Intelligent integration system of multi-source heterogeneous data for national land spatial planning based on semantic analysis

Abstract

The present invention relates to the technical field of multi-source heterogeneous data fusion for national land space planning, and specifically discloses an intelligent integration system for multi-source heterogeneous data for national land space planning based on semantic analysis. The system includes a data acquisition and recognition module, a semantic density balance analysis module, a boundary fuzzy contagion analysis module, a comprehensive reliability assessment module and a semantic calibration feedback module. By introducing two core evaluation dimensions, semantic density balance and boundary fuzzy contagion, the system accurately identifies semantic conflicts and leakage problems in multi-source data based on the overall balance of semantic distribution and local boundary stability, combines deep learning models for reliability assessment, and uses knowledge graph reasoning and transfer learning to achieve semantic dynamic calibration, forming a closed-loop feedback mechanism.

Description

Technical Field

[0001] The present invention relates to the technical field of multi-source heterogeneous data fusion for national land space planning, and in particular to an intelligent integration system for multi-source heterogeneous data for national land space planning based on semantic analysis. Background Art

[0002] National land and space planning involves the integration and analysis of multi-source heterogeneous data, including remote sensing images, GIS data, current land use maps, planning texts, and socioeconomic statistical information. Traditional data integration methods rely primarily on manual annotation, rule matching, or simple statistical similarity calculations, making it difficult to effectively address the semantic heterogeneity between multi-source data. In recent years, some studies have attempted to use natural language processing and computer vision technologies to perform semantic analysis on unstructured data, but most methods focus only on single-modal data alignment and lack a systematic assessment of cross-modal semantic consistency. In addition, existing technologies typically use static thresholds or fixed rules to determine data reliability and are unable to dynamically adapt to the complex semantic boundary ambiguity and classification system differences in national land and space planning, resulting in semantic conflicts or information loss during data integration.

[0003] The existing technology has the following deficiencies:

[0004] Traditional methods do not fully consider the differences in information distribution between different data sources under a unified semantic framework. For example, remote sensing images and planning texts may use different classification systems, resulting in local over-density or over-sparseness of semantic units after mapping, affecting the balance of data fusion. Existing technologies lack quantitative assessment of semantic density balance, making it difficult to identify semantic distribution discreteness caused by differences in classification systems. In national land space planning, land boundaries (such as urban-rural fringe areas) often have semantic ambiguity. Existing methods mostly judge the data consistency of boundary areas based on geometric overlap or simple semantic similarity, ignoring the "contagious" effect of fuzzy semantics on adjacent areas. For example, areas with gradual changes in land use types may cause semantic label leakage from adjacent data sources, but existing technologies lack the ability to dynamically model the semantic contamination radius and cannot effectively suppress the diffusion effect of boundary ambiguity.

[0005] To address these shortcomings, this paper introduces semantic density balance anomaly eigenvalues ​​and boundary fuzziness contagion eigenvalues ​​to quantify the semantic distribution dispersion and boundary leakage risk of multi-source data, respectively. Furthermore, a dynamic reliability assessment mechanism is constructed in conjunction with a deep learning model to address the hidden impact of uneven cross-modal semantic distribution and the dynamic contagion effect of boundary fuzziness. Summary of the Invention

[0006] The purpose of the present invention is to provide a multi-source heterogeneous data intelligent integration system for national land space planning based on semantic analysis to solve the problems in the above background.

[0007] The purpose of the present invention can be achieved through the following technical solutions:

[0008] The intelligent integration system of multi-source heterogeneous data for national land space planning based on semantic analysis includes:

[0009] A data collection and recognition module is used to obtain national land space planning data and identify and select semantic density balance and boundary fuzzy contagion as core evaluation dimensions;

[0010] A semantic density balance analysis module, which calculates the information entropy differences of various data sources under a unified semantic framework to generate semantic density balance anomaly feature values ​​for evaluating the degree of information distribution dispersion in different classification systems;

[0011] Boundary fuzzy contagion analysis module, which analyzes the semantic gradient changes of the boundary fuzzy area, calculates its semantic contamination radius to the adjacent data area, and generates a boundary fuzzy contagion feature value for evaluating whether there is semantic leakage in the data of national land space planning;

[0012] A comprehensive reliability assessment module, which constructs a comprehensive reliability assessment vector based on the semantic density balance anomaly eigenvalue and the boundary fuzzy contagion eigenvalue, and inputs it into a deep learning model for analysis. Based on the analysis results, the national land space planning data is divided into reliable data and unreliable data;

[0013] A semantic calibration feedback module is provided, which calibrates the national land space planning data based on unreliable data.

