A stone statue-based migration path planning and positioning measurement system

Abstract

The application relates to the technical field of stone statue migration measurement, and discloses a stone statue migration path planning and positioning measurement system. A path planning prediction module of the system monitors geological signals and obtains geological deformation parameters, collects historical environment and time parameters to generate prediction path offset parameters; a positioning measurement screening module uses a laser scanner to collect terrain point cloud data, identifies actual path offset parameters according to the prediction path offset parameters and a positioning measurement classifier group; a migration dynamic optimization module triggers path reconstruction to generate optimization instructions according to the actual path offset parameters, collects real-time obstacle distribution data by using an industrial measuring instrument, and drives an obstacle avoidance classifier group to output actual obstacle distribution parameters; and a positioning calibration feedback module generates a path correction amount in combination with the two actual parameters, synchronously updates the geological deformation parameter weight coefficient of the path planning prediction module, and improves the safety and efficiency of stone statue migration.

Description

Technical Field

[0001] This invention relates to the field of stone statue migration measurement technology, specifically a stone statue migration path planning and positioning measurement system. Background Technology

[0002] In the protection and relocation of stone statues, these statues typically possess high historical, cultural, and artistic value. They are large in size and heavy, and the stability and safety requirements during relocation are extremely stringent. Traditional stone statue relocation operations rely heavily on manual experience combined with basic topographical survey data for route planning. This method struggles to fully capture the dynamic changes in the geological environment of the relocation area. For example, some relocation areas may have hidden geological instability issues, such as loose soil layers or underground cavities. Manual surveys or conventional geological monitoring methods cannot accurately and in real time obtain this geological deformation information, leading to deviations in the pre-planned relocation route during actual execution and increasing the risk of the statues tipping over or being damaged.

[0003] Traditional positioning and measurement methods often employ single measuring devices, such as total stations, for fixed-point measurements. This approach is not only inefficient and unable to quickly acquire complete topographic data of the relocation area, but its accuracy is also easily affected in complex terrain environments, such as areas with significant elevation differences and numerous obstacles. When the relocation path deviates and needs adjustment, traditional methods lack an effective dynamic optimization mechanism, often requiring the relocation operation to be paused and the path replanned and measured again, extending the relocation cycle and increasing manpower and time costs. Furthermore, during the traditional relocation of stone statues, the identification and avoidance of obstacles rely heavily on manual observation, making it difficult to grasp the distribution and changes of obstacles along the relocation path in real time. For example, temporary construction equipment or fallen stones may not be identified in time, potentially hindering the relocation process or even causing collision damage to the stone statues.

[0004] Traditional stone statue relocation systems lack effective feedback and integration mechanisms for data at each stage. Geological monitoring data, positioning measurement data, and obstacle information are independent of each other, failing to form a closed-loop parameter adjustment system. When deviations occur between the actual relocation path and the planned path, the deviation information cannot be promptly fed back to the path planning stage to adjust the weights of geological deformation parameters. This results in subsequent path planning still being based on inaccurate parameters, further exacerbating the path deviation problem and making it difficult to guarantee the safety and stability of the stone statue relocation process. Summary of the Invention

[0005] The purpose of this invention is to provide a system for planning and positioning the migration path of stone statues, in order to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides a system for planning and positioning a stone statue migration path, the system comprising:

[0007] The path planning and prediction module monitors geological signals in the target stone statue migration area, obtains geological deformation parameters, and collects historical environmental parameters and migration time parameters of the migration path to generate predicted path offset parameters.

[0008] The positioning measurement and filtering module collects terrain point cloud data of the target stone statue migration area using a laser scanner, filters the positioning measurement classifier group according to the predicted path offset parameters, and identifies and obtains the actual path offset parameters.

[0009] The migration dynamic optimization module triggers migration path reconstruction based on the actual path offset parameters, generates migration path optimization instructions, and collects real-time obstacle distribution data of the migration path through an industrial measuring instrument. Based on the migration path optimization instructions, it drives the obstacle avoidance classifier group to output the actual obstacle distribution parameters.

[0010] The positioning calibration feedback module combines the actual path offset parameters and the actual obstacle distribution parameters to generate a migration path correction amount, and synchronously updates the geological deformation parameter weight coefficients of the path planning prediction module.

[0011] Preferably, the path planning and prediction module includes:

[0012] The geological deformation monitoring unit monitors geological signals in the area where the target stone statue has migrated, and obtains geological deformation parameters.

[0013] The environmental parameter analysis unit collects historical environmental parameters and migration time parameters of the migration path and calculates the path environmental interference index.

[0014] The path offset prediction unit constructs a path offset predictor based on the random forest algorithm, inputs the geological deformation parameters and the path environment disturbance index, and outputs the predicted path offset parameters.

[0015] Preferably, the positioning measurement and screening module performs the following:

[0016] Topographic point cloud data of the target stone statue relocation area were collected using a laser scanner;

[0017] Based on the numerical range of the predicted path offset parameters, the corresponding terrain classifier group is matched;

[0018] The terrain point cloud data is input in parallel into multiple terrain classification paths in the terrain classifier group, and a set of terrain matching results is output.

[0019] The high-frequency offset patterns in the terrain matching result set are statistically analyzed to generate actual path offset parameters.

[0020] Preferably, the migration dynamic optimization module includes:

[0021] The path reconstruction unit triggers migration path topology reconstruction based on the actual path offset parameters and generates migration path optimization instructions.

