A land space planning data quality inspection method and system

Abstract

The embodiment of the application discloses a kind of land space planning data quality inspection method and system, method includes: using FME processing engine module, obtains original land space planning data, original land space planning data is preprocessed, and preprocessed land space planning data is obtained;Using FME processing engine module, the quality inspection of preprocessed land space planning data is carried out, and first quality inspection result is obtained;Using FME processing engine module, in the case where it is determined that preprocessed land space planning data needs to call spatial semantic AI engine module based on preset strategy, preprocessed land space planning data is transmitted to spatial semantic AI engine module;Using spatial semantic AI engine module, the semantic conflict inference of preprocessed land space planning data is carried out, and second quality inspection result is obtained, and second quality inspection result is transmitted to FME processing engine module;Using FME processing engine module, based on first quality inspection result and second quality inspection result, first initial quality inspection report is generated.

Description

Technical Field

[0001] This application relates to the interdisciplinary field of geographic information systems and artificial intelligence, and involves, but is not limited to, a method and system for quality inspection of land spatial planning data. Background Technology

[0002] Detailed land use planning is the legal basis for implementing land use control and development, protection, and construction, and its data quality is of paramount importance. Currently, the quality inspection methods in related technologies are based on tools such as FME (Feature Manipulation Engine), which achieve automated inspection by manually pre-writing clear and fixed rules (such as topology rules and attribute value range rules).

[0003] However, the quality inspection methods in related technologies have shortcomings such as a lack of ability to process complex spatial business logic and a lack of deep semantic consistency checks. Summary of the Invention

[0004] In view of this, embodiments of this application provide a method and system for quality inspection of land and space planning data, which at least solves the problems of lack of complex spatial business logic processing capabilities and lack of deep semantic consistency checks.

[0005] The technical solution of this application embodiment is implemented as follows: In a first aspect, embodiments of this application provide a method for quality inspection of land spatial planning data, applied to a land spatial planning data quality inspection system. The system includes: a human-computer collaborative interaction module, an FME processing engine module, and a spatial semantic AI engine module. The method includes: Using the FME processing engine module, the original land spatial planning data submitted by the user through the human-computer collaborative interaction module is obtained, and the original land spatial planning data is preprocessed to obtain preprocessed land spatial planning data. The preprocessed land spatial planning data is quality checked using the FME processing engine module to obtain the first quality check result; Using the FME processing engine module, when it is determined based on a preset strategy that the preprocessed land spatial planning data needs to be checked by calling the spatial semantic AI engine module, the preprocessed land spatial planning data is transmitted to the spatial semantic AI engine module. Using the spatial semantic AI engine module, semantic conflict reasoning is performed on the preprocessed land spatial planning data to obtain a second quality inspection result, and the second quality inspection result is transmitted to the FME processing engine module. Using the FME processing engine module, a first initial quality inspection report is generated based on the first quality inspection result and the second quality inspection result.

[0006] Secondly, embodiments of this application provide a land spatial planning data quality inspection system, the system comprising: a human-computer collaborative interaction module, an FME processing engine module, and a spatial semantic AI engine module, wherein: The FME processing engine module is used to acquire the original land spatial planning data submitted by the user through the human-computer collaborative interaction module, and to preprocess the original land spatial planning data to obtain preprocessed land spatial planning data. The FME processing engine module is used to perform quality checks on the preprocessed land spatial planning data and obtain the first quality check result; The FME processing engine module is used to transmit the preprocessed territorial spatial planning data to the spatial semantic AI engine module when it is determined, based on a preset strategy, that the preprocessed territorial spatial planning data needs to be checked by calling the spatial semantic AI engine module. The spatial semantic AI engine module is used to perform semantic conflict reasoning on the preprocessed land spatial planning data, obtain a second quality inspection result, and transmit the second quality inspection result to the FME processing engine module. The FME processing engine module is used to generate a first initial quality inspection report based on the first quality inspection result and the second quality inspection result.

[0007] The beneficial effects of the technical solutions provided in this application include at least the following: By constructing a collaborative quality inspection architecture for the FME processing engine module and the spatial semantic AI engine module, this approach breaks through the limitations of traditional single-rule quality inspection, which can only cover surface-level explicit errors. It achieves full-scenario coverage from basic error verification to deep semantic conflict detection, solving both the regular error inspection needs of clearly defined rules in land spatial planning data and accurately identifying semantic conflicts with implicit complex logic. Simultaneously, it clarifies the standardized process of data acquisition, preprocessing, rule-based quality inspection, AI-powered intelligent quality inspection, and report generation, ensuring the standardization and repeatability of quality inspection operations, effectively reducing human error. Furthermore, it dynamically determines whether to call the AI ​​engine through preset strategies, avoiding resource waste caused by indiscriminate AI calls, balancing quality inspection efficiency and detection depth. Finally, by fusing the first and second quality inspection results to generate an initial quality inspection report, it provides more comprehensive and accurate data support for subsequent decision-making, improves the processing capabilities of complex spatial business logic, and realizes deep semantic consistency checks based on AI semantic reasoning. Attached Figure Description

[0008] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort, wherein: Figure 1 A flowchart illustrating a method for quality inspection of land and space planning data provided in this application embodiment; Figure 2 A general architecture block diagram of a land spatial planning data quality inspection system provided in this application embodiment; Figure 3 This is a schematic diagram of the composition structure of a land spatial planning data quality inspection system provided in an embodiment of this application. Detailed Implementation

[0009] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. The following embodiments are used to illustrate this application, but are not intended to limit the scope of this application. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0010] In the following description, references are made to “some embodiments,” which describe a subset of all possible embodiments. However, it is understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0011] It should be noted that the terms "first, second, and third" used in the embodiments of this application are merely to distinguish similar objects and do not represent a specific ordering of objects. It is understood that "first, second, and third" can be interchanged in a specific order or sequence where permitted, so that the embodiments of this application described herein can be implemented in an order other than that illustrated or described herein.

[0012] It will be understood by those skilled in the art that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which the embodiments of this application pertain. It should also be understood that terms such as those defined in general dictionaries should be understood to have a meaning consistent with their meaning in the context of the prior art, and should not be interpreted in an idealized or overly formal sense unless specifically defined as herein.

[0013] This application provides a method for quality inspection of land and space planning data, applied to electronic devices. The electronic devices include, but are not limited to, mobile phones, laptops, tablets, handheld internet devices, multimedia devices, streaming media devices, mobile internet devices, wearable devices, or other types of electronic devices. The functions implemented by this method can be achieved by a processor in the electronic device calling program code. The program code can be stored in a computer storage medium; therefore, the electronic device includes at least a processor and a storage medium. The processor can be used to perform processing based on a land and space planning data quality inspection process, and the memory can be used to store the data required and generated during the land and space planning data quality inspection process.

[0014] Figure 1 A flowchart illustrating a land spatial planning data quality inspection method provided in this application embodiment is applied to, for example... Figure 2 The illustrated land spatial planning data quality inspection system includes a human-computer collaborative interaction module, an FME processing engine module, and a spatial semantic AI engine module, as shown below. Figure 1 As shown, the method includes at least the following steps: Step S110: Using the FME processing engine module, the original land spatial planning data submitted by the user through the human-computer collaborative interaction module is obtained, and the original land spatial planning data is preprocessed to obtain preprocessed land spatial planning data. Among them, territorial spatial planning data is a standardized spatial data system supporting the entire process of territorial spatial planning, including compilation, approval, implementation, and supervision. It is a digital description of various natural resources, artificial facilities, and planning constraints within the territorial space, serving as core data support for territorial spatial development and protection, land use control, and project approval. FME is a professional geospatial data processing and integration tool. The FME processing engine module utilizes and configures FME functions related to territorial spatial planning data quality inspection scenarios. Users submit multi-source, heterogeneous detailed territorial spatial planning data as input data through a human-computer collaborative interaction module, such as data in DWG, GDB, SHP, and Excel formats. The FME processing engine module cleans and standardizes the input data, including coordinate system conversion to CGCS2000, preliminary topology error correction, data format standardization, and cleaning and regularization of basic attribute fields. This step outputs clean and standardized data (i.e., preprocessed territorial spatial planning data) through data preprocessing, providing high-quality input for subsequent intelligent analysis.