[0014] As a further solution of the present invention: the evaluation of the information distribution dispersion of different classification systems specifically includes:

[0015] According to the identification and selection of semantic density balance, the information entropy difference of each data source under the unified semantic framework is calculated, and the abnormal characteristic value of the semantic density balance is calculated. It is judged whether the abnormal characteristic value of the semantic density balance is greater than or equal to the preset threshold. If so, the information distribution of different classification systems is dispersed. If not, the information distribution of different classification systems is concentrated.

[0016] As a further solution of the present invention: the process of obtaining the abnormal feature value of the semantic density balance is:

[0017] The semantic units in multi-source heterogeneous land space planning data are mapped to a unified semantic representation space through the CLIP model to achieve cross-modal semantic alignment; then the kernel density estimation method is combined with the Gaussian kernel function and the dynamic bandwidth selection strategy based on reinforcement learning to perform local density modeling of the semantic distribution of each data source; based on the local density function, a probability distribution is constructed and the information entropy is calculated; at the same time, a semantic relationship graph is constructed, and the graph attention network is used to perform weighted modeling of the semantic similarity between different data sources. On this basis, the calculation method of Kullback-Leibler divergence is improved to evaluate the semantic distribution deviation between data sources; finally, the semantic density balance anomaly scoring function is constructed by combining information entropy and KL divergence, and the semantic density balance anomaly feature value is output.

[0018] As a further solution of the present invention, the evaluation data of land space planning data for semantic leakage specifically includes:

[0019] According to the identified and extracted boundary fuzzy contagion, by analyzing the semantic gradient changes of the boundary fuzzy area, its semantic pollution radius to the adjacent data area is calculated, and the boundary fuzzy contagion characteristic value is generated. It is judged whether the boundary fuzzy contagion characteristic value is greater than or equal to the preset threshold. If so, there is semantic leakage in the national land space planning data; if not, there is no semantic leakage in the national land space planning data.

[0020] As a further solution of the present invention: the process of obtaining the boundary fuzzy infectious characteristic value is:

[0021] Firstly, the boundary areas between different land types are identified through the semantic segmentation model and defined as semantic fuzzy areas. Then, the semantic similarity gradient change rate between adjacent land types is calculated in the fuzzy area, and the sliding window method is used to extract the local semantic gradient direction and amplitude. Then, a semantic pollution propagation model is established based on the gradient change to simulate the impact range of semantic leakage, and the semantic pollution radius is solved by the diffusion equation. Finally, a boundary fuzzy contagion scoring function is constructed by combining the pollution radius and the semantic similarity decrease rate to generate the boundary fuzzy contagion feature value.

[0022] As a further solution of the present invention, the semantic density balance abnormality feature value and the boundary fuzzy infectiousness feature value are used to construct a comprehensive reliability evaluation vector, and input it into the deep learning model for analysis, specifically including:

[0023] Obtain the semantic density balance anomaly eigenvalues ​​and the boundary fuzzy infectious eigenvalues, and construct a comprehensive reliability evaluation vector with the semantic density balance anomaly eigenvalues ​​and the boundary fuzzy infectious eigenvalues ​​as the input of the deep learning model. The deep learning model is trained, and the confidence score of the national land space planning data is output according to the trained deep learning model. The deep learning model uses a convolutional neural network model.

[0024] As a further solution of the present invention: the training process of the deep learning model is:

[0025] The comprehensive reliability assessment vector is used as an input sample and input into a convolutional neural network model for training. The convolutional neural network model consists of several one-dimensional convolutional layers, pooling layers and fully connected layers, and is used to extract local correlation and high-order interaction information in the feature vector. The cross-entropy loss function is used in the training process, and manually labeled reliable or unreliable samples are used as supervision signals. The network parameters are continuously optimized through the backpropagation algorithm to improve the accuracy of the model's semantic consistency judgment.

[0026] As a further solution of the present invention: the division of national land space planning data into reliable data and unreliable data specifically includes:

[0027] The threshold interval is set according to the confidence score output by the deep learning model. If the score is higher than the first threshold, it is judged as highly reliable data; if the score is between the first and second thresholds, it is judged as generally reliable data; if the score is lower than the second threshold, it is judged as unreliable data.

