A map generation method under support of a customized large model

Abstract

The present application relates to the technical field of map making of geographic spatial data expression, and particularly relates to a map generation method supported by a customized large model, wherein the method comprises: establishing a cartographic demand corpus set and a structured cartographic instruction set according to cartographic demand, and constructing a map making corpus; fine-tuning the customized large model by using a fine-tuning mechanism and a cross-entropy loss function of the customized large model; collecting basic geographic information data satisfying a preset large-scale condition, and constructing vector tiles corresponding to multi-scale map making data; constructing at least one map making function comprising a data selection function, a symbol configuration function, a note placement function, a legend generation function, a scale placement function and a border decoration function, and generating a map supported by the customized large model.The present application can realize professional map instruction output of the customized large model for different map demands, and dispatches a cartographic engine to perform automatic map making, thereby greatly reducing map making time.

Description

Technical Field

[0001] This invention relates to the field of mapmaking technology for geospatial data representation, and in particular to a map generation method supported by a customized large model. Background Technology

[0002] With the advent of the information age, cartography technology has evolved from traditional manual simulation to digitalization and informatization, becoming an important tool supporting urban planning, military operations, emergency disaster relief, and public transportation. Under the wave of artificial intelligence, the cartography industry urgently needs to integrate AI technology to transform cartographic productivity and further enhance the high-quality spatiotemporal services that maps provide to various industries.

[0003] The development of cartography has evolved from manual drawing to digital mapping. However, despite continuous technological advancements, current cartography still faces numerous bottlenecks. For example, the process suffers from sparse data relationships, inefficient tools, high labor costs, and long update cycles. Furthermore, with the explosive growth of data volume, efficiently processing and utilizing massive amounts of geographic data, and establishing efficient mechanisms for representing massive crowdsourced data, have become major challenges. These issues highlight the necessity of introducing data- and knowledge-driven artificial intelligence technologies.

[0004] However, existing customized large-scale models lack an understanding of professional cartographic needs. While these models, pre-trained on massive amounts of general text corpora, can handle common natural language problems such as translation, writing, and editing, their response to professional cartographic needs remains at a superficial descriptive level and cannot support cartographic work. Current cartographic software has a high professional threshold and heavily relies on human-computer interaction. In traditional cartographic work, manually using tools to perform tasks such as data scheduling, symbolization, annotation placement, conflict resolution, and map styling is a necessary part of the process. This work mode limits the possibility of non-professionals participating in cartography, hindering the dissemination of cartographic technology; it also significantly increases cartographic work time, making it difficult to meet emergency response requirements, and urgently needs improvement. Summary of the Invention

[0005] This invention provides a map generation method supported by a customized large-scale model to address the problem that existing customized large-scale models lack understanding of professional cartographic needs. Existing customized large-scale models are pre-trained on massive amounts of general text corpora and can handle common natural language problems such as translation, writing, and editing. However, their response to professional cartographic needs remains at a superficial descriptive level and cannot support map creation. Current cartographic software has a high professional threshold and heavily relies on human-computer interaction. In traditional cartographic work, manually calling tools to perform tasks such as data scheduling, symbolization, annotation placement, conflict resolution, and map sheet finishing are necessary processes. This work mode limits the possibility of non-professionals participating in cartography, hindering the dissemination of cartographic technology. Furthermore, it significantly increases the time required for map creation, making it difficult to meet emergency response requirements.

[0006] This invention provides a map generation method supported by a customized large-scale model, comprising the following steps: establishing a cartographic requirement corpus and a structured cartographic instruction set according to map cartographic needs, and constructing a map cartographic corpus based on the cartographic requirement corpus and the structured cartographic instruction set; based on the map cartographic corpus, fine-tuning the customized large-scale model using a fine-tuning mechanism and a cross-entropy loss function until the fine-tuning stopping condition is met, thereby generating a fine-tuned customized large-scale model; based on the fine-tuned customized large-scale model, collecting basic geographic information data that meets preset large-scale conditions, and processing the data... Multi-scale derivation processing is performed on basic geographic information data with preset large-scale conditions to generate multi-scale map mapping data, and vector tiles corresponding to the multi-scale map mapping data are constructed. Based on the vector tiles, at least one map mapping function is constructed, including data selection function, symbol configuration function, annotation placement function, legend generation function, scale placement function, and border decoration function. A map mapping function library is formed according to the at least one map mapping function, and a map template library is generated according to different map types. The map mapping function library and the map template library are coupled to generate a map supported by the customized large model.