[0022] The obstacle measurement unit collects real-time obstacle distribution data along the migration path using an industrial measuring instrument;

[0023] The obstacle avoidance unit activates the corresponding obstacle classifier group based on the migration path optimization instruction, inputs the real-time obstacle distribution data into multiple obstacle classification paths, and outputs the actual obstacle distribution parameters.

[0024] Preferably, the positioning calibration feedback module performs:

[0025] The actual path offset parameters are spatiotemporally aligned with the actual obstacle distribution parameters;

[0026] Calculate the path correction vector and generate the migration path correction amount;

[0027] The weighting coefficients of the geological deformation parameters in the path planning and prediction module are adjusted in reverse based on the migration path correction amount.

[0028] Preferably, the path offset prediction unit of the path planning prediction module adopts:

[0029] Historical path database stores historical geological deformation parameters and historical path environmental disturbance indexes for the migration of similar stone statues.

[0030] The dynamic weight allocator adjusts the contribution ratio of geological deformation parameters in the path offset predictor based on the weight coefficients returned by the positioning calibration feedback module.

[0031] Preferably, the terrain classifier group of the positioning measurement and screening module is constructed as follows:

[0032] Multiple convolutional neural network classifiers were trained based on historical topographic data of the migration area;

[0033] Each classifier group is bound to a specific path offset parameter range and contains multiple independent terrain classification paths;

[0034] The terrain classification path outputs a binary result showing the matching degree between terrain features and a preset offset pattern.

[0035] Preferably, the obstacle avoidance unit of the migration dynamic optimization module adopts:

[0036] A distributed obstacle classifier library, pre-stores classifier groups corresponding to different migration path optimization instructions;

[0037] A dynamic path loader calls the target classifier group from the classifier library according to the migration path optimization instructions;

[0038] The obstacle classification path outputs a binary classification result of the obstacle distribution density level.

[0039] Preferably, the positioning calibration feedback module includes:

[0040] The spatiotemporal alignment engine registers the spatial coordinates of the actual path offset parameters with the timestamps of the actual obstacle distribution parameters.

[0041] The vector synthesizer calculates the direction and magnitude of the path correction vector based on the registered parameters.

[0042] The weighted feedback channel maps the magnitude of the path correction vector to the weighted adjustment coefficient of the geological deformation parameters.

[0043] Preferably, the system further includes:

[0044] The migration path preloading module loads terrain reference data from the adjacent migration area database according to the migration path optimization instructions generated by the migration dynamic optimization module.

[0045] The protocol compatibility verification module verifies the coordinate transformation compatibility of the preloaded terrain reference data based on the terrain point cloud data format of the positioning measurement filtering module.

[0046] When the protocol compatibility verification fails, the path conflict resolution engine triggers the path planning and prediction module to regenerate the predicted path offset parameters.

[0047] Compared with the prior art, the beneficial effects of the present invention are:

[0048] This stone statue relocation path planning and positioning measurement system, through the inclusion of a path planning prediction module, can monitor geological signals in the target stone statue relocation area, acquire geological deformation parameters, and simultaneously collect historical environmental parameters and relocation time parameters of the relocation path to generate predicted path offset parameters. This process no longer relies on single-dimensional human experience or conventional monitoring methods, but rather, through the integrated analysis of multi-dimensional parameters, it anticipates potential path offsets in advance. This makes path planning more forward-looking, effectively addressing dynamic changes in the geological environment of the relocation area, reducing the risk of path deviation due to incomplete geological information, and providing a more comprehensive and accurate basis for the preliminary planning of the stone statue relocation path.

[0049] The positioning and measurement screening module collects topographic point cloud data of the target stone statue's migration area using a laser scanner. Compared to traditional single-measurement devices, the laser scanner can quickly and efficiently acquire complete and high-precision topographic data, covering information such as elevation changes and topographic details within the migration area, providing a solid data foundation for subsequent positioning measurements. Simultaneously, this module filters positioning and measurement classifier groups based on predicted path offset parameters, enabling the targeted selection of measurement methods suitable for the current path offset prediction. This avoids the problems of insufficient accuracy or low efficiency of traditional measurement methods in complex terrain. Through the synergistic effect of the classifier group, the actual path offset parameters are more accurately identified, ensuring a more accurate judgment of the path offset and providing reliable data support for subsequent path adjustments.

[0050] The migration dynamic optimization module triggers migration path reconstruction based on actual path offset parameters, generating migration path optimization instructions. This breaks the limitation of traditional migration operations where path adjustments require pausing and replanning, enabling dynamic adjustment of the migration path. It can correct offset paths promptly without interrupting the migration process, shortening downtime caused by path adjustments and improving overall migration efficiency. Simultaneously, the module collects real-time obstacle distribution data of the migration path using an industrial measuring instrument and drives the obstacle avoidance classifier group to output actual obstacle distribution parameters. This allows for real-time and comprehensive monitoring of obstacle distribution changes on the migration path, eliminating reliance on manual observation and preventing unrecognized obstacles due to human error. This ensures timely obstacle avoidance during migration, reducing obstacles that hinder the migration process and the possibility of collision damage to the statue.

