A system for dynamically adjusting a cadastral survey route

Abstract

This invention belongs to the field of cadastral surveying technology and is a system for dynamically adjusting cadastral surveying routes. The system includes a cadastral database, a field interaction module, a feature change confirmation module, and a route planning and optimization module. The field interaction module is used to collect multimodal data on feature changes in real time. The feature change confirmation module is used to perform rapid AI verification on the collected multimodal change data and combine it with historical time-series tracing to achieve accurate confirmation of feature changes. The route planning and optimization module is used to calculate the surveying priority based on the confirmed feature change data, and combine an improved obstacle avoidance algorithm with real-time fine-tuning of the cadastral spatial model to dynamically and intelligently adjust the surveying route.

Description

Technical Field

[0001] This invention belongs to the field of cadastral surveying technology, and in particular to a system for dynamically adjusting cadastral surveying routes. Background Technology

[0002] Cadastral surveying is a core foundational technology for land ownership management and urban planning and construction. Its core requirement is to obtain the spatial location, attributes, and change information of features in the area to be surveyed through accurate mapping. Currently, cadastral surveying route planning largely relies on static formulation based on historical cadastral data. That is, a fixed route is generated based on historical feature information in the database before the survey, and the operation is only carried out according to the preset route during the survey. With the acceleration of urban renewal, such as old area renovation, new resettlement housing construction, and road widening, the features in the area to be surveyed change frequently, and the traditional static route planning model is no longer suitable for dynamic on-site scenarios.

[0003] In existing technologies, when surveyors discover changes in ground features, such as the demolition of houses or the construction of new buildings, the surveying process needs to be manually interrupted. The change information is recorded on paper or manually entered, and then the historical data is retrieved again through the back-end system to plan the route. The entire process relies on manual intervention, which not only prolongs the surveying cycle but is also prone to distortion of change information transmission due to errors in manual recording. At the same time, although some systems have attempted to introduce simple change feedback functions, they lack a mechanism to verify the authenticity of change information, which can easily misjudge temporary obstructions, such as construction fences, as permanent changes in ground features, leading to inaccurate route adjustments.

[0004] Therefore, there is an urgent need for a system that dynamically adjusts cadastral survey routes to solve the above problems. Summary of the Invention

[0005] This invention proposes a system for dynamically adjusting cadastral survey routes.

[0006] A system for dynamically adjusting cadastral survey routes includes a cadastral database, a field interaction module, a feature change confirmation module, and a route planning and optimization module. The field interaction module is used to collect multimodal data on feature changes in real time. The feature change confirmation module performs rapid AI verification on the collected multimodal change data and combines historical time-series tracing to achieve accurate confirmation of feature changes. The route planning and optimization module calculates surveying priorities based on the confirmed feature change data and performs dynamic intelligent adjustment of the surveying route by combining an improved obstacle avoidance algorithm with a large cadastral spatial model for real-time fine-tuning.

[0007] Preferably, the cadastral database is used to store historical land feature data, multimodal change data collected by the field interaction module, valid change data confirmed by the land feature change confirmation module, and surveying route data generated by the route planning and optimization module. It also includes real-time data updates and historical time series retrieval, and transmits the data to the land feature change confirmation module and the route planning and optimization module.

[0008] Preferably, the multimodal change data collected by the on-site interaction module includes image data and voice semantic data, including real-time photo annotation and voice semantic recognition. The real-time photo annotation function collects images of the areas of ground feature change and marks them with bounding boxes, and the voice semantic recognition function extracts keywords related to ground feature change. The image data and semantic data are then fused to form multimodal change data.

[0009] Preferably, the AI-based rapid verification of the land feature change confirmation module includes feature matching verification and semantic consistency verification. It extracts image features from multimodal change data, matches them with historical image features in the cadastral database, verifies the consistency between the voice description and the image changes through a semantic matching verification algorithm, and outputs the result of determining the authenticity of the land feature change.

[0010] Preferably, the land feature change confirmation module includes feature matching verification, which extracts land feature feature points from multimodal change data and compares them with historical land feature database. When the similarity reaches a preset threshold, it is determined to be a valid change feature; otherwise, it is determined to be temporary occlusion or recording error.

[0011] Preferably, the feature change confirmation module includes a historical time-series tracing function, which retrieves historical feature data and compares it with the currently collected multimodal change data.

[0012] Preferably, the route planning and optimization module also includes a change adaptation planning module, which determines the mapping priority of the changed area based on the real ground feature change data output by the ground feature change confirmation module, uses an obstacle avoidance algorithm to bypass obstacles formed by demolished or newly built ground features, and generates an initial optimized route that adapts to the ground feature changes.

[0013] Preferably, the route planning and optimization module also includes a large model real-time response module, which uses real-world feature change data as incremental training data to fine-tune the cadastral spatial large model in real time, shortens the model response time through a fast inference algorithm, and achieves instant output of the optimized route; Incremental training includes a parameter freezing strategy, which updates only the parameters of the top fully connected layer of the model, while fast inference employs model pruning optimization techniques.

[0014] Preferably, the on-site interaction module transmits the collected multimodal change data to the feature change confirmation module, the feature change confirmation module transmits the confirmed feature change data to the route planning and optimization module, and the route planning and optimization module feeds back the adjusted survey route to the on-site interaction module in real time.

[0015] Preferably, the route planning and optimization module also includes an intelligent recommendation module for alternative routes, which generates several different optimized route schemes based on constraints such as surveying efficiency and measurement accuracy, and uses a multi-objective decision-making algorithm to comprehensively evaluate each scheme and automatically select the optimal route; In this multi-objective decision-making method, a weighted summation method is used. Efficiency and accuracy weights are set according to the requirements of the surveying and mapping task. The comprehensive score of the schemes is calculated, and the scheme with the highest score is the optimal route. The present invention has the following beneficial effects: 1. This invention utilizes the multimodal acquisition function of the on-site interaction module, including real-time photo annotation and voice semantic recognition, to capture details of ground feature changes in real time. Through AI verification and historical time-series tracing of the ground feature change confirmation module, interference factors such as temporary obstruction and recording errors are eliminated, ensuring the authenticity and completeness of ground feature change information. Compared with the traditional manual recording and background verification mode, the acquisition and confirmation of change information can be completed without interrupting the surveying process, greatly improving the on-site response speed and change confirmation accuracy.

[0016] 2. This invention, through the dynamic adjustment capability of the route planning and optimization module, can automatically calculate surveying priorities, avoid obstacle areas, and generate optimal routes based on real-world changes in land features. Simultaneously, the real-time fine-tuning and rapid inference technology of the large cadastral spatial model ensures that route adjustment response time is controlled within seconds, avoiding the time loss caused by traditional manual replanning. Furthermore, the multi-module interactive linkage mechanism, including change confirmation data-driven route adjustments and real-time feedback of route adjustment results to the site, can adapt to the complex and ever-changing surveying scenarios in urban renewal, reducing invalid surveying mileage and improving overall surveying efficiency. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the structure of a system for dynamically adjusting cadastral survey routes according to the present invention; Figure 2 This is a schematic diagram of an embodiment of a system for dynamically adjusting cadastral survey routes according to the present invention; Figure 3 This is a schematic diagram of an embodiment of a system for dynamically adjusting cadastral survey routes according to the present invention. Detailed Implementation

[0018] To enable those skilled in the art to better understand the technical solutions of this invention, the technical solutions in the embodiments of this invention will be clearly described below in conjunction with the examples.

[0019] Example 1 like Figure 1-3As shown, this invention proposes a system for dynamically adjusting cadastral survey routes, specifically including the following modules: a cadastral database, used to store historical cadastral survey data and land feature change-related data of the area to be surveyed; a field interaction module, used to collect land feature change information during the survey process and provide feedback for route optimization; a land feature change confirmation module, used to verify the authenticity of land feature change information; and a route planning and optimization module, used to generate and optimize the survey route based on real land feature change data.

[0020] The cadastral database includes a real-time change data update module and a fast retrieval and adaptation module. The real-time change data update module, based on an incremental storage mechanism, stores only the difference data corresponding to changes in land features, recording the time-series trajectory of these changes to achieve dynamic database updates. Incremental storage employs a data sharding strategy, dividing the area under test into shards according to a spatial grid. Each shard corresponds to a fixed range of latitude and longitude. Change data is associated with historical data in the corresponding shard using a unique land feature identifier. This unique identifier consists of an administrative division code, a land feature type code, and a sequential number, ensuring data traceability. The data sharding uses a quadtree grid partitioning algorithm, recursively dividing the area under test into several square grids with sides of 100 meters. Each grid is assigned a unique shard identifier, and a mapping relationship is established between the shard identifier and the land feature identifier, facilitating rapid location of data storage.