[0015] Step S120: Use the FME processing engine module to perform quality inspection on the preprocessed land spatial planning data to obtain the first quality inspection result; Specifically, predefined quality control rules can be applied to the preprocessed data to generate a first quality control result. Specifically, based on a pre-configured FME converter and workflow, a first type of check, namely a rule-based check, is performed on the standardized preprocessed land spatial planning data. This type of check targets rules with existing clear standards, including attribute checks (such as field non-emptiness, value range validity, code compliance, etc.), topological checks (such as whether features overlap, gaps, self-intersections, etc.), and basic spatial relationship checks (such as whether point features are located within polygon features, whether line features coincide with polygon feature boundaries, etc.). This step outputs the first quality control result.

[0016] Step S130: Using the FME processing engine module, if it is determined based on a preset strategy that the preprocessed land spatial planning data needs to be checked by calling the spatial semantic AI engine module, the preprocessed land spatial planning data is transmitted to the spatial semantic AI engine module. Specifically, based on preset strategies, it can determine which data or types of checks require calling the spatial semantic AI engine module. For example, it is necessary to call it when dealing with implicit and uncertain complex spatial business rules such as consistency between land use and landmarks, coordination of spatial features, and rationality of facility service radius.

[0017] Step S140: Using the spatial semantic AI engine module, perform semantic conflict reasoning on the preprocessed land spatial planning data to obtain a second quality inspection result, and transmit the second quality inspection result to the FME processing engine module. Semantic conflict reasoning refers to understanding the deep logic of preprocessed land spatial planning data and identifying hidden conflicts that do not conform to planning rules. Semantic conflicts may include: a plot of land is designated as "R2 residential land" but the supporting facilities are "large industrial plants," violating the semantic rule that "polluting industrial facilities shall not be built on residential land"; a building in a historical preservation area has a height exceeding 80 meters, violating the control requirement that "the height of buildings in historical blocks shall not exceed 15 meters"; and a garbage transfer station is less than 50 meters away from a nearby primary school, violating the "safe distance requirements between public service facilities and polluting facilities."

[0018] The spatial semantic AI engine module is an intelligent semantic analysis module designed specifically for land spatial planning data and based on artificial intelligence (AI) technology. The spatial semantic AI engine module can perform deep intelligent reasoning on the preprocessed land spatial planning data, find semantic conflicts that the FME processing engine module cannot handle, generate a second quality inspection result, and then send the second quality inspection result back to the FME processing engine module.

[0019] Step S150: Using the FME processing engine module, a first initial quality inspection report is generated based on the first quality inspection result and the second quality inspection result.

[0020] The FME processing engine module can receive and integrate the first quality inspection result obtained based on explicit quality inspection rules and the second quality inspection result obtained based on AI semantic conflict reasoning, perform unified structured processing, and automatically generate a comprehensive quality inspection report (i.e., the first initial quality inspection report) with the ability to identify deep semantic conflicts and verify compliance with complex spatial rules.

[0021] In the above embodiments, by constructing a collaborative quality inspection architecture of the FME processing engine module and the spatial semantic AI engine module, the limitations of traditional single-rule quality inspection, which can only cover surface-level explicit errors, are overcome. This achieves full-scenario coverage from basic error verification to deep semantic conflict detection, solving both the regular error quality inspection needs of clearly defined rules in land spatial planning data and accurately identifying semantic conflicts with implicit complex logic. Simultaneously, a standardized process for data acquisition, preprocessing, rule-based quality inspection, AI-powered intelligent quality inspection, and report generation is established, ensuring the standardization and repeatability of quality inspection operations, effectively reducing human error. Furthermore, a preset strategy dynamically determines whether to call the AI ​​engine, avoiding resource waste caused by indiscriminate AI calls, balancing quality inspection efficiency and detection depth. Finally, by fusing the first and second quality inspection results to generate an initial quality inspection report, a more comprehensive and accurate data support is provided for subsequent decision-making, improving the processing capability of complex spatial business logic and realizing deep semantic consistency checks based on AI semantic reasoning.

[0022] In some embodiments, the FME processing engine module includes a data access and preprocessing submodule, a rule quality inspection submodule, a quality inspection scheduling submodule, and a quality inspection report generation and output submodule; the spatial semantic AI engine module includes a spatial semantic embedding model submodule and an AI service interface submodule. Step S110, "Using the FME processing engine module, acquiring the original land spatial planning data submitted by the user through the human-computer collaborative interaction module, and preprocessing the original land spatial planning data to obtain preprocessed land spatial planning data," includes: Step S1101: Using the data access and preprocessing submodule, obtain the original land space planning data submitted by the user through the human-computer collaborative interaction module, and preprocess the original land space planning data to obtain preprocessed land space planning data. The data access and preprocessing submodule provides a unified data access interface that supports parsing planning data in various formats such as DWG, GDB, SHP, and Excel. It is also responsible for coordinate system transformation, preliminary topological relationship repair, data format standardization, and attribute field cleaning and regularization. It can perform data preprocessing based on various converters in FME Workspace, such as the coordinate transformation converter Reprojector, the geometry validation converter GeometryValidator, and the attribute manager AttributeManager. The data format standardization can be handled by FME's own format parsing and output capabilities.

[0023] Step S120, "Using the FME processing engine module to perform quality inspection on the preprocessed land spatial planning data and obtain a first quality inspection result," includes: Step S1201: Use the rule quality inspection submodule to perform quality inspection on the preprocessed land and space planning data to obtain the first quality inspection result; The rule-based quality inspection submodule primarily executes predefined, explicit quality inspection rules. It is equipped with a flexibly configurable rule library, covering attribute rules (such as non-empty validation and value range restrictions), topology rules (such as overlap detection and gap identification), and basic spatial relationship rules (such as inclusion judgment and adjacency verification). Its basic framework is a visual quality inspection workflow based on FME. This workflow is constructed from multiple core converters, including AttributeValidator for attribute rule checking, TopologyBuilder and TopologyValidator for topology rule checking, SpatialRelator for basic spatial relationship checking, and the general condition judgment tool Tester.

[0024] Step S130, "Utilizing the FME processing engine module, when it is determined based on a preset strategy that the preprocessed land spatial planning data needs to be checked by calling the spatial semantic AI engine module, the preprocessed land spatial planning data is transmitted to the spatial semantic AI engine module," includes: Step S1301: Using the quality inspection scheduling submodule, if it is determined based on a preset strategy that the preprocessed land spatial planning data needs to be checked by calling the spatial semantic AI engine module, the preprocessed land spatial planning data is transmitted to the spatial semantic embedding model submodule through the AI ​​service interface submodule. In some embodiments, the preset strategy includes: determining the task type and data characteristics to be inspected based on the preprocessed land spatial planning data; and determining that the spatial semantic AI engine module needs to be invoked for inspection when the task type characterization involves complex logic requiring semantic association analysis, or when the data characteristic characterization involves special land use types. The task types include checks on land use nature and indicator consistency, landscape coordination, and spatial relationship rationality; the data characteristics include land use attributes, supporting facility attributes, spatial location attributes, and building height attributes.

[0025] If the task type involves complex logic requiring semantic association analysis (such as constraint matching between land use and supporting facilities, and judgment of the harmony between architectural style and surrounding environment), or if the data features correspond to special types such as mixed land use or historical protection land, or if reasoning is required based on multi-factor contextual information, then it is determined that the spatial semantic AI engine module needs to be called for inspection. By pre-setting judgment strategies for task types and data features, intelligent adaptation of the quality inspection mode is achieved. For complex logic requiring semantic association analysis (such as the rationality of facility allocation) or special land use types (such as historical protection land), AI semantic reasoning is prioritized to ensure the depth of quality inspection. For simple rule-based errors (such as missing attributes or topological overlap), only FME rule-based quality inspection is needed, effectively balancing quality inspection efficiency and detection depth. Simultaneously, this strategy avoids blindly calling the AI ​​engine for all data, reducing unnecessary computational resource consumption and improving the overall system operating efficiency. Furthermore, the pre-set strategy can flexibly adjust the judgment criteria for task types and data features, adapting to the quality inspection needs of land and space planning in different regions and at different levels, thus enhancing the system's versatility.