[0051] The positioning calibration feedback module combines actual path offset parameters with actual obstacle distribution parameters to generate migration path corrections, and simultaneously updates the geological deformation parameter weighting coefficients of the path planning and prediction module, constructing a closed-loop parameter adjustment system. This mechanism ensures that the data from each stage are no longer independent, but rather form effective feedback and integration. When path offsets or obstacle interference occur during the actual migration process, relevant parameters can be promptly fed back to the path planning and prediction stage, dynamically adjusting the weights of the geological deformation parameters. This allows subsequent path planning to be based on the updated accurate parameters, continuously optimizing the accuracy of predicted path offset parameters, further improving the accuracy and reliability of path planning, forming a virtuous cycle, continuously ensuring the safety and stability of the statue migration process, reducing manpower and time costs in the migration operation, and improving the overall quality of the migration project. Attached Figure Description

[0052] Figure 1 This is a timing diagram of the stone statue migration path planning and positioning measurement system described in this invention;

[0053] Figure 2 A flowchart illustrating the working principle of the path planning and prediction module;

[0054] Figure 3 A flowchart illustrating the working principle of the migration dynamic optimization module;

[0055] Figure 4 This is a flowchart illustrating the working principle of the path offset prediction unit in the path planning and prediction module. Detailed Implementation

[0056] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0057] Please see Figure 1 This invention provides a system for planning and locating a stone statue migration path, the system comprising:

[0058] The system employs the coordinated operation of multiple modules to achieve precise planning and dynamic adjustment of the stone statue relocation process. During system initialization, the path planning and prediction module is activated, monitoring geological signals in the target stone statue relocation area. It collects geological vibration, settlement, and displacement data through a ground-based sensor network, and comprehensively processes this data to obtain geological deformation parameters, including deformation rate and direction vector. Simultaneously, this module extracts environmental parameters of the relocation path from historical databases, such as temperature, humidity, and wind speed, as well as relocation time parameters, such as operation duration and intervals. Through data fusion calculations, it generates predicted path offset parameters, which are represented in vector form as the degree and direction of the expected path's deviation from the baseline.

[0059] The positioning and measurement filtering module is then activated, using a high-precision laser scanner to perform a 3D scan of the target area, generating terrain point cloud data containing elevation, slope, and surface feature information. Based on the numerical range of the predicted path offset parameters, this module selects a matching terrain classifier group from a pre-trained classifier group. Each classifier group consists of multiple parallel-processed classification paths, each corresponding to an offset pattern. The terrain point cloud data is input into these classification paths, and a pattern recognition algorithm outputs a set of terrain matching results. After statistically analyzing frequently occurring offset patterns, actual path offset parameters are generated, quantifying the path deviation caused by the current terrain. Upon receiving the actual path offset parameters, the migration dynamic optimization module triggers a path reconstruction mechanism, recalculating the topology of the migration path based on graph theory algorithms and generating migration path optimization instructions. These instructions include a sequence of path point coordinates and adjustment strategies. Simultaneously, the module uses industrial measuring instruments, such as total stations or radar, to collect real-time obstacle distribution data on the migration path, including obstacle location, size, and type. Based on the migration path optimization command, the module activates the corresponding obstacle avoidance classifier group. This group consists of multiple classification paths. After inputting real-time obstacle distribution data, it outputs actual obstacle distribution parameters, which are represented as a density matrix of obstacle clustering areas. The positioning calibration feedback module executes last, aligning the actual path offset parameters and actual obstacle distribution parameters spatiotemporally to ensure data processing within the same coordinate system and timestamp. The module calculates the path correction vector, determines the correction direction and magnitude through a vector synthesis algorithm, and generates the migration path correction amount. This correction amount is used to adjust the weight coefficients of the geological deformation parameters in the path planning and prediction module, updating the weight allocation through the feedback channel to optimize subsequent prediction accuracy. The entire system forms a closed-loop control, achieving continuous optimization of the migration path.

[0060] Example 1: See Figure 2 In the implementation of the stone statue relocation path planning and positioning measurement system, the geological deformation monitoring unit of the path planning and prediction module was the first to be activated. This unit deployed a multi-type sensor network at key nodes around and within the target stone statue relocation area, including a high-precision microelectromechanical system (MEMS) seismograph, a dual-axis tilt sensor, and a differential GPS receiver module. The seismograph captured surface vibration signals at a 1000Hz sampling rate, the tilt sensor continuously monitored changes in ground tilt angle, and the GPS module recorded displacement data with millimeter-level precision. This raw monitoring data was wirelessly transmitted to edge computing nodes, where random noise interference was eliminated using a Kalman filter algorithm, and representative geological deformation parameters were extracted using multi-source data fusion technology. These parameters were structured and stored as a three-dimensional vector group containing time series data, where each vector contained key indicators such as deformation rate, cumulative displacement, and trend index.

[0061] The environmental parameter analysis unit synchronously accesses the historical meteorological database and equipment operation log database of the migration area. This unit extracts temperature and humidity records, wind speed and direction data, and solar radiation intensity curves for the same season within the past three years. It also obtains time parameters such as operation time periods, duration of single operations, and intervals recorded in historical migration tasks. After standardized preprocessing, this heterogeneous data is input into a comprehensive evaluation model based on the entropy weight method. By calculating the influence weights of each environmental factor on path stability, the model ultimately outputs a path environmental interference index ranging from 0 to 1. A higher index value indicates a greater potential interference from environmental conditions on the migration path.