[0021] The rapid retrieval and adaptation module constructs a spatial index of changed areas, enabling rapid location of historical data corresponding to changed features and establishing a mapping between multimodal data and cadastral data. The spatial index employs an R-tree structure, using the bounding rectangle of the feature as the index key. The latitude and longitude coordinates of the feature are stored in the nodes of the R-tree, and rapid spatial queries are achieved through hierarchical traversal of the R-tree. The mapping between multimodal data and cadastral data is achieved by establishing a related data table. This table includes fields for feature identification, multimodal data type, data storage path, data acquisition time, and data verification status. The multimodal data type field distinguishes between image data and speech / semantic data, while the data verification status field indicates three states: unverified, verified, and verified failed. The feature identification links the multimodal data with historical attribute data and spatial data in the cadastral database, ensuring the accuracy and efficiency of data retrieval.

[0022] The on-site interaction module includes a multimodal change feedback module, a change visualization feedback module, and an interactive command quick trigger module. The multimodal change feedback module acquires images of areas of ground feature change through real-time image capture and annotation, and extracts relevant keywords based on ground feature changes through speech semantic recognition. The image data and semantic data are then fused to form multimodal change data. The real-time image capture and annotation function allows surveyors to capture images of areas of ground feature change using the camera of a mobile surveying device. The resolution of the captured images must be no less than 1920×1080 pixels. The bounding box marking uses a rectangular bounding box algorithm based on OpenCV. Surveyors draw rectangles on the image through touch operation, and the system automatically records the pixel coordinates of the rectangles and converts them into corresponding latitude and longitude coordinate ranges to form the spatial positioning information of the change area.

[0023] Speech semantic recognition employs a Transformer-based semantic understanding model, which consists of an encoder and a decoder. The encoder uses a 6-layer Transformer encoder structure, with each layer containing 8 attention heads, and the hidden layer dimension is set to 512. The decoder uses a 4-layer Transformer decoder structure, achieving semantic association with the encoder through a cross-attention mechanism. The input speech signal is first preprocessed, including pre-emphasis, framing, and windowing operations. Pre-emphasis uses a first-order high-pass filter. ,in The framing process uses a frame length of 20ms and a frame shift of 10ms, with Hanning windowing applied. Mel-frequency feature extraction is then performed, converting the preprocessed speech signal into Mel-frequency cepstral coefficients (MFCCs). 13-dimensional MFCC features and first- and second-order difference features are extracted to form a 39-dimensional feature vector. Finally, through the encoding and decoding process of the Transformer model, the speech features are converted into structured change description information, which includes fields describing the type, location, and extent of land cover changes.

[0024] The change visualization feedback module overlays real-time collected ground feature change data with historical data in the cadastral database to generate a visual comparison interface. It then classifies the differences into levels according to preset rules and provides pop-up prompts. The image overlay processing employs a pixel-level fusion algorithm. First, the real-time collected images are registered with historical images using a feature-point-based registration method. SIFT feature points are extracted from the two images, and FLANN matching algorithm is used for feature point matching. Based on the matched feature points, the homography matrix is ​​calculated. The formula is as follows:

[0025] in, The homography matrix describes the projection transformation relationship between two images. The matrix elements are obtained by solving for the coordinates of four pairs of matched feature points.

[0026] After registration, a weighted fusion algorithm is used to overlay the images. The fusion formula is as follows:

[0027] in, These are the pixel values ​​after overlay. These are the pixel values ​​of the real-time image. These are the pixel values ​​of the historical image. These are the weighting coefficients.

[0028] This algorithm visually presents changes in land features. The classification of differences is based on the area percentage of the changed area and the type of change. A change area percentage greater than 30% or a change type of new construction or demolition is classified as Level 1 (significant change); a change area percentage between 10% and 30% or a change type of reconstruction is classified as Level 2 (moderate change); and a change area percentage less than 10% or a change type of local modification is classified as Level 3 (minor change). Different levels of difference correspond to different colored pop-up notifications: Level 1 differences are indicated by a red pop-up, Level 2 by a yellow pop-up, and Level 3 by a blue pop-up.

[0029] The interactive command quick trigger module supports surveyors to define custom voice commands. These preset commands trigger data synchronization and route adjustment requests, enabling real-time transmission of multimodal change data to subsequent modules. The custom command setting process includes three steps: command input, command verification, and command storage. Surveyors input commands such as "adjust route" and "synchronize data" via voice. The system performs voice recognition and semantic analysis on the input commands, verifying their uniqueness and validity. Verified commands are stored in a preset command library, which uses a key-value pair storage format. The key is the command keyword, and the value is the corresponding trigger operation.

[0030] Custom commands employ a command word matching algorithm, which recognizes commands by calculating the similarity between the input command and a preset command library. The similarity calculation includes:

[0031] in, For instruction similarity, The first in the preset instructions The weight of each keyword, For the first input command The semantic similarity between each keyword and the preset keywords. For the number of keywords, This refers to the keyword overlap rate of the instruction. , To match the number of keywords, The mean semantic similarity. , This is the regularization coefficient, with a value of 0.1.

[0032] After a successful command match, the system triggers the corresponding operation. If a data synchronization operation is triggered, the multimodal change data is transmitted to the cadastral database via the HTTP / 2 protocol. During the transmission process, the data is encrypted using the AES-256 encryption algorithm to ensure the security of data transmission. If a route adjustment request is triggered, the system sends a change data transmission request to the feature change confirmation module to initiate the feature change verification process.

[0033] The land feature change confirmation module includes an AI rapid verification module, a historical time-series tracing module, and a manual review triggering module. The AI ​​rapid verification module uses the SIFT algorithm to extract image features from multimodal change data, matches them with historical image features in the cadastral database, verifies the consistency between the voice description and the image changes through a semantic matching verification algorithm, and outputs the land feature change authenticity judgment result.

[0034] The SIFT algorithm's feature extraction process includes four steps: keypoint detection, keypoint localization, orientation assignment, and feature vector generation. First, a Gaussian difference pyramid is constructed by convolving the image with Gaussian kernels of different scales to obtain multi-scale images. Images at adjacent scales are subtracted to obtain a Gaussian difference image, where local extrema are detected as keypoints. Next, keypoints are precisely localized by fitting a three-dimensional quadratic function to remove low-contrast keypoints and edge response points. Then, the gradient orientation histogram of pixels in the keypoint's neighborhood is calculated, and the direction corresponding to the peak value in the histogram is taken as the principal direction of the keypoint, achieving rotation invariance of the feature. Finally, a 16×16 neighborhood window is extracted centered on the keypoint and divided into 4×4 sub-windows. Gradient histograms in eight directions are calculated for each sub-window, forming a 128-dimensional feature vector.

[0035] Feature matching uses weighted normalized Euclidean distance to calculate feature vector similarity, including:

[0036] in, Weighted normalized Euclidean distance is applied to the feature vectors. For the real-time image feature vector, the first dimensional components, For the historical image feature vector, the first dimensional components, The dimension of the feature vector is 128. For the first Importance weights of dimensional features , .

[0037] The semantic matching verification algorithm achieves consistency by comparing the structured description information output by speech semantic recognition with the image feature matching results. The type and location of land feature changes in the structured description information must be consistent with the changes reflected in the image feature matching. For example, when the speech description is "house demolition," the image feature matching should show that the feature points at the original house location have disappeared or changed significantly. If the speech description and the image feature matching results are inconsistent, the degree of inconsistency is calculated. , ,in, For semantic matching inconsistency, "Number of Consistent Fields" refers to the number of fields in the structured description information that match the image feature matching results, and "Total Number of Fields" refers to the total number of fields contained in the structured description information; when When the value is greater than 0.3, the semantic matching is considered to have failed. A semantic match is considered successful when the score is less than or equal to 0.3. Only when both image feature matching and semantic matching are successful does the AI ​​fast verification module output a result indicating that the ground feature change is true; otherwise, it outputs a result indicating that the ground feature change is questionable, and calculates the verification confidence level. The confidence level is calculated using the following formula:

[0038] in, To enable AI to quickly verify confidence levels, the "image matching pass rate" is the proportion of successfully matched feature vector pairs to the total number of feature vector pairs. This represents the semantic matching inconsistency level.

