Method and system for generating GIS buffer based on fuzzy semantics and geographic constraints

Abstract

The application provides a GIS buffer generation method and system based on fuzzy semantics and geographic constraints, and belongs to the technical field of geographic information systems. The method comprises the following steps: inputting fuzzy description words, membership values and geographic entity types into a pre-trained language-space mapping model to obtain a basic distance parameter; querying a spatial database based on geographic entity keywords to obtain geometric features associated with the geographic entity and calculate the feature length corresponding to the geometric features; correcting the basic distance parameter based on the intention direction expressed by the fuzzy description words, the membership values, the geographic entity types and the feature length to obtain a geographic distance threshold; generating an initial two-dimensional buffer polygon based on the geographic distance threshold and the spatial geometric features of the geographic entity; and querying a digital elevation model based on the initial two-dimensional buffer polygon to obtain terrain data, and generating a three-dimensional buffer surface through a slope-sensitive interpolation algorithm. The application improves the accuracy of GIS buffer generation.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of geographic information system, and in particular to a GIS buffer zone generation method and system based on fuzzy semantics and geographic constraints. BACKGROUND

[0002] In the field of GIS spatial analysis, buffer zone generation technology is crucial, as it can effectively define the influence range of geographic entities and is widely used in urban planning, resource management, and many other fields. Known buffer zone generation methods struggle when faced with fuzzy semantics such as "near" and "around" expressed by users in natural language. Since these methods highly depend on precise distance parameters, they are difficult to convert fuzzy semantics into precise spatial parameters, resulting in a significant semantic gap between user demand and system execution, which severely hinders efficient human-computer interaction.

[0003] At the same time, existing methods often overlook key factors such as the scale of geographic entities and terrain features. Different geographic entities, such as roads of varying lengths and cities of different areas, have significant scale differences, and diverse terrains such as mountains and plains have a significant impact on the actual coverage of buffer zones. However, known methods do not fully consider these factors when generating buffer zones, resulting in significant deviations between the generated results and the real geographic scene.

[0004] With the increasing demand for intelligent interaction, how to accurately convert fuzzy semantics into precise spatial parameters and fully integrate geographic constraints to generate more realistic buffer zones has become a core and critical issue in improving the efficiency of GIS applications. Therefore, a GIS buffer zone generation method and system based on fuzzy semantics and geographic constraints are needed to better understand user natural language instructions and improve the accuracy of GIS buffer zone generation. SUMMARY

[0005] To solve the above technical problems, the present application provides a GIS buffer zone generation method and system based on fuzzy semantics and geographic constraints.

[0006] The first aspect of the embodiments of the present application provides a GIS buffer zone generation method based on fuzzy semantics and geographic constraints, comprising:

[0007] extracting the geographic entity keywords, geographic entity types, and fuzzy descriptors in the user's natural language instructions through a semantic analysis engine, and analyzing the fuzzy descriptors to obtain membership values corresponding to the fuzzy descriptors;

[0008] inputting the fuzzy descriptors, membership values, and geographic entity types into a pre-trained language-space mapping model to obtain a basic distance parameter, the basic distance parameter being a reference value with a distance dimension or a proportion factor relative to the typical scale of the entity;

[0009] query a spatial database based on the geographic entity keyword to obtain geometric features associated with the geographic entity, and calculate a feature length corresponding to the geometric features;

[0010] correct the basic distance parameter based on the intention direction, the membership value, the geographic entity type, and the feature length expressed by the fuzzy description word to obtain a geographic distance threshold;

[0011] generate an initial two-dimensional buffer polygon based on the geographic distance threshold and the spatial geometric features of the geographic entity;

[0012] query a digital elevation model based on the initial two-dimensional buffer polygon to obtain terrain data, and generate a three-dimensional buffer curved surface through a slope-sensitive interpolation algorithm.

[0013] In a second aspect, the embodiment of the application provides a GIS buffer generation system based on fuzzy semantics and geographic constraints, which comprises:

[0014] a semantic analysis module configured to extract a geographic entity keyword, a geographic entity type, and a fuzzy description word in a natural language instruction of a user through a semantic analysis engine, and analyze the fuzzy description word to obtain a membership value corresponding to the fuzzy description word;

[0015] a data mapping module configured to input the fuzzy description word, the membership value, and the geographic entity type into a pre-trained language-space mapping model to obtain a basic distance parameter, the basic distance parameter being a reference value with a distance dimension or a proportional factor relative to a typical scale of an entity;

[0016] a data calculation module configured to query a spatial database based on the geographic entity keyword to obtain geometric features associated with the geographic entity, and calculate a feature length corresponding to the geometric features;

[0017] a data correction module configured to correct the basic distance parameter based on the intention direction, the membership value, the geographic entity type, and the feature length expressed by the fuzzy description word to obtain a geographic distance threshold;

[0018] a first generation module configured to generate an initial two-dimensional buffer polygon based on the geographic distance threshold and the spatial geometric features of the geographic entity;

[0019] a second generation module configured to query a digital elevation model based on the initial two-dimensional buffer polygon to obtain terrain data, and generate a three-dimensional buffer curved surface through a slope-sensitive interpolation algorithm.

[0020] In a third aspect, an electronic device is provided, which includes a memory, a processor, and a computer program stored in the memory and running on the processor, and the processor implements the steps of the above-mentioned GIS buffer generation method based on fuzzy semantics and geographic constraints when running the computer program.

[0021] In a fourth aspect, a computer readable storage medium is provided, which stores a computer program, and the computer program implements the steps of the above-mentioned GIS buffer generation system based on fuzzy semantics and geographic constraints when executed by a processor.

[0022] The GIS buffer generation method and system based on fuzzy semantics and geographic constraints provided by the embodiments of the present application have the following beneficial effects: The embodiments of the present application extract fuzzy description words in natural language and parse membership values through a semantic analysis engine, and obtain basic distance parameters through a pre-trained language-space mapping model, effectively filling the gap between fuzzy semantics and precise spatial parameters, and improving the accuracy and efficiency of human-computer interaction. Secondly, the feature length is calculated based on the geometric features of geographic entities, and then a scale correction function is used to obtain a scale correction coefficient, so that the geographic distance threshold is fully adapted to the scale of the entity itself, and the result deviation caused by ignoring the entity difference in the traditional method is reduced. A three-dimensional buffer surface is generated based on a digital elevation model and a terrain adaptive algorithm, taking into account the terrain factor, so that the buffer is more suitable for the real geographic scene, and the application efficiency of GIS in the fields of city planning, resource management and the like is significantly improved. BRIEF DESCRIPTION OF DRAWINGS

[0023] Figure 1 A flowchart of the GIS buffer generation method based on fuzzy semantics and geographic constraints provided by an embodiment of the present application is shown in the figure;

[0024] Figure 2 A structural block diagram of the GIS buffer generation system based on fuzzy semantics and geographic constraints provided by an embodiment of the present application is shown in the figure;

[0025] Figure 3 A schematic block diagram of the electronic device provided by an embodiment of the present application is shown in the figure. DETAILED DESCRIPTION

[0026] In the following description, specific details are set forth such as particular system configurations, techniques, etc., in order to provide a thorough understanding of the embodiments of the present application. However, it should be apparent to those skilled in the art that the present application can be practiced in other embodiments that depart from these specific details. In other instances, detailed descriptions of well-known systems, devices, circuits, and methods are omitted so as not to obscure the description of the present application with unnecessary detail.