[0007] Optionally, in one embodiment of the present invention, the step of fine-tuning the customized large model using the fine-tuning mechanism of the customized large model and the cross-entropy loss function until the fine-tuning stopping condition is met, and generating the fine-tuned customized large model, includes: adjusting the model weights based on the low-rank adaptation mechanism of the customized large model through incremental learning of the original weights of the customized large model to generate adjusted weights; measuring the difference between the output instructions of the customized large model and the structured mapping instructions using the cross-entropy loss function based on the adjusted weights; and fine-tuning the customized large model according to the fine-tuning mechanism and the cross-entropy loss function based on the difference to generate the fine-tuned customized large model.

[0008] Optionally, in one embodiment of the present invention, the formula for calculating the original weights is:

[0009]

[0010] in, Let be the set of real numbers. and These are the row and column dimensions of the weight matrix, respectively.

[0011] The formula for calculating the cross-entropy loss function is as follows:

[0012]

[0013] in, For structured drawing instructions. The probability of the mapping instructions output by the model. To train the number of mapping instructions. For the first One drawing instruction.

[0014] Optionally, in one embodiment of the present invention, after collecting the basic geographic information data that meets the preset large-scale conditions, the method further includes: determining the basic geographic information data that meets the preset large-scale conditions based on at least one map element among administrative divisions, buildings, roads, water systems, and annotations; and uniformly managing the geographic information data that meets the preset large-scale conditions in a preset data format to generate basic geographic information data that meets the target large-scale conditions.

[0015] Optionally, in one embodiment of the present invention, the step of performing multi-scale derivation processing on the basic geographic information data that meets the preset large-scale conditions to generate multi-scale map mapping data includes: merging, typicalizing, and simplifying the building data in the basic geographic information data that meets the preset large-scale conditions to generate mapping data that meets the first preset large-scale conditions; and selecting and simplifying the road data in the basic geographic information data that meets the preset large-scale conditions to generate mapping data that meets the second preset large-scale conditions.

[0016] The water system data in the basic geographic information data that meets the preset large-scale conditions are processed by merging lakes, centralizing double-line rivers, selecting water systems, and simplifying outlines to generate mapping data that meets the third preset large-scale condition. The annotations in the basic geographic information data that meets the preset large-scale conditions are processed by deleting related information and checking feature derivation consistency to generate mapping data that meets the fourth preset large-scale condition. Based on the mapping data that meets the first preset large-scale condition, the second preset large-scale condition, the third preset large-scale condition, the fourth preset large-scale condition, the preset medium-scale condition, and the preset small-scale condition, the multi-scale map mapping data is generated.

[0017] Optionally, in one embodiment of the present invention, after constructing a map-making function including at least one of a data selection function, a symbol configuration function, an annotation placement function, a legend generation function, a scale placement function, and a border finishing function, the method further includes: scheduling the multi-scale map-making data using the data selection function and planning the hierarchical order of the multi-scale map-making data; configuring the expression symbols for different map elements using the symbol configuration function based on the hierarchical order; and annotating at least one element including administrative divisions, roads, and water systems using the annotation placement function based on the expression symbols. The process involves generating placed annotations; based on these annotations, a map legend is generated using the legend generation function according to map element information, and the layout and position of the map legend are calculated based on map representation information; based on the layout and position of the map legend, a map scale is calculated using the scale placement function according to the actual size of the map data and the size of the representation window, and the scale style of the map scale is placed to generate a placed scale style; based on the placed scale style, different map decoration borders are generated using the border decoration function according to the size of the map representation content window.

[0018] Optionally, in one embodiment of the present invention, the step of generating a map template library according to different map types includes: determining the base map elements of the map based on the administrative divisions of the map, and generating an administrative division map in the map based on the base map elements and the order of buildings-water systems-roads-annotations; emphasizing the highways, railways, or water systems in the map mapping requirements with symbols to generate important route maps in the map; determining the base map of the map based on high-resolution remote sensing imagery, and expressing the map by overlaying roads, water systems, and annotations based on the base map to generate a remote sensing image map in the map; determining the base map of the map based on the shading expression of digital elevation model data, and expressing the map by overlaying roads, water systems, and annotations based on the base map to generate a topographic map in the map; and generating the map template library based on the administrative division map, the important route maps, the remote sensing image map, and the topographic map.

[0019] This invention addresses the challenges of mapping requirements by summarizing and categorizing them, constructing structured mapping instructions, and fine-tuning a customized large-scale model. This allows the model to output professional map instructions for different map needs, supporting relevant mapping functions to generate maps from geographic information data. Furthermore, this invention supports a natural language-driven map generation workflow, where users can verbally or by typing mapping requirements. The customized large-scale model then parses these requirements and schedules a mapping engine for automated map generation. This workflow facilitates map creation for non-specialist users and significantly reduces mapping time. This addresses the shortcomings of existing customized large-scale models, which are primarily trained on massive amounts of general natural language samples. The lack of specialized mapping data in the training corpus hinders the models' ability to provide standardized parsing and responses to mapping requirements. Current map creation systems suffer from significant deficiencies in requirement understanding and response speed. The lack of effective intelligent requirement understanding methods and automated mapping frameworks means the mapping process still relies on manual intervention, leading to low efficiency and difficulty in meeting diverse user needs.

