An open street map based land cover classification sample automatic generation method and system

Abstract

The application discloses a land cover classification sample automatic generation method and system based on an open street map, the method first acquires open street map data, then constructs a mapping relationship between the open street map data and general geometric models (namely simple geographic feature models) and general land cover classification categories, finally proposes a conversion method from geographic object latitude and longitude coordinates to in-map coordinates, and finally realizes automatic batch generation of land cover classification samples. Compared with an existing manual sample labeling method, the method has absolute advantages in efficiency, cost and quality control.

Description

Technical Field

[0001] This invention belongs to the fields of geographic information technology and computer vision processing, and specifically relates to an automated sample generation method for a deep learning model for remote sensing land cover classification. Background Technology

[0002] With the rapid development of artificial intelligence (AI) technology and its widespread application in remote sensing, remote sensing deep learning models have seen rapid performance improvements in land cover classification tasks. Despite the continuous advancements in deep learning technology, mainstream data-driven artificial neural network methods still rely on large amounts of high-quality labeled sample data. This labeled sample data is typically obtained through manual annotation, a method that suffers from high costs, long processing times, and difficulty in guaranteeing quality. Therefore, the automated and low-cost acquisition of massive amounts of labeled samples is one of the key bottlenecks in current AI-based remote sensing land cover classification tasks. In recent years, volunteer geographic information, exemplified by Open Street Maps (OSM), has experienced unprecedented growth. OSM collects geographic big data generated by cartographers from various sources, helping to fill numerous data gaps in basic maps. For example, after the devastating earthquake in Haiti in 2010 and Hurricane Kenneth in Mozambique in 2019, millions of buildings and roads were added to OSM, providing crucial mapping data for disaster relief efforts. This vast amount of geographic big data from OSM, rich in detailed semantic information, has become a promising source of massive, free training data, fully capable of addressing the bottleneck of a lack of high-quality remote sensing training samples. Exploring and utilizing the rich semantic information in OSM to automatically construct high-quality remote sensing land cover classification label samples is of great significance for the further development of remote sensing artificial intelligence. Summary of the Invention

[0003] The technical problem to be solved by this invention is to address the urgent need in the fields of remote sensing and computer science for massive land cover classification samples by proposing a method for automatically generating a large number of high-precision samples.

[0004] To address the above problems, this invention provides a method for automatically generating land cover classification samples based on open street maps. This method mainly includes the following steps:

[0005] Step 1: Obtain OSM data for the target area;

[0006] Step 2: Construct a mapping table from the OSM geometric model to the simple geographic feature geometric model, and convert the OSM data into a common geospatial data format;

[0007] Step 3: Construct a mapping table between OSM tags and general land cover classification tags, and convert OSM tags into general geographic feature tags;

[0008] Step 4: Convert the latitude and longitude coordinates of the OSM object to coordinates within the label map;

[0009] Step 5: Rasterize the OSM objects with in-map coordinates into in-map label samples.

[0010] Furthermore, in step 2, the OSM geometric model defines three types of geometric object models: Node, Way, and Relation, while the simple geographic feature geometric model uses multiple geometric objects to define different geographic features, including Point, Polyline, Polygon, MultiPoint, and MultiPolygon.

[0011] Furthermore, the mapping table between the OSM geometric model and the simple geographic feature geometric model is shown in Table 1;

[0012] Table 1. Mapping Table from OSM Geometric Model to Simple Geographic Feature Geometric Model

[0013]

[0014] The above mapping table converts OSM data into a common geospatial data format.

[0015] Furthermore, common geospatial data formats include GeoJSON.

[0016] Furthermore, the mapping table between OSM labels and general land cover classification labels is shown in Table 2;

[0017] Table 2. Mapping Table of OSM Labels and General Land Cover Classification Labels

[0018]

[0019] The above mapping table converts OSM labels into generic geographic feature labels.

[0020] Furthermore, the specific implementation method of step 4 is as follows:

[0021] 1) Obtain the latitude and longitude coordinates of the starting and ending points of the target area ( ), ( ); 2) Calculate the latitude and longitude span of each pixel on the map based on the expected label map size (H, W) data. The calculation method is shown in Formula 1; 3) Traverse each coordinate point of the OSM object ( ), converting its latitude and longitude coordinates into coordinate points on the map ( The calculation method is shown in Formula 2;

[0022]

[0023] .