[0062] The path offset prediction unit employs an ensemble learning framework to construct its prediction model. The unit's historical training dataset contains over 2000 historical migration cases, each including a complete sequence of geological deformation parameters, an environmental disturbance index, and actual recorded path offset data. The random forest predictor consists of 150 decision trees of depth 12, each trained using bootstrap sampling to select a subset from the feature space. When new geological deformation parameters and environmental disturbance indices are input, each decision tree independently performs inference and prediction. Finally, a voting mechanism is used to synthesize the outputs of all trees to generate the predicted path offset parameters. These parameters are represented in polar coordinates, containing two dimensions: offset angle and expected offset distance.

[0063] After the positioning and measurement screening module is activated, a pulsed ground laser scanning system is used to perform a three-dimensional scan of the target area. The scanner acquires surface elevation data at a density of 500,000 points per second within a 120-degree field of view. Each point cloud data point contains three-dimensional coordinate information, echo intensity value, and color information, forming a detailed terrain model with a density of up to 500 points per square meter. After denoising and registration processing, the point cloud data is divided into 10cm × 10cm grid cells for subsequent analysis.

[0064] Based on the range of predicted path offset parameters output by the path offset prediction unit, the system automatically selects a matching terrain classifier group from a pre-trained classifier library. For example, when the predicted offset angle is in the range of 5-10 degrees and the expected offset distance exceeds 2 meters, the system calls classifier group C, which is specifically designed for medium-slope terrain. This classifier group consists of five parallel convolutional neural network classification paths, each using the ResNet-18 architecture and pre-trained on a large amount of historical terrain data. Each classification path focuses on identifying the correlation between specific types of terrain features and offset patterns. For example, classification path C-1 mainly detects the correlation between surface roughness and lateral offset, while classification path C-2 analyzes the mapping relationship between slope changes and longitudinal offset.

[0065] After the terrain point cloud data is converted into a 256×256 pixel depth image, it is simultaneously input into each classification path of the classifier group. The output layer of each classification path uses the sigmoid activation function to generate a matching score between 0 and 1. The system sets a matching threshold of 0.7. When the matching score of a grid cell exceeds the threshold, it is determined that the cell has a preset offset pattern. The judgment results of all classification paths are summarized to form a terrain matching result set. The system uses a density clustering algorithm to identify spatially continuous high-frequency matching regions. Offset patterns in these regions that are simultaneously confirmed by three or more classification paths are accepted, and finally, actual path offset parameters are generated. These parameters not only include offset and direction information but also include a confidence index, providing a quantitative basis for subsequent path optimization.

[0066] Example 2: See Figure 3 During the implementation of the migration dynamic optimization module, the path reconstruction unit immediately initiates the path topology reconstruction algorithm upon receiving the actual path offset parameters transmitted by the positioning measurement and screening module. This unit models the migration area as a weighted directed graph, where each node represents a geographic coordinate point, and edges represent feasible path segments. The edge weights comprehensively consider multiple factors such as terrain slope, surface bearing capacity, and historical offset data. The actual path offset parameters are converted into constraints in the graph structure, and the optimal path is calculated using an improved Dijkstra algorithm. During algorithm execution, a dynamic heuristic function is set to evaluate path feasibility; when offset parameters indicate a high-risk area, the weight value of nodes in that area is automatically increased. The final generated migration path optimization instruction contains a series of ordered path point coordinates, recommended movement speed curves, and expected transit times for each path segment. This data is encapsulated in JSON format and transmitted to downstream units.

[0067] The obstacle measurement unit employs a fusion perception system composed of multiple millimeter-wave radars and stereo vision sensors, scanning the environment along the planned path at a frequency of 10Hz. The radar equipment operates in the 77GHz band, with an effective detection range of 200 meters and an angular resolution of 0.5 degrees, enabling precise measurement of the radial distance and relative velocity of obstacles. The vision system is equipped with a global shutter camera and a structured light projector, acquiring the three-dimensional contour information of obstacles through triangulation. After spatiotemporal registration, the multi-source perception data is fused using Kalman filtering to generate real-time obstacle distribution data containing obstacle position, size, motion state, and surface characteristics. This data is stored in point cloud format, with each point containing coordinate information, reflection intensity, and dynamic markers.

[0068] The obstacle avoidance unit calls the corresponding obstacle classifier group from the distributed classifier library based on the risk level parameters specified in the migration path optimization instruction. When the instruction includes the "high-density dynamic obstacle" identifier, the system loads classifier group D for complex dynamic environments. This group contains four parallel-working support vector machine classifiers, each using a different kernel function to process obstacle feature data. The classifier input features include obstacle density gradient, motion consistency coefficient, and spatial distribution entropy value, and the output is a binary classification result indicating the obstacle threat level. The outputs of each classifier are integrated through a weighted voting mechanism to finally generate the actual obstacle distribution parameters, which are represented in the form of a grid map, with each grid containing a threat probability value and a recommended avoidance direction.

[0069] The spatiotemporal alignment engine of the positioning calibration feedback module employs a feature-matching-based registration method to unify the spatial coordinate system of the actual path offset parameters with the temporal reference of the actual obstacle distribution parameters. The engine first extracts a set of feature points from the actual path offset parameters; these feature points correspond to key turning points in the path. Simultaneously, it extracts spatial feature points at the same time from the obstacle distribution data. An iterative nearest-point algorithm is used to calculate the transformation matrix between the two sets of feature points, which includes rotation and translation components, achieving precise alignment of the spatial coordinate systems. For temporal alignment, a sliding window correlation analysis method is used to find the optimal time offset, ensuring that the two types of parameters are synchronized in the time dimension. The vector synthesizer receives the aligned parameter data and calculates the path correction vector using the following formula:

[0070]

[0071] in: This represents the final path correction vector. The number of sampling points. It is the first The weighting coefficient of each sampling point is determined by the terrain stability and the reliability of historical data for that point. Indicates the first The actual observed offset vector of each sampling point The predicted offset vector corresponding to the same spatiotemporal location, It is the obstacle influence coefficient vector at that point. This indicates element-wise multiplication. Weighting coefficients. The calculation takes into account the sensor accuracy level and environmental interference at that point, ensuring that highly reliable observation data receives greater weight. Obstacle influence coefficient vector. It is derived from obstacle distribution parameters, its modulus reflects obstacle density, and its direction indicates the avoidance direction.

