Island chain identification method under support of Grignard tower principle

By combining Gestalt principles and clustering degree clustering methods with convex hull algorithm and proximity graph construction, the overall and local features of island chains are identified, solving the problem of identifying island chains with uneven distribution and large shape differences, and achieving more accurate island chain identification.

CN119004032APending Publication Date: 2024-11-22Chinese People's Liberation Army Cyberspace Force Information Engineering University
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411106349.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-13
Publication Date
2024-11-22

AI Technical Summary

Technical Problem

Existing technologies struggle to effectively identify island chains, especially in scenarios where natural surface elements such as islands and lakes are unevenly distributed, have significant shape differences, and are locally clustered. Existing methods cannot accurately identify their linear patterns.

Method used

Using Gestalt principles, this method identifies locally clustered island clusters through clustering degree clustering. It then uses the convex hull algorithm for substitution and constructs a proximity graph. Combining the principles of proximity, directionality, and wholeness, it designs feature parameters and screening rules to identify island chains.

Benefits of technology

It effectively identifies the overall continuity and local fragmentation characteristics of island chains, improving the accuracy of identification and conforming to human cognition.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119004032A_ABST
    Figure CN119004032A_ABST
Patent Text Reader

Abstract

The invention belongs to the technical field of map element spatial pattern recognition, and discloses an island chain recognition method under the support of a Grignard tower principle, which comprises the following steps of: firstly, recognizing a locally aggregated island cluster by adopting an aggregation degree clustering method, and performing overall element replacement by utilizing a convex hull algorithm; then, according to the principles of proximity, directivity and integrity of the Grignard principle, constructing an island chain identification model which takes the minimum spanning tree as a proximity graph and takes local direction difference, global direction difference and width-to-length ratio as main parameters; and finally, constructing an adjacent graph by taking the island convex hull and the cluster convex hull as units, identifying the island chain structure, and combining two island chain identification results according to the principle of'large-range priority '. The island chain identification method effectively solves the problem that the island chain identification effect is poor under the overall continuous and local broken and dispersed distribution characteristics, and can effectively identify the island chain from the overall and local levels.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of spatial pattern recognition of map elements, and in particular to an island chain recognition method supported by the Gestalt principle. Background Art

[0002] The spatial distribution pattern of map elements is an important form of reflecting the geographical characteristics of the mapping area and is one of the objective bases for determining map synthesis. The identification of spatial distribution patterns is an important means to mine implicit spatial knowledge and enhance existing data, and has become a necessary preparation for map synthesis. Natural elements such as islands are important components of map elements and are usually represented by surface elements at medium and large scales. In a certain spatial area, clustered natural surface elements form a linearly arranged spatial structure, forming unique geographical features, such as island chains. Island chains reflect not only the spatial structure of specific landforms, but also the geological laws such as geological structure and landform evolution, which are of great value to geological and geomorphological research. Identifying island chains is not only conducive to familiarizing with the geographical characteristics of the mapping area and improving the quality of map synthesis, but also conducive to enhancing the organization and expression capabilities of spatial data and promoting spatial reasoning research.

[0003] The recognition of spatial patterns of map elements mainly includes spatial clustering and pattern extraction. Spatial clustering methods usually use characteristic parameters such as distance and density as the basis for measuring the proximity of map elements, so as to achieve the purpose of hierarchical classification of map elements, such as K-Means algorithm and DBSCAN algorithm. These clustering algorithms play an important role in exploring the clustering characteristics of point groups, but surface elements have a certain spatial range, and simply using parameters such as distance and density cannot accurately describe the proximity and clustering of elements. To this end, by constructing a constrained Delanuay triangulation, the proximity relationship of surface elements is more accurately described and represented in the form of a proximity graph. Based on the above proximity graph, the clustering algorithms of face groups mainly develop the minimum spanning tree method based on distance, the MSSCP algorithm and NPGCA algorithm based on distribution density, and the GCNN algorithm based on feature learning. This type of algorithm fully considers the spatiality of face elements and can mine the clustering characteristics of face groups of various distribution forms.