[0039] The historical time-series tracing module calls upon multiple versions of historical cadastral data from the cadastral database to perform cross-time period comparative analysis on changed land features. Based on the temporal change patterns, it establishes a trend model to determine the rationality of land feature changes and outputs the trend analysis results. The calling of multiple versions of historical data is based on land feature identifiers and time ranges. Surveyors can specify the time span to be traced; the system defaults to tracing historical data for the past 5 years, selecting cadastral data from one time point each year, with June 30th of each year forming a multi-version historical dataset. The cross-time period comparative analysis combines image differencing and attribute comparison. Image differencing calculates the differences between historical images and real-time images from different years, while attribute comparison analyzes changes in land feature ownership, area, and usage information.

[0040] The trend model uses the ARIMA model from time series analysis, and the model expression is:

[0041] in, For the first Indicators of the degree of change in land features over the years These are the autoregressive coefficients. The moving average coefficient is... It is a white noise sequence. Let the order be the autoregressive order. This represents the order of the moving average.

[0042] The ARIMA model was fitted using historical data from multiple versions to determine the model parameters. Difference order , Based on the fitted model, the system predicts the natural change trend of land features. If the real-time detected changes in land features are consistent with the predicted trend, the changes in land features are deemed reasonable; otherwise, the changes in land features are deemed abnormal, and the change trend analysis results are output.

[0043] When the confidence level of the AI ​​rapid verification result falls below a preset confidence threshold of 0.7, the manual review trigger module automatically triggers a critical change warning, pushes a review request to relevant staff, receives the manual review result, and quickly transmits it back to subsequent modules to complete the final confirmation of the feature change information. The critical change warning is implemented through two methods: mobile terminal APP push and SMS notification. The push information includes feature identifiers, coordinates of the changed area, multimodal change data links, and a summary of the AI ​​verification result. Staff can click the link to view detailed change data and conduct reviews using manual visual interpretation and attribute verification. The manual review result includes a review conclusion and a review explanation. The review conclusion includes "true," "false," and "requires further verification." The review result is transmitted back to the system through an encrypted channel. The system associates and stores the review result with the corresponding feature change data and updates the data verification status field.

[0044] The route planning and optimization module includes a change adaptation planning module, a large model real-time response module, and an alternative route intelligent recommendation module. The change adaptation planning module determines the mapping priority of the changed area based on the real ground feature change data output by the ground feature change confirmation module, and uses an obstacle avoidance algorithm to bypass obstacles formed by demolished or newly built ground features, thereby generating an initial optimized route that adapts to the ground feature changes.

[0045] Surveying priority calculation includes:

[0046] in, For surveying priority, The weights are affected by the changes. , For data importance weights, , This is the ownership adjustment factor. , To determine the degree of impact of changes in terrain features on the original route, To emphasize the importance of cadastral data on changed land features, This is a correction factor for the magnitude of change. .

[0047] The obstacle avoidance algorithm employs an improved A algorithm, designating obstacle areas formed by demolished or newly constructed features as impassable zones and avoiding traversing these zones during path search. The heuristic function of the A algorithm is designed as follows:

[0048] in, For nodes Heuristic distance to the target node As a heuristic weight, a value of 1.2 is set. For nodes coordinates The coordinates of the target node.

[0049] An improved A algorithm is used to search for the optimal path from the current mapping location to the target mapping location. This path must avoid obstacle areas and pass through change areas with high mapping priority. An initial optimized route is generated and stored in the form of a latitude and longitude coordinate sequence, which includes information such as the coordinates of each inflection point on the route, the distance between adjacent inflection points, and the estimated mapping time.

[0050] The large-scale model real-time response module uses real-world feature change data as incremental training data to fine-tune the cadastral spatial large-scale model in real time. It shortens the model response time through a fast inference algorithm, enabling immediate output of optimized routes. The cadastral spatial large-scale model is built on a Transformer architecture, containing a 12-layer encoder and a 6-layer decoder, with a hidden layer dimension of 1024 and 16 attention heads. The model's pre-training data includes massive amounts of cadastral spatial data, historical measurement route data, and feature change data.

[0051] The incremental training process employs mini-batch gradient descent with a batch size of 32. A cosine annealing scheduling strategy is used for the learning rate, with an initial learning rate of 1e-5 and a minimum learning rate of 1e-7. The training iterations consist of 10 epochs. A hybrid loss function is used for incremental training. ,in, For mixed loss values, The mean squared error loss is used to regress the route coordinates. Cross-entropy loss is used to classify route types. Weights for mean squared error loss. The weights are determined by cross-entropy loss. During training, only the parameters of the top three fully connected layers are updated, while the parameters of the bottom feature extraction layers are frozen, reducing computation and improving training speed.

[0052] Fast inference employs an optimization technique combining model pruning and quantization. Model pruning utilizes a structured pruning algorithm to prune 40% of the fully connected layers, removing redundant neurons and connections. Model quantization uses INT8 quantization, converting the model's weights and activation values ​​from 32-bit floating-point numbers to 8-bit integers, reducing model storage capacity and computational complexity. Through these optimization techniques, the model's inference speed is improved by over 60%, ensuring an optimized path is output within one second.

[0053] The intelligent route recommendation module generates 2-3 different optimized route schemes based on surveying efficiency and measurement accuracy constraints. A multi-objective decision-making algorithm is used to comprehensively evaluate each scheme and automatically select the optimal route. Surveying efficiency is measured by two indicators: total route length and estimated surveying time. Measurement accuracy is measured by two indicators: the completeness of the area covered by the route and the density of survey points. The weights of each indicator are determined using the analytic hierarchy process (AHP), constructing a judgment matrix and calculating a weight vector. Surveying efficiency has a weight of 0.4, and measurement accuracy has a weight of 0.6. Specifically, the weights for total route length (0.15), estimated surveying time (0.25), coverage completeness (0.35), and survey point density (0.25) are all calculated.

[0054] Multi-objective decision-making employs a weighted summation method to calculate the comprehensive score of each option. ,include:

[0055] in, To determine the overall score for the proposal, For the first The weight of each indicator, For the first The standardized scores of each indicator were calculated using min-max standardization. , The actual values ​​of the indicators in the plan. The minimum value of the index among all options. This represents the maximum value of the index across all options.

[0056] The route with the highest overall score is the optimal route. If the difference in overall scores between two routes is less than 0.05, the scores of the measurement accuracy index are further compared, and the route with the higher measurement accuracy score is the optimal route.

[0057] The on-site interaction module, the feature change confirmation module, and the route planning and optimization module form an interactive linkage mechanism. Multimodal change data collected by the on-site interaction module directly triggers the AI-based rapid verification process of the feature change confirmation module through a data transmission interface. This data transmission interface uses a RESTful API design to support synchronous data transmission. The real feature change data output by the feature change confirmation module drives the route generation process of the route planning and optimization module in real time through a message queue. The message queue uses Kafka to ensure high throughput and low latency in data transmission.

[0058] The optimized route output by the route planning and optimization module is presented to surveyors in a multimodal format (voice, text, and map annotations) through the change visualization feedback module of the on-site interaction module. Voice feedback utilizes TTS (Text-to-Speech) technology to convert the route description text into a speech signal. Text feedback displays the coordinates of key inflection points, estimated surveying time, and key surveying areas. Map annotations on the electronic map show the optimized route as a solid red line and marked the changed areas in yellow. When abnormal changes occur, the abnormal change determination is based on the change amplitude output by the feature change confirmation module exceeding a preset threshold (>80%). All modules collaboratively trigger an early warning through interactive warning signals. The warning signal includes the abnormal change feature identifier, the coordinates of the changed area, and the warning level. The on-site interaction module displays a red warning pop-up and plays the warning audio. The feature change confirmation module marks the abnormal change data as key verification data. The route planning and optimization module pauses route generation and awaits manual instructions, achieving seamless integration of change feedback, authenticity confirmation, and route adjustment.

[0059] The feature change confirmation module, cadastral database, and route planning and optimization module form an interactive linkage mechanism. The verified change data output by the feature change confirmation module automatically triggers the cadastral database update process through the database update interface. The database update interface uses a JDBC interface, supports transaction processing, and ensures the atomicity of data updates. The cadastral database accurately pushes the latest related data to the route planning and optimization module based on the change area index. Data push adopts a subscription and publish model. The route planning and optimization module subscribes to change data update topics and receives the latest data pushed by the database in real time. The latest data includes verified feature change data, related historical data, and change trajectory data.