[0026] The quality inspection scheduling submodule serves as a bridge connecting the FME processing engine module and the spatial semantic AI engine module. Its core responsibilities include scene recognition, data encapsulation and retrieval, and result reception and relay. The scene recognition stage determines, based on preset strategies, which data or types of inspections require invoking the spatial semantic AI engine module, such as when complex logic involving land use consistency and indicator checks, or landscape inspections. The data encapsulation and retrieval stage extracts specific element sets and spatial context information from the generated preprocessed land spatial planning data according to actual quality inspection needs, such as all plots, control lines, and public service facilities within a block area. It then serializes the data to be inspected and its context information into standard formats such as GeoJSON, and integrates the encapsulated data packets with the quality inspection request using the built-in HTTPCaller converter, in a RESTful manner. The API approach actively calls the corresponding services of the spatial semantic AI engine based on the integrated content. The result receiving and relaying stage is responsible for receiving the JSON format quality inspection results (i.e., the second quality inspection results) returned by the spatial semantic AI engine module, converting them into FME internal feature attributes, and passing them to the spatial semantic AI engine module (providing feedback to the spatial semantic AI engine module). The implementation of this quality inspection scheduling sub-module depends on FME HTTPCaller (used for data encapsulation and calling), JSONFragmenter (used for result receiving and relaying, such as the decomposition and format conversion of the second quality inspection results), and custom scheduling logic.

[0027] The AI ​​service interface submodule is responsible for encapsulating the AI ​​functions of the spatial semantic AI engine module into a lightweight, reusable RESTful API service. In its daily work, it receives data packets sent by the FME processing engine module, calls the underlying model to perform calculations, and then returns the final second quality inspection result to the FME processing engine module. The AI ​​service interface submodule is the key to achieving loosely coupled integration with the FME processing engine module. It is built by using Python Web frameworks such as FastAPI and Flask (the core framework for building Web services) and RESTful API (used to define the communication protocol and interface specification of the service).

[0028] Step S140, "Using the spatial semantic AI engine module, performing semantic conflict reasoning on the preprocessed land spatial planning data to obtain a second quality inspection result, and transmitting the second quality inspection result to the FME processing engine module," includes: Step S1401: Using the spatial semantic embedding model in the spatial semantic embedding model submodule, perform semantic conflict reasoning on the preprocessed land spatial planning data to obtain a second quality inspection result, and transmit the second quality inspection result to the quality inspection scheduling submodule through the AI ​​service interface submodule. The spatial semantic embedding model, a core component of the spatial semantic embedding model submodule, is the heart of intelligent quality inspection. As the core algorithm model, it employs a Graph Neural Network (GNN) architecture, mapping planning elements (such as individual plots) and their spatial context (such as neighboring elements and inclusions) into high-dimensional vectors (i.e., embedding vectors). These high-dimensional vectors implicitly encode the geometric features, attribute information, and business semantics of the elements. In terms of training, the spatial semantic embedding model submodule uses historical compliance data and comparative learning for training, ensuring that semantically related elements are close to each other in the vector space, while semantically unrelated elements are far apart. In terms of inference, the quality inspection process calculates the similarity or distance between the vectors of the elements to be inspected, quantifying the logical consistency between elements using a numerical "semantic conflict loss value." This process is supported by deep learning frameworks such as Graph Neural Networks (GNN), PyTorch / TensorFlow (providing a runtime environment and toolkit for the model or algorithm), and comparative learning algorithms.

[0029] Specifically, the spatial semantic embedding model constructs triples (anchor point, positive sample, negative sample) for comparative learning training. The positive sample is semantically related to the anchor point element (such as adjacent plots or supporting facilities), while the negative sample contains artificially constructed semantic conflict cases (such as industrial land being associated with a primary school). The model uses a graph neural network to map elements into high-dimensional vectors. By optimizing the triple loss function, the vector distance between the anchor point and the positive sample is made less than the distance between the anchor point and the negative sample by at least a boundary value α. This encodes complex spatial business logic into geometric relationships in vector space, enabling automated semantic conflict detection based on vector similarity calculation.

[0030] Step S150, "using the FME processing engine module, based on the first quality inspection result and the second quality inspection result, to generate a first initial quality inspection report," includes: Step S1501: Using the quality inspection report generation and output submodule, obtain the first quality inspection result transmitted by the rule quality inspection submodule and the second quality inspection result transmitted by the quality inspection scheduling submodule, and generate a first initial quality inspection report based on the first quality inspection result and the second quality inspection result.

[0031] The quality inspection report generation and output submodule integrates results from both the rule-based quality inspection submodule and the quality inspection scheduling submodule, summarizing, classifying, and categorizing the results to ultimately generate a structured comprehensive quality inspection report. The report supports multiple formats, including HTML and PDF, and includes visual links for quick error location. Its implementation relies on FMEHTMLReportGenerator (for generating structured HTML reports), Excel Writer (for exporting quality inspection results to Excel spreadsheets, complementing the quality inspection report function), and custom template technology (for defining the unified format and content layout of the quality inspection report).

[0032] In one embodiment, the quality inspection report generation and output submodule receives and merges the first and second quality inspection results, performs unified structured processing, and automatically generates a comprehensive quality inspection report containing the error level, spatial location (plot ID), detailed description, and AI-powered intelligent repair suggestions. The report supports HTML, PDF, and other formats and provides a visual link, allowing for one-click location of erroneous plots within the GIS (Geographic Information System) platform.

[0033] In the above embodiments, by further decomposing the FME processing engine module and the spatial semantic AI engine module into specialized sub-modules such as data access and preprocessing, rule quality inspection, and quality inspection scheduling, the responsibilities of each module are clearly defined and loosely coupled. This not only facilitates the upgrading, maintenance, or replacement of individual sub-modules, effectively improving the system's scalability and adaptability to meet the iterative quality inspection needs in different scenarios, but also clearly defines the data transmission paths and collaboration logic between sub-modules, ensuring a smooth and stable collaborative process between rule quality inspection and AI semantic reasoning. At the same time, through precise division of labor, the rule quality inspection sub-module focuses on traditional explicit rule verification, the spatial semantic embedding model sub-module specializes in deep semantic conflict reasoning, and the AI ​​service interface sub-module ensures cross-module data compatibility and transmission. Each sub-module performs its own duties, significantly improving the overall efficiency and accuracy of quality inspection.

[0034] In some embodiments, the spatial semantic AI engine module further includes a knowledge graph management submodule; Step S1401, "using the spatial semantic embedding model in the spatial semantic embedding model submodule to perform semantic conflict reasoning on the preprocessed land spatial planning data to obtain the second quality inspection result," includes: Step S14011: Using the spatial semantic embedding model submodule, if it is determined that domain knowledge needs to be supplemented based on the preprocessed land spatial planning data, the target knowledge is obtained from the knowledge graph of the knowledge graph management submodule. The knowledge graph management submodule is mainly responsible for the construction, storage, querying and maintenance of the knowledge graph in the field of land and space planning. Based on graph databases such as Neo4j, the knowledge graph clearly defines the core concepts in the planning field (such as land use, facilities, control lines), entity attributes and business relationships between entities (such as supporting facilities, prohibited facilities, and belonging facilities). Its implementation relies on graph database technology, ontology construction technology and Cypher query language (Neo4j's dedicated query language).

[0035] Step S14012: Using the spatial semantic embedding model in the spatial semantic embedding model submodule and the target knowledge, semantic conflict reasoning is performed on the preprocessed land spatial planning data to obtain the second quality inspection result.

[0036] Among them, after receiving the quality inspection request from the FME processing engine module, the spatial semantic AI engine module calls the knowledge graph management submodule to obtain domain knowledge, performs reasoning through the spatial semantic embedding model submodule, and generates a second quality inspection result.

[0037] In the above embodiments, by adding a knowledge graph management submodule to the spatial semantic AI engine module, professional knowledge support in the field of land and space planning is provided for the spatial semantic embedding model. This effectively solves the reasoning bias problem caused by the lack of domain common sense in the AI ​​model, making semantic conflict detection more accurate and professional. At the same time, a dual-driven quality inspection mode of data and knowledge is realized, so that AI reasoning not only relies on the characteristics of the data itself, but can also perform logical deduction based on the planning business rules in the knowledge graph. This successfully identifies implicit conflicts where the data itself is error-free but violates domain rules. Moreover, the knowledge graph can uniformly manage and update the planning domain rules, adapting to the planning quality inspection needs of different regions and types without modifying the core logic of the AI ​​model, greatly improving the flexibility and adaptability of the system.

[0038] In some embodiments, the human-computer collaborative interaction module includes a result review submodule and a feedback collection and push submodule, and the method further includes: Step S160: Using the quality inspection report generation and output submodule, the first initial quality inspection report is sent to the result review submodule; Step S170: Using the result review submodule, display the first initial quality inspection report, and in response to the user's revision operation on the first initial quality inspection report, generate a target quality inspection report and send the target quality inspection report to the feedback collection and push submodule; The human-machine collaborative interaction module is a unified portal for the system to interact with users (such as quality inspectors and planning experts), ensuring that people are in the loop. It has two core sub-modules: a result review sub-module and a feedback collection and push sub-module.