[0027] For the purpose, technical solutions and advantages of the present application to be clearer, the following will be combined with the accompanying drawings to be explained. Figures 1-3 The present application is described by specific embodiments.

[0028] Please refer to Figure 1 , Figure 1 The flowchart of the GIS buffer generation method based on fuzzy semantics and geographic constraints provided by an embodiment of the present application, which comprises:

[0029] S101: extracting the geographic entity keywords, geographic entity types and fuzzy description words in the natural language instructions of the user through the semantic analysis engine, and analyzing the fuzzy description words to obtain the membership values corresponding to the fuzzy description words; In this embodiment, the semantic analysis engine includes a bidirectional long short-term memory network and a conditional random field model, which can extract geographic entity keywords, geographic entity types and fuzzy description words; wherein the geographic entity keywords are, for example, XX river and YY mountain, etc.; the geographic entity types are, for example, linear water body and planar terrain, etc.; the fuzzy description words are, for example, nearby, around, remote area, etc.

[0031] Based on the preset fuzzy semantic dictionary, the membership values corresponding to the fuzzy description words are calculated by the membership function, wherein the value range of the membership value is 0-1, for example, very close corresponds to 0.9, and far corresponds to 0.3. The preset fuzzy semantic dictionary includes common fuzzy space terms in actual life.

[0032] S102: inputting the fuzzy description words, membership values and geographic entity types into the pre-trained language-space mapping model to obtain the basic distance parameters, which are the reference values with distance dimension or the scale factors relative to the typical scale of the entity;

[0033] In this embodiment, the language-space mapping model adopts a neural network model based on the Transformer architecture, wherein the input layer is a fusion module for receiving the fuzzy description words, membership values and geographic entity types; the output layer is a double-channel structure, the first channel outputs the absolute distance value of the basic distance parameter, and the second channel outputs the scale factor of the basic distance parameter.

[0034] The language-space mapping model of this embodiment uses a large-scale geographic text-space corresponding data set for training in the pre-training stage, and the training samples include: fuzzy space relationship phrases described in natural language, corresponding geographic entity types, and labeled real basic distance parameters (reference distance values or reference scale factors determined according to the context and geographic background).

[0035] Wherein, the training process adopts a weighted loss function: for samples predicted as absolute distance values, the mean square error of the predicted distance and the true distance is calculated; for samples predicted as scale factors, the mean square error of the predicted scale factor and the true scale factor is calculated, and the length of the geographical entity feature corresponding to the sample is additionally considered as a weight factor to amplify the influence of the scale error under a larger entity scale.

[0036] In this embodiment, the geographical entity type is mapped to a high-dimensional embedding vector before being input into the language-space mapping model; the membership value is normalized to the interval [0, 1]; and the feature length is logarithmically transformed or standardized before being input into the scale correction function, so as to improve the stability and generalization ability of the model training.

[0037] In the inference application of the language-space mapping model of this embodiment, according to a pre-defined rule or a gating mechanism inside the model, the first channel (outputting absolute distance) or the second channel (outputting scale factor) is selected as the final basic distance parameter. The selection logic is based on the inherent characteristics of the geographical entity type (such as point entities tend to use absolute distance, and linear or planar entities tend to use scale factor) or the implicit inclination of the user instruction.

[0038] In this embodiment, the fusion module adopts a multi-head attention mechanism to weight and fuse the fuzzy description word, the membership value and the geographical entity type, including:

[0039] The fuzzy description word, the membership scalar and the geographical entity type are spliced into a joint input vector; the correlation weights of the fuzzy description word, the membership scalar and the geographical entity type are respectively calculated through a multi-head self-attention layer; after the output of the attention layer, the original joint input vector is superimposed to form a residual structure to retain the original feature information; and a learnable gated unit (Gated Linear Unit, GLU) is used to dynamically adjust the contribution proportion of the word embedding and the category embedding to the final fused feature.

[0040] S103: Query the spatial database based on the geographical entity keyword to obtain the geometric features associated with the geographical entity, and calculate the feature length corresponding to the geometric features;

[0041] In this embodiment, the spatial database adopts a spatio-temporal index structure (the spatial database supports spatial data engines such as PostGIS and OracleSpatial), which supports millisecond-level geometric feature retrieval; the geometric feature extraction includes a topological relationship verification module to automatically correct abnormal nodes. In this embodiment, the feature length is calculated by a spatial analysis algorithm: for point entities, the equivalent radius within the coordinate accuracy range is taken; for linear entities, the arc length or the total length of the polyline is calculated; and for planar entities, the feature value of the longest axis length or the boundary perimeter is extracted. At the same time, the spatial reference system of the entity is recorded to ensure the consistency of subsequent distance calculation.

[0042] S104: Correct the basic distance parameter based on the intention direction expressed by the fuzzy description word, the membership value, the geographic entity type and the feature length to obtain a geographic distance threshold;

[0043] In this embodiment, the intention direction is confirmed by direction word vector encoding, and the distance weight is adjusted using a direction attenuation factor. The scale correction function of this embodiment can be designed as a piecewise function, for example, linear correction when the feature length is <100 meters, exponential correction when the feature length is in the range of 100-1000 meters, and logarithmic correction when the feature length is >1000 meters. A dynamic weight matrix is added when calculating the feature length, and the proportion of the basic distance and the correction coefficient is allocated according to the entity type.

[0044] S105: Generate an initial two-dimensional buffer polygon based on the geographic distance threshold and the spatial geometric features of the geographic entity.

[0045] In this embodiment, the offset algorithm is used to process the sharp corner problem of the line entity buffer to generate an initial two-dimensional buffer, which improves the offset accuracy. The surface entity buffer supports bidirectional expansion inward / outward, and the overlapping area is processed by a buffer fusion algorithm (for example, Boolean operation of adjacent polygon buffers); the point entity buffer provides multiple shape options such as circle, rectangle, and sector, and the sector buffer can set the opening angle through the intention direction parameter.

[0046] S106: Query the digital elevation model to obtain terrain data based on the initial two-dimensional buffer polygon, and generate a three-dimensional buffer surface through a slope-sensitive interpolation algorithm.

[0047] In this embodiment, the digital elevation model uses 10-meter resolution DEM data (terrain elevation data), and multi-scale terrain loading is achieved through a tile pyramid structure.

[0048] This embodiment uses a slope-sensitive interpolation algorithm to extend the two-dimensional buffer to three dimensions based on terrain data. In areas with steep slopes, the elevation change of the buffer surface is more significant; in flat areas, the elevation change is relatively gentle, thereby generating a three-dimensional buffer surface that can reflect the actual terrain undulations. This makes the buffer not only meet the requirements in the plane range, but also truly reflects the influence of terrain factors, providing more comprehensive and accurate spatial information support for geographic analysis, planning decisions, etc.

[0049] In another possible embodiment, this embodiment queries the digital elevation model to obtain terrain data based on the initial two-dimensional buffer polygon, and generates a three-dimensional buffer surface through a terrain adaptive algorithm.

[0050] In the embodiment, the terrain adaptive algorithm includes three core modules: slope threshold judgment (>25° region starts elevation correction), slope direction attenuation factor (20% increase in distance along the slope direction, 30% decrease in distance against the slope direction), and elevation scaling coefficient (adjustment based on contour density); the three-dimensional curved surface is generated in the form of irregular triangle mesh and / or grid, and the triangle mesh nodes are encrypted in steep terrain areas (slope >45°) to ensure that the curved surface accuracy reaches the 1-meter error range.