[0060] The route planning and optimization module retrieves historical data and change trajectory data from the cadastral database through a data retrieval interface based on planning requirements. The data retrieval interface supports multi-condition searches based on feature identifiers, spatial ranges, and time ranges. Historical data assists in the route optimization decision-making process; for example, it adjusts the density of surveying points based on surveying difficulty information in historical data, ensuring the reliability and accuracy of route adjustments.

[0061] The route planning and optimization module, the on-site interaction module, and the measurement execution process form an interactive linkage mechanism. The optimized route output by the route planning and optimization module is synchronized to the on-site interaction module in a multimodal manner. Surveyors then execute the mapping task according to the optimized route. The on-site interaction module collects the surveyors' current location information in real time via GPS positioning at a frequency of once per second, and transmits this information back to the route planning and optimization module in real time. Based on the comparison between the current location information and the optimized route, the route planning and optimization module calculates the mapping execution progress. ,include:

[0062] in, To ensure the progress of the surveying and mapping work, This represents the length of the survey route already completed by the surveyors. To optimize the total length of the route.

[0063] The effectiveness of route adjustments is evaluated based on the execution progress. If the deviation between the execution progress and the expected progress is less than 10%, the route adjustment is considered effective. If the deviation is greater than 10%, the cause of the deviation is analyzed and the route parameters are adjusted. When surveyors provide feedback on secondary changes through the on-site interaction module, the collection process for secondary changes is the same as that for the first change. The system quickly triggers a new round of route adjustment. This new round of adjustment reuses the aforementioned logic for confirming changes in ground features and optimizing routes, but shortens the number of iterations for rapid AI verification and the number of rounds for incremental training of the large model, thereby improving response speed and achieving a closed-loop efficiency improvement in surveying execution.

[0064] The on-site interaction module, cadastral database, and feature change confirmation module form an interactive linkage mechanism. Multimodal change data collected by the on-site interaction module is written to the cadastral database in real time via a data writing interface. This interface supports batch writing, improving data writing efficiency. Based on the location information of changed features, the cadastral database retrieves associated historical data through a change area index. This associated historical data includes historical attribute data, historical image data, and historical change records of the features, supporting the AI-driven rapid verification and historical time-series tracing process of the feature change confirmation module.

[0065] The final confirmation result of the land feature change confirmation module is updated by labeling the corresponding data in the cadastral database through the data labeling interface. The fields for labeling and updating include data verification status, change confirmation result, confirmation time, and confirmation personnel identification. A data consistency verification algorithm ensures the consistency between multimodal data and cadastral data. The data consistency verification uses a hash value comparison method, calculating the SHA-256 hash values ​​of the multimodal data and the cadastral data. If the two hash values ​​match, the data is considered consistent; if they do not match, a data synchronization correction process is triggered to resynchronize the multimodal data to the cadastral database, improving the accuracy of land feature change confirmation.

[0066] Example 2 A system for dynamically adjusting cadastral survey routes includes the following modules: a cadastral database for storing historical cadastral survey data and land feature change data for the area to be surveyed; a field interaction module for collecting land feature change information during the survey process and providing feedback to optimize the route; a land feature change confirmation module for verifying the authenticity of land feature change information; and a route planning and optimization module for generating and optimizing the survey route based on real land feature change data.

[0067] This embodiment applies to a cadastral survey scenario in an urban renewal area. The area to be surveyed is XX Street in the old city of a certain city, with a range of 116°23′-116°25′ east longitude and 39°54′-39°56′ north latitude, covering an area of ​​approximately 2 square kilometers. It includes three main land use types: residential land, commercial land, and public facilities land. Historical cadastral survey data covers cadastral data up to June 30th of each year from 2018 to 2023, stored in the cadastral database. This area is currently undergoing urban renewal, with numerous changes in land features such as demolition of old houses, construction of new resettlement housing, and road widening. Dynamic adjustments to the surveying route are necessary to ensure the accuracy and efficiency of the cadastral survey.

[0068] The cadastral database adopts a distributed storage architecture, consisting of three data nodes and one management node. The data nodes use solid-state drives for storage, and the management node is responsible for data sharding scheduling and load balancing. The real-time change data update module is based on an incremental storage mechanism, storing only the difference data corresponding to changes in land features. The difference data includes attribute difference data before and after the change, as well as spatial difference data. Attribute difference data records changes in attributes such as ownership, use, and area of ​​land features, while spatial difference data records changes in spatial information such as location, shape, and boundaries of land features.

[0069] The time-series trajectory of ground feature changes is recorded using timestamp fields. Each change data point includes three time fields: creation timestamp, update timestamp, and effective timestamp. The creation timestamp records the data entry time, the update timestamp records the last modification time, and the effective timestamp records the actual time the ground feature change occurred. The data sharding storage strategy divides the 2 square kilometer area into 200 100m × 100m square grids using a quadtree grid. Each grid is assigned a unique sharding identifier, which consists of an area code, a grid row number, and a grid column number. For example, "XXJD-01-05" represents the grid in the 1st row and 5th column of XX street.

[0070] Change data and historical data are linked and tagged using unique feature identifiers. These unique feature identifiers are 18-digit codes: the first 6 digits represent the administrative division code, the middle 4 digits represent the feature type (e.g., 0001 for residential land, 0002 for commercial land, 0003 for public facilities land), and the last 8 digits are a sequential number. For example, "110101-0001-00000001" represents the first residential land feature in Dongcheng District, Beijing. This association allows for rapid tracing of historical changes in features, ensuring data traceability.

[0071] The R-tree spatial index constructed by the fast retrieval adaptation module has a root node containing 4 branch nodes, and each branch node contains 8 leaf nodes. The leaf nodes store the bounding rectangle coordinates of the feature and the data shard identifier. When it is necessary to retrieve historical data of a changed feature, the system traverses the R-tree index based on the latitude and longitude coordinates of the changed feature, finds the leaf node corresponding to the bounding rectangle containing the coordinates, obtains the data shard identifier, and then retrieves the historical data from the corresponding database shard. The retrieval response time is controlled within 50 milliseconds.

[0072] The mapping between multimodal data and cadastral data is achieved through a related data table, which is stored in the management node and includes fields for feature identification, multimodal data type, data storage path, data collection time, and data verification status.

[0073] Multimodal data types: 01 represents image data, 02 represents speech and semantic data; data verification status: 00 represents unverified, 01 represents verified, 02 represents verified failed.

[0074] For example, the image data storage path corresponding to the feature identifier "110101-0001-00000001" is " / data / image / 110101 / 0001 / 00000001_202405201430.jpg", the data collection time is "2024-05-20 14:30:25", and the data verification status is "01". Through this associated data table, the association query between multimodal data and cadastral data can be quickly realized.

[0075] The on-site interaction module is deployed on a portable mobile surveying device. The device runs on Android 12 and is equipped with an 80-megapixel rear camera, a high-definition microphone, a 4G / 5G network module, and a GPS positioning module. It supports touch operation and voice interaction. The multimodal change feedback module features real-time image annotation. When surveyors detect changes in terrain features, they can activate the device's camera to capture images of the changed area. During the capture, the device automatically records the current GPS coordinates as the initial location information of the changed area.

[0076] The bounding box marking uses a rectangular bounding box algorithm based on OpenCV. Surveyors draw a rectangle on the touchscreen with their fingers to select the core part of the changing area, and the system automatically records the pixel coordinates of the top left and bottom right corners of the rectangle. and And by using the mapping relationship between pixel coordinates and latitude and longitude coordinates, the pixel coordinates are converted into corresponding latitude and longitude coordinates. and This forms the latitude and longitude range of the changing area. The mapping relationship is determined by the camera calibration parameters of the device. The camera calibration adopts the Zhang Zhengyou calibration method to obtain the camera's intrinsic and extrinsic parameter matrices, establish a conversion model between pixel coordinates and world coordinates, and then combine GPS positioning data to convert the world coordinates into latitude and longitude coordinates.

[0077] The speech semantic recognition uses a Transformer model, which is pre-trained on publicly available speech recognition datasets and cadastral-specific datasets through transfer learning. The cadastral-specific dataset contains 100,000 speech data related to changes in land features, covering various types of changes such as house demolition, new construction, reconstruction, and road widening. The model's encoder is a 6-layer Transformer encoder, with 8 attention heads per layer and a hidden layer dimension of 512. The FeedForward network has a hidden layer dimension of 2048. The decoder is a 4-layer Transformer decoder with the same structure as the encoder.