[0004] In terms of pattern extraction, most existing methods follow the principles of proximity, similarity, and directionality in the Gestalt principle, select specific metric parameters for the distribution characteristics of the target pattern, and use feature rules to "prune" the proximity graph in the spatial clustering results, thereby realizing the extraction of the target spatial pattern. For the linear pattern recognition problem of building complexes, the current mainstream technical routes include rule setting method and data-driven method. The rule setting method usually selects area, main direction, rectangularity, arrangement direction, etc. as homogeneity metric parameters, and formulates rules and sets thresholds based on these metric parameters to achieve pattern extraction. The data-driven method constructs feature data with a graph structure based on the proximity graph, and automatically establishes a mapping relationship between features and distribution patterns based on deep learning methods, reducing manual participation in pattern recognition. However, the data-driven method requires a large number of data samples as the basis for model training, and the selection of feature parameters in these data samples still requires a lot of professional knowledge to support it. At the same time, since the training model usually outputs probability, the prediction accuracy of non-fixed distribution patterns is unstable and the recognition effect is poor. In terms of the distribution pattern of natural surface groups, Steiniger believes that the distribution pattern of island groups can be divided into three scales: micro, meso and macro. Among them, the proximity of islands at the micro scale is strongly correlated [Steiniger, Stefan. Enabling pattern-aware automated map generalization [D]. Zurich University, 2007.]. In response to the problem of island chain identification, Huang Yafeng borrowed the research ideas of linear pattern recognition of building groups, constructed a proximity map of island groups through the Delanuay triangulation, and evaluated the similarity based on the characteristics of proximity distance, size, direction, etc., to achieve the purpose of island chain identification [Huang Yafeng. Research on comprehensive data model and method of island (reef) element map [D]. Wuhan University, 2011.]. It can be seen that the distribution pattern recognition method based on the principles of proximity and similarity can extract island chain structures with similar size and uniform distribution. However, natural surface elements such as islands and lakes usually vary greatly in shape and size, and have distribution characteristics of overall dispersion, local aggregation, and varying degrees of aggregation. There are also multiple groups with uneven distribution density and inconsistent distribution directions within their linear patterns, and the above method cannot effectively identify them. Summary of the invention

[0005] In view of the above problems, the present invention is based on the Gestalt principle and proposes an island chain recognition method (Island Linear Pattern Recognition, ILPR) supported by the Gestalt principle based on the principles of proximity, directionality and integrity.

[0006] In order to achieve the above object, the present invention adopts the following technical solutions:

[0007] The present invention proposes an island chain identification method based on Gestalt principle, comprising:

[0008] Step 1: Use the aggregation degree clustering method to identify the local cluster structure in the natural surface element data, merge the adjacent clusters according to the target scale, construct the cluster convex hull using the convex hull algorithm, replace its internal surface elements, and identify the linear pattern of the island with the cluster convex hull as the basic unit;

[0009] Step 2: According to the correspondence between the cluster convex hull and the natural surface elements, the surface elements corresponding to the cluster convex hull are marked as "identified". Then, the linear pattern recognition of the islands is performed with the natural surface elements as the basic unit to obtain the preliminary recognition results. The surface elements marked as "identified" are removed, and the surface element groups that can still maintain the linear pattern characteristics are retained;

[0010] Step 3: Combine the linear pattern recognition results of the islands corresponding to step 1 and step 2, and output the final recognition result of the linear pattern of the islands.

[0011] Furthermore, in step 1, the degree of aggregation is calculated according to the following formula:

[0012]

[0013] Where φ(i) is the first-order neighboring polygon element of polygon element i and itself, S(j) is the area of ​​polygon element j; S vor (j) is the Voronoi diagram area of ​​surface element j; d(i,j) is the shortest distance between the boundaries of surface elements i and j; d c Represents the cutoff distance.

