A geodetic survey report generation system based on artificial intelligence

Abstract

The present application relates to geodetic surveying, geographic information engineering technical field, specifically point to a kind of geodetic surveying report generation system based on artificial intelligence, surveying specification analysis engine output specification quantization rule set synchronous transmission to each module;Multi-source surveying data acquisition module transmits multi-source spatial data and acquisition full parameter to adaptive pose solution module and penetrating responsibility traceability module;Adaptive pose solution module exports point cloud data to compliance check and report generation module, unqualified data back transmission correction;Qualified data are sent into perception meta-learning module, and its processing result is checked again, and unqualified back transmission, and the report content and full-link compliance check record synchronous transmission to penetrating responsibility traceability module.The present application quantizes and disassembles surveying specification and uniformly executes, adaptive pose solution adapts project accuracy, meta-learning multi-task processing depth binding specification, through process check back transmission and full-link traceability, clear error component and responsibility attribution.

Description

Technical Field

[0001] This invention relates to the fields of geodetic surveying and geographic information engineering technology, specifically to an artificial intelligence-based geodetic surveying report generation system. Background Technology

[0002] Geodetic surveying is a fundamental task in fields such as natural resource investigation, land spatial planning, and engineering construction surveying. The quality of its results must strictly comply with relevant national standards for geodetic surveying, industry standards for the acceptance of surveying results quality, and the grade requirements of corresponding projects. The surveying report is the legally mandated presentation of surveying results, and its compliance directly affects the legality and reliability of its applications in the relevant fields.

[0003] The existing geodetic surveying technology system has the following objective defects in practical applications: 1. The implementation of surveying and mapping standards relies on manual interpretation and process control. Different operators and different work procedures have different understandings and execution standards of the standard clauses, which makes it impossible to achieve unified and quantitative implementation of the standard requirements throughout the entire surveying and mapping operation process, and easily leads to defects in the compliance of the results.

[0004] 2. Pose calculation of multi-source surveying data often adopts a fixed-weight fusion calculation model, which cannot dynamically adjust the calculation strategy according to the accuracy requirements of the corresponding project specifications and the project level. This can easily lead to problems such as the calculation accuracy not meeting the specifications or excessive pursuit of accuracy resulting in reduced work efficiency.

[0005] 3. The surveying and mapping data processing procedures such as point cloud denoising, land feature classification, and terrain parameter extraction are independent of each other. The processing process is not directly constrained by surveying and mapping standards. The processing results need to be manually verified multiple times before they can be transferred. The process is fragmented and compliance control is difficult.

[0006] 4. The compliance verification of surveying and mapping results often adopts the final manual verification after the report is generated, without setting up a compliance verification and correction mechanism in the corresponding work process, which easily leads to the accumulation of errors in multiple processes; at the same time, the data items in the report are not linked to the original data, processing process, and standard clauses, making it impossible to accurately locate the process and define responsibilities when compliance issues occur. Summary of the Invention

[0007] The purpose of this invention is to provide an artificial intelligence-based geodetic surveying report generation system, which realizes the quantitative breakdown of surveying and mapping specifications and the unified execution of the entire operation process, constructs a dynamic pose calculation mechanism adapted to specification requirements, establishes a multi-task data processing system under specification constraints, realizes compliance verification and correction within the operation process, and the corresponding binding of report data with original data and specification clauses.

[0008] To solve the above-mentioned technical problems, the technical solution provided by the present invention is as follows: An AI-based geodetic surveying report generation system includes a surveying specification parsing engine, a multi-source surveying data acquisition module, an adaptive pose calculation module, a perceptron learning module, a compliance verification and report generation module, and a penetrating responsibility traceability module. The specification quantification rule set output by the surveying specification parsing engine is synchronously transmitted to the multi-source surveying data acquisition module, the adaptive pose calculation module, the perceptron learning module, and the compliance verification and report generation module. The multi-source surveying data acquisition module synchronously transmits the acquired multi-source spatial data and all acquisition parameters to the adaptive pose calculation module and the penetrating responsibility traceability module. The module transmits the processed point cloud data to the compliance verification and report generation module. The compliance verification and report generation module transmits the point cloud data that fails the verification back to the adaptive pose calculation module. After completing the compliance verification of the point cloud data based on the standardized quantization rule set, the module transmits the qualified point cloud data to the perceptual meta-learning module. The perceptual meta-learning module transmits the processed result to the compliance verification and report generation module. The compliance verification and report generation module transmits the processing result that fails the verification back to the perceptual meta-learning module, and transmits the content of the verified report and the full-link compliance verification record synchronously to the penetrating responsibility traceability module.

[0009] Furthermore, the surveying and mapping specification parsing engine receives national standards, industry specifications, and project-level requirements for geodetic surveying and mapping. It breaks down these texts into accuracy constraint rules, content compilation rules, compliance verification rules, and responsibility attribution rules, forming a quantifiable set of specification rules that matches the project level. This set of quantifiable specification rules is then synchronously transmitted to the multi-source surveying and mapping data acquisition module, adaptive pose calculation module, perceptual element learning module, compliance verification and report generation module, and penetrating responsibility tracing module. The level adaptation calculation formula for the quantifiable specification rule set is: ; in, The total weight is determined by a set of standardized quantitative rules that match the project level. To adapt weight coefficients to the level of precision constraint rules, To assign weight coefficients to the levels of content compilation rules, To adapt the weighting coefficients to the levels of compliance verification rules, To assign weight coefficients to the levels of the liability attribution rules, The quantized feature vector of the precision constraint rule. Quantitative feature vectors for content compilation rules This is the quantized feature vector of the compliance verification rules. This is the quantitative feature vector of the responsibility attribution rule.

