Intelligent generation method of surveying and mapping technical scheme for scene semantic understanding

Abstract

The application discloses a surveying and mapping technical scheme intelligent generation method for scene semantic understanding, and particularly relates to the technical field of surveying and mapping intelligence, and comprises the following steps: constructing a task description vector to extract task characteristic parameters, acquiring dynamic environment perception data to identify semantic feature labels, generating a structured representation, and then constructing a task-driven semantic vector field based on the task description vector and the semantic feature labels, assigning a gravitational weight and calculating a semantic weight entropy, dynamically adjusting a priority, and finally determining a main semantic path according to a subject semantic ordering structure to generate a surveying and mapping technical scheme matched with a current task; the application realizes standardized expression of a surveying and mapping task target and dynamic correlation modeling of scene semantics by constructing a task description vector and a semantic vector field, introduces a semantic weight entropy regulation and control mechanism, solves the problems of semantic priority misjudgment and scheme deviation, and improves the accuracy, stability and intelligent level of surveying and mapping technical scheme generation.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of intelligent mapping technology, more specifically, the present application relates to a mapping technical scheme intelligent generation method for scene semantic understanding. BACKGROUND

[0002] With the rapid evolution of urban space and the continuous growth of the demand for intelligent application of geographic information, mapping operations are gradually shifting from single-target collection in static environments to new operational paradigms of multi-source perception, dynamic response, and intelligent decision-making. In the new generation of mapping tasks, not only do we need to meet the task-side requirements such as operational purpose, operational accuracy, and delivery time, but we also need to take into account the complex and frequently changing semantic environment in the survey area. In particular, in post-disaster reconstruction, urban emergency, smart construction sites, underground space, and other scenarios, there are higher requirements for the immediacy, scene adaptability, and decision-making intelligence of mapping schemes.

[0003] Existing mapping scheme design mainly relies on expert experience or scheme templates constructed based on fixed rules, and often uses a static rule mapping strategy, i.e., one-to-one matching of task types and pre-set scene schemes. However, in actual operations, there are often multiple complex semantics coexisting in mapping scenarios. For example, a certain area contains both water bodies and a large number of temporary buildings, accompanied by high traffic density and land development activities. In such dynamic scenarios, the system needs to make primary and secondary judgments among multiple concurrent semantic features to determine the focus of mapping and generate the most suitable technical scheme accordingly. However, existing schemes generally lack dynamic reasoning capabilities for semantic primary paths, especially in two typical problem scenarios:

[0004] First, the problem of "dynamic environment target semantic priority misordering" arises due to the lack of dynamic modeling mechanisms for the correlation between mapping task targets and scene semantics. As a result, secondary semantics with higher occurrence frequency but lower actual priority (such as temporary buildings or water bodies) are often misjudged as core elements, leading to the development of technical schemes around secondary features, resulting in missing data in the primary target area, severe precision index deviations, and even failure of mapping results to pass quality acceptance, causing data re-sampling or task failure.

[0005] Second, the problem of "semantic ambiguity causing logical path splitting" arises. Some scene semantics have multiple attribution properties (e.g., "slope stability" can be classified as geological monitoring or engineering safety supervision), and without clear instructions or primary path guidance, the system often triggers multiple conflicting paths in the reasoning process, generating several mutually exclusive or redundant technical schemes. This not only significantly increases the cost of manual judgment but also easily causes strategic misjudgment, reduces operational efficiency, and even causes systematic risks to the legality and applicability of subsequent mapping data.

[0006] In addition, the prior art lacks a deep structured expression of scene semantics, cannot uniformly model the feature form, spatial relationship and dynamic attributes in the scene, and cannot dynamically focus on the most representative semantic features under the task target dimension of surveying and mapping. There is a lack of a technical framework that integrates task demand driving, scene semantic understanding and dynamic reasoning and decision-making capabilities. Therefore, the present application proposes a surveying and mapping technical scheme intelligent generation method for scene semantic understanding to solve the above problems. SUMMARY

[0007] To achieve the above object, the present application provides the following technical scheme:

[0008] The surveying and mapping technical scheme intelligent generation method for scene semantic understanding comprises the following steps:

[0009] The task target analysis step constructs a task description vector according to the target requirements, job type and precision index of the to-be-executed surveying and mapping task, and extracts task feature parameters affecting the generation of the surveying and mapping technical scheme;

[0010] The scene semantic feature extraction step acquires dynamic environment perception data of the surveying and mapping area, identifies semantic feature labels in the scene, and generates a structured representation containing multiple semantic features;

[0011] The task-driven semantic vector field construction step constructs a task-driven semantic vector field based on the task description vector and the semantic feature labels, assigns an attractive weight to each semantic feature, and uses the attractive weight to represent the correlation degree between the semantic feature and the task target;

[0012] The dynamic semantic weight entropy calculation and regulation step calculates the entropy value of the attractive weight of each semantic feature in the semantic vector field, acquires a dynamic semantic weight entropy index, and when the semantic weight entropy index exceeds a preset standard threshold, dynamically adjusts the priority weight of the semantic feature according to the attractive direction and distribution trend in the semantic vector field, and updates the main semantic ranking structure;

[0013] The surveying and mapping technical scheme generation step determines a main control semantic path according to the updated main semantic ranking structure, calls a technical knowledge base or historical technical scheme case, and generates a surveying and mapping technical scheme matching the current task and scene.

[0014] In a preferred embodiment, the construction of the task description vector specifically comprises the following steps:

[0015] Collecting original task input information related to the surveying and mapping task, the original task input information including the job purpose, job object, job area, precision requirement, delivery form and execution period, and converting it into a standard structured data format to construct a unified task information input model;

[0016] Based on the preset task information input model, a task feature dimension system is established, each item of input information is mapped into a predefined feature dimension, the feature dimension includes a survey purpose type dimension, a work area complexity dimension, a data precision level dimension, a time constraint dimension and a result delivery form dimension, and an initial task description vector is formed through dimension mapping;

[0017] The task description vector is subjected to feature weight evaluation and parameter simplification processing. According to the statistical weight value of the influence degree of each feature dimension on the scheme generation in the historical survey task, the key task feature parameters influencing the generation of the survey technical scheme are extracted. The task feature parameters are used to reflect the task attributes that should be given priority in the survey scheme reasoning process of the current task;

[0018] The key task feature parameters are stored in a structured vector form as basic input data for subsequent process of survey task target modeling and related reasoning processing.