[0078] During speech signal preprocessing, a first-order high-pass filter is used for pre-emphasis. ,in This is used to enhance the high-frequency components of the speech signal; framing uses a 20ms frame length and a 10ms frame shift to divide the speech signal into several frames; windowing uses a Hanning window. ,in, For the Hanning window's weight, For intra-frame sampling point index, The number of sampling points corresponding to the frame length is set to reduce inter-frame interference. Mel spectral feature extraction converts the preprocessed speech signal into Mel frequency cepstral coefficients (MFCCs). First, a Fourier transform is performed on the speech frame to obtain the power spectrum. Then, the power spectrum is converted into a Mel spectrum using a Mel filter bank. Finally, logarithmic operations and discrete cosine transforms are performed on the Mel spectrum to extract 13-dimensional MFCC features. Then, first-order and second-order difference features are calculated to form a 39-dimensional feature vector.

[0079] Through the encoding and decoding process of the Transformer model, the 39-dimensional feature vector is converted into structured change description information. The structured information includes fields describing the type of land feature change, the location of the change, and the range of the change. For example, if a surveyor takes a picture of a house to be demolished and inputs "35 XX Road, 3-story old building to be demolished" in voice, the system will output the following structured description information: land feature change type "demolition", change location "35 XX Road", change range "3-story building, covering an area of ​​approximately 120 square meters".

[0080] The change visualization feedback module overlays real-time acquired images with historical images from the cadastral database to generate a visual comparison interface. First, the two images are registered using a feature-point-based registration method, extracting SIFT feature points from the real-time and historical images. Then, the FLANN matching algorithm is used for feature point matching, employing a KD-tree index with a matching threshold of 0.7, retaining feature point pairs with a matching degree higher than the threshold. Based on the matched feature point pairs, the homography matrix is ​​solved using the Random Sample Consensus (RANSAC) algorithm. Abnormal matching points are eliminated to ensure registration accuracy. Homography matrix The solution is obtained by establishing a system of equations using the coordinates of four pairs of matching feature points, and solving for the matrix elements to achieve pixel-level alignment between real-time images and historical images, and the homography matrix. include:

[0081] in, It is a homography matrix. These are matrix elements.

[0082] After registration, a weighted fusion algorithm is used to overlay the images. The fusion formula is as follows:

[0083] in, These are the pixel values ​​after overlay. These are the pixel values ​​of the real-time image. These are the pixel values ​​of the historical image. These are the weighting coefficients.

[0084] The superimposed image is displayed on the device screen. The real-time image is overlaid on the historical image with high transparency, intuitively presenting the changes in land features. The difference level classification is based on the area ratio of the changed area and the type of change. The area ratio of the changed area is determined by the ratio of the area calculated from the latitude and longitude range of the changed area to the area of ​​the original land features. The area of ​​the original land features is obtained from the historical attribute data of the cadastral database.

[0085] For example, if the historical area of ​​a residential feature is 120 square meters, and the real-time detected area of ​​change is 40 square meters, the change area accounts for 33.3% of the total area. If the change type is reconstruction, it is classified as a Level 2 difference. The system will display a yellow pop-up message with the message "Level 2 difference: feature reconstruction, change area accounts for 33.3%, priority surveying is recommended".

[0086] The interactive command quick trigger module supports surveyors to customize voice commands. The process for setting up custom commands is as follows: Surveyors select the "Custom Command" function through the device's settings interface and enter the command content via voice, such as "Adjust Route," "Synchronize Data," or "Mark Key Points." The system performs voice recognition on the entered command, converts it into a text command, performs semantic parsing, and determines the uniqueness of the command to avoid duplication with existing commands in the preset command library. After uniqueness verification, the surveyor assigns a corresponding trigger operation to the command, such as "Adjust Route" for triggering a route adjustment request, and "Synchronize Data" for triggering a data synchronization operation. Finally, the command is stored in the preset command library.

[0087] The preset command library is stored in a Redis database, storing the mapping relationship between command keywords and trigger operations in key-value pairs. The key is the hash value of the command keyword, and the value is the code of the trigger operation, such as 001 representing a route adjustment request and 002 representing a data synchronization operation. When a surveyor issues a voice command, the system first converts the voice command into a text command, calculates the keyword hash value of the text command, and queries the Redis database for the corresponding trigger operation code. If the corresponding code is found, the corresponding operation is triggered; if not, the similarity with commands in the preset command library is calculated using a command word matching algorithm. The similarity calculation includes:

[0088] in, For instruction similarity, The first in the preset instructions The weight of each keyword, For the first input command The semantic similarity between each keyword and the preset keywords. For the number of keywords, This refers to the keyword overlap rate of the instruction. , To match the number of keywords, The mean semantic similarity. , This is the regularization coefficient, with a value of 0.1.

[0089] For example, the preset command is "adjust surveying route", the keywords are "adjust", "surveying", and "route", and the corresponding weights are... , , The surveyor's voice command was "Optimize surveying route," and the keywords in the converted text command were "optimize," "survey," and "route." The semantic similarity between these keywords and the preset keywords was... The semantic similarity between "optimize" and "adjust". The semantic similarity between "surveying" and "measuring" The semantic similarity between "route" and "line"; the number of matching keywords. ,but Mean semantic similarity Substituting into the formula, we can calculate... , If the command matches successfully, a route adjustment request is triggered.

[0090] Upon successful command matching, if a data synchronization operation is triggered, the multimodal change data is transmitted to the cadastral database via the HTTP / 2 protocol. The transmission process employs the AES-256 encryption algorithm, and the encryption key is stored through the device's hardware security module to ensure its confidentiality. The data packet contains a header and a body. The header includes feature identifiers, data types, data length, and timestamp fields. The body contains the encrypted multimodal data; image data is compressed in JPEG format, and voice semantic data is stored in JSON format. If a route adjustment request is triggered, the system sends a change data transmission request to the feature change confirmation module. The request data packet includes feature identifiers, coordinates of the changed area, and the multimodal data storage path fields, initiating the feature change verification process.

[0091] The land cover change confirmation module is deployed on a cloud server, employing a distributed computing architecture to support parallel processing of multiple land cover change verification requests. The SIFT algorithm feature extraction process of the AI-based rapid verification module first constructs a Gaussian difference pyramid, then convolves the image with Gaussian kernels of different scales. ,in Using the Gaussian kernel standard deviation, images at 6 scales are obtained. Subtracting images at adjacent scales yields 5 layers of Gaussian difference images. In the Gaussian difference image, each pixel is compared with its 8 neighboring pixels and the 9 pixels in the layers above and below. If a pixel is a local extremum, it is used as a candidate keypoint.

[0092] Precisely locate candidate keypoints by fitting a three-dimensional quadratic function.

[0093] in, For the difference of Gaussians image in coordinates ,scale Pixel value at that location, Equal to the second-order partial derivatives, The first-order partial derivatives are used to determine the precise location of the extreme points and eliminate low-contrast values. Key points and edge response points .

[0094] A gradient direction histogram of pixels within the neighborhood of a keypoint is calculated, with gradient directions ranging from 0° to 360°, divided into 36 intervals of 10° each. The sum of the gradient magnitudes in each interval is calculated, and the direction corresponding to the peak value in the histogram is taken as the principal direction of the keypoint. If other directions have peak values ​​greater than 80% of the principal peak value, they are used as auxiliary directions to achieve rotation invariance of the features. Finally, a 16×16 neighborhood window is extracted centered on the keypoint, and it is divided into 4×4 sub-windows. The gradient histogram of each sub-window is calculated for 8 directions, forming a 128-dimensional feature vector. The feature vector is then normalized to enhance the illumination invariance of the features.

[0095] Feature matching uses weighted normalized Euclidean distance to calculate feature vector similarity, including:

[0096] in, Weighted normalized Euclidean distance is applied to the feature vectors. For the real-time image feature vector, the first dimensional components, For the historical image feature vector, the first dimensional components, The dimension of the feature vector is 128. For the first Importance weights of dimensional features , .

[0097] For example, a real-time image extracts 150 key points, generating 150 128-dimensional feature vectors; a corresponding historical image extracts 140 key points, generating 140 128-dimensional feature vectors. Feature vector matching is performed using the FLANN matching algorithm, calculating the weighted normalized Euclidean distance for each pair of feature vectors, and setting a matching threshold. A total of 92 feature vectors were successfully matched, with an image matching success rate of [percentage missing]. Image feature matching was successful.