[0010] Furthermore, the adaptive pose calculation module receives the standard quantization rule set transmitted by the surveying and mapping standard analysis engine and the multi-source spatial data transmitted by the multi-source surveying and mapping data acquisition module. It completes time alignment of the multi-source data based on the timestamps of the GNSS data. Based on the project level and accuracy requirements in the standard quantization rule set, it dynamically adjusts the calculation weights of the multi-source data, completing three-level pose calculations (inter-frame, intra-frame, and global) and global error closed-loop compensation. The point cloud data with unified spatiotemporal markers is then transmitted to the compliance verification and report generation module. The objective function for dynamic weighted pose calculation is: ; in, The total residual for pose calculation, For dynamic weighting coefficients of GNSS observation data, For dynamic weighting coefficients of inertial navigation observation data, For the dynamic weighting coefficients of lidar point cloud observation data, To standardize the dynamic weighting coefficients of constraint terms, the weighting coefficients are dynamically adjusted according to the item level of the standardized quantification rule set. The residual between the GNSS pose observations and the calculated pose is... The residuals between the pre-integrated observations and the calculated pose in inertial navigation. The residuals for inter-frame matching pose and pose calculation of lidar point cloud are used. To solve the matching residual between pose and the precision constraint rules in the standardized quantization rule set.

[0011] Furthermore, the perceptual meta-learning module receives the standardized quantification rule set transmitted by the surveying and mapping standard parsing engine and the verified point cloud data transmitted by the compliance verification and report generation module. It employs a multi-task meta-learning network with embedded standardized causal intervention. This network shares a backbone feature extraction network and adjusts the feature extraction weights using the standardized quantification rule set as causal intervention terms. Simultaneously, it outputs point cloud denoising results, land cover classification results, terrain parameter results, and scene classification results. The total loss of the multi-task meta-learning network is calculated as follows: ; in, This represents the total loss value during network training. The balancing weights for the loss term in the point cloud denoising task. The balancing weights for the loss terms in the land cover classification task. The balancing weights for the loss term in the terrain parameter inversion task. Balanced weights for the loss terms in the scene classification task. To standardize the balancing weights of the regularized loss term for causal intervention, To standardize the balancing weights of the level-adaptive meta-learning loss term, The L2 loss value for point cloud denoising task. The cross-entropy loss value for the land cover classification task. This represents the smoothed L1 loss value for the terrain parameter inversion task. The cross-entropy loss value for scene classification tasks. To standardize the regularized loss value for causal intervention, To standardize the meta-learning loss value for level adaptation.

[0012] Furthermore, based on the output scene classification results and the standardized quantization rule set, the perceptual meta-learning module dynamically adjusts the balance weights of each loss term in the multi-task meta-learning network. For mountain scenes, it increases the loss weight of the terrain parameter inversion task; for urban scenes, it increases the loss weight of the ground feature classification task; and for linear engineering scenes, it increases the loss weight of the point cloud denoising task. The processed results with adjusted weights are then transmitted to the compliance verification and report generation module.

[0013] Furthermore, the compliance verification and report generation module receives the set of standardized quantitative rules transmitted by the surveying and mapping standard parsing engine, and performs single-stage compliance verification on the point cloud data transmitted by the adaptive pose calculation module and the processing results transmitted by the perceptual element learning module. Data that passes verification is transmitted to the next stage, while data that fails verification is transmitted back to the corresponding module for parameter re-optimization. The deviation calculation formula for single-stage compliance verification is: ; in, The relative deviation between the output value and the standard value of the process. This is the calculated output value for the current stage. To standardize the standard thresholds corresponding to the set of quantitative rules.

[0014] Furthermore, the compliance verification and report generation module maps the qualified processing results into structured feature vectors that conform to the standardized quantitative rule set, and binds them one by one with the corresponding standard clauses and precision requirements in the standardized quantitative rule set, generating report content directly related to the standard clauses. After performing the final compliance verification on the generated report content, the verified report content and the full-link compliance verification record are synchronously transmitted to the penetrating responsibility traceability module.

[0015] Furthermore, the penetrating responsibility tracing module receives the standardized quantitative rule set transmitted by the surveying and mapping standard parsing engine, the raw data and full collection parameters transmitted by the multi-source surveying and mapping data acquisition module, and the full-link compliance verification records and report content transmitted by the compliance verification and report generation module. It constructs a full-link error propagation chain and a responsibility attribution chain, assigning a unique association identifier embedded with a unique number corresponding to the standard clause to each data item in the report content. This unique association identifier is bound one-to-one with the raw data, processing parameters, standard clauses, compliance verification records, and responsibility attribution information of the corresponding data item. The module outputs a geodetic surveying compliance report with a full-link tracing identifier. The calculation formula for the full-link error propagation chain is: ; in, To report the total error of the data items, This refers to the error components in the data acquisition process. For the error components in the spatiotemporal pose calculation process, This refers to the error component in the feature extraction process. This refers to the error components in the report generation process.

[0016] Furthermore, the multi-source mapping data acquisition module receives the set of standardized quantitative rules transmitted by the mapping specification analysis engine. Based on the accuracy constraint rules and project level therein, it adjusts the point cloud acquisition frequency, scanning angle, and flight path parameters, and synchronously transmits the acquired multi-source spatial data and all acquisition parameters to the adaptive pose calculation module and the penetrating responsibility traceability module. The multi-source spatial data includes lidar data, GNSS data, and inertial navigation data.

[0017] The advantages of this invention compared to the prior art are: This invention uses a surveying and mapping specification analysis engine to quantitatively break down and adapt national surveying and mapping standards, industry specifications, and project level requirements, thereby achieving unified execution of surveying and mapping specification requirements throughout the entire surveying and mapping operation process and eliminating the execution scale deviation caused by manual interpretation of specifications.