[0014] Furthermore, the linear pattern of the islands is identified as follows:

[0015] 1) Minimum spanning tree construction

[0016] Taking the surface element as the basic unit, constructing a constrained Delanuay triangulation, identifying its neighboring relationship, taking the shortest distance between the boundaries of the neighboring surface elements as a constraint condition, constructing a minimum spanning tree of the surface elements, and obtaining a neighboring relationship set {R}; the surface element is a natural surface element or a cluster convex hull;

[0017] 2) Linear pattern feature screening

[0018] Step A: Read the polygon feature C with a neighbor relationship count of 1 i , and its only neighboring polygon feature C j , forming the first set of neighboring relations, and C i and C j Add the line pattern candidate set {Gc}, the centroid line L(C i , Cj ) Add the trajectory set {Lc} and execute step B;

[0019] Step B: C j As the central element, traverse its neighboring relationship R(C j ), get the value of C i Area elements other than C k , construct C j With C k The center of gravity line L(C j , C k ), L(C i , C j ) as the reference line segment, measure L(C i , C j) With L(C j , C k ) of the angle θ(i, j, k), if in C j The neighboring relationship R(C j ), take the adjacent polygon element as C k When θ(i, j, k) is the maximum value, and θ(i, j, k) ≥ θ local , where θ local is the local direction threshold, then execute step C; otherwise, execute step D;

[0020] Step C: Calculate {Lc} and L(C j , C k )’s overall directional variance σ(Lc), if σ(Lc)≤σ 0 , where σ 0 is the overall direction variance threshold, then C k Join {Gc}, L(C j , C k ) add {Lc}, and at the same time add C k As the central element, L(C j , C k ) as a reference line segment, continue to step B; otherwise, proceed to step D;

[0021] Step D: Construct the minimum area rectangle of {Gc} and calculate the width-to-length ratio γ(Gc) of the rectangle. If the number of surface elements num(Gc) in {Gc} ≥ N 0 , where N 0 is the threshold value of the number of polygon features; and γ(Gc)≤γ 0 , where γ 0 is the width-to-length ratio threshold, then {Gc} is recorded as the linear mode, {Lc} is its corresponding trajectory, and step E is executed; otherwise, step E is also executed;

[0022] Step E: Remove the neighboring relationship contained in {Lc} from {R}, and add {Gc} to the line pattern recognition result set {G res}, and clear {Gc} and {Lc} at the same time. If {R} does not have a face element with a neighbor relationship of 1, the recognition result is output and the algorithm ends; otherwise, continue to execute step A.

[0023] Furthermore, θ(i, j, k) is calculated according to the following formula:

[0024] θ(i,j,k)=|A(C i , C j )-A(C i , C k )|

[0025] Among them, A(C i ,C j ) means C i Center of gravity to C j The angle of the centroid vector, A(C i ,C k ) means C i Center of gravity to C k The angle of the centroid vector.

[0026] Furthermore, σ(Lc) is calculated according to the following formula:

[0027]

[0028] Among them G c is the linear pattern natural face group to be screened, and n represents the number of vector angles of the face group.

[0029] Furthermore, γ(Gc) is calculated according to the following formula:

[0030] γ(Gc)=l wid / l len

[0031] Among them l wid is the width of the minimum circumscribed rectangle of the face group, l len is the length of the minimum circumscribed rectangle of the face group.

[0032] Compared with the prior art, the present invention has the following beneficial effects:

[0033] In view of the overall continuity and local fragmentation and dispersion of the distribution characteristics of island chains, the present invention proposes an island chain identification method supported by the Gestalt principle that takes into account both the overall and local distribution characteristics. First, in order to reduce the adverse effects of uneven island distribution on island chain identification, the aggregation degree clustering method is used to identify locally clustered island clusters, and the convex hull algorithm is used to perform overall element replacement on the clusters; then, according to the proximity, directionality and integrity principles of the Gestalt principle, the corresponding feature parameters and screening rules are designed to achieve the identification of island chains; finally, the proximity graph is constructed with islands (natural surface elements) and cluster convex hulls as units for island chain identification, and the island chain identification results are merged in accordance with the "large range priority" principle, which effectively solves the problem of poor island chain identification under the overall continuity and local fragmentation and dispersion distribution characteristics. Experiments have shown that the method of the present invention can overcome the negative impact of the dispersed distribution of islands on island chain identification, effectively identify island chains from the overall and local levels, and the experimental results are more in line with human cognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 This is a flow chart of an island chain identification method supported by the Gestalt principle according to an embodiment of the present invention;