[0024] This invention also provides an automatic land cover classification sample generation system based on open street maps, comprising the following modules:

[0025] The data acquisition module is used to acquire OSM data for the target area;

[0026] The data conversion module is used to build a mapping table from OSM geometric models to simple geographic feature geometric models, and to convert OSM data into a common geospatial data format.

[0027] The label conversion module is used to build a mapping table between OSM labels and general land cover classification labels, and to convert OSM labels into general geographic feature labels.

[0028] The coordinate transformation module is used to convert the latitude and longitude coordinates of OSM objects into coordinates within the label map;

[0029] The sample generation module is used to rasterize OSM objects with in-map coordinates into in-map label samples.

[0030] Furthermore, the mapping table between the OSM geometric model and the simple geographic feature geometric model is shown in Table 1;

[0031] Table 1. Mapping Table from OSM Geometric Model to Simple Geographic Feature Geometric Model

[0032]

[0033] The above mapping table converts OSM data into a common geospatial data format.

[0034] Furthermore, the mapping table between OSM labels and general land cover classification labels is shown in Table 2;

[0035] Table 2. Mapping Table of OSM Labels and General Land Cover Classification Labels

[0036]

[0037] The above mapping table converts OSM labels into generic geographic feature labels.

[0038] Furthermore, the specific implementation method of the coordinate transformation module is as follows:

[0039] 1) Obtain the latitude and longitude coordinates of the starting and ending points of the target area ( ), ( ); 2) Calculate the latitude and longitude span of each pixel on the map based on the expected label map size (H, W) data. The calculation method is shown in Formula 1; 3) Traverse each coordinate point of the OSM object ( ), converting its latitude and longitude coordinates into coordinate points on the map ( The calculation method is shown in Formula 2;

[0040]

[0041] .

[0042] Compared with existing manual sample creation methods, this invention proposes an automated method for creating remote sensing land cover classification samples. Based on open-source street map data, it constructs a mapping relationship between open street maps and simple geographic feature geometric models, eliminating the discrepancy in the definition of the geometric models and providing a physical model foundation for the generation of general land cover classification samples. It also constructs a semantic model mapping relationship between open street maps and general land cover classifications, clarifying the correspondence between weakly normalized labels on open street maps and general land cover category labels, thus providing a semantic model foundation for the generation of general land cover classification samples. Furthermore, it proposes a method for converting latitude and longitude coordinate data from open street maps to coordinates within the sample map. Based on these key technologies, it achieves automated, large-scale, and high-precision land cover classification sample generation.

[0043] This invention enables the rapid generation of massive amounts of labeled samples with extremely low time and labor costs, thereby promoting the development of large-scale artificial intelligence model technology in the field of remote sensing and having extremely high application value in the fields of remote sensing technology and computer vision. Attached Figure Description

[0044] Figure 1 This is a flowchart illustrating an embodiment of the present invention;

[0045] Figure 2 This is an example of an OSM data building object of the present invention;

[0046] Figure 3 This is an example result of mapping the OSM geometric model 'way' to the simple geographic feature geometric model 'polygon' in this invention;

[0047] Figure 4 This is an example of the conversion of latitude and longitude coordinates to coordinates within a map according to the present invention;

[0048] Figure 5 This is an example of a land cover classification sample automatically generated based on an open street map according to the present invention. Detailed Implementation

[0049] The invention will now be further described with reference to the accompanying drawings.

[0050] like Figure 1 As shown in the figure, this invention proposes an automatic method for generating land cover classification label samples based on open street maps, including the following steps:

[0051] Step 1: Download the OSM data for the target area from the official OSM website (https: / / www.openstreetmap.org / export) or the OSM open-source API (http: / / overpass-api.de / ). An example of the data format is shown below. Figure 2 As shown, building number 2595 is represented in OSM by the OSM geometry object type 'way', which corresponds to the simple geographic feature geometry type 'polygon' and consists of multiple nodes 'node' (corresponding to the simple geographic feature geometry type 'point').