[0020] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0021] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein:

[0022] Figure 1 A flowchart illustrating a map generation method supported by a customized large model according to an embodiment of the present invention;

[0023] Figure 2This is an overall flowchart of a map generation method supported by a customized large model according to an embodiment of the present invention. Detailed Implementation

[0024] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0025] The following description, with reference to the accompanying drawings, illustrates a map generation method supported by a customized large model, according to an embodiment of the present invention. The existing customized large-scale models mentioned in the background are mainly trained based on massive amounts of general natural language samples. Because the training corpus lacks content specific to the cartographic field, these models struggle to provide standardized analysis and responses to cartographic requirements. Current cartographic systems suffer from significant deficiencies in requirement understanding and response speed. Due to the lack of effective intelligent requirement understanding methods and automated cartographic frameworks, the cartographic process still relies on manual intervention, resulting in low efficiency and difficulty in meeting diverse user needs. This invention provides a map generation method supported by a customized large-scale model. This method involves summarizing and organizing cartographic requirements, constructing structured cartographic instructions, and fine-tuning the customized large-scale model to output professional map instructions for different map requirements. These instructions can support relevant cartographic functions to generate maps from geographic information data. This invention supports a natural language-driven map generation workflow, where users can verbally or manually input cartographic requirements. The customized large-scale model then analyzes these requirements and schedules a cartographic engine for automated map generation. This workflow facilitates participation in map creation by non-specialist personnel and significantly reduces map creation time. This addresses the shortcomings of existing customized large-scale models, which are primarily trained on massive amounts of general natural language samples. Because the training corpus lacks content specific to the cartographic field, these models struggle to provide standardized analyses and responses to cartographic requirements. Furthermore, current cartographic systems suffer from significant deficiencies in requirement understanding and response speed. The lack of effective intelligent requirement understanding methods and automated cartographic frameworks means the cartographic process still relies on manual intervention, leading to low efficiency, difficulty in meeting diverse user needs, high technical barriers to entry, and poor flexibility.

[0026] Specifically, Figure 1 This is a flowchart illustrating a map generation method supported by a customized large model, as provided in an embodiment of the present invention.

[0027] like Figure 1 As shown, the map generation method supported by this customized large model includes the following steps:

[0028] In step S101, a cartographic requirement corpus and a structured cartographic instruction set are established according to the map cartographic requirements, and a map cartographic corpus is constructed based on the cartographic requirement corpus and the structured cartographic instruction set.

[0029] It is understandable that, such as Figure 2 As shown, in the inference application stage of this invention, the map generation method supported by a customized large model interprets the mapping requirements using the customized large model, outputs structured mapping instructions, schedules mapping data according to the mapping instructions, and simultaneously calls mapping tools to symbolically represent the mapping data, thus completing the map mapping action. The underlying construction stage involves four stages: mapping corpus creation, large model fine-tuning, mapping data organization, and mapping function construction. The following will describe the map generation method supported by a customized large model, focusing on these four stages of underlying construction.

[0030] In practical implementation, embodiments of the present invention can create a mapping corpus. First, mapping requirements are summarized and a mapping requirement corpus is built, which is used as the input dataset for pre-training customized large models. Then, structured mapping instructions corresponding to the mapping requirements are designed, and a set of structured mapping instructions corresponding to the mapping requirements is established, which is used as the output labels for fine-tuning training of customized large models.

[0031] Specifically, embodiments of the present invention can organize descriptive corpora of all mapping requirements. Taking the drawing of an administrative division map of Wuhan City as an example, firstly, the descriptive statement of the mapping requirement, "draw an administrative division map of Wuhan City," is obtained; then, the sentence components are segmented, and a general descriptive formula for expressing the mapping requirement is summarized as {cartographic action: drawing; mapping scope: Wuhan City; map type: administrative division map}; finally, the mapping corpus is enhanced based on the general descriptive formula. For example, the description of the mapping action can be replaced with "create," "produce," "generate," etc.; the description of the mapping scope can be replaced with "Hubei Province," "Guangzhou City," "Wuchang District," etc.; and the map type can be replaced with "administrative division map," "topographic map," "remote sensing image map," etc. Therefore, from the mapping requirement of "drawing an administrative division map of Wuhan City," through the arrangement and combination of sentence components, descriptive corpora of mapping requirements such as "generating an administrative division map of Wuhan City," "drawing a topographic map of Guangzhou City," and "producing a remote sensing image map of Hubei Province" can be established, and a mapping requirement corpus set can be constructed according to this rule.