[0035] Figure 2 A schematic diagram of typical features of an island chain provided by an embodiment of the present invention;

[0036] Figure 3 A schematic diagram of cluster identification and substitution provided by an embodiment of the present invention;

[0037] Figure 4 The principle of the linear pattern recognition method provided by the embodiment of the present invention;

[0038] Figure 5 An example diagram of the island chain identification process and results provided by an embodiment of the present invention;

[0039] Figure 6 This is an example diagram of island chain identification results using different methods provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0040] The present invention will be further explained below with reference to the accompanying drawings and specific embodiments:

[0041] Example 1

[0042] like Figure 1 As shown, an island chain identification method supported by a Gestalt principle of the present invention includes:

[0043] S101: Use the aggregation degree clustering method to identify the local cluster structure in the natural surface element data, merge the adjacent clusters according to the target scale, construct the cluster convex hull using the convex hull algorithm, replace the internal surface elements, and identify the linear pattern of the island with the cluster convex hull as the basic unit;

[0044] S102: According to the correspondence between the cluster convex hull and the natural surface elements, the surface elements corresponding to the cluster convex hull are marked as "identified", and then the linear pattern recognition of the islands is performed with the natural surface elements as the basic unit to obtain a preliminary recognition result, from which the surface elements marked as "identified" are eliminated, and the surface element groups that can still maintain the linear pattern characteristics are retained;

[0045] S103: The linear pattern recognition results of the islands corresponding to S101 and S102 are combined to output the final recognition result of the linear pattern of the islands.

[0046] The method of the present invention can overcome the negative impact of dispersed island distribution on island chain identification and effectively identify island chains from the overall and local levels.

[0047] Example 2

[0048] On the basis of the above embodiment, another island chain identification method supported by the Gestalt principle of the present invention includes:

[0049] First, the cluster structure of local aggregation in the natural surface element data is identified by using convex hull substitution, and the linear pattern recognition algorithm is used to identify the linear pattern of the islands with the convex hull as the basic unit. According to the correspondence between the convex hull and the natural surface elements, these surface elements are marked as "identified". Then, the linear pattern recognition of the islands is performed with the natural surface elements as units to obtain preliminary recognition results, from which the surface elements marked as "identified" are eliminated, and the surface element groups that can still maintain the linear pattern characteristics are retained. Finally, the convex hull linear pattern and the surface linear pattern recognition results are merged to output the final recognition result of the linear pattern of the island.

[0050] It is worth noting that the constituent units of the linear pattern can be either surface elements or locally aggregated surface element clusters, so it is necessary to identify the linear patterns of islands composed of different units in steps. The theory of overall priority of visual perception reflects that human visual cognition follows the principle of "large range priority", and the order of island linear pattern recognition should also follow this principle, and be identified step by step in the order of cluster unit linear pattern to surface element unit linear pattern.

[0051] The method specifically includes:

[0052] 1. Typical characteristics and measurements of island chains

[0053] Island chains, also known as linear patterns of islands, are usually composed of several relatively adjacent natural surfaces, which are distributed continuously along a gentle curve as a whole. In terms of geometric characteristics, the surface elements in the linear pattern are irregular in shape, non-uniform in size, and have obvious differences. They do not have geometric similarities like similar building complexes. In terms of distribution characteristics, the linear pattern is mainly manifested in: ① Proximity. The linearly arranged surface elements are located adjacent to each other, but the distribution density is not uniform, and there is a distribution feature of local aggregation; ② Directionality. The main body of the surface elements is distributed along a natural and smooth trajectory curve, but there are surface elements that deviate from the main direction of the trajectory curve; ③ Integrity. The overall distribution of surface elements is long and narrow, and the distribution has obvious directionality, such as Figure 2 shown.