[0028] As a further solution of the present invention: the calibration of national land space planning data based on unreliable data specifically includes:

[0029] For data marked as unreliable, the contextual information in the semantic relationship graph is called up, combined with the knowledge graph reasoning mechanism to locate the most likely semantic attribution category; the multimodal CLIP model is used to re-encode the national land space planning data and compare and match it with the benchmark vectors in the standard semantic framework; a transfer learning strategy is adopted to transfer the semantic mapping rules in the reliable data to the unreliable data to complete the semantic label correction; the calibrated data is re-entered into the system for verification until the reliability assessment criteria are met.

[0030] Beneficial effects of the present invention:

[0031] (1) This paper introduces two core evaluation dimensions, "semantic density balance" and "boundary fuzziness contagion", to construct a systematic and multi-level semantic consistency analysis framework, which can deeply characterize the semantic differences and potential conflicts between multi-source heterogeneous data in national land space planning from both macro and micro levels. Among them, semantic density balance focuses on the concentration or dispersion characteristics of the overall semantic distribution, uses the CLIP model to achieve cross-modal semantic alignment, and combines kernel density estimation and information entropy theory to improve the traditional Kullback-Leibler divergence calculation method, quantify the distribution deviation degree of different data sources in a unified semantic space, and effectively identify the semantic conflict risk caused by differences in classification systems. Boundary fuzziness contagion focuses on the local semantic stability of land class boundaries, identifies fuzzy areas through semantic segmentation, and combines semantic similarity gradient extraction and diffusion propagation model to construct a scoring function of semantic pollution radius and similarity decline rate, accurately evaluating whether there is semantic leakage in the boundary area and its impact range. The above-mentioned two-dimensional fusion analysis mechanism not only realizes the comprehensive perception of the semantic quality of multi-source data, but also provides key feature inputs for subsequent comprehensive reliability assessment, significantly improving the fusion accuracy, expression consistency and application credibility of heterogeneous data under a unified semantic framework, and has strong engineering practicality and promotion value.

[0032] (2) Based on the comprehensive evaluation results of the semantic density balance abnormal feature values ​​and boundary fuzziness contagion feature values ​​by the deep learning model, the system can realize the intelligent judgment of the reliability of land space planning data and automatically identify unreliable data with problems such as semantic conflict, distribution offset or boundary fuzziness. Once abnormal data is found, the system will immediately trigger the semantic calibration process, using the contextual association information in the semantic relationship graph and combining the semantic reasoning mechanism driven by the knowledge graph to accurately locate the source of the semantic attribution deviation. On this basis, the multimodal CLIP model is further introduced to semantically re-encode the original data and perform high-dimensional space matching with the reference vector in the standard semantic framework to improve the accuracy of semantic mapping. At the same time, the system adopts a transfer learning strategy to extract verified semantic mapping rules from the trusted data set and adapt them to the unreliable data to achieve dynamic correction and consistency enhancement of semantic labels. The calibrated data will be re-entered into the system for verification, forming a closed-loop optimization mechanism of "evaluation-identification-correction-feedback". This mechanism not only improves the system's adaptability in complex semantic environments, but also significantly enhances the robustness and maintainability of national land space planning data in the multi-source fusion process, providing solid technical support for building a high-quality and sustainably updated spatial planning database. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] The present invention will be further described below with reference to the accompanying drawings.

[0034] Figure 1 It is a flowchart of the intelligent integration system of multi-source heterogeneous data of national land space planning based on semantic analysis of the present invention. DETAILED DESCRIPTION

[0035] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts shall fall within the scope of protection of the present invention.

[0036] See also Figure 1 As shown, the present invention is an intelligent integration system for multi-source heterogeneous data of national land space planning based on semantic analysis, including:

[0037] A data collection and recognition module is used to obtain national land space planning data and identify and select semantic density balance and boundary fuzzy contagion as core evaluation dimensions;

[0038] A semantic density balance analysis module, which calculates the information entropy differences of various data sources under a unified semantic framework to generate semantic density balance anomaly feature values ​​for evaluating the degree of information distribution dispersion in different classification systems;

[0039] Boundary fuzzy contagion analysis module, which analyzes the semantic gradient changes of the boundary fuzzy area, calculates its semantic contamination radius to the adjacent data area, and generates a boundary fuzzy contagion feature value for evaluating whether there is semantic leakage in the data of national land space planning;

[0040] A comprehensive reliability assessment module, which constructs a comprehensive reliability assessment vector based on the semantic density balance anomaly eigenvalue and the boundary fuzzy contagion eigenvalue, and inputs it into a deep learning model for analysis. Based on the analysis results, the national land space planning data is divided into reliable data and unreliable data;

[0041] A semantic calibration feedback module is provided, which calibrates the national land space planning data based on unreliable data.