[0032] Furthermore, embodiments of the present invention can construct structured mapping instructions for each descriptive prediction, i.e., output labels during the fine-tuning of the large model and output templates during the inference process. Taking the mapping requirement of "drawing an administrative division map of Wuhan City" as an example, the labels corresponding to this corpus can be parsed into a JSON dictionary: {function: draw_map, map_type: administrative division map, area: Wuhan City, color: default, feature_sorted: default}. In this dictionary, "cttype" specifies the type of mapping action, including actions such as creating a new map (draw_map), modifying map features (modify_feature), and changing map color (change_mapcolor). This allows the construction of a structured mapping instruction set, and simultaneously establishes the correspondence between mapping requirements and structured mapping instructions.

[0033] This invention supports a natural language-driven map generation workflow, where users can verbally or by typing their mapping requirements. The requirements are then analyzed using a customized large-scale model, and a mapping engine is scheduled to automate the map generation process. This workflow facilitates map creation for users without a professional background and significantly reduces map generation time. Furthermore, this invention can respond to user mapping needs in real time, enabling the provision of customized and personalized map generation services.

[0034] In step S102, based on the map cartographic corpus, the customized large model is fine-tuned using the fine-tuning mechanism of the customized large model and the cross-entropy loss function until the fine-tuning stopping condition is met, and the fine-tuned customized large model is generated.

[0035] It is understood that the customized large model in the embodiments of the present invention can be fine-tuned and trained using map cartography vertical domain data on the basis of the pre-trained large model, and some parameters of the model can be adjusted to adapt the model to the map cartography task.

[0036] In practical implementation, embodiments of the present invention can perform customized large-scale model fine-tuning based on a cartographic corpus to meet the needs of cartographic professionals. This process includes three parts: design of a large-scale model fine-tuning mechanism, design of a loss function, and model fine-tuning training. The present invention can use the customized large-scale model fine-tuning mechanism and cross-entropy loss function to fine-tune the customized large-scale model until the fine-tuning stopping condition is met, thus generating the fine-tuned customized large-scale model.

[0037] This invention can summarize and organize map-making requirements, construct structured map-making instructions, fine-tune and train customized large models, and achieve professional map instruction output for different map requirements by customized large models. These instructions can support relevant map-making functions to obtain maps from geographic information data.

[0038] It's important to note that customized large-scale models, trained through large-scale natural language processing tasks, are capable of deeply understanding geographic semantics and generating structured output. This technology offers dual advantages in the field of cartography: firstly, its powerful natural language processing capabilities can accurately analyze user cartographic needs and automatically generate spatial relationship descriptions of map elements; secondly, through professional-domain fine-tuning, it can intelligently invoke professional cartographic functions such as geographic coordinate transformation and topological relationship verification. This capability will drive cartography towards intelligent understanding of needs, customized data expression, and flexible tool application, significantly improving the personalization and automation levels of cartography.

[0039] Optionally, in one embodiment of the present invention, the customized large model is fine-tuned using a fine-tuning mechanism and a cross-entropy loss function until the fine-tuning stopping condition is met, thereby generating a fine-tuned customized large model. This includes: adjusting the model weights based on the incremental learning of the original weights of the customized large model using a low-rank adaptation mechanism to generate adjusted weights; measuring the difference between the output instructions of the customized large model and the structured mapping instructions using a cross-entropy loss function based on the adjusted weights; and fine-tuning the customized large model based on the difference using the fine-tuning mechanism and the cross-entropy loss function to generate a fine-tuned customized large model.

[0040] Understandably, pre-trained large models learn the grammar, semantics, pragmatics, and other aspects of language by training on massive amounts of general text data. They can generate natural and fluent text content and solve general natural language processing problems, but they are difficult to accurately answer questions in specialized fields.

[0041] In actual implementation, embodiments of the present invention may include the following steps:

[0042] (1) Design of a fine-tuning mechanism for large models. This invention employs a low-rank adaptation mechanism (LoRA) for large models. This mechanism adjusts the model weights through incremental learning of the original weights. In one embodiment of this invention, the original weights are calculated using the following formula:

[0043]

[0044] in, Let be the set of real numbers. and These represent the row and column dimensions of the weight matrix, respectively.

[0045] Furthermore, the incremental weights for learning are expressed as: ,in i and j These are the row and column dimensions of the weight matrix, respectively. It can be expressed as the product of two matrices:

[0046]

[0047] in, A and B All are matrices. and , can be obtained .

[0048] Freeze the weights of the pre-trained large model during fine-tuning training. ,right A and B Matrix expansion training is sufficient; the updated output will then be obtained. h Expressed as:

[0049]

[0050] in, A and B All are matrices. For the original weights, Incremental weights for learning, The output of the previous stage is the embedding vector for the mapping requirements in the input layer.