[0051] From the above, the application extracts the fuzzy description words in natural language and parses the membership values through the semantic analysis engine, and obtains the basic distance parameters through the pre-trained language-space mapping model, effectively filling the gap between fuzzy semantics and accurate spatial parameters, and improving the accuracy and efficiency of human-computer interaction. Secondly, the feature length is calculated based on the geometric features of the geographic entity, and then the scale correction coefficient is obtained through the scale correction function, so that the geographic distance threshold is fully adapted to the scale of the entity itself, reducing the result deviation caused by ignoring the entity difference in traditional methods. Based on the digital elevation model and the terrain adaptive algorithm, a three-dimensional buffer curved surface is generated, taking into account the terrain factor, making the buffer more suitable for real geographic scenes, and significantly improving the application efficiency of GIS in the fields of urban planning, resource management, etc.

[0052] In an embodiment of the present application, the membership value corresponding to the fuzzy description word is parsed, including:

[0053] The preset fuzzy semantic dictionary is queried based on the fuzzy description word to obtain the corresponding membership value, or a deep learning-based semantic understanding model is used to calculate the membership value corresponding to the fuzzy description word according to the context of the fuzzy description word in the natural language instruction.

[0054] The fuzzy semantic dictionary stores a plurality of fuzzy description words and their corresponding membership functions or membership value intervals.

[0055] In the embodiment, the fuzzy semantic dictionary is constructed through a large amount of geographic space-related natural language data sorting and analysis, and stores a plurality of common fuzzy description words and the membership functions or membership value intervals corresponding to each fuzzy description word. For example, for the fuzzy description word "very close", the dictionary stores the corresponding membership function, which can represent the mapping relationship between distance and membership. When the distance is 0-100 meters, the membership gradually decreases from 1.0 to 0.5. Alternatively, the membership value interval can be directly given, for example, 0.8-1.0. When the fuzzy description word in the user instruction is extracted, the corresponding membership information can be quickly obtained by searching the fuzzy semantic dictionary, and then the specific membership value can be determined.

[0056] If the fuzzy descriptor is not included in the preset fuzzy semantic dictionary or its semantics is greatly affected by the context, a semantic understanding model based on deep learning can be used for calculation. The semantic understanding model can deeply understand the meaning of words in different contexts through training on a large amount of natural language texts. After inputting the natural language instruction containing the fuzzy descriptor into the semantic understanding model, the semantic understanding model can analyze the fuzziness expressed by the fuzzy descriptor in combination with the context, and then calculate the corresponding membership value. For example, in the instruction "walk down from the top of the mountain for a short distance", the meaning of "not far" will be affected by the geographical entity "mountain top" and the action "walk down", and the deep learning model can combine these context information to more accurately calculate the membership value corresponding to "not far".

[0057] Specifically, when the semantic understanding model based on deep learning is used, the model uses a BERT pre-training model, the input layer includes the fuzzy descriptor itself and the window text composed of the previous and next three context words, the context features are extracted through a 12-layer Transformer encoder, and the output layer uses a full connection network to map to the membership value in the interval [0, 1]. The model training data set includes a large number of labeled samples, each sample is annotated by a geographical information field expert with a membership reference value of the fuzzy descriptor in a specific context, and the training process uses an MSE loss function, and the average error on the validation set is controlled within ±0.05.

[0058] In actual application, the hybrid analysis strategy of dictionary priority + model completion can be used in the embodiment: when the matching degree of the fuzzy descriptor in the dictionary is ≥ 90%, the membership function value in the dictionary is directly called for calculation; when the matching degree is < 90% or it is an unregistered word, the semantic understanding model is automatically triggered for context calculation, for example, when processing a complex description such as a park not far from the school, the model can correct the membership value of not far based on the typical scale of the school.

[0059] In summary, the embodiment provides two ways to analyze the membership value of the fuzzy descriptor, the fuzzy semantic dictionary can quickly obtain the membership value by using the existing knowledge reserve, and the semantic understanding model based on deep learning can more accurately calculate the membership value according to the context, the combination of the two improves the accuracy and adaptability of the membership value analysis, and provides more reliable basic data for subsequent buffer generation.

[0060] In an embodiment of the present application, the corresponding membership value is obtained by querying the preset fuzzy semantic dictionary based on the fuzzy descriptor, comprising:

[0061] The preset fuzzy semantic dictionary is queried, if the fuzzy descriptor exists in the dictionary, the membership value corresponding to the fuzzy descriptor is calculated by using a semantic understanding model based on deep learning as a first membership value;

[0062] querying the preset fuzzy semantic dictionary based on the fuzzy descriptor to obtain a corresponding membership value as the second membership value;

[0063] determining a final membership value based on the first membership value, the second membership value, and a difference between the first membership value and the second membership value;

[0064] If the fuzzy descriptor is not included in the dictionary, a semantic understanding model based on deep learning is used to calculate a membership value corresponding to the fuzzy descriptor as the final membership value.

[0065] In this embodiment, specifically, when the fuzzy descriptor exists in the preset fuzzy semantic dictionary, the calculation process of the first membership value is consistent with the aforementioned calculation method of the semantic understanding model based on deep learning, that is, it is derived from the context. The second membership value is obtained from the corresponding membership function or membership value interval in the fuzzy semantic dictionary. If it is a membership function, the specific value is calculated by substituting the relevant parameters of the geographic entity. If it is a membership value interval, the interval median is taken as the second membership value.

[0066] Specifically, when querying the preset fuzzy semantic dictionary, it is first determined whether the fuzzy descriptor exists in the dictionary. If the fuzzy descriptor exists in the dictionary, two calculation methods are started simultaneously. On the one hand, the semantic understanding model based on deep learning is used to process the fuzzy descriptor, and the corresponding membership value is calculated by combining the context in the natural language instruction, which is taken as the first membership value. This process can fully consider the influence of the context on the fuzzy semantics, making the result more consistent with the user's real intention. On the other hand, the preset fuzzy semantic dictionary is queried based on the fuzzy descriptor, and the corresponding membership value is obtained from the membership function or membership value interval stored in the dictionary, which is taken as the second membership value. It relies on the standardized data in the dictionary and has a certain stability.

[0067] Then, the final membership value is determined based on the first membership value, the second membership value, and the difference between the two. For example, if the difference is small, it means that the results obtained by the two methods are highly consistent, and the average of the two can be used as the final value. If the difference is large, further analysis is needed, and factors such as the use scenario of the fuzzy descriptor and the type of geographic entity are used to allocate weights, giving higher weight to the party that is more consistent with the actual situation, and then calculating the final membership value through weighting to balance flexibility and stability.

[0068] If the fuzzy description word is not included in the preset fuzzy semantic dictionary, at this time, the reference data provided by the dictionary is lacking, and the membership value corresponding to the fuzzy description word is directly calculated by using the semantic understanding model based on deep learning, and the membership value is taken as the final membership value. In this case, the model can flexibly cope with the uncollected fuzzy description word by virtue of its understanding ability of complex context, and ensure the effective acquisition of the membership value.

