configured to: In step 3.1, merge the simple straight line patterns identified from various building groups into the set PatternSet. Let ListSet be the linked list set of complex combined straight line patterns formed from simple straight line patterns, and proceed to step 3.

2. Step 3.2, select any two patterns P from the set PatternSet A and P B , if P A and P B Does not contain the same elements and satisfies the conditions of formula 4.4, then P A With P B Can form complex combination straight line pattern, A With P B Add to the linked list array and execute step 3.3; Where, PO orient 、S pattern , Dis, and Deg are mode direction difference, mode interval distance ratio, mode distance, and mode alignment, respectively. orient , δ space , δ dis and δ deg are the thresholds for pattern direction difference, separation distance ratio, pattern distance, and pattern alignment, respectively; Step 3.3, traverse any linked list List in the collection ListSet i , if List i If the intersection with array is not empty, insert the non-intersecting elements in array into List i In; if List i If the intersection with array is empty, add array to ListSet; return to step 3.2 until all pattern combinations are traversed; The third identification module is specifically used for: Step 4.1, take a simple straight line pattern group from the pattern list set to create a list of objects to be matched i {P0,P1,…,P n }, select the matching object P with the least number of nodes i As the starting reference object, P i The progressive matching chain of adjacent objects to be matched is constructed as the starting point {P i ,P i-1 ,…,P0} and {P i ,P i+1 ,…,P n }; Step 4.2: Perform progressive node matching according to the established matching chain. The node matching algorithm steps for the reference object and the object to be matched are as follows: Step 4.2.1: Take the node representing one end of the non-hanging building in the reference object as the starting point. If there is no node representing a non-hanging building, take the first endpoint of the reference object as the starting point and record it and the node closest to the object to be matched as a matching pair. Then, sequentially take the nodes in the reference object that are directly adjacent to the object to be matched and directly adjacent to the previous set of corresponding matching pairs as matching pairs. The matching ends when the node representing the hanging building in the reference object is first taken. If there is no node representing a hanging building, take all nodes and add all matching pairs to Set1; then execute step 4.2.2; In step 4.2.2, take the node representing one end of the non-hanging building in the reference object as the starting point. If no node representing a non-hanging building exists, take the first endpoint in the reference object as the starting point. Record it and the next closest node in the object to be matched as a matching pair. Similarly, obtain other matching pairs and add them to Set2; then proceed to step 4.2.

3. Step 4.2.3: Compare the matching degrees of the matching results of step 4.2.1 and step 4.2.2, and take the result with the higher matching degree as the final matching result. If the matching degrees are equal, Set1 is taken as the final matching result. Step 4.3: Merge the matching pairs in order to obtain a precisely identified combined structure as a pattern unit. Remove the unmatched buildings from the simple straight line pattern to obtain a group of simple straight line patterns that can form a complex combined straight line pattern. If the influence domain of the complex combined straight line pattern intersects with other buildings, the pattern will be deleted.

5. The device for recognizing complex combined linear patterns of buildings by combining feature classification with node matching according to claim 4 is characterized in that: The classification module is specifically used for: The k-means++ clustering algorithm is used to classify the building complexes. The area and main direction features of the buildings are mapped to the Euclidean distance space, and the Euclidean distance is used to measure the similarity between buildings.

6. The device for recognizing complex combined linear patterns of buildings by combining feature classification with node matching according to claim 4, characterized in that: The first identification module is specifically configured to: A weak proximity search strategy is used to construct the spatial proximity relationship of building groups, and the search range is set to buildings within the second-order proximity; Combined with the constructed structural parameters, the extraction rules are established as shown in formula 4.2: Where R lenth 、R width 、S building LD orient ,OD orient They represent the ratio of adjacent major axis lengths, the ratio of adjacent minor axis lengths, the ratio of the spacing distances between building groups within the pattern, the local direction difference, and the overall direction difference, respectively. lenth , δ width , δ space , δ orient are the thresholds for major axis length ratio, minor axis length ratio, interval distance ratio and pattern direction difference, respectively.

Images