[0042] In the data collection and recognition module, the data collection and recognition module is used to obtain national land space planning data, and identify and select semantic density balance and boundary fuzzy contagion as core evaluation dimensions, specifically including:

[0043] The data collection and recognition module is used to acquire multi-source heterogeneous data related to national land space planning and to perform preliminary identification and analysis of the semantic features within the data. Specifically, the module connects to various data sources, including remote sensing image databases, geographic information system (GIS) platforms, current land use maps, planning text files, and socioeconomic statistics datasets, to achieve automated collection and structured processing of the basic data required for national land space planning. During the data collection process, metadata parsing technology is used to extract semantic description information for each type of data. Combining natural language processing with image semantic recognition algorithms, the module converts unstructured or semi-structured data into unified and comparable semantic units.

[0044] Furthermore, the data collection and recognition module performs semantic consistency analysis on the collected semantic units, identifying and selecting "semantic density balance" and "boundary fuzziness contagion" as core evaluation dimensions for subsequent semantic quality analysis. Semantic density balance measures the degree of concentration or dispersion of semantic distribution under different classification systems, while boundary fuzziness contagion detects semantic leakage within land boundary areas. This module outputs initial evaluation results for the two types of feature values ​​and passes them to the next analysis module to support intelligent judgment and calibration of the overall semantic consistency of national land space planning data.

[0045] In the semantic density balance analysis module, the semantic density balance analysis module calculates the information entropy difference of each data source under the unified semantic framework to generate a semantic density balance abnormality feature value for evaluating the degree of information distribution dispersion of different classification systems, specifically including:

[0046] According to the identification and selection of semantic density balance, the information entropy difference of each data source under the unified semantic framework is calculated, and the abnormal characteristic value of the semantic density balance is calculated. It is judged whether the abnormal characteristic value of the semantic density balance is greater than or equal to the preset threshold. If so, the information distribution of different classification systems is dispersed. If not, the information distribution of different classification systems is concentrated.

[0047] The process of obtaining the abnormal feature value of the semantic density balance is as follows:

[0048] The semantic units in multi-source heterogeneous land and space planning data are mapped to a unified semantic representation space through the CLIP model to achieve cross-modal semantic alignment. Subsequently, a kernel density estimation method is used in combination with a Gaussian kernel function and a dynamic bandwidth selection strategy based on reinforcement learning to perform local density modeling of the semantic distribution of each data source.

[0049] The probability distribution is constructed based on the local density function, and the information entropy of each data source is calculated to measure the degree of concentration or dispersion of the semantic distribution within each data source. The calculation expression of the information entropy is: ;in, Represents a data source The information entropy of Represents the data source, Represents a data source The number of semantic units contained in Indicates the The data source The probability of a semantic unit, represents the logarithm of natural numbers, Indicates the semantic units, Indicates the data sources;

[0050] Simultaneously build a semantic relationship graph , the graph attention network is used to perform weighted modeling of the semantic similarity between different data sources, and the attention coefficient between nodes is calculated as follows: ;in, Representation node and nodes The attention coefficient between Represents a normalization function that maps each element in a vector to interval, and the sum of these elements is equal to , represents a learnable vector used to calculate the score of the feature combination between nodes, Represents a vector The transpose of Represents a learnable weight matrix for linear transformation of node features. Representation node The eigenvector of Representation node The eigenvector of Represents a concatenation operation, which connects two vectors into a longer vector. and Represents two different nodes;

[0051] On this basis, we improve the calculation method of Kullback-Leibler divergence to evaluate the semantic distribution deviation between data sources. The specific improved calculation expression is: ;in, Indicates that the data source To the data source The Kullback-Leibler divergence (KL divergence), Indicates that at the sample point Data source and The semantic weight factor between Indicates that at the sample point Data source The probability density function value of Represents a data source The probability density function value of and Represents two different data sources, Indicates the data source sample point;

[0052] Finally, the semantic density balance anomaly scoring function is constructed by combining information entropy and KL divergence, and the semantic density balance anomaly feature value is output to reflect the consistency level of semantic distribution between different classification systems. The calculation expression of the semantic density balance anomaly scoring function is: ;in, Indicates the abnormal feature value of semantic density balance, and is the preset weight factor, represents the average information entropy of all data sources, It represents the average KL divergence between any two data sources.