[0051] (2) In this embodiment of the invention, a loss function for the customized large model fine-tuning process can be designed. Here, the cross-entropy loss function is used to measure the difference between the large model output instructions and the structured plotting instructions. The calculation formula of the cross-entropy loss function is:

[0052]

[0053] in, For structured drawing instructions. The probability of the mapping instructions output by the model. To train the number of mapping instructions. For the first One drawing instruction.

[0054] (3) This invention can perform customized large-scale model fine-tuning. This invention can select DeepSeek-R1-Distill-Qwen-7B as the pre-trained large language model, combine the LoRA (Low-Rank Adaptation, fine-tuning large pre-trained models) fine-tuning mechanism and the cross-entropy loss function, adopt the adaptive optimization Adam (Adaptive Moment Estimation, deep learning optimization algorithm) function, and use LLaMA-Factory (large language model factory) to carry out customized large-scale model fine-tuning with the understanding of cartographic requirements. Specifically, large-scale model fine-tuning involves continuing training on the pre-trained large model using a small amount of vertical domain data for map cartography, adjusting some parameters to adapt the model to map cartography tasks.

[0055] The embodiments of the present invention can carry out customized vertical fine-tuning training of large models, enabling them to understand professional cartographic requirements, thereby achieving rapid and intelligent understanding and response to map cartographic needs.

[0056] In step S103, based on the fine-tuned customized large model, basic geographic information data that meets the preset large scale conditions are collected, and multi-scale derivation processing is performed on the basic geographic information data that meets the preset large scale conditions to generate multi-scale map mapping data, and vector tiles corresponding to the multi-scale map mapping data are constructed.

[0057] Specifically, embodiments of the present invention can organize the cartographic database hierarchically. First, large-scale basic geographic information data is collected and cleaned; then, multi-scale derivation processing is performed on the large-scale geographic information data; finally, vector tiles of multi-scale, multi-element map cartographic data are constructed.

[0058] Optionally, in one embodiment of the present invention, after collecting basic geographic information data that meets the preset large-scale conditions, the method further includes: determining basic geographic information data that meets the preset large-scale conditions based on at least one map element among administrative divisions, buildings, roads, water systems, and annotations; and uniformly managing the geographic information data that meets the preset large-scale conditions in a preset data format to generate basic geographic information data that meets the target large-scale conditions.

[0059] It is understood that the target large-scale condition in the embodiments of the present invention can be the condition that the attributes are correct and the elements are complete.

[0060] In practical implementation, this invention can collect and clean large-scale cartographic data. Large-scale basic geographic information data is obtained from sources such as the National Geographic Information Public Service Platform and OpenStreetMap. The map elements used in this invention are administrative divisions, buildings, roads, water systems, and annotations. Large-scale cartographic data is uniformly managed using the GeoJSON (a format for encoding geographic data structures) data format. Simultaneously, issues such as data integrity, data attribute ambiguity, data duplication, and annotation association are detected and corrected. Through these steps, large-scale cartographic data with correct attributes and complete elements is constructed.

[0061] It should be noted that the preset large-scale conditions and the target large-scale conditions can be set by those skilled in the art according to the actual situation, and no specific restrictions are imposed here.

[0062] Optionally, in one embodiment of the present invention, multi-scale derivation processing is performed on basic geographic information data that meets preset large-scale conditions to generate multi-scale map mapping data, including: merging, typicalizing, and simplifying the building data in the basic geographic information data that meets the preset large-scale conditions to generate mapping data that meets the first preset large-scale conditions; selecting and simplifying the road data in the basic geographic information data that meets the preset large-scale conditions to generate mapping data that meets the second preset large-scale conditions; and merging lakes and simplifying the water system data in the basic geographic information data that meets the preset large-scale conditions. The process involves centralizing the river, selecting water systems, and simplifying outlines to generate mapping data that meets the third preset large-scale condition. Then, annotations in the basic geographic information data that meet the preset large-scale condition are deleted based on their relevance, and feature derivation consistency is checked to generate mapping data that meets the fourth preset large-scale condition. Based on the mapping data that meet the first, second, third, and fourth preset large-scale conditions, the mapping data that meet the preset medium-scale condition, and the mapping data that meet the preset small-scale condition, multi-scale map mapping data is generated.

[0063] In actual implementation, the embodiments of the present invention can perform multi-scale derivation processing on large-scale mapping data, including merging, typicalizing, and simplifying the outline of building data; selecting and simplifying road data; merging lakes, centralizing double-line rivers, selecting water systems, and simplifying the outline of water system data; deleting annotations based on their relevance; and performing derivation operations and checks such as element derivation consistency detection. Finally, medium-scale mapping data and small-scale mapping data are established.