[0019] In a preferred embodiment, the operation of performing feature weight evaluation and parameter simplification processing on the task description vector includes the following steps:

[0020] According to the preset task feature weight reference system, the initial weight value of the influence degree of each task feature dimension on the generation of the survey scheme is determined. The initial weight value is based on statistical analysis of historical survey task result data and domain knowledge rule setting, covering different purpose types, spatial complexity levels and result requirements and other factors;

[0021] The initial weight value is compared with each dimension feature in the current task description vector, and the effective weight index of each feature dimension is calculated by using a weighted sorting method. The effective weight index is used to quantify the contribution degree of the feature dimension to the generation of the survey scheme under the current task background;

[0022] Based on the effective weight index and the preset weight threshold value, the secondary feature dimensions with a weight index lower than the weight threshold value are removed, and only the dimension features with an effective weight index higher than the weight threshold value are retained as key feature parameters to form the parameter simplification result of the task description vector.

[0023] In a preferred embodiment, the scene semantic feature extraction includes the following steps:

[0024] Obtain the dynamic environment perception data of the survey area and perform feature extraction and fusion analysis according to the preset strategy, identify the significant feature shape, spatial structure relationship and dynamic change attribute in the scene, and assign semantic labels according to the preset task context semantic model to obtain a plurality of scene element sets with semantic attributes;

[0025] The scene element set is structured and organized, a multi-dimensional structured representation matrix is constructed, and the representation matrix is used to describe the feature distribution of the current surveying and mapping area in the spatial, attribute, and semantic dimensions.

[0026] In a preferred embodiment, constructing the task-driven semantic vector field means:

[0027] Based on the feature dimensions contained in the task description vector, the dimension structure and coordinate system reference of the semantic vector space are defined, the semantic feature labels are mapped according to their attribute types and task feature dimensions, the initial position coordinates of each semantic feature in the semantic vector space are determined, the semantic labels are distributed in the vector field according to the correlation between the task features and the semantic features to form a clustering area, and finally the construction of the task-driven semantic vector field is completed.

[0028] In a preferred embodiment, assigning the gravity weight to each semantic feature means:

[0029] The correlation score between each feature dimension in the task description vector and the semantic feature label is calculated, and then the correlation score is normalized as the initial value of the gravity weight of the semantic feature in the task-driven semantic vector field;

[0030] Based on the spatial adjacency relationship and type hierarchy relationship between the semantic features, the initial value of the gravity weight is optimized and adjusted to form the final gravity weight distribution, and the optimized gravity weight is bound to each semantic feature label.

[0031] In a preferred embodiment, the correlation score calculation process includes the following steps:

[0032] Each feature dimension in the task description vector is matched with the corresponding attribute dimension in the semantic feature label to form a set of feature correspondence relationships; for each set of feature correspondence relationships, a feature similarity value is calculated, the similarity value is calculated by using a vector angle cosine value, a feature value difference ratio, or a matching score based on a rule table, and is used to measure the adaptation degree of the task demand and the semantic attribute in a certain dimension; the similarity values of the feature dimensions are weighted and summarized according to the preset weight to form an overall semantic adaptation score, which is used as the correlation score between the task description vector and the semantic feature label.

[0033] In a preferred embodiment, optimizing and adjusting the initial value of the gravity weight means:

[0034] Based on the actual spatial coordinate information of the semantic feature label in the surveying and mapping area, the spatial adjacency degree between any two semantic features is calculated by using Euclidean distance or spatial overlap ratio, and a spatial adjacency weight matrix is constructed;

[0035] Based on the preset semantic type hierarchy structure, different hierarchical correction coefficients are set for the superior type, subordinate type and sibling type relationship, and a hierarchical relationship correction matrix is ​​constructed to measure the conceptual correlation between different types of semantic features;

[0036] A fusion weighting method is adopted, which multiplies the corresponding elements of the spatial adjacency weight matrix and the hierarchical relationship correction matrix and normalizes them. This normalized result is used as a fine-tuning factor to adjust the initial gravity weights. Finally, the adjusted weight values ​​are updated to the final gravity weights of each semantic feature.

[0037] In a preferred embodiment, when calculating and controlling the dynamic semantic weight entropy, the gravitational weight values ​​of all semantic features in the semantic vector field are statistically analyzed, and the entropy value of the semantic weight distribution is calculated using the information entropy formula. The information entropy calculation is based on the weighted sum of the proportion of each semantic feature weight and its logarithmic product, which is used to measure the degree of dispersion of the overall weight distribution.

[0038] The calculated entropy value is compared with a preset standard threshold. When the entropy value is higher than the standard threshold, it is determined that there is significant uncertainty or disorder in the current semantic gravity distribution, and priority weight adjustment is required.

[0039] Obtain the gravitational direction vector of each semantic feature and calculate the angle between it and the gravitational center direction of the current main task as the gravitational direction offset. The smaller the offset, the more concentrated the semantic feature direction is. Count the number of adjacent semantic features in the vector field coordinate space and calculate the density value of semantic features in a unit space. The higher the density value, the more aggregated the spatial distribution of semantic features is.

[0040] According to the set enhancement and attenuation control rules, when the gravitational direction offset is less than the preset angle threshold and the distribution density is higher than the preset density threshold, the enhancement strategy is executed to increase the gravitational weight of the semantic feature; when the gravitational direction offset is greater than the angle threshold or the distribution density is lower than the density threshold, the weight reduction strategy is executed to reduce its gravitational weight, dynamically update the gravitational weight distribution, and form a focused and stable master semantic ranking structure.

[0041] In a preferred embodiment, when generating the surveying and mapping technical solution, the semantic features that are ranked first are selected in sequence according to the updated main semantic sorting structure, and a main control semantic path is constructed. The main control semantic path is used to characterize the semantic requirements and logical structure that need to be prioritized in the surveying and mapping task.

[0042] Using the main semantic path as the retrieval input, the technical knowledge base and historical technical solution case library are invoked to perform a matching retrieval operation and filter the candidate technical solution structure template with the highest similarity to the current path.

[0043] According to the current task description vector and the semantic elements in the master semantic path, the candidate scheme structure template is adjusted in parameters and filled in content to generate a surveying and mapping technical scheme matched with the current task and scene, and finally the generated surveying and mapping technical scheme is taken as an intelligent output result and submitted for task execution or manual review and verification.

[0044] Technical effects and advantages of the present application:

[0045] The present application constructs a task description vector by structurally analyzing the surveying and mapping task target, operation type and precision index, and further extracts key task feature parameters, so that the surveying and mapping task information can be converted from the original, fuzzy and natural language description into a standard vector expression format that can be recognized and calculated by downstream logic. This vector expression not only improves the information analysis efficiency, but also facilitates the unified processing and comparison of different types of surveying and mapping tasks in the subsequent process. Compared with the traditional method of relying on manual experience for task understanding and technical scheme formulation, the present application can significantly reduce subjective judgment bias and improve the consistency and professionalization of scheme generation, especially in the scene of high-frequency concurrent multi-source tasks and complex task targets.