[0053] It should be noted that CLIP is a multimodal pre-training model developed by OpenAI. It aims to map text and images into the same embedding space through contrastive learning, thereby achieving cross-modal understanding and matching.

[0054] In the boundary fuzzy contagion analysis module, the boundary fuzzy contagion analysis module analyzes the semantic gradient changes of the boundary fuzzy area, calculates its semantic contamination radius to the adjacent data area, and generates a boundary fuzzy contagion feature value for evaluating whether there is semantic leakage in the land space planning data. Specifically, it includes:

[0055] According to the identified and extracted boundary fuzzy contagion, by analyzing the semantic gradient changes of the boundary fuzzy area, its semantic pollution radius to the adjacent data area is calculated, and the boundary fuzzy contagion characteristic value is generated. It is judged whether the boundary fuzzy contagion characteristic value is greater than or equal to the preset threshold. If so, there is semantic leakage in the national land space planning data; if not, there is no semantic leakage in the national land space planning data.

[0056] The process of obtaining the boundary fuzzy infectious characteristic value is as follows:

[0057] First, the boundary areas between different land types are identified through the semantic segmentation model and defined as semantic fuzzy areas. The semantic similarity gradient change rate between adjacent land types is calculated in the fuzzy area, and the sliding window method is used to extract the local semantic gradient direction and amplitude. Based on the gradient change, a semantic pollution propagation model is established to simulate the influence range of semantic leakage, and the semantic pollution radius is solved by the diffusion equation.

[0058] Finally, the contamination radius and the semantic similarity decrease rate are combined to construct a boundary fuzzy infectiousness scoring function and generate the boundary fuzzy infectiousness feature value;

[0059] The calculation expression of semantic similarity decrease rate is: ;in, represents the rate of semantic similarity decrease, represents the average semantic similarity within the fuzzy area, represents the average semantic similarity outside the fuzzy area, Indicates distance;

[0060] The calculation expression of the boundary fuzzy contagiousness scoring function is: ;in, represents the boundary fuzzy infectious eigenvalue, and represents the preset weight factor, represents the semantic pollution radius, Indicates the rate of semantic similarity decrease.

[0061] In the comprehensive reliability assessment module, the module constructs a comprehensive reliability assessment vector based on the semantic density balance abnormality feature value and the boundary fuzzy contagion feature value, and inputs it into the deep learning model for analysis. Based on the analysis results, the national land space planning data is divided into reliable data and unreliable data, specifically including:

[0062] Obtain the semantic density balance anomaly eigenvalues ​​and the boundary fuzzy infectious eigenvalues, and construct a comprehensive reliability evaluation vector with the semantic density balance anomaly eigenvalues ​​and the boundary fuzzy infectious eigenvalues ​​as the input of the deep learning model. The deep learning model is trained, and the confidence score of the national land space planning data is output according to the trained deep learning model. The deep learning model uses a convolutional neural network model.

[0063] The training process of the deep learning model is:

[0064] The comprehensive reliability assessment vector is used as an input sample and input into a convolutional neural network model for training. The convolutional neural network model consists of several one-dimensional convolutional layers, pooling layers and fully connected layers, and is used to extract local correlation and high-order interaction information in the feature vector. During the training process, the cross-entropy loss function is adopted, and manually labeled reliable or unreliable samples are used as supervision signals. The network parameters are continuously optimized through the back propagation algorithm to improve the accuracy of the model's semantic consistency judgment. Finally, after the model training is completed, the trained convolutional neural network model is used to predict the new national land space planning data and output the corresponding confidence score. The confidence score represents the probability that the national land space planning data is judged as reliable data under the current semantic framework.

[0065] The threshold interval is set according to the confidence score output by the deep learning model. If the score is higher than the first threshold, it is judged as highly reliable data; if the score is between the first and second thresholds, it is judged as generally reliable data; if the score is lower than the second threshold, it is judged as unreliable data; for unreliable data, it is marked as an object to be calibrated, and its data source, geographical area and semantic conflict type are recorded to form a data set to be calibrated.