[0039] The results review submodule presents potential errors identified by the spatial semantic AI engine module through a clear and user-friendly interface. The content includes the error location, error type, confidence level, and reasoning basis, making it convenient for experts to confirm, reject, or correct these potential errors. Its implementation relies on web front-end technologies such as Vue.js and React, as well as map visualization libraries such as Leaflet and Mapbox GL JS.

[0040] Step S180: Using the feedback collection and push submodule, the target quality inspection report is converted into a structured training sample, and transmitted to the spatial semantic embedding model submodule and the knowledge graph management submodule through the AI ​​service interface submodule, respectively, to perform incremental learning of the spatial semantic embedding model and update of the knowledge graph.

[0041] The feedback collection and push submodule automatically transforms the results confirmed by experts (especially new error patterns) into structured training samples and pushes them to the sample library of the spatial semantic AI engine module. This provides data support for the incremental learning of the spatial semantic embedding model and is the core link that enables the system to evolve. Its implementation relies on form technology responsible for structured feedback and message queue technology such as RabbitMQ and Kafka responsible for reliable data transmission, or API call technology responsible for rapid data connection.

[0042] Users review the initial quality inspection report through the result review submodule of the human-computer collaborative interaction module. The system provides a human-computer interaction interface, allowing experts to review, confirm, or correct the second quality inspection results. The feedback collection and push submodule automatically converts the quality inspection results confirmed by experts, especially previously unknown new error types, into structured training samples and pushes them to the spatial semantic AI engine module via RabbitMQ or Kafka. This is used for incremental learning and knowledge graph expansion and optimization of the spatial semantic embedding model, thereby achieving continuous self-improvement and evolution of the entire system.

[0043] In the above embodiments, a complete closed loop of AI quality inspection, manual review, sample feedback, and model optimization / knowledge graph update is constructed through the result review submodule and feedback collection and push submodule in the human-computer collaborative interaction module. Users can revise the initial quality inspection report to correct conflicting cases of AI misjudgment and omission, ensuring the accuracy of the final target quality inspection report and avoiding quality inspection risks caused by AI's single decision. At the same time, the feedback collection and push submodule transforms the revised target quality inspection report into structured training samples, which are used for incremental learning of the spatial semantic embedding model and updating of the knowledge graph, respectively. This allows the system to continuously accumulate experience as it is used, continuously improve the inference accuracy of the AI ​​model and the completeness of the knowledge graph, realize the self-evolution of the system, and gradually enhance the quality inspection capability over long-term use, effectively reducing subsequent maintenance costs.

[0044] In some embodiments, step S14011, "using the spatial semantic embedding model submodule, and determining that domain knowledge needs to be supplemented based on the preprocessed territorial spatial planning data, to obtain target knowledge from the knowledge graph of the knowledge graph management submodule," includes: Step S201: Using the spatial semantic embedding model submodule, the preprocessed land spatial planning data is parsed to obtain multiple elements in the preprocessed land spatial planning data and the attribute value corresponding to each element; based on the attribute value corresponding to each element, a query request for the corresponding element against the knowledge graph is generated, and each query request is transmitted to the knowledge graph management submodule. Step S202: Using the knowledge graph management submodule, search for the business definition, constraints and relationships of the entities associated with the corresponding elements in the knowledge graph according to each query request; Step S203: Using the spatial semantic embedding model submodule, the business definitions, constraints, and relationships of the entities corresponding to all elements are determined as target knowledge.

[0045] In the above embodiments, the preprocessed land spatial planning data is parsed through the spatial semantic embedding model submodule. Based on the elements and their corresponding attribute values, targeted knowledge graph query requests are generated, avoiding the blind acquisition of irrelevant knowledge and significantly improving the efficiency of knowledge retrieval and the targeting of AI reasoning. At the same time, it is clear that three types of core knowledge—business definitions, constraints, and relationships—are extracted from the knowledge graph, which fully supports the semantic conflict reasoning of the AI ​​model. This ensures that the reasoning process has clear domain rules as its basis, reduces meaningless reasoning, and by accurately binding data elements with entities and rules in the knowledge graph, AI reasoning is not only based on data feature similarity but also combined with business logic for judgment, greatly improving the rationality and interpretability of conflict detection.

[0046] In one embodiment, after receiving a quality inspection request, the spatial semantic AI engine module will carry out a series of core operations to achieve reasoning and conflict detection: first, it will call the built-in knowledge graph of the land and space planning domain to obtain the business definitions, constraints and relationships of related entities such as "R2 residential land" and "kindergarten", so as to provide domain knowledge support for subsequent semantic analysis.

[0047] In some embodiments, step S14012, "using the spatial semantic embedding model in the spatial semantic embedding model submodule and the target knowledge to perform semantic conflict reasoning on the preprocessed land spatial planning data to obtain a second quality inspection result," includes: Step S204: Using the spatial semantic embedding model in the spatial semantic embedding model submodule, each element and its corresponding spatial relationship are converted into a corresponding high-dimensional feature vector. Step S205: Based on the constraints and relationships in the target knowledge and the spatial topological relationships in the preprocessed land planning data, select the element pairs that need to be semantically conflict detected; Step S206: For each pair of elements, determine the cosine similarity or Euclidean distance between the corresponding high-dimensional feature vectors; Step S207: Based on the cosine similarity or Euclidean distance of each element pair, generate the semantic conflict loss value of the corresponding element pair. Step S208: Based on the semantic conflict loss value corresponding to each element pair, generate the conflict determination result and confidence level of the corresponding element pair; based on the business definition and constraints of the elements contained in each element pair in the target knowledge, determine the conflict type of the corresponding element pair. Step S209: Based on the conflict determination result, conflict type, and confidence level of each element pair, generate a second quality inspection result.

[0048] The received planning elements can be input into a pre-trained spatial semantic embedding model. This model uses a graph neural network (GNN) as its core architecture, which can transform each element (node) and its spatial relationship (edge) into a high-dimensional feature vector. The vector implicitly encodes the geometric features, attribute semantics, and spatial context information of the elements, realizing the transformation of unstructured spatial information into a computable vector form. Finally, in the high-dimensional semantic vector space, a "semantic conflict loss value" is generated by calculating the cosine similarity or Euclidean distance between key element pairs. For example, the similarity between a single "land parcel" vector and its associated "planning indicator" vector, or the distance between the "industrial land parcel" vector and the "residential land parcel" vector, can be calculated to identify potential semantic conflicts.

[0049] After the spatial semantic AI engine module completes the reasoning, it will send the reasoning results, including "whether there is a conflict", "conflict type" and "confidence level", back to the FME processing engine module in JSON format. This feedback result is the second quality inspection result generated by this intelligent quality inspection process.

[0050] In the above embodiments, the elements and spatial relationships are converted into high-dimensional feature vectors through a spatial semantic embedding model. A "semantic conflict loss value" is generated by calculating cosine similarity or Euclidean distance, transforming the ambiguous "semantic conflict" into a quantifiable numerical indicator, making conflict judgment more objective and accurate, and avoiding subjective assumptions. At the same time, based on domain knowledge constraints and spatial topological relationships, the element pairs that need to be conflict detected are selected, avoiding indiscriminate pairwise comparisons of all elements, which significantly reduces the amount of computation and improves the efficiency of AI reasoning. In addition, the generated second quality inspection result includes the conflict judgment result, conflict type, and confidence level. It not only clarifies "whether there is a conflict" but also provides detailed information on the "conflict type" and "judgment confidence level," providing more comprehensive reference information for user review and subsequent rectification, and significantly improving the practicality of the quality inspection report.

[0051] In some embodiments, the method further includes: Step S121: Using the FME processing engine module, if it is determined based on a preset strategy that the preprocessed land spatial planning data does not need to be checked by calling the spatial semantic AI engine module, a second initial quality inspection report is generated based on the first quality inspection result.

[0052] In the above embodiments, by optimizing the quality inspection process branches, for simple quality inspection scenarios that do not require AI semantic reasoning, a second initial quality inspection report can be directly generated based on the first quality inspection result. This eliminates redundant steps such as AI calls and data transmission, significantly shortening the quality inspection cycle and greatly improving the quality inspection efficiency in simple scenarios. At the same time, this design enables the system to adapt to diverse quality inspection needs, satisfying both the in-depth quality inspection needs of complex data and the rapid quality inspection needs of simple data, thus improving the system's practicality and user experience. In addition, it reduces unnecessary AI module startup and cross-module data transmission, lowering the system's computing and communication resource consumption and improving the system's operational economy.