[0018] This invention uses an adaptive pose calculation module to dynamically adjust the calculation weights of multi-source data based on a standardized quantization rule set, completing three levels of pose calculation and global error compensation: inter-frame, intra-frame, and global. This achieves direct adaptation of the pose calculation process to the project's accuracy requirements and ensures that the calculation results meet the standard constraints.

[0019] This invention achieves simultaneous processing of multiple tasks, including point cloud denoising, ground feature classification, terrain parameter inversion, and scene classification, by embedding a multi-task meta-learning network with standardized causal intervention. At the same time, it can dynamically adjust task weights based on scene type and standard requirements, thus achieving deep binding between the data processing process and standard requirements.

[0020] This invention achieves compliance verification and parameter adjustment for corresponding processes in surveying and mapping operations through a single-process compliance verification and non-conforming data feedback correction mechanism. By calculating the error transmission throughout the entire process and constructing a unique associated identifier, it realizes a one-to-one correspondence between surveying and mapping report data items and original data, processing parameters, and standard clauses, and clarifies the error components and responsibility attribution for the corresponding processes. Attached Figure Description

[0021] The accompanying drawings, which form part of this application, are used to provide a further understanding of the application and to make other features, objects, and advantages of the application more apparent. The illustrative embodiments and descriptions of this application are used to explain the application and do not constitute an undue limitation of the application.

[0022] In the attached diagram: Figure 1 This is a system framework diagram of the artificial intelligence-based geodetic mapping report generation system in Example 1. Detailed Implementation

[0023] The following detailed description of the embodiments is used to illustrate the principles of this application, but should not be used to limit the scope of this application. That is, the artificial intelligence-based geodetic mapping report generation system of this application is not limited to the described embodiments.

[0024] The present invention will be further described below with reference to embodiments.

[0025] Example 1 like Figure 1 As shown, an AI-based geodetic surveying report generation system includes a surveying specification parsing engine, a multi-source surveying data acquisition module, an adaptive pose calculation module, a perceptron learning module, a compliance verification and report generation module, and a penetrating responsibility traceability module. The specification quantification rule set output by the surveying specification parsing engine is synchronously transmitted to the multi-source surveying data acquisition module, the adaptive pose calculation module, the perceptron learning module, and the compliance verification and report generation module. The multi-source surveying data acquisition module synchronously transmits the acquired multi-source spatial data and all acquisition parameters to the adaptive pose calculation module and the penetrating responsibility traceability module. The calculation module transmits the processed point cloud data to the compliance verification and report generation module. The compliance verification and report generation module transmits the point cloud data that fails the verification back to the adaptive pose calculation module. After completing the compliance verification of the point cloud data based on the standardized quantization rule set, the module transmits the qualified point cloud data to the perceptual meta-learning module. The perceptual meta-learning module transmits the processed result to the compliance verification and report generation module. The compliance verification and report generation module transmits the processing result that fails the verification back to the perceptual meta-learning module, and transmits the content of the verified report and the full-link compliance verification record synchronously to the penetrating responsibility traceability module.

[0026] In a specific embodiment, the system is deployed on an industrial-grade ground processing workstation supporting airborne mapping. The workstation is configured with an Intel Xeon W-1370 processor, 32GB ECC running memory, an NVIDIA RTX A2000 graphics card, and runs the Ubuntu20.04 LTS operating system. The modules rely on the topic communication mechanism of the ROS Noetic framework to complete data interaction. The data transmission uses the PCD point cloud format and the TXT parameter format, and the transmission baud rate is set to 115200.

[0027] After semantic decomposition and quantization processing, the mapping specification analysis engine generates a specification quantization rule set, which is synchronously sent to the multi-source mapping data acquisition module, the adaptive pose calculation module, the perception element learning module, and the compliance verification and report generation module in the form of an XML configuration file. After receiving the rule set, each module automatically loads it into the local configuration directory as the execution criterion for module operations, unifying the specification execution standards for the entire process.

[0028] The multi-source mapping data acquisition module transmits the multi-source spatial data and acquisition full parameters collected on-site to the adaptive pose calculation module and the penetration-type responsibility traceability module in real time through a 4G industrial router. After receiving the data, the penetration-type responsibility traceability module immediately completes double backups on the local and cloud sides, and the retention format of the original data is the same as the acquisition format.

[0029] After the adaptive pose calculation module completes the processing of the point cloud data, it transmits the point cloud data with spatio-temporal tags to the compliance verification and report generation module in the PCD format. This module completes automatic verification according to the specification quantization rule set. The unqualified point cloud data is transmitted back to the adaptive pose calculation module in the original format, and at the same time, a verification deviation detail document is attached. The calculation module adjusts the calculation parameters according to the details and processes them again. The qualified point cloud data is transmitted to the perception element learning module in the standardized PCD format for feature extraction and processing operations.

[0030] The processing result of the perception element learning module is transmitted to the compliance verification and report generation module in the JSON format. The unqualified result is transmitted back to the perception element learning module. The module adjusts the network weights according to the details and recalculates. The data that passes the verification generates a report in the DOCX format and is transmitted to the penetration-type responsibility traceability module together with the full-process compliance verification record. The traceability module completes the association and binding of the report and the record after receiving it.

[0031] Preferably, the surveying and mapping specification parsing engine receives national standards, industry specifications, and project-level requirements for geodetic surveying and mapping. It then breaks down the text into accuracy constraint rules, content compilation rules, compliance verification rules, and responsibility attribution rules, forming a quantifiable set of specification rules that matches the project level. This set of quantifiable specification rules is then synchronously transmitted to the multi-source surveying and mapping data acquisition module, the adaptive pose calculation module, the perceptual element learning module, the compliance verification and report generation module, and the penetrating responsibility tracing module. The level adaptation calculation formula for the quantifiable specification rule set is: ; in, The total weight is determined by a set of standardized quantitative rules that match the project level. To adapt weight coefficients to the level of precision constraint rules, To assign weight coefficients to the levels of content compilation rules, To adapt the weighting coefficients to the levels of compliance verification rules, To assign weight coefficients to the levels of the liability attribution rules, The quantized feature vector of the precision constraint rule. Quantitative feature vectors for content compilation rules This is the quantized feature vector of the compliance verification rules. This is the quantitative feature vector of the responsibility attribution rule.