[0052] Step 2 involves converting the OSM geometric model into a simple geographic feature geometric model. The OSM geometric model defines three types of geometric objects: nodes, ways, and relations. The simple geographic feature specification, however, uses multiple geometric objects to define different geographic features, including points, lines, polygons, multipoints, and multipolygons. Therefore, to address the discrepancy in their geometric model definitions, the OSM geometric model needs to be converted into a generic simple geographic feature geometric model. This invention addresses this issue by constructing a mapping table between the OSM geometric model and the simple geographic feature geometric model, as shown in Table 1.

[0053] Table 1. Mapping Table from OSM Geometric Model to Simple Geographic Feature Geometric Model

[0054]

[0055] Based on the mapping table from OSM geometric models to simple geographic feature geometric models constructed in Table 1, OSM data is converted into a common geospatial data format, such as GeoJSON. An example of the converted data format is shown below. Figure 3 As shown.

[0056] Step 3: Convert OSM tags into generic geographic feature tags. OSM uses a free tagging system, meaning users can freely define and edit the names and attributes of geographic objects to describe their features. This loosely structured tag definition format leads to semantic discrepancies. For example, user A defines grassland as "grass," user B defines it as "greenfield," and user C defines it as "scrub." Although all three define the same feature category—grass—the semantic tags differ, making it difficult to create generic feature classification tag samples. Therefore, this invention constructs a mapping table between OSM tags and generic land cover classification tags, as shown in Table 2.

[0057] Table 2. Mapping Table of OSM Labels and General Land Cover Classification Labels

[0058]

[0059] In step 3, the attribute values ​​of all features on the OSM tile are retrieved. Generally, an OSM object has multiple key-value pairs, such as... Figure 2 As shown, a building has 6 key-value pairs (6 k and 6 corresponding v). The primary key (k="building") is taken as the key of the primary label, and its value (v="yes") is retrieved as the label of the OSM object. Then, referring to Table 2, the land cover classification label category (building) corresponding to the value is obtained.

[0060] Step 4: Convert OSM object latitude and longitude coordinates to in-map coordinates. Vector data downloaded from OSM is encoded using latitude and longitude coordinates for spatial location. However, existing deep learning models, influenced by convolutional neural networks, generally require input image sizes of 2 to the power of N, such as 256*256 or 512*512. Therefore, it is necessary to convert the OSM object encoded with latitude and longitude coordinates into a raster image under an in-map coordinate system to achieve high-precision label sample creation. The specific steps are: 1) Obtain the latitude and longitude coordinates of the starting and ending points of the target region (…). ), ( ); 2) Calculate the latitude and longitude span of each pixel on the map based on the expected label map size (H, W) data. The calculation method is shown in Formula 1; 3) Traverse each coordinate point of the OSM object ( ), converting its latitude and longitude coordinates into coordinate points on the map ( The calculation method is shown in Formula 2.

[0061]

[0062]

[0063] Iterate through each simple geographic feature object and convert latitude and longitude coordinates into map coordinates according to formulas (1) and (2).

[0064] Step 5, Vector OSM Object Rasterization: Traverse all OSM objects and rasterize them into a series of pixel combinations within the image based on the object's geometric feature type (e.g., polygon), ultimately generating the label sample image. That is, rasterize OSM objects with in-image coordinates into in-image label samples, as shown in the image. Figure 5 As shown.

[0065] This invention also provides an automatic land cover classification sample generation system based on open street maps, comprising the following modules:

[0066] The data acquisition module is used to acquire OSM data for the target area;

[0067] The data conversion module is used to build a mapping table from OSM geometric models to simple geographic feature geometric models, and to convert OSM data into a common geospatial data format.

[0068] The label conversion module is used to build a mapping table between OSM labels and general land cover classification labels, and to convert OSM labels into general geographic feature labels.

[0069] The coordinate transformation module is used to convert the latitude and longitude coordinates of OSM objects into coordinates within the label map;

[0070] The sample generation module is used to rasterize OSM objects with in-map coordinates into in-map label samples.

[0071] The specific implementation methods of each module are the same as those of each step, and will not be described in this invention.