[0069] In summary, the embodiment simultaneously calculates the value by using the deep learning model and the dictionary query value for the fuzzy description word existing in the dictionary, and determines the final value according to the relationship between the two, which can integrate the advantages of the two methods. When the difference between the two is small, the weighted average value can fuse the information of the two, improve the accuracy; when the difference is large, the final value is further determined according to the complexity analysis of the natural language instruction, so that the result is more in line with the actual situation, and the robustness of the method is enhanced.

[0070] In an embodiment of the present application, the final membership value is determined based on the first membership value, the second membership value, and the difference between the first membership value and the second membership value, comprising:

[0071] If the difference between the first membership value and the second membership value is less than a preset difference threshold, a weighted average value of the first membership value and the second membership value is calculated as the final membership value;

[0072] If the difference between the first membership value and the second membership value is greater than or equal to the preset difference threshold, the complexity value of the natural language instruction is obtained by complexity analysis, and the final membership value is determined according to the complexity value.

[0073] In the embodiment, when determining the final membership value, first, the difference between the first membership value and the second membership value is calculated. When the difference is less than or equal to the first preset difference threshold, it indicates that the difference between the calculation results of the two methods is small, and the arithmetic average of the two is taken as the final membership value.

[0074] When the difference is greater than the first preset difference threshold and less than the second preset difference threshold, the difference is moderate, and the weighted average is used, wherein the weight of the first membership value belongs to the first range value, and the weight of the second membership value belongs to the second range value; wherein the value in the second range value is less than the value in the first range value; the second preset difference threshold is the preset difference threshold; the third preset difference threshold is greater than the second preset difference threshold, and the second preset difference threshold is greater than the first preset difference threshold.

[0075] When the difference is greater than or equal to the second preset difference threshold and less than the third threshold, the difference is large, and the complexity value of the natural language instruction is obtained by complexity analysis at this time; the final membership value is determined according to the complexity value.

[0076] When the difference is greater than or equal to the third preset difference threshold, the difference is very large, and a manual verification mechanism is started, and an expert determines the final membership value according to specific context and geographical entity characteristics, and feeds back the result to the system, and updates the fuzzy semantic dictionary and the training sample library of the semantic understanding model to improve the accuracy of subsequent calculation.

[0077] When the difference is greater than or equal to the second preset difference threshold and less than the third threshold, the natural language instruction complexity analysis is expanded from three dimensions: one is the syntax complexity, the average sentence length and the number of nested clauses are calculated by dependency syntax analysis; two is the semantic complexity, the number of geographical entities and the number of fuzzy description words are counted; three is the context correlation degree, the correlation strength of the fuzzy description word and the entity word before and after the text is calculated by the cosine similarity of the word vector.

[0078] For example, if the difference between the first membership value and the second membership value is less than the preset difference threshold, it indicates that the results calculated by the two methods have a small difference and high consistency. At this time, the weighted average value of the first membership value and the second membership value is calculated as the final membership value. The weight here can be set according to the actual application scene and the trust degree of the two calculation methods, for example, if the deep learning model is preferred for understanding the context, the first membership value can be given a higher weight of 0.6, and the second membership value can be given a weight of 0.4, and the final result is obtained by weighted calculation, which not only combines the advantages of the two methods, but also reflects the emphasis on a specific method.

[0079] If the difference between the first membership value and the second membership value is greater than or equal to the preset difference threshold, it indicates that there is a large difference between the two calculation results, which needs to be further processed. At this time, the complexity analysis of the natural language instruction is performed to obtain a complexity value. The complexity analysis can be performed from the number of geographical entities, the number of fuzzy description words and the nesting degree, the complexity of the syntax structure of the sentence, etc. For example, a long sentence containing multiple geographical entities and multiple fuzzy description words has a high complexity value; while a short sentence involving only a single geographical entity and a simple fuzzy description word has a low complexity value. According to the obtained complexity value, the final membership value is determined, if the complexity value is low, it indicates that the instruction semantics is relatively clear, and the first membership value calculated by the deep learning model can be used first, because it can better combine the simple context; if the complexity value is high, the instruction semantics is complex, and the standardized data in the dictionary is more referential, the second membership value can be used, or the final value is determined after the two are comprehensively adjusted combined with more auxiliary information.

[0080] In summary, the embodiment adopts different processing methods according to the difference between the first and second membership values. When the difference is small, the two values are reasonably fused through weighted average, and when the difference is large, the instruction complexity is analyzed, which can more flexibly cope with different situations. For complex instructions, the first membership value calculated by the deep learning model is more dependent, because it has a deeper understanding of the context, which can improve the accuracy of the membership value.

[0081] In an embodiment of the present application, the final membership value is determined according to the complexity value, comprising:

[0082] If the complexity value is greater than or equal to the complexity threshold, the first membership value is taken as the final membership value.

[0083] If the complexity value is less than the complexity threshold, the first membership value is taken as the final membership value.

[0084] In this embodiment, the complexity threshold is determined based on a large number of natural language instruction processing cases, aiming to unify the decision standard and improve the processing efficiency. This setting is based on the first membership value calculated by the deep learning model combined with the specific context, which can better fit the user's real intention in a specific instruction. Whether it is a simple low-complexity instruction or a high-complexity instruction containing multiple semantics, the first membership value can reflect the degree of fuzzy descriptor in the current context more accurately by the model's deep understanding of the context.

[0085] Specifically, the complexity value is calculated in percentage, with three dimensions accounting for 40%, 40%, and 20%, respectively. The calculation method is: complexity value = 0.4 x syntax score + 0.4 x semantic score + 0.2 x correlation score, where each score is mapped to the 0-100 interval through standardization processing.

[0086] If the complexity value is greater than or equal to the complexity threshold, the first membership value is taken as the final membership value. This is because when the complexity of the natural language instruction reaches or exceeds the set threshold, it means that the information contained in the instruction is complex, and there may be multiple geographic entities associated with each other, nested use of fuzzy descriptors, etc. The semantic understanding model based on deep learning can better capture the true meaning of the fuzzy descriptor in this complex context, and through deep analysis of the context, the first membership value that better fits the user's intention is obtained, so it is selected as the final result.

[0087] If the complexity value is less than the complexity threshold, the first membership value is also used as the final membership value. Even if the complexity of natural language instructions is low and the semantics are relatively simple and clear, deep learning-based semantic understanding models can still accurately understand the meaning of vague descriptive words in simple contexts, and the calculated first membership value can accurately reflect the user's vague description of the spatial range. In this case, using the first membership value as the final result can ensure the consistency and accuracy of the results in processing instructions of different complexities.

[0088] For example, when the complexity value is less than 30, the first membership value is used as the final result because the model calculation is more reliable in a simple context; when the complexity value is less than 70, a second context analysis is initiated, and geographical entity encyclopedia knowledge is added to the original model for recalculation, and the average value of the recalculated value and the second membership value is taken; when the complexity value is greater than or equal to 70, the human-computer interaction mechanism is triggered, and the system displays the buffer preview effect corresponding to the two membership values ​​to the user, allowing the user to choose the result that better meets expectations, and records the user's selection preference for model optimization.

[0089] In summary, this embodiment uses the first membership value calculated by the deep learning model as the standard for instructions with high complexity, which can better handle complex semantic situations; for instructions with low complexity, the first membership value is also used, which ensures the consistency and simplicity of the method and helps to improve the rationality and reliability of the membership value determination.