[0032] In a specific embodiment, the input to the surveying and mapping specification parsing engine is the currently valid legal text applicable to Class B 1:2000 urban cadastral surveying and mapping projects, including GB / T 7930-2008, GB / T 24356-2009, CH / T 8024-2011, and the Class B operation level requirements specified in the project approval document. The input text is imported into the engine in PDF format. The engine has a built-in PDF text parsing plugin that can automatically recognize the text format and extract the content.

[0033] The engine uses a BERT-base-Chinese pre-trained semantic parsing network. After fine-tuning the surveying and mapping industry standard text, the network performs sentence segmentation, entity recognition, clause extraction, and quantification on the input text. The text is decomposed into four core rules: accuracy constraint rules, content compilation rules, compliance verification rules, and responsibility attribution rules. For each rule, core elements such as constraint thresholds, scope of application, and execution requirements are extracted and converted into 128-dimensional quantified feature vectors that can be recognized and calculated by computers. Then, the level is combined with the project level to complete the level adaptation and form a set of standardized quantitative rules. The rule set is synchronously distributed to all modules in the entire process in the form of an XML configuration file, while the original file is stored locally on the surveying and mapping standard parsing engine.

[0034] This embodiment adopts The total weight of the set of standardized quantitative rules, matched to the project level, is used to integrate the quantitative characteristics of the four rule categories. , , , The weight coefficients representing the level adaptation of the four rule categories are respectively determined according to the weight allocation requirements of the quality elements of surveying and mapping results in GB / T 24356-2009. For this Class B 1:2000 urban cadastral surveying and mapping project, the following settings are established: , , , .use , , , The four types of rules are represented by quantitative feature vectors, and each dimension of the vector corresponds to the specific constraint requirements in the normative clauses, thereby realizing the quantitative expression and calculation of the normative requirements.

[0035] Preferably, the adaptive pose calculation module receives the standard quantization rule set transmitted by the surveying and mapping standard analysis engine and the multi-source spatial data transmitted by the multi-source surveying and mapping data acquisition module. It completes time alignment of the multi-source data based on the timestamps of the GNSS data. Based on the project level and accuracy requirements in the standard quantization rule set, it dynamically adjusts the calculation weights of the multi-source data, completing three-level pose calculation (inter-frame, intra-frame, and global) and global error closed-loop compensation. The point cloud data with unified spatiotemporal markers is then transmitted to the compliance verification and report generation module. The objective function for dynamic weighted pose calculation is: ; in, The total residual for pose calculation, For dynamic weighting coefficients of GNSS observation data, For dynamic weighting coefficients of inertial navigation observation data, For the dynamic weighting coefficients of lidar point cloud observation data, To standardize the dynamic weighting coefficients of constraint terms, the weighting coefficients are dynamically adjusted according to the item level of the standardized quantification rule set. The residual between the GNSS pose observations and the calculated pose is... The residuals between the pre-integrated observations and the calculated pose in inertial navigation. The residuals for inter-frame matching pose and pose calculation of lidar point cloud are used. To solve the matching residual between pose and the precision constraint rules in the standardized quantization rule set.

[0036] In a specific embodiment, the adaptive pose calculation module is built on the ROS Noetic framework and integrates the LIO-SAM laser inertial odometry calculation framework. After receiving the standard quantification rule set issued by the surveying and mapping standard analysis engine and the multi-source spatial data transmitted by the multi-source surveying and mapping data acquisition module, the module starts the data preprocessing process. Using the PPS second pulse timestamp of the GNSS data as a reference, the module performs time alignment of the lidar data and inertial navigation data through linear interpolation. The interpolation step size is set to 0.001s to control the time synchronization error within 1ms. After the time alignment is completed, the module reads the project level and accuracy requirement indicators in the standard quantification rule set, dynamically adjusts the calculation weight of the multi-source data according to the indicators, and sequentially performs inter-frame pose calculation, intra-frame pose calculation, and global pose calculation. During the calculation process, global error closed-loop compensation is performed simultaneously. The compensation algorithm adopts the Kalman filter algorithm to correct systematic errors and random errors in real time. Finally, the module outputs point cloud data with WGS84 geodetic coordinate system spatiotemporal markers and transmits it to the compliance verification and report generation module in standardized PCD format.

[0037] This embodiment adopts Characterizing the total residual of pose calculation, the entire calculation process is based on Minimization is the core objective, and the following approach is adopted. , , , These represent the dynamic weighting coefficients for GNSS observation data, inertial navigation observation data, lidar point cloud observation data, and specification constraints, respectively. The coefficient values ​​are dynamically adjusted according to the project level and accuracy requirements. For this Class B 1:2000 urban cadastral surveying project, the following parameters are set: , , , ,use , , , These parameters represent the matching residuals between each observation and the calculated pose. These parameters can intuitively reflect the degree of deviation between the calculation results from each data source and the ideal pose. When the residual exceeds the threshold, the module automatically increases the weight of the corresponding specification constraint.