[0072] The weighted feedback channel maps the magnitude of the path correction vector to the weight adjustment coefficients of the geological deformation parameters. The mapping function employs an S-shaped curve characteristic, resulting in smooth weight changes when the correction vector magnitude is small and significant weight adjustments when the magnitude exceeds a threshold. The adjustment coefficients are expressed in normalized form, ranging from 0 to 1, where 0 indicates complete dependence on environmental parameters and 1 indicates complete dependence on geological deformation parameters. The channel incorporates a built-in hysteresis compensation mechanism to prevent frequent oscillations of the weight coefficients near critical values. The updated weight coefficients are transmitted via a dedicated data bus to the dynamic weight allocator in the path planning and prediction module, completing the entire feedback loop.

[0073] Example 3: See Figure 4 The historical path database in the path planning and prediction module uses a distributed architecture to store data, comprising three main data partitions: the geological data partition stores seismic waveform data, surface subsidence monitoring data, and geotechnical parameters collected during previous migration missions; the environmental data partition stores meteorological monitoring records such as temperature, humidity, and wind speed, as well as vibration frequency and load distribution data of the migration equipment during operation; and the path data partition records the coordinate sequence, velocity curve, and offset correction records of the actual migration trajectory. Each data partition uses a time-series database to store the raw data, while also establishing a multi-dimensional index to support fast querying. The data extraction process employs a sliding time window mechanism, using the time characteristics of the current task as a benchmark to automatically retrieve historical data records under similar geological conditions and at the same time of year, ensuring the relevance of the training samples.

[0074] After receiving the weight adjustment coefficients from the positioning calibration feedback module, the dynamic weight allocator initiates a weight reallocation procedure. This procedure first parses the feature vector of the current prediction task, including data quality scores for geological deformation parameters, completeness indicators for environmental parameters, and task urgency parameters. Based on these feature values ​​and the feedback weight coefficients, fuzzy logic reasoning is used to determine the final weight ratio of each input feature. The weight adjustment of geological deformation parameters considers not only the absolute value of the feedback coefficients but also analyzes their changing trends. When multiple consecutive feedbacks indicate a need to increase the weight, the system adopts a more aggressive adjustment strategy. The weight allocation results are updated in the path offset prediction unit's memory in the form of a configuration file, ensuring that the real-time prediction task immediately uses the new weight settings.

[0075] The random forest predictor in the path offset prediction unit employs an online learning mechanism to maintain model performance. Whenever a new migration task is completed, the system automatically adds new training samples, composed of the geological deformation parameters, environmental disturbance index, and actual offset parameters of that task, to the historical dataset. The model update process uses incremental learning, retraining only a portion of the decision trees to avoid the computational overhead of complete retraining. During training, the feature sampling probability is adjusted according to the weight ratios provided by the dynamic weight allocator, giving geological deformation-related features a higher probability of selection during decision tree construction. The model performance monitoring module continuously tracks changes in prediction error and automatically triggers a model reconstruction process when a performance degradation is detected.

[0076] The terrain classifier group construction process of the positioning and measurement screening module adopts a multi-stage training strategy. The original training data comes from a fusion dataset of satellite remote sensing imagery, airborne laser scanning data, and field survey reports. After coordinate unification and resolution standardization preprocessing, this data is converted into 256×256 pixel terrain feature maps. The training of the convolutional neural network classifier adopts a transfer learning method. It is first pre-trained on a large natural terrain dataset, and then fine-tuned on specific transfer area data. The network architecture uses depthwise separable convolutions to reduce computational cost, while incorporating an attention mechanism to enhance the perception of key terrain features.

[0077] Each set of terrain classifiers corresponds to a path offset parameter range, a relationship derived through analysis of extensive historical data. The system establishes a mapping table between offset parameter ranges and terrain features; for example, when the expected offset angle is within the range of 5-15 degrees, it is typically associated with gravel strata with a moderate slope. Training data is grouped according to these mapping relationships, and a dedicated set of classifiers is trained for each offset range. Each set of classifiers contains 3-5 independently running convolutional neural networks, each focusing on identifying different types of terrain feature patterns. For example, network A primarily identifies surface texture features, network B focuses on elevation change patterns, and network C analyzes geological structural clues.

[0078] The terrain classification path implementation employs a parallel inference architecture. When new terrain point cloud data is input, the data preprocessing pipeline first converts the point cloud data into multi-resolution terrain elevation maps, and then extracts terrain features at different scales. These feature maps are simultaneously fed into all classification paths of the classifier group, with each path containing a complete convolutional neural network inference process. The output layer of each network uses a sigmoid activation function to generate matching probability values, and the system sets a dynamic threshold mechanism to automatically adjust the judgment threshold based on current environmental conditions and data quality. The output results of all classification paths are sent to the decision fusion module, which uses a weighted voting mechanism to synthesize the judgment results of each path, where the weights are dynamically adjusted based on the historical accuracy of each classification path.