[0090] In one embodiment of this application, calculating the feature length corresponding to the geometric feature includes:

[0091] Based on the geographic entity type and the intent expressed by the fuzzy descriptive words, the corresponding calculation strategy is selected from the preset feature length calculation rule base;

[0092] Among them, the preset feature length calculation rule library defines the feature length calculation methods corresponding to different entity types or intent directions;

[0093] The feature length corresponding to the geometric feature is calculated based on the computational strategy.

[0094] In this embodiment, the preset feature length calculation rule base is constructed based on a large number of geographic entity features and actual application scenarios, which clearly defines the feature length calculation methods corresponding to different entity types or intent directions. Specifically, a two-dimensional index structure is adopted. The horizontal dimension is divided into three major categories according to geographic entity type: point entities, line entities, and polygon entities, with each major category further subdivided into multiple subcategories. The vertical dimension is divided into three major categories according to intent direction: omnidirectional, unidirectional, and multidirectional. The feature length calculation rule base includes multiple calculation rules, and each calculation rule includes four elements: applicable entity type, direction range, calculation algorithm, and precision parameters.

[0095] For example, when the geographic entity type is a linear river, and the intended direction expressed by the fuzzy descriptor is along the upstream and downstream of the river, the corresponding calculation strategy is to calculate the total length of the river; if the intended direction is the two sides of a certain river segment, the calculation strategy is the length of that specific river segment. For an isometric urban park, if the intended direction is the overall area surrounding the park, the calculation strategy is the perimeter of the park; if the intended direction is near the park entrance, the average distance from the entrance point to the park boundary is used as the basis for calculating the feature length. Then, the feature length corresponding to the geometric feature is calculated based on the selected calculation strategy. After obtaining the geometric feature data of the geographic entity, the calculation is performed according to the steps and methods specified in the calculation strategy. For example, in the case of using the strategy of calculating the total length of the river, the coordinates of the linear geometric features of the river are extracted from the spatial database, and the total length of the river is calculated by accumulating the distances between the coordinate points, which is the corresponding feature length; for the calculation strategy of calculating the perimeter of the park, the boundary coordinates of the isometric geometric features are used to calculate the distances between each adjacent boundary point in sequence and accumulate them to obtain the perimeter of the park as the feature length. In this way, it can be ensured that the calculated feature length matches the geographic entity type and the user's intent direction, providing accurate data support for subsequent geographic distance threshold correction.

[0096] In summary, this embodiment selects a feature length calculation strategy based on the geographic entity type and the intent direction of the fuzzy descriptive terms, making the feature length calculation more consistent with the actual situation of geographic entities and user needs. Different calculation methods correspond to different types of geographic entities and different intent directions, which can more accurately reflect the spatial characteristics of geographic entities, provide more precise parameters for subsequent buffer generation, and improve the accuracy of buffer generation.

[0097] In one embodiment of this application, a corresponding calculation strategy is selected from a preset feature length calculation rule base based on the geographic entity type and the intent direction expressed by the fuzzy descriptive words, including:

[0098] Based on the geographic entity type, a set of candidate calculation strategies applicable to the geographic entity type is selected from the preset feature length calculation rule library;

[0099] Parse the intent direction of fuzzy descriptive words and generate corresponding intent direction semantic vectors;

[0100] The cosine similarity between the semantic vector of the intent direction and all standard policy semantic vectors in the preset feature length calculation rule base is calculated; the standard policy semantic vector is generated based on the standard intent description label associated with each calculation policy defined in the preset feature length calculation rule base;

[0101] The final computation strategy is selected from the candidate computation strategy set, based on the computation strategy corresponding to the standard strategy semantic vector that has the highest cosine similarity to the semantic vector of the user intent direction.

[0102] In this embodiment, the preset feature length calculation rule base has corresponding calculation strategies for different geographic entity types. For example, when the geographic entity type is a linear highway, the calculation strategies applicable to highways in the rule base include calculating the total length of the highway, the length of a specific road segment, and the extension length corresponding to the widest point of the highway. These strategies together constitute a set of candidate calculation strategies.

[0103] This embodiment uses natural language processing technology to perform deep analysis on the directional information expressed by vague descriptive words, extract semantic features, and then convert these semantic features into a computer-recognizable vector form, namely, the intention direction semantic vector. For example, for the intention direction of "both sides of a road," the generated semantic vector after analysis will contain relevant semantic information such as "road," "both sides," and "surrounding the road and biased towards both sides."

[0104] This embodiment calculates the cosine similarity between the intent direction semantic vector and all standard policy semantic vectors in a preset feature length calculation rule base. The standard policy semantic vectors are generated based on the standard intent description tags associated with each calculation strategy defined in the preset feature length calculation rule base. Each calculation strategy has a corresponding standard intent description tag, such as calculating the total length or calculating the length of a specific road segment. The standard policy semantic vector is generated based on these tags. By calculating the cosine similarity, the degree of similarity between the intent direction semantic vector and each standard policy semantic vector can be quantified. The higher the similarity, the more closely the intent directions expressed by the two are aligned.

[0105] In this embodiment, the calculation strategy corresponding to the standard strategy semantic vector with the highest cosine similarity to the semantic vector of the user's intent direction is selected as the final calculation strategy from the candidate calculation strategy set. Since the candidate calculation strategy set has already limited the scope applicable to the current geographic entity type, selecting the calculation strategy with the highest similarity ensures that the strategy meets the requirements of the geographic entity type and matches the intent direction expressed by the user's vague descriptive words to the greatest extent, thus providing an accurate and demand-appropriate method for calculating feature length.

[0106] Specifically, when filtering the set of candidate computation strategies based on geographic entity type, the geographic entity type is first matched hierarchically. For example, when the entity type is highway, the rule base will automatically filter out all computation strategies marked as line entity - traffic road type, forming a set of candidate computation strategies including 5-8 strategies. The filtering process is achieved through hash mapping of type labels.

[0107] The semantic vector generation process for parsing intent direction employs BERT word vector encoding, transforming the directional description text into a 768-dimensional dense vector. Directional words, distance words, and range words are reinforced through attention weights. For descriptions without a clear direction, a default omnidirectional semantic vector is generated.

[0108] The standard policy semantic vector is generated based on 3-5 standard intent description tags bound to each policy in the rule base. For example, the policy associated tags for "line entity - unidirectional projection length" include "on one side", "on one side", and "extending along a certain direction". The standard vector is obtained by averaging the word vectors of these tags.

[0109] Cosine similarity calculation employs vector normalization, with the formula: cosθ=(A・B) / (|A|・|B|), where A is the semantic vector of the intent direction and B is the semantic vector of the standard strategy. After calculation, the strategy with the highest similarity is selected from the candidate strategy set as the final solution; if the highest similarity is less than a preset similarity threshold, a strategy fusion mechanism is triggered, taking a weighted combination of the top two highly similar strategies, where the weights are allocated according to the similarity ratio.

[0110] In summary, this embodiment selects a computational strategy by calculating the cosine similarity between the semantic vector of the intent direction and the semantic vector of the standard strategy. This quantifies the degree of matching between the user's intent and the strategies in the rule base, thereby selecting the computational strategy that best matches the user's intent. This method based on semantic vector similarity has high scientific validity and accuracy, and can improve the rationality and intelligence level of feature length calculation strategy selection.