[0038] Preferably, the perceptual meta-learning module receives the standardized quantification rule set transmitted by the surveying and mapping standard parsing engine and the verified point cloud data transmitted by the compliance verification and report generation module. It employs a multi-task meta-learning network with embedded standardized causal intervention. This multi-task meta-learning network shares a backbone feature extraction network, adjusts the feature extraction weights using the standardized quantification rule set as causal intervention terms, and simultaneously outputs point cloud denoising results, land feature classification results, terrain parameter results, and scene classification results. The total loss of the multi-task meta-learning network is calculated as follows: ; in, This represents the total loss value during network training. The balancing weights for the loss term in the point cloud denoising task. The balancing weights for the loss terms in the land cover classification task. The balancing weights for the loss term in the terrain parameter inversion task. Balanced weights for the loss terms in the scene classification task. To standardize the balancing weights of the regularized loss term for causal intervention, To standardize the balancing weights of the level-adaptive meta-learning loss term, The L2 loss value for point cloud denoising task. The cross-entropy loss value for the land cover classification task. This represents the smoothed L1 loss value for the terrain parameter inversion task. The cross-entropy loss value for scene classification tasks. To standardize the regularized loss value for causal intervention, To standardize the meta-learning loss value for level adaptation.

[0039] In a specific embodiment, the perceptual meta-learning module is built on the PyTorch 1.12 framework and adopts a multi-task meta-learning network with the PointNet++ point cloud feature extraction network as the backbone. After training with 100,000 sets of surveying point cloud data, the model accuracy and generalization ability meet the requirements of Class B 1:2000 urban cadastral surveying. The module receives point cloud data that has been deemed qualified by the compliance verification and report generation module, and simultaneously reads the locally loaded standardized quantization rule set. The network is set to share the backbone feature extraction network, and voxel downsampling is used to perform dimensionality reduction processing on the input point cloud. The voxel size is set to 0.1m. The standardized quantization rule set is used as the causal intervention item, and the feature extraction weight allocation is dynamically adjusted through the channel attention mechanism to match the accuracy and focus required by the standard. The module simultaneously outputs point cloud denoising results, land feature classification results, terrain parameter results, and scene classification results. The four types of results are encapsulated in JSON format to realize the parallel execution of multiple surveying data processing tasks.

[0040] This embodiment adopts The total loss value representing network training and inference is used to iteratively optimize network parameters throughout the process. Minimize execution. , , , The weights of the loss terms for the four processing tasks—point cloud denoising, ground feature classification, terrain parameter inversion, and scene classification—are respectively represented, and the following methods are used: , The initial settings represent the balance weights of the normative causal intervention regularization and normative level adaptation meta-learning loss terms, respectively, for this Class B 1:2000 urban cadastral surveying project. , , , , , ,use , , , , , Each of these represents a quantified value of a loss term, where L2 loss is used for calculation. and Calculated using cross-entropy loss, Using smoothed L1 loss calculation, each value can directly reflect the execution deviation of the corresponding processing task and the degree of fit between the specification constraints and the level adaptation.

[0041] Preferably, the perceptual meta-learning module dynamically adjusts the balance weights of each loss term in the multi-task meta-learning network based on the output scene classification results and the standardized quantization rule set. For mountain scenes, the loss weight of the terrain parameter inversion task is increased; for urban scenes, the loss weight of the ground feature classification task is increased; and for linear engineering scenes, the loss weight of the point cloud denoising task is increased. The processed results with adjusted weights are then transmitted to the compliance verification and report generation module.

[0042] In a specific embodiment, after the perceptual meta-learning module outputs the scene classification results, it reads the scene-specific accuracy requirement indicators from the standardized quantization rule set and dynamically adjusts the weights of the loss terms in the multi-task meta-learning network based on the classification results. The weight adjustment is completed by the module's built-in weight optimization plugin, and the adjustment results are immediately loaded into the network and applied to the inference process. In mountainous surveying scenarios, terrain elevation, slope, and aspect are the core output indicators; the module increases the loss weights for terrain parameter inversion tasks. In urban scenarios, where feature boundary and category recognition are the core, the module increases the loss weights for feature classification tasks. Linear engineering scenarios have higher requirements for point cloud purity; the module increases the loss weights for point cloud denoising tasks.

[0043] Adjustments were made to the urban built-up area sub-scenario of this project. , , Adjustments were made for the mountainous and hilly sub-scenes. , , For the linear engineering sub-scenario of road pipelines, adjustments were made. , , The weights of the remaining loss items remain unchanged. After the weights are adjusted, the module will transmit the processing results to the compliance verification and report generation module in JSON format, so that the data processing results match the surveying and mapping operation requirements and standard accuracy requirements of different scenarios.

[0044] Preferably, the compliance verification and report generation module receives the set of standardized quantitative rules transmitted by the surveying and mapping standard parsing engine, performs single-stage compliance verification on the point cloud data transmitted by the adaptive pose calculation module and the processing results transmitted by the perceptual element learning module, respectively, transmits the data that passes the verification to the next stage, and transmits the data that fails the verification back to the corresponding module to complete parameter re-optimization. The deviation calculation formula for single-stage compliance verification is as follows: ; in, The relative deviation between the output value and the standard value of the process. This is the calculated output value for the current stage. To standardize the standard thresholds corresponding to the set of quantitative rules.

[0045] In a specific embodiment, the compliance verification and report generation module is built on Python 3.8 and integrates the GDAL 3.4.1 geographic information data processing library. After receiving the standardized quantitative rule set, the module parses the verification standard thresholds for each process and stores them in the local verification parameter library. The module performs independent single-process compliance verification on the point cloud data transmitted by the adaptive pose calculation module and the processing results transmitted by the perceptual meta-learning module. The two types of data adopt a parallel processing mechanism, matching the corresponding standardized thresholds in the verification parameter library. The point cloud data verification extracts three core indicators: point cloud density, plane mean square error, and elevation mean square error. The processing result verification extracts three core indicators: land feature classification accuracy, terrain parameter error, and point cloud denoising rate. The built-in deviation calculation plugin completes the batch calculation of relative deviation.