[0079] The generation process of actual path offset parameters includes a spatial consistency check. The system performs spatial clustering analysis on the preliminary results output by the classifier group to identify matching areas that are continuously distributed in geographic space. For isolated matching points, the system re-evaluates them in conjunction with surrounding terrain features to eliminate mismatches caused by data noise. The final generated parameters include not only offset and direction information but also confidence assessment results. This confidence level is calculated based on the size and continuity of the matching area and the consistency of classifier voting. The entire processing flow ensures the reliability of the output parameters, providing accurate input for subsequent path optimization. The system continuously improves the performance of the terrain classifier group through a continuous learning mechanism. After each migration task, the system compares and analyzes the actual observed offset data with the classifier prediction results, calculating the accuracy and recall of each classification path. These performance metrics are used to adjust the voting weights of each classification path in subsequent tasks and also serve as a reference for model retraining. When a classification path is found to be consistently underperforming, the system automatically triggers the retraining process for that path, updating the network parameters with the latest collected data to maintain the overall performance of the classifier group.

[0080] Example 4: During the implementation of the obstacle avoidance unit in the migration dynamic optimization module, the distributed obstacle classifier library organizes classifier resources using a hierarchical storage architecture. The library establishes a three-dimensional index structure based on geographical region, migration type, and risk level, with each classifier group associated with an environmental scene descriptor. When a migration path optimization instruction arrives, the instruction parsing engine first extracts scene feature parameters, including terrain complexity index, visibility level, and expected movement speed range. These parameters are converted into 128-bit feature hash codes and quickly matched with available resources in the classifier library using a Bloom filter. Successfully matched classifier groups are marked as candidates, and then the loader selects the most suitable classifier group for the current hardware environment for instantiation based on the classifier group's model size and computational complexity score.

[0081] The dynamic path loader employs a lazy loading mechanism to optimize resource utilization. It maintains a cache pool of recently used classifier groups, managed based on a least recently used algorithm. When a new classifier group needs to be loaded, the loader first checks if an instance of the classifier group with the same scene descriptor exists in the cache pool. If a cache hit occurs, the existing instance is reused; otherwise, the model file is loaded from persistent storage. The loading process includes an integrity check step, using the SHA-256 algorithm to verify the integrity of the model file and checking the model version's compatibility with the current system version. After loading, the classifier group is injected into a dedicated inference engine, which allocates independent computing resources to each classification path.

[0082] The obstacle classification path execution adopts a pipelined architecture. Each classification path comprises three main stages: feature extraction, dimensionality reduction, and classification decision. The feature extraction stage extracts multi-dimensional features from real-time obstacle distribution data, including obstacle spatial distribution density, size distribution statistics, motion trend indicators, and material reflectivity. The dimensionality reduction stage uses principal component analysis (PCA) to compress high-dimensional features into a low-dimensional space, retaining 90% of the original feature variance. The classification decision stage employs a support vector machine (SVM) algorithm, calculating the obstacle threat level based on radial basis function kernels. The output of each classification path is a binary classification result, indicating whether there are high-density obstacles that need to be avoided in the corresponding area.

[0083] The spatiotemporal alignment engine of the positioning calibration feedback module employs multimodal sensor fusion technology to achieve parameter alignment. Internally, the engine maintains a unified spatiotemporal reference frame, under which all input parameters are transformed and processed. Spatial alignment uses an improved iterative nearest-point algorithm, calculating coordinate transformation parameters through feature point matching. Temporal alignment employs dynamic time warping technology to address the issue of inconsistent sampling frequencies between different data sources. During alignment, the engine generates detailed transformation parameter records, including rotation matrices, translation vectors, and time offsets, which are used for subsequent vector synthesis calculations.

[0084] The vector synthesizer receives aligned parameter data and generates a path correction vector using a hierarchical weighted fusion strategy. The synthesis process consists of three layers: spatial, temporal, and semantic. The spatial layer focuses on the geospatial consistency of the parameters, assigning higher weights to spatially adjacent sampling points. The temporal layer analyzes the trends of parameter changes over time, giving greater weight to parameters that maintain stable values. The semantic layer incorporates domain knowledge, giving priority to parameters closely related to the safety of the stone statue relocation. The weighted results from each layer are ultimately combined linearly to generate a comprehensive path correction vector, which contains two main components: direction angle and correction magnitude.

[0085] The weight feedback channel implements a non-linear mapping function for weight coefficients. Internally, the channel uses a lookup table mechanism to store the mapping relationships, and the construction of the lookup table is based on statistical analysis results of a large amount of historical data. The mapping process considers multiple factors such as the magnitude of the correction vector, directional consistency, and temporal persistence. When a new path correction vector is received, the channel first calculates the vector's overall confidence score, and then finds the corresponding weight adjustment coefficient in the lookup table based on the score. The adjustment coefficient is output in differential form, containing the difference between the current weight value and the target weight value, allowing for smooth weight adjustment. The table below shows the configuration parameters of the classifier group in different scenarios:

[0086] Table 1: Obstacle Classifier Group Configuration Parameters

[0087]

[0088] The parameter registration accuracy of the spatiotemporal alignment engine directly affects the effect of subsequent processing. The engine adopts a multi-stage registration strategy. First, coarse registration is performed to eliminate obvious spatiotemporal biases, and then fine registration is performed to optimize the alignment accuracy. In the coarse registration stage, feature descriptor matching technology is used to extract ORB feature points from the actual path offset parameters and match them with feature points from the actual obstacle distribution data. In the fine registration stage, optical flow is used to calculate the sub-pixel displacement field to further improve the registration accuracy. The transformation parameters generated during the registration process are recorded in metadata for subsequent quality assessment. The vector synthesizer uses a sliding time window mechanism to process time-series data. The window size is dynamically adjusted according to the data sampling frequency to ensure that each window contains a sufficient number of sampling points for statistical analysis. For each time window, the synthesizer independently calculates the local correction vector, and then integrates the results of each window into a global correction vector through time series analysis technology. This processing method ensures both real-time computation and stability of results, avoiding abnormal output due to transient interference.