[0098] The semantic matching verification algorithm compares the consistency between the structured description information output by speech semantic recognition and the image feature matching results. The structured description information includes three fields: feature change type, change location, and change range. The image feature matching results include the location, shape, and size information of the changed area. For example, the structured description information shows that the feature change type is "demolition," the change location is "No. 35, XX Road," and the change range is "a 3-story building, covering an area of ​​approximately 120 square meters." The image feature matching results show that most of the feature points corresponding to the original 3-story building at No. 35, XX Road have disappeared, the shape of the changed area is rectangular, and the area is approximately 118 square meters. The number of fields consistent with the structured description information is 3, and the total number of fields is 3, resulting in a discrepancy of 3%. The semantic matching is deemed successful. The AI-powered rapid verification module outputs a determination of the authenticity of the ground feature changes, verifying the confidence level. .

[0099] The historical time-series tracing module calls upon the historical data of the land feature from 2018 to 2023 in the cadastral database to form a multi-version historical dataset. Cross-time period comparison analysis employs a combination of image differencing and attribute comparison. Image differencing calculates the difference between historical images and real-time images for each year from 2018 to 2023, quantifying the degree of difference by calculating the sum of pixel values ​​in the differencing images. Attribute comparison analysis examines changes in the land feature's ownership, area, and usage information. Historical attribute data for the land feature shows that from 2018 to 2023, ownership was private, area was 120 square meters, and usage was residential, with no historical change records.

[0100] The trend model adopts the ARIMA model, which fits the model based on the land cover change index from 2018 to 2023. When there is no change, the index value is 0, and the model parameters are determined. , , The model expression is:

[0101] in, For the first Indicators of the degree of change in land features over the years For the first Indicators of the degree of change in land features over the years For the first White noise of the year For the first The white noise for 1 year is 0.8, the autoregressive coefficient is 0.8, and the moving average coefficient is 0.3.

[0102] The model predicted a land feature change index value of 0 for 2024, while the real-time detected land feature change index value was 0.613, which is inconsistent with the predicted trend. However, in combination with the old area renovation planning documents of the region, it was determined that the land feature change was a normal change within the plan, and a reasonable analysis result of the change trend was output.

[0103] The manual review trigger module automatically triggers a critical change warning when the confidence level of the AI ​​rapid verification result is below 0.7. For example, the AI ​​rapid verification result for a certain feature shows an image matching pass rate of 58% and a semantic matching inconsistency rate of... Verification confidence level The system automatically triggers a critical change warning and sends a verification request to staff via mobile app push notification and SMS notification. The push information includes the feature identifier "110101-0002-00000015", the coordinates of the changed area "116°24′12″E, 39°55′08″N", the multimodal data link "http: / / cloudserver / data / 110101000200000015", and the AI ​​verification result summary "Image matching pass rate 58%, semantic matching inconsistency 35%, confidence level 61.5%, feature change is questionable".

[0104] Staff logged into the cloud-based verification platform via a link to view detailed multimodal change data and AI verification process data. Verification was conducted using a combination of manual visual interpretation and attribute checking. Manual visual interpretation compared real-time images with historical imagery to confirm the authenticity of the changed areas. Attribute checking retrieved historical attribute data of land features and relevant planning documents to confirm the rationality of the changes. After verification, staff entered the verification conclusion "Authentic" and the verification explanation "The features on this commercial land were partially demolished due to road widening, which complies with the regional redevelopment plan" on the platform. The verification result was transmitted back to the system via an encrypted channel. The system associated and stored the verification result with the corresponding land feature change data and updated the data verification status field to "01," indicating successful verification.

[0105] The route planning and optimization module is deployed on a cloud server and interacts with the feature change confirmation module via an internal local area network, with data transmission latency controlled within 100 milliseconds. The change adaptation planning module determines the mapping priority of the changed area based on the actual feature change data output by the feature change confirmation module. The mapping priority calculation includes:

[0106] in, For surveying priority, The weights are affected by the changes. , For data importance weights, , This is the ownership adjustment factor. , To determine the degree of impact of changes in terrain features on the original route, To emphasize the importance of cadastral data on changed land features, This is a correction factor for the magnitude of change. .

[0107] For example, the land ownership of a newly built resettlement housing site is state-owned public facility land. ,but , The change range was 60%. The overlapping area between the original route and the changed area square meters, the area corresponding to the total length of the original route. square meters, The land use type is public facilities land. Substituting into the formula, we can calculate... .

[0108] The other land where the old houses and buildings were demolished was collective residential land. , , The change was 40%. The overlapping area between the original route and the changed area square meters, square meters, The land use type is residential. Substituting into the formula, we can calculate... The mapping priority of the demolished old buildings and features is higher than that of the newly built resettlement buildings and features. The system will prioritize mapping the changed areas corresponding to the demolished old buildings and features.

[0109] The obstacle avoidance algorithm employs an improved A algorithm, designating obstacle areas formed by demolished or newly constructed features as impassable zones. The boundary of the obstacle zone is determined by extending it outward by 5 meters from the latitude and longitude range of the changed area, ensuring a safe distance between the surveying route and the obstacle zone. The heuristic function of the A algorithm... include:

[0110] in, For nodes Heuristic distance to the target node As a heuristic weight, a value of 1.2 is set. For nodes coordinates The coordinates of the target node.

[0111] This heuristic function guides the path search towards the target node, improving search efficiency. For example, the original survey route runs from point A (116°23′10″E, 39°54′20″N) to point B (116°24′30″E, 39°55′10″N), passing through an area of ​​obstacles where old houses and structures have been demolished (116°23′50″E-116°24′05″E, 39°54′40″N-39°54′55″N).

[0112] The improved A algorithm avoids obstacle areas and generates a new optimized route when searching for a path. The new route starts from point A, passes through point C (116°23′45″E, 39°54′30″N), and point D (116°24′10″E, 39°54′35″N) to reach point B. The total length of the route is about 80 meters longer than the original route, but it avoids obstacle areas, ensuring the smooth progress of the surveying task.

[0113] The large-scale model real-time response module uses real-world land cover change data as incremental training data to fine-tune the cadastral spatial large-scale model in real time. The cadastral spatial large-scale model is built on the Transformer architecture, containing a 12-layer encoder and a 6-layer decoder, with a hidden layer dimension of 1024 and 16 attention heads. The pre-training data includes 1 million cadastral spatial data points, 500,000 historical measurement route data points, and 300,000 land cover change data points.

[0114] Incremental training employs mini-batch gradient descent with a batch size of 32. The learning rate is scheduled using cosine annealing with an initial learning rate of 1e-5 and a minimum learning rate of 1e-7. The training iterations are 10 epochs. The loss function for incremental training is... ,in, For mixed loss values, The mean squared error loss is calculated by comparing the predicted route coordinates with the actual route coordinates. The cross-entropy loss is calculated by comparing the predicted route type with the actual route type. , During training, only the parameters of the top 3 fully connected layers of the model are updated, while the parameters of the bottom 9 encoder layers and 6 decoder layers are frozen to reduce computation and keep training time within 5 minutes.

[0115] Fast inference employs an optimization technique combining model pruning and quantization. Model pruning utilizes a structured pruning algorithm, pruning 40% of the fully connected layers by calculating neuron importance scores and removing the 40% of neurons and their corresponding connections with the lowest scores. Model quantization uses INT8 quantization, converting the model's weights and activation values ​​from 32-bit floating-point numbers to 8-bit integers. The quantization process employs min-max quantization, calculating the minimum and maximum values ​​of weights and activation values ​​and mapping them to the INT8 value range [-128, 127]. The mapping formula is as follows:

[0116] in, It is the quantized 8-bit integer. For the original 32-bit floating-point number, The minimum value of the original data. The maximum value of the original data. This is a rounding function. Through the above optimization techniques, the model's storage capacity was reduced from 2GB to 500MB, and the inference speed was improved from 2.5 seconds / inference to 0.8 seconds / inference, ensuring that the optimized route is output within 1 second.

[0117] The intelligent route recommendation module generates three different optimized route schemes based on surveying efficiency and measurement accuracy constraints. Of the two surveying efficiency metrics, the total route length is obtained by calculating the sum of the straight-line distances between all inflection points on the route. The straight-line distance calculation uses the Haversine formula:

[0118] in, The distance between two points on the Earth's surface. The radius of the Earth is taken as 6,371,000 meters. The latitude and longitude of the first point. The latitude and longitude of the second point.