[0046] Qualified data is transmitted to the next workflow in its original format, while unqualified data is transmitted back to the corresponding generation module in its original format. A detailed CSV document of the verification deviation is also sent along with the data. The details include information such as deviation indicators, deviation values, and standard thresholds. The generation module adjusts its parameters based on the details and reprocesses the data. The new processing results are then sent back to the module for verification until the data meets the standard requirements.

[0047] This embodiment adopts Characterizing the relative deviation between the output value of a process and the standard value, this parameter is a core criterion for compliance verification. The module uses... Using this as the sole criterion, without introducing any human intervention, and adopting... The actual calculated values ​​characterizing the output results of the process are calculated in strict accordance with the legally mandated calculation methods of the surveying and mapping industry. Characterize the standard thresholds corresponding to the specification quantization rule set. The threshold sources are national standards and industry specifications such as GB / T 7930-2008 and GB / T 24356-2009. For this Class B 1:2000 urban cadastral surveying and mapping project, the are set as the point cloud density of 16 points / ㎡, the plane medium error of 0.2m, and the elevation medium error of 0.1m. When all the verification indicators are less than 10%, it can be determined that the verification is qualified.

[0048] Preferably, the compliance verification and report generation module maps the processed results that pass the verification into structured feature vectors that conform to the specification quantization rule set, binds them one by one with the corresponding specification clauses and accuracy requirements in the specification quantization rule set, generates report content directly associated with the specification clauses, and after performing the final compliance verification on the generated report content, synchronously transmits the report content that passes the verification and the full-link compliance verification record to the penetrative responsibility traceability module.

[0049] In a specific embodiment, the compliance verification and report generation module maps the processed results transmitted by the perception element learning module and passing the verification into structured feature vectors that meet the requirements of the specification quantization rule set through the GDAL 3.4.1 geographic information data processing library. The dimension of the feature vector is set to 64 dimensions, each dimension corresponding to specific result indicators in the surveying and mapping specifications. The vector values are consistent with the accuracy and units required by the specifications, and are bound one by one with the corresponding clauses and accuracy requirements in the specification quantization rule set. The binding adopts an association method combining clause numbers and indicator values, so that the processed results are accurately corresponding to specific specification clauses and legal bases.

[0050] Based on the bound structured feature vectors, the module calls the python-docx library to automatically generate a geodetic surveying report. The report format follows the requirements of GB / T 24356-2009, including core chapters such as project overview, surveying and mapping basis, result data, accuracy analysis, and compliance verification. Each data result in the report is marked with the corresponding specification clause number, establishing a direct association between the report content and the specification clauses.

[0051] After the report is generated, the module performs the final compliance verification on the report. The verification adopts a full-index coverage method, covering four core dimensions: report content integrity, data accuracy compliance, clause correspondence accuracy, and format standardization. Corresponding verification items are set for each dimension. The report that passes the final verification is stored in DOCX format and transmitted to the penetrative responsibility traceability module together with the full-process compliance verification record. The full-process compliance verification record includes information such as verification time, verification indicators, verification results, and deviation details, constituting the verification traceability record of the report results.

[0052] Preferably, the penetrating responsibility tracing module receives the standardized quantitative rule set transmitted by the surveying and mapping standard parsing engine, the raw data and full collection parameters transmitted by the multi-source surveying and mapping data acquisition module, and the full-link compliance verification records and report content transmitted by the compliance verification and report generation module. It constructs a full-link error propagation chain and a responsibility attribution chain, assigning a unique association identifier embedded with a unique number corresponding to a standard clause to each data item in the report content. This unique association identifier is bound one-to-one with the raw data, processing parameters, standard clauses, compliance verification records, and responsibility attribution information of the corresponding data item. The module outputs a geodetic surveying compliance report with a full-link tracing identifier. The calculation formula for the full-link error propagation chain is: ; in, To report the total error of the data items, This refers to the error components in the data acquisition process. For the error components in the spatiotemporal pose calculation process, This refers to the error component in the feature extraction process. This refers to the error components in the report generation process.

[0053] In a specific embodiment, the penetrating responsibility tracing module is built on a MySQL 8.0 relational database. The module receives, through a database interface, the quantitative rule set of the surveying and mapping specification parsing engine, the raw data and full collection parameters uploaded by the multi-source surveying and mapping data acquisition module, and the full-process compliance verification records and final report content output by the compliance verification and report generation module. All data is structured and stored in the database, and a data association table is established according to the work process. Based on the full-process data flow and the data association table in the database, the module constructs a full-process error propagation chain and a responsibility attribution chain. The error propagation chain adopts a tree structure to represent the path of error generation, propagation, and accumulation in each work process. The responsibility attribution chain, based on surveying and mapping industry work specifications, maps the work responsibility of each process to specific work positions and equipment. The module assigns a unique association identifier to each data item in the report. The identifier uses a 16-bit encoding rule combining the project number, specification clause number, data item number, and timestamp, ensuring uniqueness and non-repeatability. The identifier embeds the corresponding specification clause number and is bound one-to-one with the raw collection data, processing parameters of each process, matching specification clauses, full-process compliance verification records, and responsibility attribution information in the database. This identifier allows you to retrieve all relevant information about a data item from data collection to report generation.

[0054] The module ultimately outputs a geodetic surveying compliance report with a unique identifier. This identifier is a QR code placed in the header of each page of the report; scanning the code allows access to the full-process traceability information of the corresponding page's data items. This embodiment uses... This parameter characterizes the total error of the reported data items. It directly reflects the final accuracy of the data items, and its value is the cumulative value of the error components from each process step. , , , The module characterizes the error components of the four workflows: data acquisition, spatiotemporal pose calculation, feature extraction, and report generation. It uses legally mandated surveying error calculation methods to... The process involves precise breakdown, quantifying the impact of each workflow on the final accuracy of the data items, and incorporating the breakdown results into the traceability information.