[0064] Furthermore, embodiments of the present invention can construct vector tiles for multi-scale mapping data. The "vector tiling" function provided by QGIS (Geographic Information System) is used to construct vector tiles for mapping data at various levels, generating a unique identifier for each tile, and storing the generated multi-scale mapping data vector tiles in the file system.

[0065] It should be noted that the first preset large-scale condition, the second preset large-scale condition, the third preset large-scale condition, the fourth preset large-scale condition, the preset medium-scale condition, and the preset medium-scale condition can be set by those skilled in the art according to the actual situation, and no specific restrictions are imposed here.

[0066] In step S104, based on vector tiles, a map cartographic function is constructed that includes at least one of the following: data selection function, symbol configuration function, annotation placement function, legend generation function, scale placement function, and border finishing function. A cartographic function library is formed based on at least one map cartographic function, and a map template library is generated according to different map types. The cartographic function library and the map template library are coupled to generate a map supported by a customized large model.

[0067] It is understood that the embodiments of the present invention establish a natural language-driven map-making mode, supporting natural language-driven map-making. This model can accurately analyze user needs and intelligently call professional map-making tools such as projection definition, data processing, symbol expression, color matching, annotation placement, and map sheet decoration. The map generation in the present invention is different from the traditional human-computer interaction map-making mode. It integrates a large language model and a map-making engine to realize a natural language-driven map-making model. The large language model is a very large-scale artificial neural network model based on natural language text training and learning. It usually contains billions or even hundreds of billions of parameters.

[0068] In practical implementation, embodiments of this invention can design map cartographic functions. First, map cartographic functions such as data selection, symbol configuration, annotation placement, map generation, scale bar placement, and border finishing are constructed. Then, map templates are designed for different map types, and the cartographic function library and map template library are coupled to generate maps supported by customized large-scale models. This invention, by coupling a professionally customized cartographic function library with cartographic data, forms a map generation capability driven by customized large-scale models.

[0069] This invention overcomes the limitations of shallow parsing of professional cartographic commands in general-purpose customized large models by constructing a professional cartographic corpus and using command fine-tuning technology. It accurately transforms natural language requirements into structured cartographic function parameters (such as cartographic behavior and map types), enabling automated responses to professional cartographic logic such as data scheduling, symbol expression, annotation placement, and map sheet embellishment. This invention significantly improves the personalization and automation of mapmaking, solving problems such as sparse data relationships, low tool efficiency, high labor costs, and long update cycles in existing mapmaking processes. It meets diverse user needs and provides more efficient and intelligent map services for urban planning, military operations, emergency disaster relief, and public transportation.

[0070] Optionally, in one embodiment of the present invention, after constructing at least one of the following map-making functions—data selection, symbol configuration, annotation placement, legend generation, scale placement, and border embellishment—the method further includes: using the data selection function to schedule multi-scale map-making data and plan the hierarchical order of the multi-scale map-making data; using the symbol configuration function to configure the expression symbols for different map elements based on the hierarchical order; using the annotation placement function to place annotations for at least one element including administrative divisions, roads, and water systems based on the expression symbols, to generate placed annotations; using the legend generation function to generate a map legend based on the map element information and calculate the layout and position of the map legend based on the map expression information; using the scale placement function to calculate the map scale based on the actual size of the map data and the size of the expression window based on the layout and position of the map legend, and placing the scale style of the map scale to generate a placed scale style; and using the border embellishment function to generate different map embellishment borders based on the size of the map expression content window based on the placed scale style.

[0071] In this embodiment of the invention, basic map-making functions can be constructed. The data selection function `data_select_function(*)` is responsible for scheduling map data at different scales and planning the hierarchical order of map data representation; the symbol configuration function `symbol_configuration_function(*)` is responsible for configuring the representation symbols for different map elements and can adaptively represent symbols for different map types; the annotation placement function `lanel_placement_function(*)` is responsible for placing annotations for elements such as administrative divisions, roads, and water systems, and handling annotation conflicts; the legend generation function `legend_generation_function(*)` is responsible for generating map legends based on map element information and calculating appropriate legend layouts and positions based on map representation information; the scale placement function `scale_placement_function(*)` calculates the map scale based on the actual size of the map data and the size of the representation window, determines the scale style, and places it; the border finishing function `appearance_finishing_function(*)` is responsible for generating different map finishing borders based on the size of the map representation content window. The above map-making functions are based on MapBox's map-making functions. This step also calls basic spatial analysis tools provided by MapBox, such as buffer analysis and overlay calculation.