[0053] This application provides an intelligent quality inspection method, applicable to, for example... Figure 2 The illustrated land spatial planning data quality inspection system includes the following steps in its method: Users first submit detailed planning data through the human-computer collaboration module at the application layer. Then, the data access and preprocessing submodule of the FME processing engine module in the core processing layer starts working, cleaning and standardizing the submitted data. After data preprocessing is complete, the data simultaneously enters the rule-based quality inspection submodule and the quality inspection scheduling submodule.

[0054] The quality inspection scheduling submodule, acting as a bridge between the FME processing engine module and the spatial semantic AI engine module of the core processing layer, identifies whether to call the spatial semantic AI engine module for inspection based on preset strategies. For example, when complex logic such as land use consistency and landscape inspection is involved, it serializes the data to be inspected and its context information into standard formats such as GeoJSON, and then actively calls the AI ​​service interface submodule of the spatial semantic AI engine module via a RESTful API through the built-in HTTPCaller converter.

[0055] After receiving the data, the AI ​​service interface submodule will hand it over to the spatial semantic embedding model submodule for inference. If domain knowledge needs to be supplemented during the process, the spatial semantic embedding model submodule will further query the knowledge graph management submodule. After the inference is completed, the spatial semantic AI engine module will return the quality inspection results to the quality inspection scheduling submodule of the FME processing engine module.

[0056] Next, the quality inspection report generation submodule of the FME processing engine module will integrate all the results of traditional rule-based quality inspection and intelligent quality inspection to generate a structured comprehensive quality inspection report, which will then be presented to the user through the human-computer collaborative interaction module. After the user reviews the AI ​​results through the human-computer collaborative interaction module, their feedback will be collected by the feedback collection and push submodule and then pushed to the spatial semantic AI engine module to support the optimization of the spatial semantic embedding model.

[0057] This modular design ensures that the system has high cohesion and low coupling characteristics, with each module having a single responsibility. This not only facilitates development and maintenance but also enables independent upgrades, fully demonstrating the advanced nature and practicality of this application.

[0058] In some embodiments, the land and space planning data quality inspection system further includes a data and knowledge management module at the data and knowledge layer. This module, as the cornerstone of the system, is mainly responsible for the unified management of various data resources. Specifically, it covers three major categories: a raw and standard data warehouse containing multi-source planning data, a land and space planning knowledge graph database, and a model parameter library and training sample library. The raw and standard data warehouse is used to store the raw data to be inspected and the preprocessed standard data. The land and space planning knowledge graph database stores the knowledge graph managed by the module. The model parameter library and training sample library store the parameters, training samples, and feedback samples of the spatial semantic embedding model. The implementation of this module relies on relational databases such as PostgreSQL / PostGIS, graph databases such as Neo4j, and file systems or object storage.

[0059] This application belongs to the interdisciplinary field of Geographic Information Systems and Artificial Intelligence, specifically focusing on the automated quality inspection of detailed land spatial planning data. Addressing the problems of FME-based quality inspection technologies, such as the lack of complex spatial business logic processing, absence of deep semantic consistency checks, and high maintenance costs due to system rigidity, this application proposes a modular technical solution driven by both an FME processing engine and a spatial semantic AI engine. The FME processing engine module is responsible for standardized preprocessing (including coordinate transformation and topology repair) of multi-source heterogeneous data (DWG (drawing database format), GDB (geographic database format), SHP (shape file format), etc.), traditional rule quality inspection (attribute / topology / basic spatial relationship verification), and intelligent scheduling functions. It pushes complex scenario requirements to the spatial semantic AI engine via an HTTPCaller converter and a RESTful API. The AI ​​engine provides business rule support based on a knowledge graph in the field of land spatial planning built on Neo4j. Combined with a graph neural network (GNN) semantic embedding model, it maps planning elements and spatial relationships into high-dimensional vectors. It generates a "semantic conflict loss value" through cosine similarity / Euclidean distance calculation, achieving intelligent detection of implicit rules. The system simultaneously constructs a human-machine collaborative feedback loop, automatically feeding back expert-confirmed error samples for incremental learning of the AI ​​model. Finally, it outputs a structured quality inspection report (supporting HTML / PDF and GIS visualization positioning) that integrates dual-path results. This application achieves a leap in quality inspection capabilities from syntax to semantics, possesses self-evolving characteristics and strong industrial applicability, and can be quickly integrated into existing planning information workflows.

[0060] In some embodiments, a qualitative quality inspection is conducted to ensure consistency between residential plots and supporting infrastructure indicators. In the detailed land use planning of a new urban area, there is a hidden contradiction in its GDB format planning data: some R2 residential plots are labeled with a "floor area ratio of 2.5" but require the construction of a "12-class kindergarten," while the local "Technical Regulations for Land Use Planning Management" clearly state that "a 12-class kindergarten is only required for R2 plots with a floor area ratio ≥ 3.0." Traditional FME-based quality inspection can only verify whether the floor area ratio is within the "1.0-3.5" range and cannot identify the hidden conflict between the indicators and the supporting infrastructure. This problem needs to be addressed through the solution proposed in this application.

[0061] First, the system starts the data access and preprocessing submodule of the FME processing engine module, imports the planning data of the new area in GDB format, and uses the built-in Reprojector converter of FME to convert the data coordinates into the CGCS2000 national geodetic coordinate system. Then, the GeometryValidator converter automatically repairs the topological gaps existing in the boundaries of the plots. At the same time, the AttributeManager converter is used to standardize the core attribute fields such as "land use code", "plot ratio" and "supporting facility type" to ensure that the field format is uniform and the content is complete, and finally outputs clean and standardized data.

[0062] Next, the standardized data is fed into the traditional rule quality inspection submodule of the FME processing engine module. This module uses the AttributeValidator converter to verify whether the value of the "plot ratio" field is within the preset compliance range of "1.0-3.5". After confirming that the plot ratio of 2.5 of the target plot meets the basic syntax requirements, the Tester converter filters out all residential plots with "land use code = R2" and generates a preliminary quality inspection result of "no syntax errors but further semantic logic check is required".

[0063] Subsequently, the intelligent quality inspection scheduling submodule of the FME processing engine module performs scenario identification on the preliminary quality inspection results (i.e., the first quality inspection result). It finds that the combination of "R2 plot and supporting facilities" involves the correlation logic between land use nature and supporting indicators, which belongs to a complex scenario that cannot be defined by clear rules. Therefore, the intelligent quality inspection process is triggered. This submodule first extracts the geometric information (plot boundary coordinates, area), attribute information (plot ratio 2.5, supporting 12-class kindergarten), and spatial context information (land use nature and indicators of surrounding plots) of the target plot from the standardized data (i.e., preprocessed land spatial planning data). It encapsulates this information into the GeoJSON standard data exchange format, and then sends the encapsulated data packet along with the quality inspection request to the spatial semantic AI engine module through the built-in HTTPCaller converter in the form of RESTful API interaction.

[0064] After receiving the quality inspection request, the spatial semantic AI engine module first calls the knowledge graph management submodule. This submodule, based on the land spatial planning domain knowledge graph stored in the Neo4j graph database, retrieves the business rules corresponding to "R2 residential land" using the Cypher query language. It obtains the constraint that "only a kindergarten with 6 classes or less is required when the plot ratio of R2 is <3.0," providing domain knowledge support for subsequent semantic reasoning. Next, the spatial semantic embedding model submodule starts working. This module uses a graph neural network (GNN) architecture, using the target R2 plot as the core node, the required kindergarten as the associated node, and the plot ratio as the attribute association edge. After constructing a spatial semantic graph, it maps it to a high-dimensional semantic vector. This vector implicitly encodes information such as the plot's geometric shape, land use semantics, plot ratio value, and type of supporting facilities. Based on this, the spatial semantic embedding model calculates the cosine similarity between the high-dimensional vector corresponding to "R2 plot and floor area ratio 2.5" and the baseline high-dimensional vector corresponding to "12-class kindergarten construction requirements". The similarity value is 0.32, while the system's preset semantic matching threshold is 0.7. Since 0.32 < 0.7, the model generates a "semantic conflict loss value of 0.68", and determines that there is an implicit conflict of "mismatch between floor area ratio and the scale of the kindergarten".

[0065] Subsequently, the spatial semantic AI engine module returns the inference results, including conflict type, conflict location, and confidence level (92%), in JSON format to the intelligent quality inspection scheduling submodule of the FME processing engine module. This submodule uses a JSONFragmenter converter to parse the JSON format results into FME-recognizable feature attributes, and then passes them to the quality inspection report generation and output submodule. This submodule integrates the "no syntax error" results of traditional rule-based quality inspection with the "semantic conflict" results of AI inference, and generates a structured quality inspection report in HTML format using an HTMLReportGenerator converter. The quality inspection report contains the map patch ID and spatial location link of the conflicting plots. Clicking the link allows one to locate the conflicting plots in the GIS platform with one click.