[0046] In the process of constructing the task-driven semantic vector field, the present application uses the correlation degree between the task description vector and the scene semantic feature label as the gravitational weight basis to ensure that the position and intensity of each semantic feature in the vector space reflect its real dominance over the task target, thereby establishing a dynamically adjustable semantic force field model. This method enables the system to accurately quantify the attention degree of different task targets to semantic elements such as "water body", "geology", "traffic" and "structure" in the scene, promoting intelligent focusing and structural level reconstruction of semantic features. Compared with the existing fixed rule or template-based semantic recognition technology path, the present application has higher flexibility and task adaptation ability, and can significantly improve the personalization degree and scene response ability of technical scheme generation.

[0047] The present application introduces a dynamic semantic weight entropy index as an important basis for determining the stability of semantic priority distribution, breaking through the limitations of traditional scheme design that relies on manual experience and cannot adapt to dynamically changing scenes. When the semantic gravitational weight distribution is obviously dispersed or unbalanced, the present application can perform enhancement or attenuation operation based on the gravitational direction deviation and semantic feature distribution density in the vector field, and update the semantic ranking structure in real time, so as to dynamically focus on the master semantic path most related to the task target. This design effectively avoids the scheme generation deviation, path conflict or decision delay caused by semantic ambiguity and priority error, significantly improves the robustness and intelligence level of the surveying and mapping technical scheme generation process, and is especially suitable for dynamic or multi-source semantic concurrent surveying and mapping task environment. BRIEF DESCRIPTION OF DRAWINGS

[0048] For the convenience of those skilled in the art to understand, the present application will be further described below in conjunction with the drawings;

[0049] Figure 1 The principle diagram of the intelligent generation method of the scene semantic understanding-oriented surveying and mapping technical scheme in the present application. DETAILED DESCRIPTION

[0050] The technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of protection of the present application.

[0051] Reference Figure 1 The following embodiments are obtained:

[0052] Embodiment 1: The intelligent generation method of the scene semantic understanding-oriented surveying and mapping technical scheme includes the following steps: a task target analysis step, constructing a task description vector according to the target requirements, operation type and precision index of a to-be-executed surveying and mapping task, and extracting task feature parameters affecting the generation of a surveying and mapping technical scheme; standardizing the unstructured description information of the surveying and mapping task to form a high-dimensional vector expression capable of participating in intelligent reasoning. By constructing the task description vector, not only can the technical requirements of the current task be clearly expressed, but also the key task feature parameters having a decisive influence on the generation of the technical scheme can be extracted, thereby providing an accurate task context basis for subsequent semantic modeling and reasoning matching.

[0053] A scene semantic feature extraction step, acquiring dynamic environment perception data of a surveying and mapping area, identifying semantic feature labels in the scene, and generating a structured representation containing multiple semantic features; perceiving and understanding the actual environment state of the current surveying and mapping area, extracting semantic information from multiple sources of data, identifying representative scene semantic feature labels, and performing structured expression. This process converts the spatial environment information of the real world into computable semantic feature data, so that the subsequent reasoning system can understand the survey area content in the semantic dimension, thereby improving the adaptability and intelligence of the surveying and mapping scheme.

[0054] The task-driven semantic vector field construction step is based on the task description vector and the semantic feature label to construct a task-driven semantic vector field, and the semantic features are assigned with gravitational weights to represent the correlation degree between the semantic features and the task target. By associating the task requirements with the scene semantics, a semantic vector field with spatial structure and semantic logic is constructed. In the vector field, each semantic feature is assigned with a gravitational weight according to its correlation degree with the task, and the gravity size represents the importance degree of the semantic feature to the scheme generation. This mechanism can realize the quantitative sorting of multiple semantic elements in a complex scene, and lays the foundation for intelligent decision reasoning.

[0055] The dynamic semantic weight entropy calculation and regulation step calculates the entropy value of the gravitational weight of each semantic feature in the semantic vector field to obtain a dynamic semantic weight entropy index. When the semantic weight entropy index exceeds a preset standard threshold, the priority weight of the semantic feature is dynamically adjusted according to the direction and distribution trend of the gravitational force in the semantic vector field, and the subject semantic sorting structure is updated. By calculating the distribution entropy value of the semantic gravitational weight, the information focusing degree and the priority stability in the current semantic structure are monitored. When the system identifies that the weight entropy value is too high and the semantic distribution is chaotic, the gravitational weight can be dynamically adjusted according to the spatial distribution density and the direction consistency, so as to optimize the semantic sorting result. This regulation mechanism enhances the adaptive ability of the system in a complex or dynamic survey area environment, and ensures that the main control semantic logic chain is clear and reliable.

[0056] The surveying and mapping technical scheme generation step determines the main control semantic path according to the updated main semantic sorting structure, calls the technical knowledge base or historical technical scheme case, and generates a surveying and mapping technical scheme matched with the current task and scene. The main control semantic path expresses the priority processing logic in the surveying and mapping task, and the related technical scheme template is retrieved based on this. Based on the fusion of the current task feature parameters and the scene semantic structure, the scheme content is accurately matched and intelligently generated, so as to provide a technical scheme highly matched with the actual task environment, reduce the burden of manual preparation, and improve the efficiency and quality of the surveying and mapping project.

[0057] For a city scene mapping task to be performed, raw task input information related to the mapping task is collected and unified format constraint and semantic checking are performed. The raw task input information includes job purpose, job object, job area, precision requirement, delivery form and execution period. For example, the job purpose is city road reconstruction completion acceptance measurement, the job object is expressway and both sides building facade and auxiliary structure, the job area is a strip area of 10 kilometers long and 100 meters wide, the precision requirement is that the plane position error is not greater than 2 cm and the height error is not greater than 3 cm, the delivery form is a three-dimensional real scene model and a topographic map with a scale of 1:500, and the execution period is 30 days. Through field legality check, unit conversion and time expression unification, the text description, numerical description and constraint description are converted into analyzable standard structured data to construct a unified task information input model.

[0058] Based on the task information input model, a task characteristic dimension system covering mapping purpose type, job area complexity, data precision level, time constraint and result delivery form is established, and the foregoing structured data is mapped to the corresponding characteristic dimension one by one to form an initial task description vector. For example, "city road completion measurement" is mapped to a high correlation value of the mapping purpose type dimension, "a strip area of 10 kilometers long and 100 meters wide containing dense building facades" is mapped to a higher value of the job area complexity dimension, "plane position error not greater than 2 cm and height error not greater than 3 cm" is mapped to a high level value of the data precision level dimension, "completed in 30 days" is mapped to a urgent value of the time constraint dimension, and "three-dimensional real scene model and topographic map with a scale of 1:500" is mapped to a composite value of the result delivery form dimension. Thus, an initial task description vector normalized in the closed interval of zero to one is obtained, for example, the numerical values on five characteristic dimensions are 0.78, 0.85, 0.90, 0.80 and 0.75.