[0111] In one embodiment of this application, the GIS buffer generation method based on fuzzy semantics and geographic constraints further includes: if the highest similarity is lower than a preset similarity threshold, then

[0112] The intended direction is parsed into a preset basic direction category;

[0113] Select the corresponding calculation strategy based on the entity type and the basic direction category;

[0114] If the basic direction category cannot be resolved, select the default strategy based on the entity type.

[0115] The default strategies include: for linear features, select the average width strategy; for areal features, select the equivalent radius strategy; and for point features, select the fixed radius strategy.

[0116] In this embodiment, the intent direction is parsed into a preset basic direction category. The preset basic direction categories are set based on common spatial direction requirements, such as omnidirectional, unilateral, bilateral, and specific area. When the highest similarity between the intent direction semantic vector and the standard policy semantic vector does not reach a preset threshold, it indicates that the user's intent direction description is relatively vague or specific, making it difficult to directly match the standard policy in the rule base. In this case, further semantic parsing is used to categorize the user's intent direction into a preset basic direction category to simplify the subsequent policy selection process. For example, if the user's vague description is "not far from the roadside," it will be parsed and categorized into the omnidirectional basic direction category.

[0117] The preset feature length calculation rule base in this embodiment also stores calculation strategies corresponding to combinations of entity type and basic direction category. For example, for a combination of linear features and bilateral basic direction categories, the corresponding calculation strategy is to calculate the average width within a 50-meter radius on both sides of the road as the feature length; for a combination of areal features and basic direction categories near the entrance, the corresponding calculation strategy is to calculate the average length of the park boundary within a 200-meter radius centered on the park entrance as the feature length. This embodiment, through the dual constraints of entity type and basic direction category, can select a calculation strategy that better suits the current needs from the preset feature length calculation rule base.

[0118] If the basic direction category cannot be resolved—that is, even after semantic parsing, the user's intended direction cannot be categorized into any preset basic direction category—then a default strategy is selected based on the entity type. The default strategy is a pre-defined, general calculation strategy for different entity types to ensure that feature length calculations can still be completed even in extreme cases. Specifically, for linear features, the average width strategy is selected, which calculates the average width of the entire linear feature as the feature length, such as calculating the average width of a river; for areal features, the equivalent radius strategy is selected, which calculates the radius of an equivalent circle based on the area of ​​the areal feature as the feature length, such as converting the area of ​​an irregular lake into the radius of a circle with equal area; for point features, the fixed radius strategy is selected, which presets a fixed radius value as the feature length, such as fixing the feature length of a monument to 50 meters. These default strategies ensure that feature length calculations can proceed smoothly under various complex conditions, providing foundational data for subsequent geographic distance threshold correction.

[0119] In summary, this embodiment provides a further processing method when the similarity is below a preset threshold. It resolves the intent direction to a basic direction category and combines this with the entity type to select a strategy. If resolution fails, a default strategy is used. This ensures that even when a strategy cannot be accurately matched, a relatively reasonable calculation strategy can still be selected based on the general characteristics of geographic entity types, guaranteeing the method's normal operation under various circumstances and enhancing its versatility and adaptability.

[0120] In one embodiment of this application, the geographic entity types include: linear features, areal features, and point features;

[0121] The mapping rules for the feature length calculation strategy include:

[0122] When the geographic entity type is a linear feature, if the intended direction includes a description along the line, a length segmentation strategy is adopted, and the feature length is a percentage of the total length of the entity; if the intended direction includes a description along the lateral direction, an average width strategy is adopted, and the length of the minor axis of the geometric outer rectangle of the entity is calculated as the feature length.

[0123] When the geographic entity type is a isal feature, if the intended direction includes a boundary class description, the perimeter strategy is used to calculate the total length of the geometric boundary as the feature length; if the intended direction includes an interior class description, the equivalent radius strategy is used to calculate the radius of the area equivalent circle as the feature length.

[0124] When the geographic entity type is a point feature, a fixed radius strategy is adopted, and the feature length is the preset typical influence radius.

[0125] In one embodiment of this application, the basic distance parameters are modified based on the intent direction, membership value, geographic entity type, and feature length expressed by the fuzzy descriptive terms to obtain a geographic distance threshold, including:

[0126] Based on the intent direction, membership value, geographic entity type and feature length expressed by the fuzzy descriptive words, the scale correction coefficient is calculated through a preset scale correction function.

[0127] The geographic distance threshold is obtained by weighting the basic distance parameter with the scale correction coefficient.

[0128] In this embodiment, firstly, based on the intent direction, membership value, geographic entity type, and feature length expressed by the fuzzy descriptive terms, a scale correction coefficient is calculated using a preset scale correction function. The preset scale correction function is a mathematical model constructed by integrating multiple factors. Its inputs are the four key parameters mentioned above, and its output is a scale correction coefficient that reflects the combined influence of each parameter.

[0129] From the perspective of the intended direction, different directions will affect the correction coefficient. For example, if the intended direction is one side of a linear feature, the correction coefficient will be lower compared to the all-around direction, because the range of one side is usually smaller than the range of the all-around direction; while if the intended direction is pointing to the key area of ​​the feature, the correction coefficient will be smaller in order to narrow the range of the distance threshold.

[0130] The membership degree value is also closely related to the correction coefficient. The higher the membership degree value, the stronger the user's certainty about the fuzzy range, and the closer the correction coefficient is to 1, so as to reduce the adjustment of the basic distance parameter; conversely, when the membership degree value is low, the correction coefficient will expand or shrink the basic distance parameter to adapt to the needs of a more fuzzy range.

[0131] The type of geographic entity is an important factor affecting the correction coefficient. Different types of geographic entities have different spatial ranges and characteristics. For example, for densely populated residential areas, the correction coefficient for the surrounding distance threshold will be relatively small to avoid errors caused by an excessively large range; while for open nature reserves, the correction coefficient will be larger to include a wider surrounding area.

[0132] Feature length provides a quantitative basis for the correction coefficient. Geographic entities with larger feature lengths will have correspondingly larger correction coefficients, because larger entities typically require a wider surrounding area; conversely, entities with smaller feature lengths will have smaller correction coefficients.

[0133] After calculating the scale correction coefficient by integrating the above factors through the scale correction function, the basic distance parameter and the scale correction coefficient are weighted and calculated to obtain the geographic distance threshold. The weights of the weighted calculation can be set according to the importance of each parameter in the actual application scenario to ensure that the final geographic distance threshold can accurately reflect the user's fuzzy semantic needs and conform to the spatial characteristics and constraints of geographic entities.

[0134] This embodiment calculates a scale correction coefficient using a preset scale correction function, and then weights it with the basic distance parameters to obtain the geographic distance threshold. This approach comprehensively corrects the distance parameters based on multiple factors, including the intent direction of fuzzy descriptors, membership values, geographic entity types, and feature lengths. This method makes the determination of the geographic distance threshold more scientific and reasonable, more accurately reflecting the user's requirements for the buffer zone range, thereby generating a GIS buffer zone that better meets actual needs.

[0135] In one embodiment of this application, a semantic parsing engine is used to extract geographic entity keywords and fuzzy descriptive terms from the user's natural language commands, and to parse the membership values ​​corresponding to the fuzzy descriptive terms, including:

[0136] The predefined features include a geographic entity type dictionary, a fuzzy descriptor dictionary, and association syntax rules. The fuzzy descriptor dictionary stores the mapping relationship between words and basic membership intervals.