[0066] Finally, the system, through the result review submodule of the human-computer collaborative interaction module, presents the quality inspection report to planning experts using a visual interface built with Vue.js. After reviewing the conflict details and reasoning basis, the experts confirm that the conflict is a genuine error and then submit confirmation feedback through the interface. The feedback collection and push submodule, upon capturing the expert feedback, automatically transforms the conflict case into a structured training sample and pushes it to the sample library of the spatial semantic AI engine module via a RabbitMQ message queue. This sample is used for incremental learning of the spatial semantic embedding model, optimizing the semantic vector mapping accuracy between "R2 plot - floor area ratio - supporting facilities requirements," and simultaneously supplementing it to the association rules of the knowledge graph to improve the detection accuracy of similar scenarios in the future.

[0067] Through the above process, the system successfully identified 12 hidden errors in the new area where the plot ratio of R2 plots did not match the size of the supporting kindergartens. This improved the efficiency of the traditional FME quality inspection by 40%, and the error identification coverage increased from 65% to 98%. In subsequent quality inspections of similar projects, the confidence level of the judgment of such conflicts remained stable at over 90%.

[0068] In some embodiments, a collaborative quality inspection of the architectural style and land use of historical districts is conducted. In the historical district protection plan of a certain historical and cultural city, there is a contradiction between the DWG building data and the SHP land use data: within the H1 historical style protection land, there is a "modern high-rise commercial building with a height of 45m", while the "Standards for the Protection Planning of Historical and Cultural Cities" requires that "the building height of H1 land be ≤30m, the style must be traditional pitched roof / Republic of China style, and the material must be blue brick and gray tile". Traditional quality inspection can only verify height exceeding the limit and cannot identify style conflicts, so the inspection method in this application is required.

[0069] First, the system starts the data access and preprocessing submodule of the FME processing engine module, synchronously imports DWG and SHP data, uses the SpatialRelator converter to establish the spatial relationship between the building and the land, uses the AttributeManager converter to supplement attributes such as "architectural style" and "protection level", then uses the Reprojector converter to unify the coordinates to CGCS2000, and uses the GeometryValidator converter to repair the topological errors of the building outline, outputting standardized data with complete association.

[0070] Next, the standardized data enters the traditional rule-based quality inspection submodule of FME, where the AttributeValidator converter marks the height exceeding the limit error "45m>30m", and the Tester converter filters out the combination of "land use = H1 + architectural style = modern high-rise" and passes it to the intelligent quality inspection scheduling submodule.

[0071] Subsequently, the intelligent quality inspection and scheduling submodule judged the combination as a complex scenario, extracted the building's geometry (outline, height), attributes (modern style), and protection rules for the H1 land use, encapsulated them into GeoJSON, and sent them to the spatial semantic AI engine via RESTful API.

[0072] After receiving the request, the spatial semantic AI engine module first calls the knowledge graph management submodule to retrieve the "style, height, and material constraints of land use H1" from the Neo4j graph database; then it starts the spatial semantic embedding model submodule to map land use H1 (core node), modern buildings (related nodes), and style requirements (attribute edges) into high-dimensional semantic vectors using a GNN architecture; it calculates the Euclidean distance between the building vector and the style reference vector of land use H1, obtaining a value of 8.6 (preset threshold 3.0), generating a "semantic conflict loss value of 8.6", and determining a style conflict.

[0073] Afterwards, the spatial semantic AI engine module returns the "height exceeding limits and style conflict" results to the FME processing engine module in JSON format. After parsing, the results are passed to the quality inspection report generation submodule. The fusion results generate a PDF report with a MapboxGLJS visualization link. Clicking on the report allows you to locate conflicting buildings and view style comparisons.

[0074] Finally, the experts confirmed the conflict through the human-computer collaborative interaction module. The feedback information was transformed into samples by the feedback collection and push submodule and pushed to the AI ​​sample library through the Kafka message queue. This was used to expand the H1 land use rules of the Neo4j knowledge graph and optimize the "landscape semantics" encoding accuracy of the GNN model.

[0075] Through this process, the system identified eight style conflict errors, improving the integrity of historical blocks from 70% to 95%, and providing precise support for the renovation of the historical blocks.

[0076] In some embodiments, a quality inspection of the service radius of public service facilities (kindergartens) is conducted. In a special plan for education facilities in a certain urban area, the SHP data shows a layout problem: the coverage overlap rate of kindergartens A and B (both with a service radius of 300m) reaches 50%, and a newly built residential community of 3,000 households is not covered by any kindergarten. Traditional quality inspection can only verify whether the radius of a single kindergarten meets the standard, and cannot determine overlap and blind spots. This solution is required for inspection.

[0077] First, the system starts the data access and preprocessing submodule of the FME processing engine module, imports SHP data of kindergartens and residential communities, uses the Bufferer converter to generate a 300m service radius buffer, uses the SpatialRelator converter to calculate the superposition relationship between buffers and the inclusion relationship with communities, and then uses the AttributeManager converter to supplement attributes such as "buffer area" and "educational needs corresponding to the number of households in the community", and outputs standardized data containing spatial relationships.

[0078] Next, FME's intelligent quality inspection scheduling submodule identified "multiple facilities superimposed and community coverage" as a complex scenario, extracted the buffer geometry and radius attributes of kindergartens A and B, as well as the location of the community and information on 3,000 households, encapsulated them into GeoJSON, and sent them to the spatial semantic AI engine module via the HTTPCaller converter using a RESTful API.

[0079] After receiving the request, the spatial semantic AI engine module first calls the knowledge graph management submodule to retrieve the rule "kindergarten radius 300m, 3000 households need 1 kindergarten with 12 classes, overlap rate ≤30%" from the Neo4j graph database; then it starts the spatial semantic embedding model submodule to map kindergarten A / B (core nodes), buffer (spatial edges), and community (related nodes) into high-dimensional semantic vectors using a GNN architecture; the similarity of the A / B buffer vectors is calculated to be 0.85 (threshold 0.3), which is judged as "excessive overlap"; the distance between the community vector and all buffer vectors is calculated to judge as "coverage blind spot".

[0080] Afterwards, the AI ​​engine returns the results of "overlap rate 50% and coverage blind spot" to FME in JSON format. After parsing, it is passed to the quality inspection report generation submodule, which integrates the data to generate a report containing "overlap area, number of households in blind spot, and adjustment suggestions", and embeds the Leaflet interactive map to mark the location of the problem.

[0081] Finally, experts confirmed the problem through the human-machine collaborative interaction module, and the feedback information was converted into samples and pushed to the AI ​​sample library to optimize the "facilities-demand" vector calculation logic of the GNN model and expand the community-facilities matching rules of the knowledge graph.

[0082] Through this process, the system identified 3 overlaps and 2 blind spots, improving the accuracy of service radius rationality judgment from 60% to 92%, thus avoiding resource waste and service gaps.

[0083] This application employs an advanced architecture characterized by dual-core drive, loose coupling, and modularity. The system uses the FME processing engine module as the process scheduling and execution hub, and the spatial semantic AI engine module as the intelligent decision-making brain. Communication is achieved through standardized service interfaces, realizing efficient collaboration between traditional rule-driven and AI semantic-driven approaches. This system not only automates quality inspection but also achieves intelligent and self-evolving quality inspection processes. It enables the computer system to partially replace the role of experts, autonomously discovering deep-seated data contradictions and semantic errors that are difficult to describe with explicit rules, thereby achieving a fundamental leap in quality inspection capabilities from the syntactic layer to the semantic layer.

[0084] This application proposes a deep collaborative architecture combining two engines: FME (Reliable Execution) and Spatial Semantic AI (Intelligent Reasoning). FME is no longer merely the execution end, but rather the scheduling center and connector for intelligent processes; AI is no longer an isolated technology demonstration, but a deeply integrated core for business decision-making. This architecture effectively solves the bottleneck problem of relying on a single technology path.

[0085] By introducing a spatial semantic embedding model, this application elevates the computer's understanding of planning data from "symbolic logic" (processing codes and coordinates) to "geometric logic" (processing vectors and similarity). This enables the system to autonomously discover deep-seated, hidden errors that rely on human expert experience, such as "inconsistency between land use and indicators," "conflicts in spatial appearance," and "irrational functional layout." This represents a revolutionary leap in quality inspection from syntax to semantics, enhancing quality inspection capabilities.