[0059] The task description vector is subjected to feature weight evaluation and parameter reduction processing, only retaining key task feature parameters that have a decisive influence on the generation of the surveying and mapping technical scheme, and recording the source field and value basis in a traceable manner; for example, in the above example, the higher value of the job area complexity dimension, the high level value of the data accuracy level dimension, and the urgent value of the time constraint dimension are retained, while the three-dimensional real scene model in the delivery form dimension and the topographic map with a scale of one to five hundred are taken as constraint parameters to ensure that the subsequent reasoning link can focus on the real task key. Finally, the extracted key task feature parameters are stored in a structured vector form, with a timestamp, task identification, and data integrity check value, serving as the basic input data for subsequent processes; when entering the scene semantic feature extraction step, the aforementioned key task feature parameters will be used to limit the data acquisition priority and processing precision threshold, and when entering the step of constructing the task-driven semantic vector field, the aforementioned key task feature parameters will be used as the prior basis for correlation degree calculation, thereby realizing the coherent connection from task target analysis to semantic understanding to scheme reasoning.

[0060] The operation of feature weight evaluation and parameter reduction processing on the task description vector refers to: according to the preset task feature weight reference system, the task description vector that has been constructed is subjected to preliminary evaluation. This reference system is derived from the statistical data analysis of a large number of historical surveying and mapping tasks and the modeling induction of actual job result evaluation results, involving five key dimensions of surveying and mapping purpose type, job area complexity, data accuracy level, time constraint strength, and result delivery form. Each dimension is assigned an initial weight value to reflect the basic influence degree of the feature in the generation of the surveying and mapping technical scheme. For example, for a city underground space surveying and mapping task targeting high-precision modeling, the initial weight value of the data accuracy level dimension can be set to 0.30, and the initial weight value of the result delivery form dimension is set to 0.20, and the remaining dimensions are in turn purpose type 0.18, area complexity 0.17, and time constraint 0.15. The above initial weight values are automatically selected according to the specific task field category, and can also be fine-tuned after manual experience verification.

[0061] The initial weight value is compared with each dimension feature value in the current task description vector, and the effective weight index of each dimension is calculated by using the weighted product method, which is used to reflect the comprehensive importance of the dimension in the current task context. Taking the "work area complexity dimension" as an example, if the feature value of this dimension in the task description vector is 0.85, and the corresponding initial weight value is 0.17, then the effective weight index is calculated as 0.85*0.17=0.1445. For other dimensions, the same method is used. Finally, an effective weight index list can be formed, which is used to sort the importance of each feature dimension in the task. This operation not only embodies the universality of the standard weight, but also introduces the personalized adjustment mechanism of the task specific parameters, thereby realizing the fusion calculation of global experience and local actuality.

[0062] In the present application, the statistical weight value refers to the quantitative evaluation index obtained by analyzing the influence frequency and weight contribution of different task feature dimensions on the technical scheme selection result in the historical surveying and mapping task database, which is usually determined by expert evaluation, task frequency analysis or regression statistical model. The effective weight index is a comprehensive score calculated based on the current task demand matching degree, which is used to reflect the influence degree of each feature dimension on the scheme generation decision in the current task.

[0063] The weight threshold of task feature parameter screening is set, for example, the preset threshold is 0.12. In the actual projects such as smart city modeling, emergency surveying and mapping, and engineering acceptance measurement: the project unit often sets "key parameter items" (such as survey area complexity ≥0.8, modeling accuracy ≥0.9) as the technical scheme must-selected feature basis; some intelligent surveying and mapping platforms (such as South Digital, China Haidian, Space Macrograph, etc.) all use the heuristic strategy of "high weight feature priority participating in scheme assembly" in their scheme generation engine; in actual deployment, the default screening threshold is often set to the interval of 0.12-0.15 to balance the model accuracy and system overhead. The effective weight index calculated is screened, and only the feature dimensions greater than or equal to the threshold are retained as the key task feature parameters for subsequent semantic association modeling and technical scheme reasoning. Taking the example data as an example, if the effective weight indexes of the data accuracy level dimension, the work area complexity dimension and the use type dimension are 0.27, 0.14 and 0.13 respectively, which exceed the preset threshold, then the above three will be retained, while the time constraint dimension and the result delivery form dimension will be excluded because their weight indexes are lower than the threshold (0.09 and 0.10 respectively). Through this simplification mechanism, the interference of irrelevant or less influential features participating in subsequent modeling calculation is significantly reduced, and the accuracy and efficiency of subsequent semantic mapping and scheme generation are improved.

[0064] Finally, the key task feature parameters retained after screening are reorganized into a simplified structured vector representation, with the feature dimension identifier, calculation timestamp, and index number of the source field as the basic data source for the "task side" input in the semantic vector field construction stage. In practical applications, this simplified vector not only directly participates in the initial value calculation of the semantic feature weight, but also can be used for subsequent weight entropy monitoring, path reasoning guidance, and key feature anchoring in technical scheme matching, thus running through the entire process of task analysis, semantic understanding, and scheme generation, ensuring the logical closed loop and semantic consistency of the technical process.

[0065] The scene semantic feature extraction includes the following steps: based on the feature dimensions contained in the task description vector which has been constructed, the dimension structure and coordinate system reference of the semantic vector space are set. The dimension structure of the semantic vector space should cover all the task feature dimensions involved in the current surveying and mapping task, including but not limited to the complexity of the work area, the type of surveying and mapping purpose, the level of data accuracy, the degree of time constraint, and the requirement of result delivery, etc. The scene element set is structured and organized to construct a multi-dimensional structured representation matrix. The representation matrix is used to describe the feature distribution of the current surveying and mapping area in the spatial, attribute, and semantic dimensions. The multi-dimensional structured representation matrix is used to organize the scene elements in the surveying and mapping area in the spatial, attribute, and semantic dimensions, and its matrix structure is in the form of three-dimensional tensor. The first dimension represents the spatial location index of the ground object, the second dimension represents the attribute feature vector of the ground object, and the third dimension represents the semantic label index. Through this tensor structure, the linkage modeling of the same geographical unit in multiple perception dimensions can be realized.

[0066] In the setting process, the Euclidean space is taken as the basic framework, and each task feature dimension is mapped into an orthogonal axis in the semantic vector space, and the initial scale of each dimension is set by the importance weight of the task target. For example, in a certain high-precision underground facility surveying and mapping task, the importance of the data accuracy level dimension is evaluated as 0.30, which corresponds to the highest proportion of the axis range in the vector space.

[0067] In the already set semantic vector space, the scene semantic feature labels are respectively mapped to the task feature dimensions according to their attribute types. The semantic feature labels come from the scene semantic extraction step in the previous stage, and their types include ground object shape class (such as bridge, pipeline, slope), dynamic change class (such as water accumulation, water level change, traffic flow), and spatial structure class (such as road density, obstacle distribution, and region boundary). For example, the dynamic attribute of the "traffic flow" semantic label can be mapped to the time constraint dimension in the task description vector, and "underground pipeline node density" can be mapped to the region complexity dimension. After the mapping is completed, the initial position coordinates of the labels in the semantic vector space are calculated according to the correlation scores of the labels and each task dimension. The higher the score, the closer to the positive direction of the corresponding dimension.