[0055] Preferably, the multi-source surveying and mapping data acquisition module receives the standard quantification rule set transmitted by the surveying and mapping standard analysis engine. Based on the accuracy constraint rules and project level therein, it adjusts the point cloud acquisition frequency, scanning angle, and flight path parameters, and synchronously transmits the acquired multi-source spatial data and all acquisition parameters to the adaptive pose calculation module and the penetrating responsibility traceability module. The multi-source spatial data includes lidar data, GNSS data, and inertial navigation data.

[0056] In a specific embodiment, the multi-source mapping data acquisition module is mounted on the DJI M300 RTK rotary-wing mapping drone platform. The drone has a maximum payload of 2.7kg and a maximum flight time of 55 minutes. It is suitable for high-precision cadastral mapping operations in small and medium-sized areas and can meet the needs of mapping complex urban terrain.

[0057] The UAV is equipped with a Zenmuse L2 airborne lidar, a dual-frequency GNSS positioning module, and a six-axis IMU inertial navigation unit. Before operation, the UAV and each sensor are jointly calibrated. The calibration includes time synchronization calibration, spatial position calibration, and sensor accuracy calibration. The calibration results meet the requirements of CH / T 8024-2011. A CORS ground reference station is deployed in the operation area. The reference station and the UAV's GNSS module achieve real-time differential positioning to improve positioning accuracy.

[0058] The module receives a set of standardized quantitative rules through an onboard embedded system, analyzes the accuracy constraints and project level requirements, and adjusts the point cloud acquisition frequency, LiDAR scanning angle, and UAV flight path parameters accordingly to match the acquisition parameters with the project's accuracy requirements. For this Class B 1:2000 urban cadastral surveying project, the LiDAR point cloud acquisition frequency is set to 240kHz, the scanning angle to ±70°, the UAV flight path relative altitude to 120m, the lateral overlap rate to 60%, the forward overlap rate to 80%, and the flight speed to 8m / s. This parameter combination meets the project's point cloud density and accuracy requirements.

[0059] The module collects multi-source spatial data including lidar point cloud data, GNSS dual-frequency positioning data, and inertial navigation attitude data. During the acquisition process, it synchronously records all acquisition parameters, such as equipment internal parameters, track parameters, acquisition timestamps, ambient temperature and humidity, and air pressure values. The multi-source spatial data is in PCD format, and the acquisition parameters are in TXT format. The data is synchronously transmitted to the adaptive pose calculation module and the penetrating responsibility traceability module through a 4G industrial router, providing a complete spatiotemporal data source for pose calculation and preserving the original acquisition evidence for responsibility determination.

[0060] It should be noted that the combination of the technical features in this case is not limited to the combination methods described in the claims of this case or the combination methods described in the specific embodiments. All technical features described in this case can be freely combined or combined in any way, unless they contradict each other.

[0061] It should also be noted that the embodiments listed above are merely specific embodiments of the present invention. Obviously, the present invention is not limited to the above embodiments, and similar changes or modifications made thereto are those that can be directly derived or easily conceived by those skilled in the art from the content disclosed in the present invention, and should all fall within the protection scope of the present invention.

Claims

1. An artificial intelligence-based geodetic surveying report generation system, characterized in that; The system includes a surveying and mapping specification parsing engine, a multi-source surveying and mapping data acquisition module, an adaptive pose calculation module, a perceptron learning module, a compliance verification and report generation module, and a penetrating responsibility tracing module. The specification quantification rule set output by the surveying and mapping specification parsing engine is synchronously transmitted to the multi-source surveying and mapping data acquisition module, the adaptive pose calculation module, the perceptron learning module, and the compliance verification and report generation module. The multi-source surveying and mapping data acquisition module synchronously transmits the acquired multi-source spatial data and all acquisition parameters to the adaptive pose calculation module and the penetrating responsibility tracing module. The adaptive pose calculation module processes the point cloud data... The data is transmitted to the compliance verification and report generation module. The compliance verification and report generation module transmits the point cloud data that fails the verification back to the adaptive pose calculation module. After completing the compliance verification of the point cloud data based on the standardized quantization rule set, the point cloud data that passes the verification is transmitted to the perceptual meta-learning module. The perceptual meta-learning module transmits the processed result back to the compliance verification and report generation module. The compliance verification and report generation module transmits the processing result of the data that fails the verification back to the perceptual meta-learning module, and transmits the report content of the data that passes the verification and the full-link compliance verification record synchronously to the penetrating responsibility traceability module.

2. The artificial intelligence-based geodetic mapping report generation system according to claim 1, characterized in that: The surveying and mapping specification parsing engine receives national standards, industry specifications, and project level requirements for geodetic surveying and mapping. It breaks down the text into accuracy constraint rules, content compilation rules, compliance verification rules, and responsibility attribution rules, forming a quantifiable set of specification rules that matches the project level. This set of quantifiable specification rules is then synchronously transmitted to the multi-source surveying and mapping data acquisition module, adaptive pose calculation module, perceptual element learning module, compliance verification and report generation module, and penetrating responsibility tracing module. The level adaptation calculation formula for the quantifiable specification rule set is: ； in, The total weight is determined by a set of standardized quantitative rules that match the project level. To adapt weight coefficients to the level of precision constraint rules, To assign weight coefficients to the levels of content compilation rules, To adapt the weighting coefficients to the levels of compliance verification rules, To assign weight coefficients to the levels of the liability attribution rules, The quantized feature vector of the precision constraint rule. Quantitative feature vectors for compiling rules for content. This is the quantized feature vector of the compliance verification rules. This is the quantified feature vector of the responsibility attribution rule.