[0086] This application introduces a human-machine collaborative feedback loop. The system not only provides services but also continuously learns from expert feedback, enabling the spatial semantic AI engine module to iteratively optimize. This means the system becomes increasingly intelligent with use, constantly adapting to new specifications and discovering new error types, fundamentally reducing long-term maintenance costs and the lifecycle cost of the technology, thus achieving system self-evolution.

[0087] This application fully considers the integration with relevant technology systems (FME, GIS platform). All modules are connected through standardized interfaces (REST API), making implementation easy and easy to integrate into existing planning information workflows. It has immediate promotion and application value and enhances industrial practicality.

[0088] This application proposes a quality inspection system architecture that integrates FME and spatial semantic AI as dual engines, and a collaborative mechanism between the two through GeoJSON data encapsulation and RESTful API communication. The system includes intelligent scene recognition and AI invocation capabilities on the FME side, and semantic reasoning result feedback capabilities on the AI ​​side, thus solving the problem of missing semantic layer detection in traditional single-rule quality inspection.

[0089] This application proposes a spatial semantic reasoning technology based on knowledge graphs and GNNs to achieve intelligent detection of implicit complex logic (such as land use-indicator consistency and landscape harmony).

[0090] This application proposes a human-machine collaborative system self-evolution mechanism. Experts review the AI ​​results, and the feedback information is transformed into structured samples. These samples are then pushed to the spatial semantic AI engine module via a message queue for incremental learning of the GNN model and knowledge graph expansion. This solves the problems of rigid rules and the need for manual maintenance in traditional systems, and enables the system to adaptively update regulations and identify new error types.

[0091] In this embodiment, multi-source data standardization and dual-path quality inspection results are integrated to ensure data compatibility and the practicality of the quality inspection results.

[0092] Based on the foregoing embodiments, this application provides a land and space planning data quality inspection system. The system includes modules and sub-modules, and each unit of each sub-module can be implemented by a processor in an electronic device; of course, it can also be implemented by specific logic circuits. In the implementation process, the processor can be a central processing unit (CPU), a microprocessor (MPU), a digital signal processor (DSP), or a field-programmable gate array (FPGA), etc.

[0093] Figure 3 A schematic diagram of the composition structure of a land spatial planning data quality inspection system provided in this application embodiment is shown below. Figure 3 As shown, the system 200 includes: The modules include a human-computer collaborative interaction module 201, an FME processing engine module 202, and a spatial semantic AI engine module 203, among which: The FME processing engine module 202 is used to acquire the original land spatial planning data submitted by the user through the human-computer collaborative interaction module 201, and to preprocess the original land spatial planning data to obtain preprocessed land spatial planning data. The FME processing engine module 202 is used to perform quality inspection on the preprocessed land spatial planning data and obtain a first quality inspection result. The FME processing engine module 202 is used to transmit the preprocessed territorial spatial planning data to the spatial semantic AI engine module when it is determined, based on a preset strategy, that the preprocessed territorial spatial planning data needs to be checked by calling the spatial semantic AI engine module. The spatial semantic AI engine module 203 is used to perform semantic conflict reasoning on the preprocessed land spatial planning data, obtain a second quality inspection result, and transmit the second quality inspection result to the FME processing engine module 202. The FME processing engine module 202 is used to generate a first initial quality inspection report based on the first quality inspection result and the second quality inspection result.

[0094] In some possible embodiments, the FME processing engine module includes a data access and preprocessing submodule, a rule quality inspection submodule, a quality inspection scheduling submodule, and a quality inspection report generation and output submodule; the spatial semantic AI engine module includes a spatial semantic embedding model submodule and an AI service interface submodule. The data access and preprocessing submodule is used to acquire the original land spatial planning data submitted by the user through the human-computer collaborative interaction module, and to preprocess the original land spatial planning data to obtain preprocessed land spatial planning data. The rule-based quality inspection submodule is used to perform quality inspection on the preprocessed land and space planning data to obtain a first quality inspection result; The quality inspection scheduling submodule is used to transmit the preprocessed territorial spatial planning data to the spatial semantic embedding model submodule through the AI ​​service interface submodule when it is determined based on a preset strategy that the preprocessed territorial spatial planning data needs to be checked by calling the spatial semantic AI engine module. The spatial semantic embedding model in the spatial semantic embedding model submodule is used to perform semantic conflict reasoning on the preprocessed territorial spatial planning data to obtain a second quality inspection result, and the second quality inspection result is transmitted to the quality inspection scheduling submodule through the AI ​​service interface submodule. The quality inspection report generation and output submodule is used to obtain the first quality inspection result transmitted by the rule quality inspection submodule and the second quality inspection result transmitted by the quality inspection scheduling submodule, and generate a first initial quality inspection report based on the first quality inspection result and the second quality inspection result.

[0095] In some possible embodiments, the spatial semantic AI engine module further includes a knowledge graph management submodule; The spatial semantic embedding model submodule is used to obtain target knowledge from the knowledge graph of the knowledge graph management submodule when it is determined that additional domain knowledge is needed based on the preprocessed territorial spatial planning data. The spatial semantic embedding model submodule is used to perform semantic conflict reasoning on the preprocessed territorial spatial planning data based on the spatial semantic embedding model and the target knowledge to obtain the second quality inspection result.

[0096] In some possible embodiments, the human-computer collaborative interaction module includes a result review submodule and a feedback collection and push submodule. The quality inspection report generation and output submodule is used to send the first initial quality inspection report to the result review submodule; The result review submodule is used to display the first initial quality inspection report, and in response to the user's revision operation on the first initial quality inspection report, generate a target quality inspection report and send the target quality inspection report to the feedback collection and push submodule. The feedback collection and push submodule is used to convert the target quality inspection report into structured training samples, and transmit them to the spatial semantic embedding model submodule and the knowledge graph management submodule through the AI ​​service interface submodule, respectively, to perform incremental learning of the spatial semantic embedding model and updating of the knowledge graph.

[0097] In some possible embodiments, the spatial semantic embedding model submodule is used to parse the preprocessed land spatial planning data to obtain multiple elements in the preprocessed land spatial planning data and the attribute value corresponding to each element; based on the attribute value corresponding to each element, to generate a query request for the corresponding element against the knowledge graph, and to transmit each query request to the knowledge graph management submodule. The knowledge graph management submodule is used to find the business definition, constraints and relationships of entities associated with the corresponding elements in the knowledge graph according to each query request. The spatial semantic embedding model submodule is used to determine the business definitions, constraints, and relationships of entities corresponding to all elements as target knowledge.

[0098] In some possible embodiments, the spatial semantic embedding model in the spatial semantic embedding model submodule is used to convert each element and its corresponding spatial relationship into a corresponding high-dimensional feature vector; based on the constraints and relationships in the target knowledge, and the spatial topological relationships in the preprocessed land planning data, select element pairs that need to be semantically conflict detected; for each element pair, determine the cosine similarity or Euclidean distance between the corresponding high-dimensional feature vectors; based on the cosine similarity or Euclidean distance of each element pair, generate the semantic conflict loss value of the corresponding element pair; based on the semantic conflict loss value of each element pair, generate the conflict determination result and confidence level of the corresponding element pair; based on the business definition and constraints of the elements contained in each element pair in the target knowledge, determine the conflict type of the corresponding element pair; and based on the conflict determination result, conflict type, and confidence level of each element pair, generate a second quality inspection result.

[0099] In some possible embodiments, the preset strategy includes: determining the task type and data characteristics to be inspected based on the preprocessed land spatial planning data; and determining that the spatial semantic AI engine module needs to be invoked for inspection when the task type characterization involves complex logic requiring semantic association analysis, or the data characteristic characterization involves special land use types.

[0100] In some possible embodiments, the FME processing engine module is further configured to generate a second initial quality inspection report based on the first quality inspection result, provided that the preprocessed land spatial planning data does not require the spatial semantic AI engine module to be called for inspection based on a preset strategy.

[0101] It should be understood that the phrase "one embodiment" or "an embodiment" throughout the specification means that a specific feature, structure, or characteristic related to the embodiment is included in at least one embodiment of this application. Therefore, "in one embodiment" or "in an embodiment" appearing throughout the specification does not necessarily refer to the same embodiment. Furthermore, these specific features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. It should be understood that in the various embodiments of this application, the sequence numbers of the above-described processes do not imply a sequential order of execution; the execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application. The sequence numbers of the above-described embodiments are merely descriptive and do not represent the superiority or inferiority of the embodiments.