[0137] The core predicate structure in the instruction is identified through dependency parsing, combined with part-of-speech tagging and named entity recognition.

[0138] Extracting geographic entity keywords from noun phrases; extracting fuzzy descriptive terms from adverb / adjective phrases;

[0139] For fuzzy descriptive terms, membership degree scalarization is performed. If it is a single descriptive term, the median of the basic membership degree interval is taken as the scalar membership degree value. If it is a combined descriptive term, the membership degree aggregation function is used to calculate the scalar membership degree value. Among them, progressive combination is synthesized by multiplication, and parallel combination is synthesized by the maximum value.

[0140] The geographic entity type is determined by querying the metadata table of the spatial database based on geographic entity keywords.

[0141] In one embodiment of this application, when a geographic entity has a time attribute or is a dynamic entity:

[0142] Feature length is calculated based on the geometric shape of a geographic entity at a specific point in time or within a specific time period;

[0143] Alternatively, the feature length can be calculated based on the expected maximum range of a geographic entity or its geometry under typical conditions.

[0144] Alternatively, the feature length can be calculated based on the length of the geographic entity's movement trajectory or the spatial distribution range.

[0145] In this embodiment, the feature length is calculated based on the geometric shape of a geographic entity at a specific point in time or within a specific time period. For geographic entities with clear temporal attributes, their geometric shape changes over time, thus requiring a specific time range to determine the feature length. For example, tidal flats under the influence of tides are geographic entities with temporal attributes, and their area and boundaries will differ significantly at high and low tides. If the user is interested in the buffer zone around a beach at high tide, the geometric shape of the beach at that specific point in time, high tide, needs to be extracted, and the feature length calculated based on this, such as the maximum width of the tidal flat or the length of the shoreline at high tide. For seasonal rivers, if analyzing the surrounding area during the rainy season, the average width, length, and other geometric features of the river during that specific time period need to be used as the basis for calculating the feature length.

[0146] In this embodiment, the feature length is calculated based on the expected maximum range or the geometry of a geographic entity under typical conditions. For dynamic entities, if their range of change is regular or predictable, the feature length can be determined using the expected maximum range or the geometry under typical conditions to meet the general requirements of buffer zone generation. For example, migrating bird populations are dynamic entities whose activity range changes with their migration routes. If a buffer zone around the activity area of ​​this bird population needs to be generated, the feature length can be calculated based on the expected maximum activity range predicted from its historical migration data, such as the diameter of the maximum activity area. For temporary gathering areas in cities, their geometry under typical conditions (such as the area occupied and distribution range during regular gatherings) can be used as the basis for calculating the feature length, such as the perimeter of the gathering area under typical conditions.

[0147] In this embodiment, the feature length is calculated based on the length of the geographic entity's movement trajectory or its spatial distribution range. For dynamic entities with continuous movement trajectories, the spatial distribution characteristics during their movement are key to calculating the feature length. For example, a moving convoy is a dynamic entity, and its feature length can be determined based on the length of the convoy's movement trajectory. For a dispersing pollutant cloud, its spatial distribution range (such as the maximum radius of the diffusion area or the equivalent diameter corresponding to the coverage area) can be used as the feature length. If the cloud forms an elliptical distribution during diffusion, its major axis length can be calculated as the feature length to reflect the scale of its spatial influence range. This embodiment, through the above-described method for calculating the feature length of entities with time attributes or dynamic characteristics, ensures that when processing such special geographic entities, the value of the feature length still conforms to their actual state or change pattern, providing accurate basic data for subsequent correction of geographic distance thresholds, and thus generating a GIS buffer that meets actual needs.

[0148] Corresponding to the GIS buffer generation method based on fuzzy semantics and geographic constraints in the above embodiment, Figure 2 This is a structural block diagram of a GIS buffer generation system based on fuzzy semantics and geographic constraints, provided as an embodiment of this application. For ease of explanation, only the parts relevant to the embodiment of this application are shown. References Figure 2 The GIS buffer generation system 20 based on fuzzy semantics and geographic constraints includes: a semantic parsing module 21, a data mapping module 22, a data calculation module 23, a data correction module 24, a first generation module 25, and a second generation module 26.

[0149] The semantic parsing module 21 is used to extract geographic entity keywords, geographic entity types and fuzzy descriptive words from the user's natural language instructions through the semantic parsing engine, and to parse the fuzzy descriptive words to obtain the membership value of the corresponding fuzzy descriptive words.

[0150] The data mapping module 22 is used to input fuzzy descriptive terms, membership values ​​and geographic entity types into the pre-trained language-space mapping model to obtain basic distance parameters. The basic distance parameters are benchmark values ​​with distance dimensions or scaling factors relative to the typical scale of entities.

[0151] The data calculation module 23 is used to query the spatial database based on geographic entity keywords, obtain the geometric features associated with the geographic entities, and calculate the feature length corresponding to the geometric features.

[0152] The data correction module 24 is used to correct the basic distance parameters based on the intent direction, membership value, geographic entity type and feature length expressed by the fuzzy descriptive words to obtain the geographic distance threshold.

[0153] The first generation module 25 is used to generate an initial two-dimensional buffer polygon based on the geographic distance threshold and the spatial geometric features of geographic entities.

[0154] The second generation module 26 is used to obtain terrain data from the digital elevation model based on the initial two-dimensional buffer polygon, and generate a three-dimensional buffer surface through a slope-sensitive interpolation algorithm.

[0155] See Figure 3 , Figure 3 This is a schematic block diagram of an electronic device provided according to an embodiment of this application. Figure 3 The electronic device 300 in this embodiment may include one or more processors 301, one or more input devices 302, one or more output devices 303, and one or more memories 304. The processors 301, input devices 302, output devices 303, and memories 304 communicate with each other via a communication bus 305. The memories 304 store computer programs, including program instructions. The processors 301 execute the program instructions stored in the memories 304. Specifically, the processors 301 are configured to invoke the program instructions to perform the functions of the modules in the aforementioned device embodiments, for example... Figure 2 The semantic parsing module 21, data mapping module 22, data calculation module 23, data correction module 24, first generation module 25, and second generation module 26 are shown.

[0156] It should be understood that, in the embodiments of this application, the processor 301 may be a central processing unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0157] Input device 302 may include a touchpad, a fingerprint sensor (for collecting the user's fingerprint information and fingerprint orientation information), a microphone, etc., and output device 303 may include a display (LCD, etc.), a speaker, etc.

[0158] The memory 304 may include read-only memory and random access memory, and provides instructions and data to the processor 301. A portion of the memory 304 may also include non-volatile random access memory. For example, the memory 304 may also store device type information.

[0159] In specific implementations, the processor 301, input device 302, and output device 303 described in the embodiments of this application can execute the implementation methods described in any embodiment of the GIS buffer generation method based on fuzzy semantics and geographic constraints provided in the embodiments of this application, or they can execute the implementation methods of the electronic devices described in the embodiments of this application, which will not be repeated here.

[0160] In another embodiment of this application, a computer-readable storage medium is provided. This computer-readable storage medium stores a computer program, which includes program instructions. When executed by a processor, the program instructions implement all or part of the processes in the methods described above. Alternatively, the computer program can instruct related hardware to complete the process. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include any entity or device capable of carrying computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium, etc.