[0068] According to the correlation between the semantic feature tags and the task features, the gravitational effect of all semantic tags in the semantic vector space is calculated to form a vector direction and a gravitational amplitude. The gravitational effect is used to simulate the "semantic pull" exerted by a certain semantic tag on the task to achieve the target. The calculation of the gravitational value can be based on the aforementioned correlation score, and combined with the spatial adjacency degree and the semantic hierarchical distance for correction. For example, when the correlation score between "geological subsidence area" and "high-precision monitoring" task target is 0.88, and adjacent to the "historical landslide point" tag in space, the tag will have a stronger aggregated gravitational effect in the vector space. After the calculation of multiple sets of tags is completed, they are aggregated into several semantic clustering areas according to similar gravitational directions, and different areas represent potential directions of task semantic focus.

[0069] Finally, the complete task-driven semantic vector field is constructed by combining the coordinates of all semantic features in the space, the gravitational direction and amplitude characteristics. The semantic vector field is a multi-dimensional space structure with a dynamically evolving semantic clustering pattern, which can respond to changes in task feature parameters or environmental inputs in real time. In the subsequent process, the semantic vector field will serve as the basis for dynamic semantic regulation, priority sorting and scheme path reasoning, supporting the stability and intelligent level of the entire surveying and mapping technical scheme generation process. The semantic vector field constructed in this way has significant adaptive ability and semantic convergence, and can effectively deal with problems such as semantic conflict and master path splitting in a dynamic environment.

[0070] In actual application, in order to ensure that each semantic feature in the semantic vector field can effectively represent the correlation degree between it and the surveying and mapping task target, the embodiment proposes a gravitational weight distribution method based on the adaptive relationship between task features and semantic features. According to the multiple feature dimensions contained in the task description vector that has been constructed, such as "surveying purpose type dimension", "data precision level dimension" and "result delivery form dimension", etc., dimension matching processing is performed on each semantic feature tag. Through traversal, the attribute structure of each semantic feature is corresponded to the task feature dimension one by one to form a feature matching mapping table, laying a foundation for subsequent correlation measurement. Based on the completed matching mapping table, the correlation score between each pair of task feature dimension and semantic feature is calculated. The embodiment mainly uses vector angle cosine similarity calculation method, supplemented by feature value difference ratio evaluation and rule scoring method. Taking "data precision level" as an example, if the task feature is "0.2 meter precision" and the semantic feature tag is "high-resolution remote sensing image (0.1-0.3 meters)", the cosine value is close to 1 and the similarity is high, and the final score can be set to 0.92. All scores are normalized to the [0, 1] interval.

[0071] The correlation scores in the above multiple dimensions are integrated in a weighted manner. The dimension scores are weighted and summed according to the preset feature influence weight coefficients in the task description vector, and finally form the "initial gravity weight" of each semantic feature. For example, assuming that the scores of a semantic feature in three task dimensions are 0.85, 0.76 and 0.93, and the corresponding dimension weights are 0.3, 0.4 and 0.3, then the weighted score is: 0.85*0.3+0.76*0.4+0.93*0.3=0.841. Finally, the calculated initial gravity weight is bound to each semantic feature, and is visualized and mapped in the semantic vector space, and is labeled with the initial gravity strength, which is used as the input basis for subsequent dynamic optimization and control steps.

[0072] To ensure that the gravity weight of the semantic feature has good task adaptability, the embodiment proposes a multi-dimensional feature fusion semantic adaptation degree calculation method, which is characterized by extracting the similarity relationship between the task and the semantic feature from multiple attribute dimensions. Based on the structured feature dimensions of the task description vector, the comparable attribute information in all semantic feature labels is extracted. Taking "work area complexity" as an example, if the task input is "urban high-density building area", the elements with "dense buildings and complex road network" attributes in the semantic label will be included in the comparison set to ensure dimensional consistency.

[0073] For the constructed feature comparison set, a similarity calculation operation is performed. The embodiment mainly uses three calculation methods: one is the vector angle cosine value (such as mapping the attribute vector to an n-dimensional space and calculating the angle cosine); two is the numerical feature difference ratio (such as area size, length, error index, etc.); three is a scoring model based on a rule base (such as setting the similarity between "urban road" and "town road" to 0.85 according to industry experience). All scores are standardized. The similarity values in each dimension are multiplied by the known preset dimension weights in the task feature to form the weighted score, i.e. the overall semantic adaptation score. The preset weight refers to the standard influence factor set for each task feature dimension, which is set according to the influence of each feature on the scheme selection result in the statistical historical surveying and mapping project, and combined with the expert experience knowledge base. The weight value can be classified and set according to the purpose category (such as urban and rural planning, disaster assessment), complexity level (such as terrain relief degree), result requirement (such as resolution level), etc. and normalized in percentage form to ensure uniform reference in task feature evaluation.

[0074] Taking five dimensions as an example, if the corresponding scores are [0.8, 0.7, 0.9, 0.6, 0.85] and the weights are [0.2, 0.2, 0.25, 0.15, 0.2], the total weighted score is: 0.8*0.2+0.7*0.2+0.9*0.25+0.6*0.15+0.85*0.2=0.79. Finally, the semantic adaptation degree is taken as the final "correlation score" between the task description vector and the semantic feature label, which is used to preliminarily determine the initial value of the gravity weight of the semantic feature in the semantic vector field, as a basic input to participate in the subsequent gravity weight optimization process.

[0075] To solve the problems of spatial distribution deviation and concept hierarchy imbalance of the initial gravity weight of the semantic feature, the embodiment designs a gravity weight fine-tuning method that fuses spatial adjacency and semantic hierarchy relationship. The spatial coordinates (such as latitude and longitude, grid center, or three-dimensional point cloud position) of all semantic feature labels in the survey area are obtained. Taking any two labels as a unit, the Euclidean distance is calculated and normalized to the range of [0, 1] to construct a spatial adjacency weight matrix. The spatial adjacency weight matrix is a two-dimensional symmetric matrix, each row and each column of the matrix corresponds to a semantic feature label, and the matrix element is the adjacency value between the semantic feature and other semantic features. The adjacency value can be calculated according to the Euclidean space distance, and the numerical range is normalized to 0-1, representing the strength of spatial proximity. For example, the distance between labels A and B is 20 meters, and the maximum distance is set to 200 meters, then the adjacency value is (1-20 / 200)=0.9, representing high spatial proximity. According to the pre-defined semantic type hierarchy structure (for example: building > commercial building > commercial plaza), a correction coefficient is set for the superior-inferior relationship between each pair of semantic labels. The "superior-inferior" setting coefficient is 1.0, the "same level" setting is 0.8, and the "irrelevant" setting is 0.3. Finally, a semantic hierarchy relationship correction matrix is constructed to reflect the aggregation relationship in the concept.