[0054] (1) Proximity

[0055] The proximity of surface elements can be described from two levels: one-dimensional and two-dimensional. The distance reflects the "one-to-one" proximity relationship in the one-dimensional level, and the density reflects the "one-to-many" proximity relationship in the two-dimensional level. To this end, the present invention uses the shortest proximity distance d(i, j) and the aggregation degree δ(i) to describe the proximity of surface elements. The calculation method is as shown in Formula 1.

[0056]

[0057] Where φ(i) is the first-order neighboring polygon element of polygon element i and itself, S(j) is the area of ​​polygon element j; S vor (j) is the Voronoi diagram area of ​​surface element j; d(i,j) is the shortest distance between the boundaries of surface elements i and j; d c Represents the cutoff distance, which is generally set to the median of the shortest distances between all adjacent face boundaries.

[0058] (2) Directionality

[0059] In the linear mode, the elements are distributed along a smooth and continuous curve. From a local perspective, the distribution trajectory of the surface elements is close to a line; from a global perspective, the extension direction of the surface elements is highly consistent, such as Figure 2 To this end, the present invention uses the local direction difference θ(i,j,k) and the global direction difference σ(Lc) to describe the directionality of the face group, and the calculation method is as shown in Formula 2 and Formula 3.

[0060] θ(i,j,k)=|A(C i , C j )-A(C i , C k )| Formula 2

[0061]

[0062] Among them C i ,C j,C k Both represent area elements, and C i With C j ,C k Topological proximity, A(C i ,C j ) means C i Center of gravity to C j The angle of the centroid vector. G c is the linear pattern natural face group to be screened, and n represents the number of vector angles of the face group.

[0063] (3) Holistic

[0064] The distribution of linearly arranged face groups has a tendency as a whole, that is, the length of the face group distributed along a certain direction is significantly longer than the length of its vertical distribution. To this end, a minimum area circumscribed rectangle of the face group is constructed, and the aspect ratio of the rectangle is used as a characteristic parameter to describe the integrity of the face group, denoted as the face group aspect ratio γ(Gc), and its calculation method is shown in Formula 4.

[0065] γ(Gc)=l wid / l len Formula 4

[0066] Among them l wid is the width of the minimum circumscribed rectangle of the face group, l len is the length of the minimum circumscribed rectangle of the face group.

[0067] 2. ILPR method identification principle and process

[0068] Due to the differences in the degree of aggregation, the linear pattern can be regarded as composed of several clusters with different distribution densities. Such an overall treatment of local clusters greatly eliminates the characteristic "noise" of a few islands, highlights their main features, and contributes to the accuracy of linear pattern recognition. To this end, the present invention adopts a "bottom-up" strategy to gradually realize the recognition of linear patterns. Based on the principles of proximity, directionality, and integrity in the Gestalt principle, clusters and surface elements are used as the basic units of pattern recognition, local and global directional rules, quantity rules, and overall pattern feature rules are set, and the proximity graph is "pruned" to realize the recognition of natural face group linear patterns.

[0069] 2.1 Cluster identification and substitution

[0070] The natural face groups arranged in a linear pattern are distributed along the curve trajectory as a whole, with strong directional consistency; the local distribution is clustered, and the direction is relatively dispersed, which interferes with the direction screening in the linear pattern recognition process and has a negative impact on the "pruning" of the neighboring graph. Therefore, it is necessary to identify such locally aggregated clusters, use the principle of integration to transform them, and eliminate the adverse effects of complex neighboring relationships and local distribution direction heterogeneity. Considering that different clusters have different distribution densities and large differences in neighboring distances, the present invention uses the aggregation degree clustering method to identify locally aggregated clusters, such as Figure 3 (b). Appropriately merge neighboring clusters according to the target scale, such as Figure 3 (c) The convex hull algorithm is used to construct the cluster convex hull and replace its internal surface elements, such as Figure 3 Middle (d).