[0072] The embodiments of the present invention can rely on the deep integration of customized large models and mapping engines to establish a natural language-driven map generation chain. Users can trigger the mapping function chain to automatically execute data scheduling, symbolization and finishing processes through spoken commands (such as "generate remote sensing image map of Nanjing City"), which greatly reduces the professional threshold and improves the efficiency of emergency mapping.

[0073] Optionally, in one embodiment of the present invention, generating a map template library according to different map types includes: determining the base map elements of the map based on the administrative divisions of the map, and generating an administrative division map in the map by overlaying the base map elements and buildings-water systems-roads-annotations in that order; emphasizing highways, railways, or water systems in the map-making requirements with symbols to generate important route maps in the map; determining the base map of the map based on high-resolution remote sensing imagery, and overlaying roads, water systems, and annotations on the base map to generate a remote sensing image map in the map; determining the base map of the map based on the shading expression of digital elevation model data, and overlaying roads, water systems, and annotations on the base map to generate a topographic map in the map; and generating a map template library based on the administrative division map, important route maps, remote sensing image maps, and topographic maps.

[0074] In practical implementation, embodiments of this invention can construct a map template library composed of map mapping functions and encapsulations of different parameters. The map template library constructed by this invention includes administrative division maps, important route maps, remote sensing image maps, topographic maps, and a map atlas template library. The administrative division map template library uses administrative divisions as the base map element, superimposing them in the order of buildings-water systems-roads-annotations, with symbol representation designed according to the corresponding map mapping standards. The important route map template is based on the administrative division map template, emphasizing important highways, railways, or water systems required in the mapping needs through symbolic salience. For example, in the mapping requirement of "drawing a map along the Beijing-Guangzhou Railway," the Beijing-Guangzhou Railway is highlighted, presenting the administrative divisions it passes through. The remote sensing image map template uses high-resolution remote sensing imagery as the base map, superimposing roads, water systems, and annotations. The topographic map template uses a shaded representation of DEM (Digital Elevation Model) data as the base map, superimposing roads, water systems, and annotations. Atlases are based on maps with hierarchical or containment relationships, and are constructed through unit-based cartography. For example, for an administrative division map of Hubei Province, a Hubei Province municipal-level administrative division atlas is constructed using city-level administrative division maps as units; for maps along the Beijing-Guangzhou Railway, a provincial-level administrative division atlas along the Beijing-Guangzhou Railway is constructed using provincial-level administrative division maps along the railway as units. The above map templates are controlled and invoked by the "map_type" keyword in the cartographic instructions output from the customized large model.

[0075] The map generation method supported by a customized large-scale model proposed in this invention can summarize and generalize map-making requirements, construct structured map-making instructions, and fine-tune and train the customized large-scale model to achieve professional map instruction output for different map requirements. These instructions can support relevant map-making functions to generate maps from geographic information data. This invention supports a natural language-driven map generation operation mode, where map-making requirements are verbally stated or entered, and the customized large-scale model parses the requirements and schedules the map-making engine for automated map generation. This operation mode helps non-professionals participate in map generation and significantly reduces map generation time. Therefore, it solves the problem that existing customized large-scale models are mainly trained on massive amounts of general natural language samples. Due to the lack of content in the professional field of map-making in the training corpus, existing customized large-scale models are unable to provide standardized parsing and responses to map-making requirements. Current map-making systems have significant shortcomings in requirement understanding and response speed. Due to the lack of effective intelligent requirement understanding methods and automated map-making frameworks, the map-making process still relies on manual intervention, resulting in low efficiency and difficulty in meeting diverse user needs.

[0076] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0077] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "N" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0078] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or N executable instructions for implementing custom logic functions or processes, and the scope of preferred embodiments of the invention includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as will be understood by those skilled in the art to which embodiments of the invention pertain.