[0076] The spatial adjacency weight matrix and the semantic hierarchy relationship correction matrix are multiplied element by element and then normalized to obtain the gravity fine-tuning factor matrix. For example, the spatial adjacency value of A and B is 0.9, and the hierarchy coefficient is 1.0, so the fine-tuning factor is 0.9*1.0=0.9. If the spatial value of A and C is 0.5 and the hierarchy coefficient is 0.3, then the factor is 0.15, indicating weak correlation. Taking the initial gravity weight of the semantic feature itself as the benchmark, the clustering strength represented by the fine-tuning factor is superimposed and fused to update the final gravity weight of each semantic feature. The update formula can be set as: final weight = initial weight * (1+average fine-tuning factor). After adjustment by this method, features that are spatially concentrated and semantically close will have higher priority, effectively improving the semantic focusing ability.

[0077] In actual surveying tasks, due to the continuous change of the scene environment, the importance of different semantic features in the task background is not constant, and the gravity distribution between semantic features is often unbalanced, which leads to the deviation of the main semantic recognition from the target, thereby affecting the accurate generation of the surveying technical scheme. Therefore, the embodiment provides a dynamic semantic weight entropy calculation and regulation method based on a semantic vector field, which aims to identify the semantic weight confusion condition through the information entropy index, dynamically adjust the priority order structure of the semantic features by combining the gravity direction and distribution density factors, and realize the automatic focusing of the main control semantic.

[0078] Statistical gravity weight of all semantic features and calculation of semantic distribution entropy value, after the completion of the surveying scene construction, the gravity weight value of all semantic features in the current semantic vector field needs to be extracted, and the proportion occupied by it is calculated. The gravity weight proportion is used to reflect the importance of each semantic feature in the current task background. Then, through the weighted summation method, whether the distribution of these weight proportions is concentrated or dispersed is evaluated, so as to obtain the entropy value of the semantic weight distribution. The entropy value is used as an index to measure the stability and consistency of the semantic weight, to judge whether there is a problem of ambiguous main control semantic, scattered weight or multi-target competition in the current scene. If the entropy value exceeds the standard threshold set in advance, it means that the current semantic state is in a non-focused state, and the subsequent priority reconstruction operation needs to be performed.

[0079] Analysis of the gravity direction and spatial distribution density of semantic features, after judging that the semantic weight has a dispersion tendency, the gravity direction and spatial distribution of each semantic feature in the vector field need to be further analyzed. Specifically, based on the geometric difference relationship between the direction of the current surveying task target and the gravity direction of each semantic feature in the semantic vector field, the gravity direction offset of each semantic feature is calculated. The offset reflects the current task adaptation trend of the semantic feature. The calculation method is as follows: first, set the direction of the surveying task target in the semantic vector field as the reference direction, which can be derived from the weighted center position of the key task feature parameters in the task description vector. Then, for each semantic feature, its current gravity direction in the semantic vector field is obtained, which can be calibrated by the spatial vector between the semantic feature and the task target center. The direction angles of the reference direction and the semantic gravity direction are extracted respectively, and the included angle between them is calculated. The obtained included angle value is the gravity direction offset of the semantic feature. The smaller the offset, the more consistent it is with the task target in the current scene, and the higher the adaptation degree; the larger the offset, the more likely it is that the semantic feature deviates from the task core in direction. After completing the offset calculation, the number of other semantic labels adjacent to the semantic feature in the unit space is counted by combining the adjacency of the position of each semantic feature in the vector field space, so as to obtain its distribution density value. The higher the density value, the closer the relationship between the semantic feature and other key semantic features, and the higher the scene structure stability.

[0080] The reference direction can be derived from the weighted center position of the key task feature parameters in the task description vector. It refers to: in the implementation process, in order to obtain the reference direction of the surveying and mapping task target in the semantic vector field, it is necessary to first extract several dimensions judged as key task feature parameters based on the constructed task description vector, and give corresponding weight values according to the influence degree of each dimension in the generation of the surveying and mapping scheme. Each key task feature parameter has a clear vector position and direction attribute, which can be mapped to a point with direction attribute in the semantic vector space. Each key task feature parameter is regarded as a vector point, which is distributed to the corresponding coordinate position in the semantic vector space according to its feature dimension. For example, if the task target is the three-dimensional modeling work of complex building group, the key task feature parameters may include "high building structure complexity", "significant spatial occlusion relationship", "three-dimensional model as the achievement requirement", etc. These parameters have corresponding coordinate positions in the spatial complexity dimension, achievement accuracy dimension and other spaces.

[0081] According to the weight value corresponding to each feature parameter, the coordinate contribution in each dimension is calculated. Specifically, the coordinate value of each feature parameter in the semantic vector space is multiplied by the corresponding weight coefficient to obtain the weighted coordinate. For example, the coordinate of the key task feature parameter in the "spatial complexity" dimension is zero point eight, and the weight is zero point three, so the weighted contribution of the parameter in this dimension is zero point two four. The weighted contribution values of all key task feature parameters in each dimension are accumulated to obtain a comprehensive weighted center coordinate. The weighted center represents the direction of the current task target in the semantic vector space, that is, it can be regarded as a spatial vector direction from the coordinate origin to the weighted center position, which is used to express the main direction of the semantic attraction of the current task. This weighted center direction is taken as the reference direction of the task target, and is taken as the standard basis for subsequent calculation of the direction deviation of the semantic feature attraction. By comparing the actual attraction direction of the semantic feature with the reference direction, the consistency of each semantic feature with the task target direction can be accurately measured, so as to provide accurate geometric reference for dynamically adjusting its priority.

[0082] According to the direction deviation and spatial density, the weight enhancement and weight weakening strategies are implemented. After obtaining the direction deviation and spatial density information, corresponding regulation rules need to be developed to distinguish the semantic features that need to be enhanced or weakened. Weight enhancement strategy: when the direction deviation of a semantic feature is less than the set angle threshold (e.g. fifteen degrees), and its distribution density value is higher than the preset density threshold (e.g. more than three adjacent semantic labels per cubic unit of space), the system judges that the semantic feature has the advantages of strong direction consistency and high spatial concentration, and executes the weight enhancement strategy. The specific operation is: the current gravity weight value of the semantic feature is increased by 20%, for example, the initial gravity weight of a semantic feature is 0.18, and after enhancement, it is adjusted to 0.216, to improve its priority in subsequent master semantic sorting. Weight weakening strategy: when the direction deviation of a semantic feature exceeds the set angle threshold (e.g. thirty degrees), or its distribution density value is lower than the set density threshold (e.g. less than one adjacent semantic in each cubic unit of space), the system judges that it has serious direction deviation or spatial isolation, and executes the weight weakening strategy to decrease its gravity weight. The weakening method is to reduce the current gravity weight by 25%, for example, the original gravity weight of a semantic feature is 0.16, and after weakening, it is adjusted to 0.12, to reduce its interference to the master semantic path. Enhancement and weakening operations are executed simultaneously in all semantic features, and after execution, all gravity weights are re-normalized to ensure constant total weight.