[0071] 2.2 Extraction of Line Pattern

[0072] Line pattern recognition (such as Figure 4 The purpose of the mining process is to mine natural surface elements arranged in a linear manner. The process can be divided into two steps:

[0073] (1) Minimum spanning tree construction

[0074] Taking the surface elements (natural surface elements or cluster convex hulls) as the basic unit, the constrained Delanuay triangulation is constructed to identify its neighboring relationship. Taking the shortest distance between the boundaries of neighboring surface elements as the constraint condition, the minimum spanning tree of surface elements is constructed, and the neighboring relationship set {R} is obtained.

[0075] (2) Linear pattern feature screening

[0076] Step 1: Read the polygon feature C with a neighbor relationship count of 1 i , and its only neighboring polygon feature C j , forming the first set of neighbor relationships, and the polygon element C i and C j Add the line pattern candidate set {Gc}, the centroid line L(C i , C j )Add the trajectory set {Lc} and execute step 2.

[0077] Step 2: Place the polygon element C j As the central element, traverse its neighboring relationship R(C j ), get the value of C i Area elements other than C k Build C j With C k The center of gravity line L(C j , C k ), measure the reference line segment L(C i , C j)With line segment L(C j , C k ) is θ(i, j, k), if the following conditions are met: ① In the area element C j The neighboring relationship R(C j ), take the adjacent polygon element as C k , θ(i, j, k) is the maximum value; ②θ(i, j, k) ≥ θ local , where θ local is the local direction threshold, execute step 3; if not, execute step 4.

[0078] Step 3: Calculate the trajectory set {Lc} and the line segment L(C j , C k )’s overall directional variance σ(Lc), if σ(Lc)≤σ 0 , where σ 0 is the overall directional variance threshold, then the area element C k Add the line pattern candidate set {Gc}, line segment L(C j , C k ) is added to the trajectory set {Lc}, and the surface element C k As the central element, L(C j , C k ) as the reference line segment and proceed to step 2; if not satisfied, proceed to step 4.

[0079] Step 4: Construct the minimum area rectangle of the linear pattern candidate set {Gc} and calculate the width-to-length ratio γ(Gc) of the rectangle. If the following conditions are met: ① The number of surface elements in the linear pattern candidate set num(Gc) ≥ N 0 , where N 0 is the threshold value of the number of surface elements; ②γ(Gc)≤γ 0 , where γ 0 is the width-to-length ratio threshold, then {Gc} is recorded as the linear mode, and {Lc} is its corresponding trajectory, and step 5 is executed for both.

[0080] Step 5: Remove the neighboring relations contained in the trajectory set {Lc} from the set {R}, and add {Gc} to the line pattern recognition result set {G res}, and clear the sets {Gc} and {Lc} at the same time. If there is no face element with a neighbor relationship of 1 in the neighbor relationship set {R}, the recognition result is output and the algorithm ends; otherwise, continue to step 1.

[0081] 3. Experiment and Analysis

[0082] 3.1 Parameter settings and experimental results

[0083] The experiment uses OSM data of some island groups in southwestern Finland. The local direction threshold θlocal , the overall direction variance threshold σ 0 , quantity threshold N 0 , the width-to-length ratio threshold γ 0 They are set to 120°, 30°, 4°, and 0.5° respectively. The experimental results are shown in Figure 5 shown.

[0084] It can be found that the island chain identification method based on island units can identify island chain structures that are linearly arranged at small scales, such as Figure 5 (b); while the cluster-based island chain recognition method can recognize island chain structures that are linearly arranged at large scales. By combining the above two recognition results, the island chain recognition result based on the method of the present invention is obtained, such as Figure 5 Middle (d). This recognition result takes into account both global and local island distribution characteristics and effectively identifies island chain structures at different scales.