[0089] The mapping update mechanism of the weight feedback channel maintains the system's adaptability. The channel periodically collects evaluation data on the weight adjustment effect, including the improvement in prediction accuracy and system stability indicators. This evaluation data is used to optimize the mapping relationship in the lookup table, employing a reinforcement learning algorithm to automatically adjust the mapping strategy. The mapping relationship update process uses a gradual adjustment approach, making only small modifications each time to ensure the continuity and predictability of system behavior. Historical mapping relationship records are also retained for rollback operations when needed.

[0090] Example 5: The implementation of the migration path preloading module is based on spatial data indexing technology. Upon receiving the migration path optimization instruction generated by the migration dynamic optimization module, this module immediately initiates the neighborhood data retrieval process. The system first parses the geographic coordinate range and time window parameters contained in the instruction, constructing a spherical query space with a radius of 800 meters centered on the target migration area. The adjacent migration area database adopts a geohash index structure, organizing global migration area data into grids according to latitude and longitude. The query engine quickly locates all adjacent area data records within the query range by calculating the spatial distance between the target area and each grid in the database. This data includes historical topographic point clouds, geological exploration reports, and path logs of completed migration tasks; each data record carries detailed spatial metadata and timestamp information.

[0091] The data loading process employs a chunked transmission mechanism, dynamically adjusting the transmission strategy based on network bandwidth and data priority. High-priority data, such as real-time terrain change monitoring records and recent migration trajectory data, are transmitted first, as these typically contain the most valuable information for the current task. Data compression algorithms are used during transmission to reduce network load, and real-time verification is performed to ensure data integrity and accuracy. Loaded data is temporarily stored in a high-speed cache, organized according to spatial location and chronological order for easy access by subsequent modules.

[0092] The protocol compatibility verification module performs data format consistency checks. This module receives terrain point cloud data output from the positioning and measurement filtering module and analyzes its data format specifications and coordinate reference system information. Terrain point cloud data typically includes LAS format header information, which details the acquisition device parameters, coordinate system definitions, and data accuracy indicators. The verification engine compares this metadata item by item with the format specifications of the pre-loaded terrain reference data, focusing on checking the consistency of the coordinate reference system, the uniformity of data units, and the degree of accuracy level matching. A multi-layered checking strategy is employed during the verification process, from basic file format compatibility to in-depth coordinate transformation parameter consistency, ensuring seamless data integration.

[0093] When a difference in coordinate systems is detected, the verification module automatically initiates the coordinate transformation service. This service incorporates several commonly used coordinate transformation algorithms, including transformation models from local coordinate systems to global coordinate systems. The transformation process first identifies the types of the source and target coordinate systems, then loads the corresponding transformation parameters and datum information. For complex coordinate transformations, a step-by-step transformation method is used, first transforming to the geocentric coordinate system, and then to the target coordinate system to ensure transformation accuracy. The transformed data needs to be verified in reverse by comparing the relative positions of feature points in the two coordinate systems to confirm the accuracy of the transformation results.

[0094] The path conflict resolution engine initiates an emergency handling process when protocol compatibility verification fails. Verification failure can be caused by various reasons, including coordinate system incompatibility, data precision mismatch, or inconsistent time bases. The engine first analyzes the type and severity of the failure and selects an appropriate processing strategy based on a pre-defined decision tree. For minor, repairable incompatibility issues, the engine attempts to adapt the data using data interpolation or resampling techniques; for severe protocol incompatibility, the engine sends a regeneration command to the path planning and prediction module.

[0095] The regeneration instruction includes detailed error codes and repair suggestions to help the path planning and prediction module adjust its data processing strategy. Upon receiving the instruction, the prediction module first clears the historical data in the current cache, and then reinitializes the prediction algorithm using default parameter weights. The regeneration process employs a more conservative strategy, increasing the number of data verification steps and expanding the data retrieval scope to find more usable reference data sources. The newly generated predicted path offset parameters are accompanied by a data quality report, detailing the data sources and processing methods used in the generation process.

[0096] A comprehensive data quality monitoring system was established during system implementation to evaluate the data processing results at each stage in real time. Monitoring metrics include multiple dimensions such as data integrity, coordinate accuracy, time synchronization, and format compliance. When any metric exceeds a preset threshold, the system automatically triggers corresponding corrective measures to ensure the reliability of the entire data processing flow. Monitoring results are recorded in the system log for subsequent analysis and system optimization.