[0083] Based on the adjusted gravity weight distribution, the master semantic sorting structure is generated. After weight adjustment, the ranking of each semantic feature in the gravity value changes. The system re-ranks the priority of all current semantic features to form a new master semantic sorting structure. The top-ranked semantic features in this sorting structure represent the most critical semantic elements in the current surveying task, and subsequent semantic path construction and scheme reasoning operations will be carried out around this sorting structure. Finally, by introducing the information entropy-based regulation mechanism and semantic gravity reconstruction strategy, the reasoning instability problem caused by semantic competition, feature misfocus, scene conflict and other factors is effectively solved, the accurate identification of semantic master path and efficient alignment of task target are realized, and a stable and controllable semantic basis is provided for the intelligent generation of surveying technology scheme.

[0084] In the present application, in order to realize the automatic construction of technical solutions matched with the current surveying task and scene, after completing the main semantic sorting structure update, the technical path construction and scheme matching are carried out based on the semantic priority logic. The specific implementation is as follows: according to the dynamically adjusted main semantic sorting structure, the semantic features with high semantic priority are selected in sequence to construct the main control semantic path of the current task. The main control semantic path refers to a structured semantic link composed of multiple task key semantic elements and showing a certain logical order, which is used to describe the core semantic requirements that need to be focused on in the current surveying task and the logical dependency relationship therebetween. For example, in a municipal pipeline modeling task, the semantic features with high sorting priority are "underground traffic density", "geological soil structure" and "existing pipeline complexity", and then the main control semantic path will connect the above semantic elements in this order to form a logical path of "underground traffic density→geological soil structure→existing pipeline complexity", which is used to constrain the core focus of the scheme generation.

[0085] The constructed main control semantic path is taken as an input condition to call the pre-constructed technical knowledge base and historical surveying scheme case set, and the semantic structuring matching operation is performed. In the matching process, the semantic feature chain in the main control semantic path is taken as the core retrieval vector, and the existing path structure similarity calculation, semantic label matching degree calculation and logical consistency check in the prior art are used to perform priority screening on the multiple technical scheme structure templates stored in the existing scheme set. Finally, a group of candidate technical scheme structure templates closest to the main control semantic path in terms of semantic structure and technical content are screened out.

[0086] After obtaining the candidate technical scheme structure template, the parameter adjustment and content filling operation are performed on the candidate template in combination with the key task feature parameters contained in the current task description vector. Specifically, the feature information such as the operation type, surveying accuracy level, operation area environment attribute and delivery form requirement in the task description vector related to the candidate template is extracted to replace or optimize the general parameter fields in the template. For example, if the default surveying accuracy level in the candidate template is fifty centimeters, and the current task requires an accuracy of twenty centimeters, then the parameters related to sensor selection, sampling frequency setting, point cloud density standard and the like in the template need to be adjusted synchronously to ensure that the template parameters can meet the actual requirements.

[0087] Finally, after completing the template parameter adjustment, an automatic synthesis is formed to match the current surveying and mapping task. The scheme will include surveying and mapping targets, task element analysis, surveying and mapping method selection, data acquisition and processing flow, and achievement output specification. The generated surveying and mapping technical scheme is the final intelligent output result, which can be directly submitted for on-site operation execution, or can be used for manual review, checking or supplement by surveying and mapping personnel. The method significantly improves the generation efficiency and scene adaptability of the technical scheme through semantic-driven path construction and technical knowledge reuse, and is particularly suitable for complex and variable dynamic surveying and mapping task environments.

[0088] In the prior art, the key methods for technical scheme structure template matching and screening mainly include the following three kinds: path structure similarity calculation, semantic label matching degree calculation and logic consistency checking. These technologies are derived from knowledge graph construction, intelligent question and answer system, scene reasoning modeling and other fields, and have been more maturely applied in complex task decision or scheme reuse systems, which are briefly described as follows:

[0089] Path structure similarity calculation technology is derived from graph structure matching and subgraph isomorphism determination theory, and is commonly used in knowledge graph path similarity identification and semantic reasoning path comparison. In such methods, the path is abstracted as a set of ordered semantic nodes and their connection relationships. Common similarity evaluation methods include: path edit distance: the minimum number of transformation steps, including inserting, deleting and replacing nodes, so that two paths are converted into each other; longest common sub-path matching: calculating the longest sub-path with the same structure in two paths to evaluate the preservation degree of the core structure; structure alignment scoring model: multi-factor weighted scoring based on path length, branch node number, hierarchical depth, etc. In the generation of surveying and mapping technical scheme, the similarity between the main control semantic path and the historical template path can be quantified as a numerical value by the above methods, which can be used to guide the priority sorting. Semantic label matching degree calculation technology is commonly used in natural language processing and semantic search systems to judge the closeness of two texts or label sets in semantic concepts. Logic consistency checking is mainly used to check whether the cause-and-effect, constraint or parallel relationship between the semantic elements in the scheme structure is consistent with the current task.

[0090] The above algorithms or formulas are de-dimensioned to obtain numerical values, and the results closest to the actual situation are obtained by software simulation with a large amount of data. The preset parameters are set by a person skilled in the art according to the actual situation.

[0091] It should be understood that the size of the sequence number of each process in various embodiments of the present application does not mean the order of execution, and 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 the present application.

[0092] Those skilled in the art can clearly understand that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be realized in electronic hardware or a combination of computer software and electronic hardware. Whether the functions are executed in hardware or software depends on specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0093] Those skilled in the art can clearly understand that, for the convenience and brevity of the description, the specific working processes of the above-described devices and units can refer to the corresponding processes in the foregoing method embodiments, which will not be described here.