[0066] In the semantic calibration feedback module, the semantic calibration feedback module calibrates the national land space planning data based on unreliable data, specifically including:

[0067] For data marked as unreliable, the contextual information in the semantic relationship graph is called up, combined with the knowledge graph reasoning mechanism to locate the most likely semantic attribution category; the multimodal CLIP model is used to re-encode the original data and compare and match it with the benchmark vectors in the standard semantic framework; a transfer learning strategy is adopted to transfer the semantic mapping rules in the reliable data to the unreliable data to complete the semantic label correction; finally, the calibrated data is re-entered into the system for verification until it meets the reliability assessment criteria.

[0068] Working Principle: This invention aims to address the inconsistencies in classification systems, expression granularity, and semantic boundaries among multi-source data. The system comprises a data acquisition and recognition module, a semantic density balance analysis module, a boundary fuzzy contagion analysis module, a comprehensive reliability assessment module, and a semantic alignment feedback module. The data acquisition module integrates multi-source data, including remote sensing imagery, GIS platforms, land use maps, and planning documents, to extract unified semantic units and identify "semantic density balance" and "boundary fuzzy contagion" as core evaluation dimensions. The semantic density balance analysis module uses the CLIP model to achieve cross-modal semantic alignment. Combining kernel density estimation with information entropy calculation, it improves the KL divergence method to assess semantic distribution deviation and constructs a semantic density balance anomaly scoring function to output eigenvalues. The boundary fuzzy contagion analysis module uses semantic segmentation to identify fuzzy areas at land class boundaries, extracts the semantic gradient change rate, establishes a diffusion model to determine the semantic contamination radius, and constructs a boundary fuzzy contagion scoring function to determine whether semantic leakage exists. The comprehensive reliability assessment module integrates the two eigenvalues ​​into an evaluation vector, which is then input into a convolutional neural network model for training. The resulting vector outputs a confidence score and classifies the data reliability. The semantic calibration feedback module uses semantic graphs and knowledge reasoning mechanisms based on unreliable data, combined with transfer learning to complete semantic corrections, ultimately achieving intelligent integration and semantic consistency optimization of national land space planning data. This technical solution effectively improves the accuracy and practicality of multi-source heterogeneous data fusion, providing high-quality data support for national land space planning decision-making.

[0069] The above is a detailed description of an embodiment of the present invention. However, the content described is only a preferred embodiment of the present invention and should not be considered to limit the scope of the present invention. All equivalent changes and improvements made within the scope of the present invention should still fall within the scope of the patent coverage of the present invention.

Claims

1. The intelligent integration system of multi-source heterogeneous data of national land space planning based on semantic analysis is characterized by: include: A data collection and recognition module is used to obtain national land space planning data and identify and select semantic density balance and boundary fuzzy contagion as core evaluation dimensions; The semantic density balance analysis module calculates the information entropy differences of each data source under a unified semantic framework to generate semantic density balance anomaly feature values, which are used to evaluate the degree of information distribution dispersion of different classification systems. Specifically, it includes: According to the identification and selection of semantic density balance, the information entropy difference of each data source under the unified semantic framework is calculated, and the abnormal characteristic value of the semantic density balance is calculated. It is judged whether the abnormal characteristic value of the semantic density balance is greater than or equal to the preset threshold. If so, the information distribution of different classification systems is dispersed, otherwise, the information distribution of different classification systems is concentrated; The process of obtaining the abnormal feature value of the semantic density balance is as follows: The semantic units in multi-source heterogeneous land space planning data are mapped to a unified semantic representation space through the CLIP model to achieve cross-modal semantic alignment. Subsequently, the kernel density estimation method is combined with the Gaussian kernel function and a dynamic bandwidth selection strategy based on reinforcement learning to perform local density modeling on the semantic distribution of each data source. Based on the local density function, a probability distribution is constructed and the information entropy is calculated. At the same time, a semantic relationship graph is constructed, and a graph attention network is used to perform weighted modeling of the semantic similarity between different data sources. On this basis, the calculation method of Kullback-Leibler divergence is improved to evaluate the semantic distribution deviation between data sources. Finally, a semantic density balance anomaly scoring function is constructed by combining information entropy and KL divergence to output the semantic density balance anomaly feature value. The improved calculation method of the Kullback-Leibler divergence is specifically expressed as follows: ;in, Indicates that the data source To the data source The Kullback-Leibler divergence, Indicates that at the sample point Data source and The semantic weight factor between Indicates that at the sample point Data source The probability density function value of Represents a data source The probability density function value of and Represents two different data sources, Indicates the data source sample point; The boundary fuzzy contagion analysis module analyzes the semantic gradient changes of the boundary fuzzy area, calculates its semantic contamination radius to the adjacent data area, and generates a boundary fuzzy contagion feature value to assess whether there is semantic leakage in the land space planning data. Specifically, it includes: According to the identified and extracted boundary fuzzy contagion, by analyzing the semantic gradient change of the boundary fuzzy area, the semantic contamination radius of the boundary fuzzy contagion to the adjacent data area is calculated, and the boundary fuzzy contagion characteristic value is generated. It is judged whether the boundary fuzzy contagion characteristic value is greater than or equal to the preset threshold. If so, there is semantic leakage in the national land space planning data. If not, there is no semantic leakage in the national land space planning data. The process of obtaining the boundary fuzzy infectious characteristic value is as follows: First, a semantic segmentation model is used to identify the boundary areas between different land types and define them as semantic fuzzy areas. Within the fuzzy areas, the semantic similarity gradient change rate between adjacent land types is calculated, and the sliding window method is used to extract the local semantic gradient direction and amplitude. A semantic contamination propagation model is established based on the gradient change to simulate the impact range of semantic leakage, and the semantic contamination radius is solved using the diffusion equation. Finally, a boundary fuzzy contagion scoring function is constructed by combining the contamination radius with the semantic similarity decrease rate to generate a boundary fuzzy contagion feature value. A comprehensive reliability assessment module, which constructs a comprehensive reliability assessment vector based on the semantic density balance anomaly eigenvalue and the boundary fuzzy contagion eigenvalue, and inputs it into a deep learning model for analysis. Based on the analysis results, the national land space planning data is divided into reliable data and unreliable data; A semantic calibration feedback module is provided, which calibrates the national land space planning data based on unreliable data.