[0097] The data flow maintains strict temporal consistency throughout the implementation process. All data processing steps are timestamped, ensuring traceability of the generation process and temporal context of each data product. This temporal management mechanism is particularly important for migration tasks requiring multiple iterations, helping the system accurately identify data versions and processing history, avoiding decision-making errors caused by inconsistent data timing. The final output migration path correction undergoes multiple checks and verifications to ensure it meets the accuracy requirements of engineering applications. The system generates a detailed technical report documenting the complete processing flow from data preloading to final output, including all data transformation parameters, verification results, and decision-making basis. This report provides a complete technical archive for the migration operation and valuable data support for continuous system improvement.

[0098] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0099] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A stone statue based migration path planning and positioning measurement system, characterized in that, Comprise: The path planning prediction module monitors the geological signal of the target statue migration area, obtains the geological deformation parameter, and collects the historical environment parameter and migration time parameter of the migration path to generate a predicted path offset parameter; The positioning measurement screening module collects the topographic point cloud data of the target statue migration area through a laser scanner, screens a positioning measurement classifier group according to the predicted path offset parameter, and identifies an actual path offset parameter; The migration dynamic optimization module triggers migration path reconstruction according to the actual path offset parameter, generates a migration path optimization instruction, collects real-time obstacle distribution data of the migration path through an industrial measuring instrument, and drives an obstacle avoidance classifier group to output an actual obstacle distribution parameter according to the migration path optimization instruction; The positioning calibration feedback module generates a migration path correction amount in combination with the actual path offset parameter and the actual obstacle distribution parameter, and synchronously updates the weight coefficient of the geological deformation parameter of the path planning prediction module; The migration dynamic optimization module comprises: The path reconstruction unit triggers migration path topology reconstruction according to the actual path offset parameter, and generates a migration path optimization instruction; The obstacle measurement unit collects real-time obstacle distribution data of the migration path through an industrial measuring instrument; The obstacle avoidance unit activates the corresponding obstacle classifier group based on the migration path optimization instruction, inputs the real-time obstacle distribution data into multiple obstacle classification paths, and outputs the actual obstacle distribution parameter.

2. The lithic-based migration path planning and positioning measurement system of claim 1, wherein, The path planning prediction module comprises: The geological deformation monitoring unit monitors the geological signal of the target statue migration area to obtain the geological deformation parameter; The environmental parameter analysis unit collects the historical environment parameter and migration time parameter of the migration path, and calculates a path environment interference index; The path offset prediction unit constructs a path offset predictor based on a random forest algorithm, inputs the geological deformation parameter and path environment interference index, and outputs a predicted path offset parameter.

3. The lithic-based migration path planning and positioning measurement system of claim 2, wherein, The positioning measurement screening module performs: Collecting topographic point cloud data of the target statue migration area through a laser scanner; According to the numerical interval of the predicted path offset parameter, matching the corresponding topographic classifier group; Parallelly inputting the topographic point cloud data into multiple topographic classification paths in the topographic classifier group, and outputting a set of topographic matching results; Statistically obtaining a high-frequency offset mode in the set of topographic matching results to generate an actual path offset parameter.

4. The lithic-based migration path planning and positioning measurement system of claim 1, wherein, The positioning calibration feedback module performs: Spatiotemporal alignment of the actual path offset parameter and the actual obstacle distribution parameter; Calculating a path correction vector and generating a migration path correction amount; According to the migration path correction amount, reversely adjusting the weight coefficient of the geological deformation parameter in the path planning prediction module.

5. The lithic-based migration path planning and positioning measurement system of claim 4, wherein, The path offset prediction unit of the path planning prediction module adopts: A historical path database stores historical geological deformation parameters and historical path environment interference indexes of the same type of statue migration; A dynamic weight distributor adjusts the contribution proportion of the geological deformation parameter in the path offset predictor according to the weight coefficient returned by the positioning calibration feedback module.

6. The stone statue based migration path planning and positioning measurement system according to claim 3, wherein, The topographic classifier group of the positioning measurement screening module is constructed in the following manner: A plurality of sets of convolutional neural network classifiers are trained based on historical terrain data of the migration area; Each set of classifiers is bound to a specific path offset parameter interval and contains a plurality of independent terrain classification paths; The terrain classification paths output a binary result of the matching degree of terrain features and a preset offset mode.

7. The stone statue based migration path planning and positioning measurement system according to claim 1, wherein, The obstacle avoidance unit of the migration dynamic optimization module adopts: A distributed obstacle classifier library pre-stores classifier groups corresponding to different migration path optimization instructions; A dynamic path loader calls a target classifier group from the classifier library according to the migration path optimization instruction; The obstacle classification path outputs a binary classification result of the obstacle distribution density level.

8. The lithic-based migration path planning and positioning measurement system of claim 7, wherein, The positioning calibration feedback module includes: A space-time alignment engine aligns the spatial coordinates of the actual path offset parameter with the time stamp of the actual obstacle distribution parameter; A vector synthesizer calculates the direction and amplitude of the path correction vector according to the aligned parameters; A weight feedback channel maps the amplitude of the path correction vector to the weight adjustment coefficient of the geological deformation parameter.

9. The stone statue based migration path planning and positioning measurement system of claim 1, wherein, Further comprising: A migration path preloading module loads terrain reference data from an adjacent migration area database according to the migration path optimization instruction generated by the migration dynamic optimization module; A protocol compatibility verification module verifies the coordinate conversion compatibility of the preloaded terrain reference data based on the terrain point cloud data format of the positioning measurement screening module; A path conflict resolution engine triggers the path planning prediction module to regenerate the predicted path offset parameter when the protocol compatibility verification fails.

Images