[0094] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited to this. Any person skilled in the art can easily think of changes or replacements within the technical scope disclosed by the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A method for intelligent generation of a mapping technical solution oriented to scene semantic understanding, characterized in that, The method comprises the following steps: Task target analysis step: according to the target requirements, operation type and precision index of the to-be-executed surveying and mapping task, a task description vector is constructed, and task characteristic parameters influencing the generation of a surveying and mapping technical scheme are extracted; Scene semantic feature extraction step: dynamic environment perception data of a surveying and mapping area are acquired, semantic feature labels in the scene are identified, and a structured representation containing multiple semantic features is generated; Task-driven semantic vector field construction step: based on the task description vector and the semantic feature labels, a task-driven semantic vector field is constructed, and gravity weights are assigned to each semantic feature, which are used to represent the correlation degree between the semantic features and the task target; Dynamic semantic weight entropy calculation and regulation step: the entropy values of the gravity weights of each semantic feature in the semantic vector field are calculated, a dynamic semantic weight entropy index is obtained, when the semantic weight entropy index exceeds a preset standard threshold, the priority weights of the semantic features are dynamically adjusted according to the gravity direction and distribution trend in the semantic vector field, and the main semantic ranking structure is updated; Surveying and mapping technical scheme generation step: according to the updated main semantic ranking structure, a main control semantic path is determined, a technical knowledge base or a historical technical scheme case is called, and a surveying and mapping technical scheme matched with the current task and scene is generated; The task-driven semantic vector field construction refers to: Based on the feature dimensions contained in the task description vector, the dimension structure and coordinate system reference of the semantic vector space are defined; the semantic feature labels are mapped according to their attribute types and the task feature dimensions, the initial position coordinates of each semantic feature in the semantic vector space are determined; according to the correlation relationship between the task features and the semantic features, the semantic labels are distributed in the vector field to form a clustering area, and finally the construction of the task-driven semantic vector field is completed; Assigning gravity weights to each semantic feature refers to: The correlation scores between each feature dimension in the task description vector and the semantic feature labels are calculated, and then the correlation scores are normalized as the initial gravity weight values of the semantic features in the task-driven semantic vector field; Based on the spatial adjacency relationship and type hierarchy relationship between the semantic features, the initial gravity weight values are optimized and adjusted to form the final gravity weight distribution, and the optimized gravity weight is bound to each semantic feature label; Optimizing and adjusting the initial gravity weight values refers to: Based on the actual spatial coordinate information of the semantic feature labels in the surveying and mapping area, the spatial adjacency degree between any two semantic features is calculated by using Euclidean distance or spatial overlap ratio, and a spatial adjacency weight matrix is constructed; According to the preset semantic type hierarchy structure, different hierarchical correction coefficients are set for the superordinate type, subordinate type and same level type relationships, and a hierarchical relationship correction matrix is constructed, which is used to measure the concept correlation degree between different type semantic features; The spatial adjacency weight matrix and the hierarchical relationship correction matrix are multiplied by using a fusion weighting method, and then normalized, which is used as a fine tuning factor to weight and adjust the initial gravity weight, and finally the adjusted weight value is updated as the final gravity weight of each semantic feature; In the dynamic semantic weight entropy calculation and regulation, the gravity weight values of all semantic features in the semantic vector field are counted, and the entropy value of the semantic weight distribution is calculated by using the information entropy formula. The information entropy calculation is based on the weighted sum of the product of the proportion of each semantic feature weight and its logarithm, and is used to measure the discrete degree of the overall weight distribution; The calculated entropy value is compared with the preset standard threshold value, and when the entropy value is higher than the standard threshold value, the priority weight adjustment is performed; The gravity direction vector of each semantic feature is obtained, and the included angle between the gravity direction vector and the direction of the current main task gravity center is calculated as the gravity direction deviation. The number of adjacent semantics of the semantic feature in the vector field coordinate space is counted, and the density value of the semantic feature in the unit space is calculated; According to the regulation rules of enhancement and attenuation, when the gravity direction deviation is less than the preset angle threshold value and the distribution density is higher than the preset density threshold value, the enhancement strategy is executed; when the gravity direction deviation is greater than the angle threshold value or the distribution density is lower than the density threshold value, the weight weakening strategy is executed, and the gravity weight distribution is dynamically updated to form a stable main control semantic sorting structure.

2. The method of claim 1, wherein, The construction of the task description vector specifically includes the following steps: Collecting original task input information related to the surveying and mapping task and converting it into a standard structured data format to construct a unified task information input model; Based on the preset task information input model, a task feature dimension system is established, each item of input information is mapped to a predefined feature dimension, and an initial task description vector is formed through dimension mapping; Performing feature weight evaluation and parameter simplification processing on the task description vector, extracting key task feature parameters affecting the generation of surveying and mapping technical scheme according to the statistical weight value of each feature dimension in the historical surveying and mapping task; Storing the key task feature parameters in the form of structured vector as the basic input data for subsequent process of surveying and mapping task target modeling and related reasoning processing.

3. The method of claim 1, wherein, The operation of performing feature weight evaluation and parameter simplification processing on the task description vector includes the following steps: According to the preset task feature weight reference system, the initial weight value of each task feature dimension affecting the generation of surveying and mapping scheme is determined. The initial weight value is based on the statistical analysis of historical surveying and mapping task result data and the setting of domain knowledge rules; Each dimension feature in the initial weight value and the current task description vector is compared item by item, and the effective weight index of each feature dimension is calculated by using the weighted sorting method. The effective weight index is used to quantify the contribution degree of the feature dimension to the generation of surveying and mapping scheme under the current task background; Based on the effective weight index and the preset weight threshold value, the secondary feature dimensions with weight index lower than the weight threshold value are removed, and only the dimension features with effective weight index higher than the weight threshold value are reserved as key feature parameters to form the parameter simplification result of the task description vector.

4. The method of claim 1, wherein, The scene semantic feature extraction includes the following steps: The dynamic environment perception data of the surveying area is acquired, feature extraction and fusion analysis are performed according to a preset strategy, the significant feature shapes, spatial structure relationships and dynamic change attributes in the scene are identified, semantic label assignment is performed in combination with a preset task context semantic model, and a plurality of scene element sets with semantic attributes are obtained; The scene element sets are structured and organized, a multi-dimensional structured representation matrix is constructed, and the representation matrix is used to describe the feature distribution of the current surveying area in the spatial, attribute and semantic dimensions.

5. The method of claim 4, wherein, The correlation score calculation process includes the following steps: Each feature dimension in the task description vector is matched with the corresponding attribute dimension in the semantic feature label to form a set of feature correspondence relationships; for each set of feature correspondence relationships, a feature similarity value is calculated, the similarity value is calculated by using a vector angle cosine value, a feature value difference ratio or a rule table-based matching score method, and is used to measure the adaptation degree of the task demand and the semantic attribute in a certain dimension; the similarity values of each feature dimension are weighted and summarized according to a preset weight to form an overall semantic adaptation score, which is used as the correlation score between the task description vector and the semantic feature label.

6. The method of claim 5, wherein, When the surveying technical scheme is generated, the semantic features with high ranking in the updated main semantic ranking structure are sequentially selected, and a main control semantic path is constructed; The main control semantic path is used as a retrieval input, the technical knowledge base and the historical technical scheme case base are called, a matching retrieval operation is performed, and a candidate technical scheme structure template with the highest similarity to the current path is screened; According to the current task description vector and the semantic elements in the main control semantic path, the candidate scheme structure template is adjusted in parameters and filled in content to generate a surveying technical scheme matched with the current task and scene, and finally the generated surveying technical scheme is used as an intelligent output result.

Images