[0085] 3.2 Comparative Analysis

[0086] The present invention uses three different methods to identify island chains, including manual identification, similarity-based island chain identification method [Huang Yafeng. Research on comprehensive data model and method of island (reef) element map [D]. Wuhan University, 2011.] and the present invention method. The identification results are as follows: Figure 6 A total of 6 island chains were identified by artificial identification, and the identification results were continuous and covered a wide area. The similarity identification method is an island chain identification model based on the distance and area similarity of neighboring islands. Considering the large difference in island area in this region, the distance threshold is set to 250m and the area ratio threshold of neighboring islands is set to 0.01, and the island chain identification results are obtained, such as Figure 6 (b). Compared with the manual method, this method identifies a large number of island chains, most of which cover a small area and have poor continuity, making it difficult to reflect the overall distribution trajectory of the island chains. The ILPR method of the present invention identifies a total of 7 island chain structures, and its identification results are close to those of manual identification, with good continuity and can reflect the distribution trajectory of the island chains. Figure 6 Middle (c).

[0087] In order to better reflect the experimental effect, the present invention quantitatively analyzes the recognition results of the method of the present invention from the number of island chains, the average number of islands contained in the island chains, the global directional variance of the island chains, and the average length of the island chains, as shown in Table 1. The number of island chains and the average number of islands contained in the island chains identified by the ILPR method of the present invention are close to the manual recognition results, while the number of island chains identified by the similarity method is significantly different from the average number of islands; in terms of global directional variance, the manual method is the largest and the method of the present invention is the smallest, indicating that the island chain direction consistency of the method of the present invention is better; in terms of the average length of the island chains, the manual method is the longest, the method of the present invention is the second, and the similarity method is the shortest, reflecting that the method of the present invention also has certain advantages in identifying the continuity of island chains.

[0088] However, the number of islands included and the average length of island chains are smaller than the results of manual identification.

[0089] Table 1 Quantitative statistics of island chain identification results

[0090]

[0091] With the manual recognition results as reference, the recognition results of the similarity method and the ILPR method of the present invention are quantitatively evaluated from the accuracy, precision, recall value and F1 value, as shown in Table 2. The precision, recall and F1 value of the recognition results of the similarity method are all less than 45%, indicating that they are quite different from the manual recognition results; the values ​​of the ILPR method of the present invention are all greater than 69%, indicating that they are similar to the manual recognition results and the recognition results are more in line with human cognition.

[0092] Table 2 Quantitative evaluation of island chain identification results

[0093]

[0094] In summary, the island chain is a natural landscape formed by the combined action of the earth's internal and external forces, and is one of the typical spatial distribution patterns in the geographical environment. Identifying island chains is an important means to study geographical features and mine implicit spatial knowledge. In view of the overall continuity and local fragmentation and dispersion of the distribution characteristics of the island chain, the present invention proposes an island chain identification method supported by the Gestalt principle that takes into account both overall and local characteristics. First, the aggregation degree clustering method is used to identify the locally clustered island clusters, and the convex hull algorithm is used for overall substitution; then, following the proximity, directionality and integrity principles of the Gestalt principle, an island chain identification model is constructed with the minimum spanning tree as the proximity graph, and local directional differences, global directional differences and width-to-length ratios as the main parameters; finally, the proximity graph is constructed with the island and cluster convex hull as units, the island chain structure is identified, and the two island chain identification results are merged according to the "large range priority" principle, which effectively solves the problem of poor island chain identification under the overall continuity and local fragmentation and dispersion distribution characteristics.

[0095] The above is only a preferred embodiment of the present invention. It should be pointed out that a person skilled in the art can make several improvements and modifications without departing from the principle of the present invention. These improvements and modifications should also be regarded as within the scope of protection of the present invention.

Claims

1. An island chain identification method based on Gestalt principle, characterized in that: include: Step 1: Use the aggregation degree clustering method to identify the local cluster structure in the natural surface element data, merge the adjacent clusters according to the target scale, construct the cluster convex hull using the convex hull algorithm, replace its internal surface elements, and identify the linear pattern of the island with the cluster convex hull as the basic unit; Step 2: According to the correspondence between the cluster convex hull and the natural surface elements, the surface elements corresponding to the cluster convex hull are marked as "identified". Then, the linear pattern recognition of the islands is performed with the natural surface elements as the basic unit to obtain the preliminary recognition results. The surface elements marked as "identified" are removed, and the surface element groups that can still maintain the linear pattern characteristics are retained; Step 3: Combine the linear pattern recognition results of the islands corresponding to step 1 and step 2, and output the final recognition result of the linear pattern of the islands.