[0119] The estimated surveying time is calculated based on the total route length, the average travel speed of the surveyors (1.5 m / s), and the average measurement time at each survey point (30 seconds). The calculation formula is as follows:

[0120] in, To estimate the surveying time, This is the total length of the route. The average walking speed of the surveyors is 1.5 meters per second. The number of survey points. The average measurement time for each survey point is 30 seconds.

[0121] Of the two indicators of measurement accuracy, the completeness of the route coverage of the changed area is calculated by the overlap ratio between the latitude and longitude range of the changed area and the coverage area of ​​the route. The formula for the overlap ratio is:

[0122] in, To ensure complete coverage, The overlapping area between the route coverage area and the changed area. This represents the total area of ​​the changed region.

[0123] The density of survey points is calculated by the ratio of the total route length to the number of survey points. The formula is as follows:

[0124] in, The preset density of survey points is no less than 0.2 points / meter.

[0125] The weights of each indicator were determined using the analytic hierarchy process (AHP), and the judgment matrix was constructed as follows:

[0126] Calculate the largest eigenvalue of the judgment matrix By combining the feature vectors, we obtain the weight vectors of each indicator. The weights are: total route length (0.15), estimated surveying time (0.25), coverage integrity (0.35), and survey point density (0.25).

[0127] The actual values ​​of the indicators for the three optimized route schemes are shown in the table below:

[0128] Each indicator is standardized using the min-max method; taking the total route length as an example. , Standardized score of Scheme 1 Standardized score of Scheme 2 Standardized score of Scheme 3 Similarly, calculate the standardized scores for other indicators: Estimated surveying time: , Option 1 scored 0.42, Option 2 scored 1.0, and Option 3 scored 0.0. Completeness of coverage: , Option 1 scored 0.71, Option 2 scored 1.0, and Option 3 scored 0.0. Survey point density: , Option 1 scored 0.67, Option 2 scored 1.0, and Option 3 scored 0.0.

[0129] The weighted summation method is used to calculate the overall score of each scheme. ,include:

[0130] in, To determine the overall score for the proposal, For the first The weight of each indicator, For the first The standardized score of each indicator.

[0131] Overall score of Option 1 ; Overall score of Option 2 ; Overall score of Option 3 .

[0132] Option 2 received the highest overall score and was selected as the optimal route. The system then fed back the optimized route of Option 2 to the surveyors through the on-site interaction module.

[0133] The interactive linkage mechanism formed by the on-site interaction module, the feature change confirmation module, and the route planning and optimization module allows the multimodal change data collected by the on-site interaction module to directly trigger the AI ​​rapid verification process of the feature change confirmation module via a RESTful API interface. The RESTful API interface uses a POST request method, with the request header containing authentication information (based on a JWT token) and data format instructions (JSON format), and the request body containing feature identifiers, coordinates of the changed area, multimodal data storage path, and collection time field. The AI ​​rapid verification process of the feature change confirmation module processes multiple requests in parallel, with each request allocated independent computing resources: 2 CPU cores and 4GB of memory, ensuring a single request processing time of less than 500 milliseconds.

[0134] The real-time ground feature change data output by the ground feature change confirmation module drives the route generation process of the route planning and optimization module in real time through a Kafka message queue. The Kafka message queue topic is named "land_survey_change_data", with 3 partitions and 2 replicas to ensure data reliability. The route planning and optimization module subscribes to this topic as a consumer, with the consumer group set to "route_planning_group". An automatic offset commit mechanism is used to avoid duplicate data consumption.

[0135] The optimized route output by the route planning and optimization module is presented to the surveyors in a multimodal form through the change visualization feedback module of the on-site interaction module: the voice feedback uses TTS speech synthesis technology based on the WaveNet model to convert the route description text into a natural and fluent speech signal, with a playback speed of 150 words / minute and a sampling rate of 44.1kHz; the text feedback is displayed in the status bar at the top of the device screen, including the total route length of 920 meters, the estimated surveying time of 31.0 minutes, and the coordinates of 3 key inflection points: point A: 116°23′10″E, 39°54′20″N, point C: 116°23′45″E, 39°54′30″N, and point B: 116°24′30″E, 39°55′10″N; the map annotation shows the optimized route as a solid red line on the electronic map based on the WebGIS engine, with a line width of 3 pixels, and yellow circles marking the change areas, with a circle radius of 15 meters representing the maximum distance of the change area and a circle border width of 2 pixels.

[0136] When abnormal changes occur, such as when the change of a certain feature exceeds the preset threshold of 85% > 80%, the feature change confirmation module generates an abnormal change warning signal. The warning signal includes the abnormal feature identifier, the coordinates of the changed area, and the warning level. It is then sent to the field interaction module and the route planning and optimization module via a Kafka message queue.

[0137] Upon receiving the warning signal, the on-site interaction module displays a red warning pop-up window with the message "Level 1 Warning: Ground feature change is abnormally large (85%). Please confirm before proceeding with surveying," and plays a warning audio message. The ground feature change confirmation module marks the abnormal change data as key data for review, adds an "is_key_review" field to the database and assigns it a value of "1," adding a manual review step. Upon receiving the warning signal, the route planning and optimization module pauses the route generation process, sets the task status to "waiting_manual_confirm," and sends a "waiting for manual confirmation" text prompt to the on-site interaction module via a message queue. The route optimization process resumes only after manual confirmation, achieving seamless integration of change feedback, authenticity confirmation, and route adjustment.

[0138] The interactive linkage mechanism formed by the feature change confirmation module, cadastral database, and route planning and optimization module allows the verified change data output by the feature change confirmation module to automatically trigger the update process of the cadastral database via a JDBC interface. The JDBC interface uses MySQL Connector / J 8.0, supports transaction processing, and has an isolation level of READCOMMITTED to ensure the atomicity of data updates. If a network interruption or database anomaly occurs during the update process, the transaction is automatically rolled back to avoid data inconsistency. The update operation includes two steps: First, the "change_status" field in the feature attribute table tb_land_property is updated with the value "confirmed", and the "last_update_time" field is updated with the current timestamp. Second, the feature change record is inserted into the feature change record table tb_land_change_record to record information such as change type, change magnitude, and confirmation personnel identification.

[0139] The cadastral database accurately pushes the latest related data to the route planning and optimization module based on the change area index. The data push adopts a subscription and publish mode. The route planning and optimization module subscribes to the "land_survey_db_update" topic of Kafka through a Java client and receives the latest data pushed by the database in real time. The latest data includes verified land feature change data, related historical data, and change trajectory data. The data compression format is GZIP to reduce the transmission bandwidth usage.

[0140] The route planning and optimization module retrieves historical data and change trajectory data from the cadastral database via a data retrieval interface based on planning requirements. The data retrieval interface uses the GraphQL protocol and supports multi-condition searches based on feature identifiers, spatial ranges, and time ranges. For example, searching for historical data with "feature identifier = 110101-0001-00000001, time range = 2020-01-01 to 2024-01-01" returns all historical attribute data for that feature within that time period, such as ownership, area, and use; historical image data, including the URL of orthophotos on June 30th of each year; and historical change records. If no changes are found, an empty list is returned. Historical data assists in the route optimization decision-making process. For instance, based on surveying difficulty information in historical data, the density of surveying points along the route can be adjusted, increasing it from 0.25 points / meter to 0.3 points / meter, thus increasing the number of surveying points and ensuring the reliability and accuracy of route adjustments.

[0141] The interactive linkage mechanism formed by the route planning and optimization module, the field interaction module, and the measurement execution process allows the optimized route output by the route planning and optimization module to be synchronized to the field interaction module in a multimodal manner. After that, the surveyors perform the surveying task according to the optimized route. The field interaction module collects the current location information of the surveyors in real time through the GPS positioning module. The positioning accuracy of the GPS positioning module is ±1 meter, and the acquisition frequency is 1 time / second. The location information includes latitude and longitude coordinates, positioning time, and positioning accuracy fields. It is transmitted back to the route planning and optimization module in real time through the 4G / 5G network using the WebSocket protocol. The WebSocket connection adopts a heartbeat mechanism to ensure connection stability.

[0142] The route planning and optimization module calculates the surveying and mapping progress based on a comparison between the current location information and the optimized route. ,include:

[0143] in, To ensure the progress of the surveying and mapping work, The length of the surveyed route already completed by the surveyors is calculated using the Haversine distance between the current location and the starting point of the route. For example, if the current location is 552 meters from the starting point, then... rice, To optimize the total route length, 920 meters, the calculation yielded... .