[0161] The computer-readable storage medium can be an internal storage unit of the electronic device in any of the foregoing embodiments, such as a hard disk or memory of the electronic device. The computer-readable storage medium can also be an external storage device of the electronic device, such as a plug-in hard disk, smart media card (SMC), secure digital card (SD), flash card, etc., equipped on the electronic device. Furthermore, the computer-readable storage medium can include both internal and external storage units of the electronic device. The computer-readable storage medium is used to store computer programs and other programs and data required by the electronic device. The computer-readable storage medium can also be used to temporarily store data that has been output or will be output.

[0162] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this application.

[0163] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the electronic devices and units described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0164] In the several embodiments provided in this application, it should be understood that the disclosed electronic devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces or units, or it may be an electrical, mechanical, or other form of connection.

[0165] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of the embodiments of this application, depending on actual needs.

[0166] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0167] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for generating GIS buffers based on fuzzy semantics and geographic constraints, characterized in that, include: The semantic parsing engine extracts geographic entity keywords, geographic entity types, and fuzzy descriptive words from the user's natural language commands, and parses the fuzzy descriptive words to obtain the membership value of the corresponding fuzzy descriptive words. The fuzzy descriptive terms, membership values, and geographic entity types are input into a pre-trained language-space mapping model to obtain basic distance parameters. The basic distance parameter is a baseline value with distance dimensions or a scaling factor relative to the typical scale of the entity; the language-space mapping model adopts a neural network with a dual-channel output structure, wherein the first channel outputs a baseline value with distance dimensions, the second channel outputs a scaling factor relative to the typical scale of the entity, and the active channel is selected based on the geographic entity type through a gating mechanism. Based on the geographic entity keywords, the spatial database is queried to obtain the geometric features associated with the geographic entity, and the feature length corresponding to the geometric features is calculated. The basic distance parameters are corrected based on the intent direction, membership value, geographic entity type, and feature length expressed by the fuzzy descriptive words to obtain the geographic distance threshold. Based on the geographic distance threshold and the spatial geometric features of geographic entities, an initial two-dimensional buffer polygon is generated. Based on the initial two-dimensional buffer polygon, terrain data is obtained by querying the digital elevation model, and a three-dimensional buffer surface is generated by a slope-sensitive interpolation algorithm.

2. The GIS buffer generation method based on fuzzy semantics and geographic constraints according to claim 1, characterized in that, The membership values ​​corresponding to the parsed fuzzy descriptive terms include: Based on the fuzzy descriptor, a preset fuzzy semantic dictionary is queried to obtain the corresponding membership value, or a semantic understanding model based on deep learning is used to calculate the membership value corresponding to the fuzzy descriptor according to the context of the fuzzy descriptor in the natural language instruction. The fuzzy semantic dictionary stores multiple fuzzy descriptive words and their corresponding membership functions or membership value ranges.

3. The GIS buffer generation method based on fuzzy semantics and geographic constraints according to claim 2, characterized in that, The step of querying a preset fuzzy semantic dictionary based on the fuzzy descriptive words to obtain the corresponding membership value includes: If the fuzzy descriptive word exists in the preset fuzzy semantic dictionary, then the membership value corresponding to the fuzzy descriptive word is calculated using a deep learning-based semantic understanding model as the first membership value. Based on the fuzzy descriptive words, the corresponding membership value is obtained from the preset fuzzy semantic dictionary and used as the second membership value; The final membership value is determined based on the first membership value, the second membership value, and the difference between the first membership value and the second membership value. If the fuzzy descriptor is not included in the dictionary, the membership value corresponding to the fuzzy descriptor is calculated using a deep learning-based semantic understanding model as the final membership value.

4. The GIS buffer generation method based on fuzzy semantics and geographic constraints according to claim 3, characterized in that, The step of determining the final membership value based on the first membership value, the second membership value, and the difference between the first membership value and the second membership value includes: If the difference between the first membership value and the second membership value is less than a preset difference threshold, then the weighted average of the first membership value and the second membership value is calculated as the final membership value. If the difference between the first membership value and the second membership value is greater than or equal to a preset difference threshold, then the natural language instruction is subjected to complexity analysis to obtain a complexity value; and the final membership value is determined based on the complexity value.

5. The GIS buffer generation method based on fuzzy semantics and geographic constraints according to claim 4, characterized in that, The step of determining the final membership value based on the complexity value includes: If the complexity value is greater than or equal to the complexity threshold, then the first membership value is taken as the final membership value. If the complexity value is less than the complexity threshold, then the first membership value is used as the final membership value.

6. The GIS buffer generation method based on fuzzy semantics and geographic constraints according to claim 1, characterized in that, The calculation of the feature length corresponding to the geometric feature includes: Based on the geographic entity type and the intent direction expressed by the fuzzy descriptive words, a corresponding calculation strategy is selected from a preset feature length calculation rule base; The preset feature length calculation rule library defines the feature length calculation methods corresponding to different entity types or intent directions; The feature length corresponding to the geometric feature is calculated based on the calculation strategy.

7. A GIS buffer generation method based on fuzzy semantics and geographic constraints according to claim 6, characterized in that, The step of selecting a corresponding calculation strategy from a preset feature length calculation rule base based on the geographic entity type and the intent direction expressed by the fuzzy descriptive words includes: Based on the geographic entity type, a set of candidate calculation strategies applicable to the geographic entity type is selected from the preset feature length calculation rule base; The intent direction of the fuzzy descriptive words is analyzed to generate the corresponding intent direction semantic vector; Calculate the cosine similarity between the intent direction semantic vector and all standard policy semantic vectors in the preset feature length calculation rule base; the standard policy semantic vector is generated based on the standard intent description tag associated with each calculation policy defined in the preset feature length calculation rule base. The computational strategy corresponding to the standard strategy semantic vector with the highest cosine similarity to the semantic vector of the user intent direction in the candidate computational strategy set is selected as the final computational strategy.

8. A GIS buffer generation system based on fuzzy semantics and geographic constraints, characterized in that, include: The semantic parsing module is used to extract geographic entity keywords, geographic entity types and fuzzy descriptive words from the user's natural language commands through the semantic parsing engine, and to parse the fuzzy descriptive words to obtain the membership value of the corresponding fuzzy descriptive words; The data mapping module is used to input the fuzzy descriptive terms, membership values, and geographic entity types into a pre-trained language-space mapping model to obtain basic distance parameters. The basic distance parameters are either a baseline value with distance dimensions or a scaling factor relative to the typical scale of the entity. The language-space mapping model uses a neural network with a dual-channel output structure, where the first channel outputs a baseline value with distance dimensions, the second channel outputs a scaling factor relative to the typical scale of the entity, and the activated channel is selected based on the geographic entity type through a gating mechanism. The data calculation module is used to query the spatial database based on the geographic entity keywords, obtain the geometric features associated with the geographic entity, and calculate the feature length corresponding to the geometric features. The data correction module is used to correct the basic distance parameters based on the intent direction, membership value, geographic entity type and feature length expressed by the fuzzy descriptive words to obtain the geographic distance threshold. The first generation module is used to generate an initial two-dimensional buffer polygon based on the geographic distance threshold and the spatial geometric features of the geographic entity; The second generation module is used to obtain terrain data by querying the digital elevation model based on the initial two-dimensional buffer polygon, and generate a three-dimensional buffer surface through a slope-sensitive interpolation algorithm.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1 to 7.

Images