[0102] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0103] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods, such as: multiple units or components can be combined, or integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the various components shown or discussed can be through some interfaces, and the indirect coupling or communication connection between devices or units can be electrical, mechanical, or other forms.

[0104] The units described above as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; they may be located in one place or distributed across multiple network units; some or all of the units may be selected to achieve the purpose of the embodiments of this application according to actual needs. In addition, each functional unit in the embodiments of this application may be fully integrated into one processing unit, or each unit may be a separate unit, or two or more units may be integrated into one unit; the integrated unit may be implemented in hardware or in the form of hardware plus software functional units.

[0105] Alternatively, if the integrated units described above are implemented as software functional modules and sold or used as independent products, they can also be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the embodiments of this application, or the parts that contribute to related technologies, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause the device automatic test line to execute all or part of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, ROMs, magnetic disks, or optical disks.

[0106] The methods disclosed in the several method embodiments provided in this application can be arbitrarily combined to obtain new method embodiments without conflict. The features disclosed in the several method or device embodiments provided in this application can be arbitrarily combined to obtain new method embodiments or device embodiments without conflict.

[0107] The above description is merely an embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included 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 quality inspection of land spatial planning data, applied to a land spatial planning data quality inspection system, the system comprising: The system includes a human-computer collaborative interaction module, an FME processing engine module, and a spatial semantic AI engine module. The FME processing engine module includes a data access and preprocessing submodule, a rule quality inspection submodule, a quality inspection scheduling submodule, and a quality inspection report generation and output submodule. The spatial semantic AI engine module includes a spatial semantic embedding model submodule and an AI service interface submodule; the method includes: Using the data access and preprocessing submodule, the original land spatial planning data submitted by the user through the human-computer collaborative interaction module is obtained, and the original land spatial planning data is preprocessed to obtain preprocessed land spatial planning data. Using the rule-based quality inspection submodule, the preprocessed land and space planning data is inspected based on predefined explicit quality inspection rules to obtain the first quality inspection result; Using the quality inspection scheduling submodule, when it is determined based on a preset strategy that the preprocessed territorial spatial planning data needs to be checked by calling the spatial semantic AI engine module, the preprocessed territorial spatial planning data is transmitted to the spatial semantic embedding model submodule through the AI ​​service interface submodule. Using the spatial semantic embedding model in the spatial semantic embedding model submodule and combining it with target knowledge, semantic conflict detection is performed on the element pairs obtained by parsing and filtering the preprocessed land spatial planning data. Based on the conflict judgment result, conflict type and confidence level of each element pair, a second quality inspection result is generated. The second quality inspection result is transmitted to the quality inspection scheduling submodule through the AI ​​service interface submodule. Using the quality inspection report generation and output submodule, the first quality inspection result transmitted by the rule quality inspection submodule and the second quality inspection result transmitted by the quality inspection scheduling submodule are obtained. The first quality inspection result and the second quality inspection result are summarized, classified and categorized to generate a first initial quality inspection report.

2. The method according to claim 1, characterized in that, The spatial semantic AI engine module also includes a knowledge graph management submodule; The method further includes: Using the spatial semantic embedding model submodule, when it is determined that additional domain knowledge is needed based on the preprocessed territorial spatial planning data, the target knowledge is obtained from the knowledge graph of the knowledge graph management submodule.

3. The method according to claim 2, characterized in that, The human-computer collaborative interaction module includes a result review submodule and a feedback collection and push submodule, and the method further includes: Using the quality inspection report generation and output submodule, the first initial quality inspection report is sent to the result review submodule; Using the result review submodule, the first initial quality inspection report is displayed, and in response to the user's revision operation on the first initial quality inspection report, a target quality inspection report is generated and the target quality inspection report is sent to the feedback collection and push submodule. Using the feedback collection and push submodule, the target quality inspection report is transformed into a structured training sample, and then transmitted to the spatial semantic embedding model submodule and the knowledge graph management submodule through the AI ​​service interface submodule, respectively, for incremental learning of the spatial semantic embedding model and updating of the knowledge graph.

4. The method according to claim 2, characterized in that, The step of utilizing the spatial semantic embedding model submodule, based on the preprocessed land spatial planning data and determining that supplementary domain knowledge is needed, involves obtaining the target knowledge from the knowledge graph of the knowledge graph management submodule, including: Using the spatial semantic embedding model submodule, the preprocessed land spatial planning data is parsed to obtain multiple elements in the preprocessed land spatial planning data and the attribute value corresponding to each element; based on the attribute value corresponding to each element, a query request for the corresponding element against the knowledge graph is generated, and each query request is transmitted to the knowledge graph management submodule. Using the knowledge graph management submodule, the business definition, constraints and relationships of entities associated with the corresponding elements are found in the knowledge graph according to each query request; Using the spatial semantic embedding model submodule, the business definitions, constraints, and relationships of the entities corresponding to all elements are determined as the target knowledge.

5. The method according to claim 4, characterized in that, The spatial semantic embedding model in the spatial semantic embedding model submodule, combined with target knowledge, is used to perform semantic conflict detection on the element pairs obtained by parsing and filtering the preprocessed land spatial planning data. Based on the conflict determination results, conflict types, and confidence levels of each element pair, a second quality inspection result is generated, including: Using the spatial semantic embedding model in the spatial semantic embedding model submodule, each element and its corresponding spatial relationship are converted into a corresponding high-dimensional feature vector; Based on the constraints and relationships in the target knowledge, and the spatial topological relationships in the preprocessed territorial spatial planning data, element pairs that need to be semantically conflict detected are selected. For each pair of elements, determine the cosine similarity or Euclidean distance between the corresponding high-dimensional feature vectors; Based on the cosine similarity or Euclidean distance of each element pair, a semantic conflict loss value for the corresponding element pair is generated. Based on the semantic conflict loss value corresponding to each element pair, a conflict determination result and confidence level for the corresponding element pair are generated. Based on the business definition and constraints of the elements contained in each element pair in the target knowledge, the conflict type of the corresponding element pair is determined. Based on the conflict determination result, conflict type, and confidence level of each element pair, a second quality inspection result is generated.

6. The method according to claim 1, characterized in that, The preset strategy includes: Based on the preprocessed territorial spatial planning data, determine the types of tasks and data characteristics to be inspected; When the task type representation involves complex logic requiring semantic association analysis, or when the data feature representation involves special land use types, it is determined that the spatial semantic AI engine module needs to be invoked for inspection.

7. The method according to claim 1, characterized in that, The method further includes: Using the FME processing engine module, if it is determined based on a preset strategy that the preprocessed land spatial planning data does not need to be checked by the spatial semantic AI engine module, a second initial quality inspection report is generated based on the first quality inspection result.

8. A land spatial planning data quality inspection system, the system comprising: The system comprises a human-computer collaborative interaction module, an FME processing engine module, and a spatial semantic AI engine module. The FME processing engine module includes a data access and preprocessing submodule, a rule-based quality inspection submodule, a quality inspection scheduling submodule, and a quality inspection report generation and output submodule. The spatial semantic AI engine module includes a spatial semantic embedding model submodule and an AI service interface submodule. Wherein: The data access and preprocessing submodule is used to acquire the original land spatial planning data submitted by the user through the human-computer collaborative interaction module, and to preprocess the original land spatial planning data to obtain preprocessed land spatial planning data. The rule-based quality inspection submodule is used to perform quality inspection on the preprocessed territorial spatial planning data based on predefined explicit quality inspection rules, and obtain a first quality inspection result. The quality inspection scheduling submodule is used to transmit the preprocessed territorial spatial planning data to the spatial semantic embedding model submodule through the AI ​​service interface submodule when it is determined based on a preset strategy that the preprocessed territorial spatial planning data needs to be checked by calling the spatial semantic AI engine module. The spatial semantic embedding model in the spatial semantic embedding model submodule is used to combine target knowledge to perform semantic conflict detection on the element pairs obtained by parsing and screening the preprocessed territorial spatial planning data. Based on the conflict judgment result, conflict type and confidence level of each element pair, a second quality inspection result is generated, and the second quality inspection result is transmitted to the quality inspection scheduling submodule through the AI ​​service interface submodule. The quality inspection scheduling submodule is used to obtain the first quality inspection result transmitted by the rule quality inspection submodule and the second quality inspection result transmitted by the quality inspection scheduling submodule, summarize, classify and categorize the first quality inspection result and the second quality inspection result, and generate a first initial quality inspection report.

Images