2. The island chain identification method based on the Gestalt principle according to claim 1 is characterized in that: In step 1, the degree of aggregation is calculated according to the following formula: Where φ(i) is the first-order neighboring polygon element of polygon element i and itself, S(j) is the area of ​​polygon element j; S vor (j) is the Voronoi diagram area of ​​surface element j; d(i,j) is the shortest distance between the boundaries of surface elements i and j; d c Represents the cutoff distance.

3. The island chain identification method based on the Gestalt principle according to claim 1, characterized in that: Identify the line pattern of islands as follows: 1) Minimum spanning tree construction Taking the surface element as the basic unit, constructing a constrained Delanuay triangulation, identifying its neighboring relationship, taking the shortest distance between the boundaries of the neighboring surface elements as a constraint condition, constructing a minimum spanning tree of the surface elements, and obtaining a neighboring relationship set {R}; the surface element is a natural surface element or a cluster convex hull; 2) Linear pattern feature screening Step A: Read the polygon feature C with a neighbor relationship count of 1 i , and its only neighboring polygon feature C j , forming the first set of neighboring relationships, and C i and C j Add the line pattern candidate set {Gc}, the centroid line L(C i , C j ) Add the trajectory set {Lc} and execute step B; Step B: C j As the central element, traverse its neighboring relationship R(C j ), get the value of C i Area elements other than C k , construct C j With C k The center of gravity line L(C j , C k ), L(C i , C j ) as the reference line segment, measure L(C i , C j) With L(C j , C k ) of the angle θ(i, j, k), if in C j The neighboring relationship R(C j ), take the adjacent polygon element as C k When θ(i, j, k) is the maximum value, and θ(i, j, k) ≥ θ local , where θ local is the local direction threshold, then execute step C; otherwise, execute step D; Step C: Calculate {Lc} and L(C j , C k ), if σ(Lc)≤σ0, where σ0 is the overall directional variance threshold, then C k Join {Gc}, L(C j , C k ) add {Lc}, and at the same time C k As the central element, L(C j , C k ) as a reference line segment, continue to step B; otherwise, proceed to step D; Step D: Construct a minimum area rectangle of {Gc}, calculate the width-to-length ratio γ(Gc) of the rectangle, if the number of surface elements in {Gc} num(Gc) ≥ N0, where N0 is the threshold of the number of surface elements; and γ(Gc) ≤ γ0, where γ0 is the threshold of the width-to-length ratio, then record {Gc} as a linear mode, {Lc} as its corresponding trajectory, and execute step E; otherwise, execute step E; Step E: Remove the neighboring relationship contained in {Lc} from {R}, and add {Gc} to the line pattern recognition result set {G res }, and clear {Gc} and {Lc} at the same time. If {R} does not have a face element with a neighbor relationship of 1, the recognition result is output and the algorithm ends; otherwise, continue to execute step A.

4. The island chain identification method based on the Gestalt principle according to claim 3 is characterized in that: Calculate θ(i, j, k) according to the following formula: θ(i,j,k)=|A(C i ,C j )-A(C i ,C k )| Among them, A(C i ,C j ) means C i Center of gravity to C j The angle of the centroid vector, A(C i ,C k ) means C i Center of gravity to C k The angle of the centroid vector.

5. The island chain identification method based on the Gestalt principle according to claim 3 is characterized in that: Calculate σ(Lc) according to the following formula: Among them G c is the linear pattern natural face group to be screened, and n represents the number of vector angles of the face group.

6. The island chain identification method based on the Gestalt principle according to claim 3 is characterized in that: γ(Gc) is calculated according to the following formula: γ(Gc)=l wid / l len Among them l wid is the width of the minimum circumscribed rectangle of the face group, l len is the length of the minimum circumscribed rectangle of the face group.