[0072] The specific embodiments described herein are merely illustrative of the spirit of the invention. Those skilled in the art to which this invention pertains may make various modifications or additions to the described specific embodiments or use similar methods to substitute them, without departing from the spirit of the invention or exceeding the scope defined by the appended claims.

Claims

1. An open street map based land cover classification sample automatic generation method, characterized in that, The method comprises the following steps: Step 1, obtaining OSM data of a target area; Step 2, constructing a mapping table of OSM geometric models to simple geographic feature geometric models, and converting the OSM data into a general geospatial data format; Step 3, constructing a mapping table of OSM tags to general land cover classification tags, and converting the OSM tags into general geographic feature tags; Step 4, converting OSM object latitude and longitude coordinates into in-picture coordinates; Step 5, rasterizing the OSM object with the in-picture coordinate points into in-picture tag samples.

2. The method for automatically generating a land cover classification sample based on an open street map according to claim 1, wherein: The geometric model of OSM in step 2 defines three types of geometric object models: nodes (Node), edges (Way) and relations (Relation), while the simple geographic feature geometric model uses multiple geometric objects to define different geographic features, including points (Point), lines (Polyline), surfaces (Polygon), multi-points (MultiPoint) and multi-polygons (MultiPolygon).

3. The method for automatic generation of land cover classification samples based on open street map according to claim 2, characterized in that: The mapping table between the OSM geometric model and the simple geographic feature geometric model is shown in Table 1. Table 1. Mapping table of OSM geometric model to simple geographic feature geometric model The OSM data is converted into a general geospatial data format through the above mapping table.

4. The method for automatically generating land cover classification samples based on open street maps as described in claim 1, characterized in that: The general geospatial data format includes GeoJSON.

5. The method for automatic generation of land cover classification samples based on open street map as claimed in claim 1, wherein: The mapping table of OSM tags to general land cover classification tags is shown in Table 2. Table 2. Mapping table of OSM tags to general land cover classification tags The OSM tags are converted into general geographic feature tags through the above mapping table.

6. The method of claim 1, wherein the method further comprises: The specific implementation of step 4 is as follows: 1) Obtain the latitude and longitude coordinates of the starting and ending points of the target area ( ), ( ); 2) Calculate the latitude and longitude span of each pixel on the map based on the expected label map size (H, W) data. The calculation method is shown in Formula 1; 3) Traverse each coordinate point of the OSM object ( ), converting its latitude and longitude coordinates into coordinate points on the map ( The calculation method is shown in Formula 2; 。 7. An open street map-based land cover classification sample automatic generation system, characterized by, The method comprises the following modules: a data acquisition module for acquiring OSM data of a target area; a data conversion module for constructing a mapping table of OSM geometric models to simple geographic feature geometric models, and converting the OSM data into a general geospatial data format; a tag conversion module for constructing a mapping table of OSM tags to general land cover classification tags, and converting the OSM tags into general geographic feature tags; a coordinate conversion module for converting OSM object latitude and longitude coordinates into in-picture coordinates; a sample generation module for rasterizing the OSM object with the in-picture coordinate points into in-picture tag samples.

8. The open street map based land cover classification sample automatic generation system of claim 7, wherein: The mapping table between the OSM geometric model and the simple geographic feature geometric model is shown in Table 1. Table 1. Mapping table of OSM geometric model to simple geographic feature geometric model The OSM data is converted into a general geospatial data format through the above mapping table.

9. The system for automatic generation of land cover classification samples based on open street map as claimed in claim 7 wherein: The mapping table of OSM tags to general land cover classification tags is shown in Table 2. Table 2. Mapping table of OSM tags to general land cover classification tags The OSM tags are converted into general geographic feature tags through the above mapping table.

10. The system for automatic generation of land cover classification samples based on open street map as claimed in claim 7 wherein: The specific implementation of the coordinate conversion module is as follows: 1) Obtain the latitude and longitude coordinates of the starting and ending points of the target area ( ), ( ); 2) Calculate the latitude and longitude span of each pixel on the map based on the expected label map size (H, W) data. The calculation method is shown in Formula 1; 3) Traverse each coordinate point of the OSM object ( ), converting its latitude and longitude coordinates into coordinate points on the map ( The calculation method is shown in Formula 2; 。

Images