[0144] The effectiveness of route adjustments is evaluated based on the execution progress. If the deviation between the execution progress and the expected progress is less than 10%, the route adjustment is considered to be effective. If the deviation is greater than 10%, the reasons for the deviation are analyzed. For example, if the surveyors' walking speed is too slow (actual speed of 1.2 m / s < 1.5 m / s), the route parameters are adjusted to reduce the density of survey points in subsequent sections from 0.25 points / meter to 0.22 points / meter, thereby reducing the number of survey points and shortening the expected surveying time. If temporary obstacles are encountered, the obstacle avoidance algorithm is triggered to regenerate a locally optimized route.

[0145] When surveyors provide feedback on secondary changes through the on-site interactive module, for example, if a newly built bungalow is discovered at No. 38 XX Road during route optimization, the collection process for secondary change information is the same as that for the first change information: take an image of the newly built bungalow, select the change area (latitude and longitude range: 116°23′55″-116°24′00″ E, 39°54′32″-39°54′37″ N), and input the voice message "Newly built bungalow at No. 38 XX Road." Semantic recognition outputs a structured description: type "newly built," location "No. 38 XX Road," range "bunker, area approximately 60 square meters." The process quickly triggers a new round of route adjustment, which reuses the aforementioned logic for confirming changes in terrain features and optimizing routes. However, it shortens the number of iterations for AI rapid verification and the number of rounds for incremental training of the large model. The processing time for a single request for AI rapid verification is reduced from 500 milliseconds to 200 milliseconds, and the incremental training time of the large model is reduced from 5 minutes to 90 seconds. This ensures that a new optimized route is output within 3 seconds. A new point D is added: 116°23′52″E, 39°54′35″N. The route is adjusted to A→C→D→B, with a total length of 950 meters. The estimated surveying time is 32.5 minutes, achieving a closed-loop efficiency improvement in surveying execution.

[0146] The interactive linkage mechanism formed by the on-site interaction module, the cadastral database, and the feature change confirmation module allows the multimodal change data collected by the on-site interaction module to be written to the cadastral database in real time through a data writing interface. The data writing interface adopts the HTTP / 2 protocol, supports batch writing, and the response time for writing a single data entry is less than 100 milliseconds. The written data includes image data, voice semantic data, acquisition time, and acquisition device identification fields. The image data is stored in a distributed file system, and the storage path format is " / data / image / {administrative division code} / {feature type code} / {feature identifier}_{acquisition time}.jpg".

[0147] The cadastral database retrieves associated historical data through the change area index based on the location information of changed features. The associated historical data includes historical attribute data, historical image data, and historical change records of the features, supporting the AI-based rapid verification and historical time-series tracing process of the feature change confirmation module.

[0148] The final confirmation result of the feature change confirmation module is updated by annotating the corresponding data in the cadastral database through the data annotation interface. The fields for annotation and update include: data verification status, tb_land_multimodal.data_check_status, assigned the value "01"; change confirmation result, tb_land_multimodal.confirm_result, assigned the value "true"; confirmation time, tb_land_multimodal.confirm_time, assigned the current timestamp; and confirmation personnel identifier, tb_land_multimodal.confirm_user_id, assigned the value "user_001". The update process adopts an optimistic locking mechanism, adding a "version" field to the data table. During the update, the condition "where feature identifier=? and version=?" is used to determine whether the data has been modified. If it has been modified, the update is retried 3 times to avoid concurrent update conflicts.

[0149] The consistency between multimodal data and cadastral data is ensured through a data consistency verification algorithm. The data consistency verification uses a hash value comparison method to calculate the SHA-256 hash value of the multimodal data. The calculation formula is as follows:

[0150] in, For multimodal data, the SHA-256 hash value. A binary stream of image data. The binary stream of speech and semantic data is represented by the "+" sign, indicating the concatenation of the binary streams. Simultaneously, the SHA-256 hash value of the corresponding land feature data in the cadastral database is calculated. If the two hash values ​​match, the data is considered consistent. If they do not match, the data synchronization correction process is triggered, and the multimodal data is resynchronized to the cadastral database through the data writing interface. The synchronization retry count is 3 times, with a 5-second interval between each retry, to ensure data consistency and improve the accuracy of land feature change confirmation.

[0151] This application also provides a computer-readable storage medium storing a computer program, the computer program including program instructions, which, when executed by a computer, cause the computer to perform the system or method as described in the foregoing embodiments.

[0152] Those skilled in the art will understand that all or part of the processes in the systems described in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes described in the embodiments of the above systems. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0153] It should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A system for dynamically adjusting a cadastral survey route, characterized by, The system includes a cadastral database, a field interaction module, a feature change confirmation module, and a route planning and optimization module. The field interaction module is used to collect multimodal data on feature changes in real time. The feature change confirmation module is used to perform rapid AI verification on the collected multimodal change data and combine it with historical time-series tracing to achieve accurate confirmation of feature changes. The route planning and optimization module is used to calculate the mapping priority based on the confirmed feature change data, and combine it with an improved obstacle avoidance algorithm and real-time fine-tuning of the cadastral spatial model to dynamically and intelligently adjust the mapping route.

2. The system for dynamically adjusting cadastral survey routes according to claim 1, characterized in that, The cadastral database is used to store historical land feature data, multimodal change data collected by the field interaction module, valid change data confirmed by the land feature change confirmation module, and surveying route data generated by the route planning and optimization module. It also includes real-time data updates and historical time series retrieval, and transmits the data to the land feature change confirmation module and the route planning and optimization module.

3. The system for dynamically adjusting cadastral survey routes according to claim 1, characterized in that, The multimodal change data collected by the on-site interaction module includes image data and voice semantic data, including real-time photo annotation and voice semantic recognition. The real-time photo annotation function collects images of the areas of ground feature change and marks them with bounding boxes. The voice semantic recognition function extracts keywords related to ground feature change and merges the image data and semantic data to form multimodal change data.

4. The system for dynamically adjusting cadastral survey routes according to claim 1, characterized in that, The AI-powered rapid verification of the land feature change confirmation module includes feature matching verification and semantic consistency verification. It extracts image features from multimodal change data, matches them with historical image features in the cadastral database, verifies the consistency between the voice description and the image changes through a semantic matching verification algorithm, and outputs the result of determining the authenticity of the land feature changes.

5. A system for dynamically adjusting cadastral survey routes according to claim 1, characterized in that, The land feature change confirmation module includes feature matching verification. It extracts land feature feature points from multimodal change data and compares them with historical land feature database. When the similarity reaches a preset threshold, it is determined to be a valid change feature. If the similarity does not reach the preset threshold, it is determined to be temporary occlusion or recording error.

6. The system for dynamically adjusting cadastral survey routes according to claim 1, characterized in that, The land feature change confirmation module includes a historical time-series tracing function, which retrieves historical land feature data and compares it with the currently collected multimodal change data.

7. A system for dynamically adjusting cadastral survey routes according to claim 1, characterized in that, The route planning and optimization module also includes a change adaptation planning module, which determines the mapping priority of the changed area based on the real ground feature change data output by the ground feature change confirmation module, uses an obstacle avoidance algorithm to bypass obstacles formed by demolished or newly built ground features, and generates an initial optimized route that adapts to the ground feature changes.

8. A system for dynamically adjusting cadastral survey routes according to claim 1, characterized in that, The route planning and optimization module also includes a large model real-time response module, which uses real-world feature change data as incremental training data to fine-tune the cadastral spatial large model in real time, shortens the model response time through a fast inference algorithm, and achieves instant output of optimized routes. Incremental training includes a parameter freezing strategy, which updates only the parameters of the top fully connected layer of the model, while fast inference employs model pruning optimization techniques.

9. A system for dynamically adjusting cadastral survey routes according to claim 1, characterized in that, The on-site interaction module transmits the collected multimodal change data to the feature change confirmation module, which then transmits the confirmed feature change data to the route planning and optimization module. The route planning and optimization module then feeds back the adjusted survey route to the on-site interaction module in real time.

10. A system for dynamically adjusting cadastral survey routes according to claim 1, characterized in that, The route planning and optimization module also includes an intelligent recommendation module for alternative routes. Based on constraints such as surveying efficiency and measurement accuracy, it generates several different optimized route schemes, uses a multi-objective decision-making algorithm to comprehensively evaluate each scheme, and automatically selects the optimal route. Among them, the multi-objective decision-making adopts the weighted summation method, which sets efficiency weight and accuracy weight according to the requirements of the surveying and mapping task, calculates the comprehensive score of the scheme, and the one with the highest score is the optimal route.

Images