Claims

1. A map generation method supported by a customized large model, characterized in that, Includes the following steps: A cartographic requirement corpus and a structured cartographic instruction set are established based on the cartographic requirement corpus and the structured cartographic instruction set, and a map cartographic corpus is constructed based on the cartographic requirement corpus and the structured cartographic instruction set. Based on the aforementioned map cartographic corpus, the customized large model is fine-tuned using the fine-tuning mechanism of the customized large model and the cross-entropy loss function until the fine-tuning stopping condition is met, thereby generating the fine-tuned customized large model. The step of fine-tuning the customized large model using its fine-tuning mechanism and cross-entropy loss function until a fine-tuning stopping condition is met, and then generating the fine-tuned customized large model, includes: adjusting the model weights based on the low-rank adaptation mechanism of the customized large model through incremental learning of the original weights to generate adjusted weights; measuring the difference between the output instructions of the customized large model and the structured mapping instructions using the cross-entropy loss function based on the adjusted weights; and fine-tuning the customized large model based on the difference according to the fine-tuning mechanism and the cross-entropy loss function to generate the fine-tuned customized large model. The formula for calculating the original weights is: in, Let be the set of real numbers. and These are the row and column dimensions of the weight matrix, respectively. The formula for calculating the cross-entropy loss function is as follows: in, This is a label for structured drawing instructions. The probability of the mapping instructions output by the model. To train the number of mapping instructions. For the first One drafting instruction; Based on the finely tuned customized large model, basic geographic information data that meets the preset large scale conditions are collected, and multi-scale derivation processing is performed on the basic geographic information data that meets the preset large scale conditions to generate multi-scale map mapping data, and vector tiles corresponding to the multi-scale map mapping data are constructed. The process includes, after collecting the basic geographic information data that meets the preset large-scale conditions, determining the basic geographic information data that meets the preset large-scale conditions based on at least one map element among administrative divisions, buildings, roads, water systems, and annotations; and uniformly managing the geographic information data that meets the preset large-scale conditions in a preset data format to generate basic geographic information data that meets the target large-scale conditions. The method of uniformly managing the geographic information data that meets the preset large-scale conditions in a preset data format to generate basic geographic information data that meets the target large-scale conditions includes: uniformly managing the geographic information data that meets the preset large-scale conditions in GeoJSON data format, while detecting and correcting data integrity issues, data attribute ambiguity issues, data duplication issues, and annotation association issues, so as to construct large-scale mapping data with correct attributes and complete features. The multi-scale derivation processing of the basic geographic information data that meets the preset large-scale conditions to generate multi-scale map data includes: merging, typicalizing, and simplifying the building data in the basic geographic information data that meets the preset large-scale conditions to generate map data that meets the first preset large-scale conditions; selecting and simplifying the road data in the basic geographic information data that meets the preset large-scale conditions to generate map data that meets the second preset large-scale conditions; and merging lakes, centralizing double-line rivers, and selecting water system data in the basic geographic information data that meets the preset large-scale conditions. The mapping data is simplified by extracting and outlining to generate mapping data that meets the third preset large-scale condition; the annotations in the basic geographic information data that meets the preset large-scale condition are deleted for relevance and the feature derivation consistency is checked to generate mapping data that meets the fourth preset large-scale condition; based on the mapping data that meets the first preset large-scale condition, the mapping data that meets the second preset large-scale condition, the mapping data that meets the third preset large-scale condition, the mapping data that meets the fourth preset large-scale condition, the mapping data that meets the preset medium-scale condition, and the mapping data that meets the preset small-scale condition, the multi-scale map mapping data is generated. Based on the vector tiles, a map cartography function is constructed, which includes at least one of the following: data selection function, symbol configuration function, annotation placement function, legend generation function, scale placement function, and border decoration function. A cartography function library is formed according to the at least one map cartography function, and a map template library is generated according to different map types. The cartography function library and the map template library are coupled to generate a map supported by the customized large model. The step of generating a map template library based on different map types includes: determining the base map elements of the map based on the administrative divisions of the map, and generating an administrative division map in the map by overlaying the base map elements and buildings-water systems-roads-annotations in that order; emphasizing highways, railways, or water systems in the map-making requirements with symbols to generate important route maps in the map; determining the base map of the map based on high-resolution remote sensing imagery, and overlaying roads, water systems, and annotations on the base map to generate a remote sensing image map in the map; determining the base map of the map based on the shading representation of digital elevation model data, and overlaying roads, water systems, and annotations on the base map to generate a topographic map in the map; and generating the map template library based on the administrative division map, the important route maps, the remote sensing image map, and the topographic map. This also includes: relying on the deep integration of the customized large model and the mapping engine, establishing a natural language-driven map generation chain, where users can trigger the mapping function chain through spoken commands to automatically execute data scheduling, symbolization, and finishing processes.

2. The map generation method supported by a customized large model according to claim 1, characterized in that, After constructing at least one of the following map-making functions: data selection function, symbol configuration function, annotation placement function, legend generation function, scale bar placement function, and border finishing function, it also includes: The data selection function is used to schedule the multi-scale map data and plan the hierarchical order of the multi-scale map data. Based on the hierarchical order, the symbol configuration function is used to configure the expression symbols for different map elements; Based on the expression symbols, the annotation placement function is used to place annotations on at least one element including administrative divisions, roads, and water systems to generate the placed annotations; Based on the placed annotations, the legend generation function is used to generate a map legend according to the map element information on the map, and the layout and position of the map legend are calculated according to the map expression information; Based on the layout and position of the map legend, the map scale is calculated using the scale placement function according to the actual size of the map data and the size of the expression window, and the scale style of the map scale is placed to generate the placed scale style. Based on the placed scale style, the border decoration function generates different map decoration borders according to the size of the map content window.

Images