3. The artificial intelligence-based geodetic surveying report generation system according to claim 1, characterized in that: The adaptive pose calculation module receives the standard quantization rule set transmitted by the surveying and mapping standard analysis engine and the multi-source spatial data transmitted by the multi-source surveying and mapping data acquisition module. It completes time alignment of the multi-source data based on the timestamp of the GNSS data. Based on the project level and accuracy requirements in the standard quantization rule set, it dynamically adjusts the calculation weights of the multi-source data, completing three-level pose calculation (inter-frame, intra-frame, and global) and global error closed-loop compensation. The point cloud data with unified spatiotemporal markers is then transmitted to the compliance verification and report generation module. The objective function of the dynamic weighted pose calculation is: ； in, The total residual for pose calculation, For dynamic weighting coefficients of GNSS observation data, For dynamic weighting coefficients of inertial navigation observation data, For the dynamic weighting coefficients of lidar point cloud observation data, To standardize the dynamic weighting coefficients of constraint terms, the weighting coefficients are dynamically adjusted according to the item level of the standardized quantification rule set. The residual between the GNSS pose observations and the calculated pose is... The residuals between the pre-integrated observations and the calculated pose in inertial navigation. The residuals for inter-frame matching pose and pose calculation of lidar point cloud are used. To solve the matching residual between pose and the precision constraint rules in the standardized quantization rule set.

4. The artificial intelligence-based geodetic mapping report generation system according to claim 1, characterized in that: The perceptual meta-learning module receives the standard quantification rule set transmitted by the surveying and mapping standard parsing engine and the verified point cloud data transmitted by the compliance verification and report generation module. It employs a multi-task meta-learning network with embedded standard causal intervention. This multi-task meta-learning network shares a backbone feature extraction network, adjusts the feature extraction weights using the standard quantification rule set as causal intervention terms, and simultaneously outputs point cloud denoising results, land cover classification results, terrain parameter results, and scene classification results. The total loss of the multi-task meta-learning network is calculated as follows: ； in, This represents the total loss value during network training. The balancing weights for the loss term in the point cloud denoising task. The balancing weights for the loss terms in the land cover classification task. The balancing weights for the loss term in the terrain parameter inversion task. Balanced weights for the loss terms in the scene classification task. To standardize the balancing weights of the regularized loss term for causal intervention, To standardize the balancing weights of the level-adaptive meta-learning loss term, The L2 loss value for point cloud denoising task. The cross-entropy loss value for the land cover classification task. This represents the smoothed L1 loss value for the terrain parameter inversion task. The cross-entropy loss value for scene classification tasks. To standardize the regularized loss value for causal intervention, To standardize the meta-learning loss value for level adaptation.

5. The artificial intelligence-based geodetic surveying report generation system according to claim 4, characterized in that: The perceptual meta-learning module dynamically adjusts the weights of each loss term in the multi-task meta-learning network based on the output scene classification results and the standardized quantization rule set. For mountain scenes, it increases the loss weight of the terrain parameter inversion task; for urban scenes, it increases the loss weight of the ground feature classification task; and for linear engineering scenes, it increases the loss weight of the point cloud denoising task. The processed results with adjusted weights are then transmitted to the compliance verification and report generation module.

6. The artificial intelligence-based geodetic surveying report generation system according to claim 5, characterized in that: The compliance verification and report generation module receives the set of standardized quantitative rules transmitted by the surveying and mapping standard analysis engine. It performs single-stage compliance verification on the point cloud data transmitted by the adaptive pose calculation module and the processing results transmitted by the perceptual element learning module. Data that passes verification is transmitted to the next stage, while data that fails verification is transmitted back to the corresponding module for parameter re-optimization. The deviation calculation formula for the single-stage compliance verification is: ； in, The relative deviation between the output value and the standard value of the process. This is the calculated output value for the current stage. To standardize the standard thresholds corresponding to the set of quantitative rules.

7. The artificial intelligence-based geodetic surveying report generation system according to claim 6, characterized in that: The compliance verification and report generation module maps the qualified processing results into structured feature vectors that conform to the standardized quantitative rule set, and binds them one by one with the corresponding standard clauses and precision requirements in the standardized quantitative rule set. It generates report content that is directly related to the standard clauses. After performing the final compliance verification on the generated report content, the verified report content and the full-link compliance verification record are synchronously transmitted to the penetrating responsibility traceability module.

8. The artificial intelligence-based geodetic surveying report generation system according to claim 7, characterized in that: The penetrating responsibility tracing module receives the standardized quantitative rule set transmitted by the surveying and mapping standard analysis engine, the raw data and full collection parameters transmitted by the multi-source surveying and mapping data acquisition module, and the full-link compliance verification records and report content transmitted by the compliance verification and report generation module. It constructs a full-link error propagation chain and a responsibility attribution chain, assigning a unique association identifier embedded with a unique number corresponding to a standard clause to each data item in the report content. This unique association identifier is bound one-to-one with the raw data, processing parameters, standard clauses, compliance verification records, and responsibility attribution information of the corresponding data item. The module outputs a geodetic surveying compliance report with a full-link tracing identifier. The calculation formula for the full-link error propagation chain is: ； in, To report the total error of the data items, This refers to the error components in the data acquisition process. For the error components in the spatiotemporal pose calculation process, This refers to the error component in the feature extraction process. This refers to the error components in the report generation process.

9. The artificial intelligence-based geodetic surveying report generation system according to claim 8, characterized in that: The multi-source mapping data acquisition module receives the set of standardized quantitative rules transmitted by the mapping standard analysis engine. Based on the accuracy constraint rules and project level therein, it adjusts the point cloud acquisition frequency, scanning angle, and flight path parameters, and synchronously transmits the acquired multi-source spatial data and all acquisition parameters to the adaptive pose calculation module and the penetrating responsibility tracing module. The multi-source spatial data includes lidar data, GNSS data, and inertial navigation data.

Images