2. The intelligent integration system for multi-source heterogeneous data of national land space planning based on semantic analysis according to claim 1 is characterized in that: The above method constructs a comprehensive reliability evaluation vector by combining the semantic density balance abnormality feature value and the boundary fuzzy infectious feature value, and inputs it into the deep learning model for analysis, specifically including: Obtain the semantic density balance anomaly eigenvalues ​​and the boundary fuzzy infectious eigenvalues, and construct a comprehensive reliability evaluation vector with the semantic density balance anomaly eigenvalues ​​and the boundary fuzzy infectious eigenvalues ​​as the input of the deep learning model. The deep learning model is trained, and the confidence score of the national land space planning data is output according to the trained deep learning model. The deep learning model uses a convolutional neural network model.

3. The intelligent integration system for multi-source heterogeneous data of national land space planning based on semantic analysis according to claim 2 is characterized in that: The training process of the deep learning model is: The comprehensive reliability assessment vector is used as an input sample and input into a convolutional neural network model for training. The convolutional neural network model consists of several one-dimensional convolutional layers, pooling layers and fully connected layers, and is used to extract local correlation and high-order interaction information in the feature vector. The cross-entropy loss function is used in the training process, and manually labeled reliable or unreliable samples are used as supervision signals. The network parameters are continuously optimized through the backpropagation algorithm to improve the accuracy of the model's semantic consistency judgment.

4. The intelligent integration system for multi-source heterogeneous data of national land space planning based on semantic analysis according to claim 1 is characterized in that: The classification of national land space planning data into reliable data and unreliable data specifically includes: The threshold interval is set according to the confidence score output by the deep learning model. If the score is higher than the first threshold, it is judged as highly reliable data; if the score is between the first and second thresholds, it is judged as generally reliable data; if the score is lower than the second threshold, it is judged as unreliable data.

5. The intelligent integration system for multi-source heterogeneous data of national land space planning based on semantic analysis according to claim 1 is characterized in that: The calibration of national land space planning data based on unreliable data specifically includes: For data marked as unreliable, the contextual information in the semantic relationship graph is called up, combined with the knowledge graph reasoning mechanism to locate the most likely semantic attribution category; the multimodal CLIP model is used to re-encode the national land space planning data and compare and match it with the benchmark vectors in the standard semantic framework; a transfer learning strategy is adopted to transfer the semantic mapping rules in the reliable data to the unreliable data to complete the semantic label correction; the calibrated data is re-entered into the system for verification until the reliability assessment criteria are met.

Images