Foundation data processing method and system based on three-dimensional point cloud

Abstract

The invention provides a foundation data processing method and system based on three-dimensional point clouds, and relates to the technical field of foundation data processing, and the method comprises the steps: determining a plurality of divided regions of a foundation based on three-dimensional point cloud data before foundation excavation; determining the cross section shot video of each divided area based on the plurality of divided areas of the foundation and the cross section shot video in the foundation excavation process; generating a simulation three-dimensional foundation image of each divided area and mining information of each divided area; determining an earthwork correction coefficient of each divided area based on the simulation three-dimensional foundation image of each divided area and the mining information of each divided area; and determining all corrected earthwork result data based on the estimated earthwork data of each divided area and the correction coefficient of each divided area. The method can accurately determine earthwork data of foundation excavation.

Description

Technical Field

[0001] The present invention relates to the technical field of ground-based data processing, and in particular to a ground-based data processing method and system based on three-dimensional point clouds. Background Art

[0002] In the field of foundation construction, the accurate calculation of earthwork data volume directly affects project costs and construction safety. Traditional foundation data processing relies on single-point measurement or empirical formula simulation, which has significant technical bottlenecks. Single-point measurement can only obtain discrete data and cannot reflect the overall physical characteristics of the foundation, such as soil moisture distribution and geological structure continuity, resulting in large errors in earthwork volume estimation. Experience-based finite element simulation requires a large number of preset parameters, and has poor adaptability to complex geological conditions such as voids and collapsible soil layers. The calculation cycle is long and prone to deviations. In addition, the dynamic monitoring information generated during the construction process, such as changes in cross-sectional soil conditions, has not been effectively integrated and utilized, making it difficult to correct earthwork calculation results in real time according to actual geological conditions. Differences in soil properties often lead to significant deviations between the actual compaction volume and the estimated results.

[0003] Therefore, how to accurately determine the earthwork data of foundation excavation is a problem that needs to be solved urgently. Summary of the Invention

[0004] The main technical problem solved by the present invention is how to accurately determine the earthwork data of foundation excavation.

[0005] According to a first aspect, the present invention provides a foundation data processing method based on three-dimensional point clouds, comprising: acquiring three-dimensional point cloud data before foundation excavation, three-dimensional point cloud data after foundation excavation, and a cross-sectional video shot during foundation excavation; determining a plurality of divided areas of the foundation based on the three-dimensional point cloud data before foundation excavation; determining estimated earthwork data for each divided area based on the three-dimensional point cloud data before foundation excavation, the three-dimensional point cloud data after foundation excavation, and the plurality of divided areas of the foundation; determining a cross-sectional video shot for each divided area based on the plurality of divided areas of the foundation and the cross-sectional video shot during foundation excavation; generating a simulated three-dimensional foundation image of each divided area and excavation information of each divided area based on the cross-sectional video shot for each divided area; determining an earthwork correction coefficient for each divided area based on the simulated three-dimensional foundation image of each divided area and the excavation information of each divided area; and determining all corrected earthwork result data based on the estimated earthwork data of each divided area and the correction coefficient of each divided area.

[0006] In one possible implementation, the determining of multiple division areas of the foundation based on the three-dimensional point cloud data before the foundation excavation includes: determining multiple significant point cloud sets based on the three-dimensional point cloud data before the foundation excavation; determining multiple high-significance division points, multiple medium-significance division points, and multiple low-significance division points based on the multiple significant point cloud sets; determining multiple sets of division schemes based on the multiple high-significance division points, multiple medium-significance division points, and multiple low-significance division points; and determining multiple division areas of the foundation using a neural network model based on the multiple sets of division schemes.

[0007] In a possible implementation, the foundation excavation is a straight foundation excavation for an airport runway.

[0008] In one possible implementation, the method of determining multiple division areas of the foundation based on the multiple division schemes using a neural network model includes: constructing a scheme graph based on the multiple division schemes, the scheme graph including multiple division scheme nodes and edges between nodes, the node features of the division scheme nodes including three-dimensional point cloud data of each division area corresponding to a division scheme, and the edges between the division scheme nodes being the difference of the division schemes; processing the scheme graph based on a graph neural network to obtain a target division scheme; and dividing the foundation based on the target division scheme to obtain multiple division areas of the foundation.

[0009] According to a second aspect, the present invention provides a foundation data processing system based on a three-dimensional point cloud, comprising: a data acquisition module for acquiring three-dimensional point cloud data before foundation excavation, three-dimensional point cloud data after foundation excavation, and a cross-sectional video captured during the foundation excavation process; a region division module for determining a plurality of divided regions of the foundation based on the three-dimensional point cloud data before foundation excavation; an earthwork estimation module for determining estimated earthwork data for each divided region based on the three-dimensional point cloud data before foundation excavation, the three-dimensional point cloud data after foundation excavation, and the plurality of divided regions of the foundation; a video processing module for determining a cross-sectional video captured for each divided region based on the plurality of divided regions of the foundation and the cross-sectional video captured during the foundation excavation process; an image generation module for generating a simulated three-dimensional foundation image for each divided region and excavation information for each divided region based on the cross-sectional video captured for each divided region; a correction coefficient determination module for determining an earthwork correction coefficient for each divided region based on the simulated three-dimensional foundation image for each divided region and the excavation information for each divided region; and an earthwork correction module for determining corrected total earthwork result data based on the estimated earthwork data for each divided region and the correction coefficient for each divided region.

[0010] In one possible implementation, the area division module is also used to: determine multiple significant point cloud sets based on the three-dimensional point cloud data before the foundation excavation; determine multiple high-significance division points, multiple medium-significance division points, and multiple low-significance division points based on the multiple significant point cloud sets; determine multiple sets of division schemes based on the multiple high-significance division points, multiple medium-significance division points, and multiple low-significance division points; and determine multiple division areas of the foundation using a neural network model based on the multiple sets of division schemes.

[0011] In a possible implementation, the foundation excavation is a straight foundation excavation for an airport runway.

[0012] In a possible implementation, the area division module is also used to: construct a scheme map based on the multiple sets of division schemes, the scheme map includes multiple division scheme nodes and edges between nodes, the node features of the division scheme nodes include three-dimensional point cloud data of each division area corresponding to a division scheme, and the edges between the division scheme nodes are the differences of the division schemes; process the scheme map based on a graph neural network to obtain a target division scheme; divide the foundation based on the target division scheme to obtain multiple division areas of the foundation.

[0013] According to a third aspect, an embodiment of the present invention provides an electronic device, comprising: a processor; a memory; and a computer program; wherein the computer program is stored in the memory and configured to be executed by the processor to implement the method as described above, the method comprising: obtaining three-dimensional point cloud data before foundation excavation, three-dimensional point cloud data after foundation excavation, and a cross-sectional video shot during foundation excavation; determining multiple divided areas of the foundation based on the three-dimensional point cloud data before foundation excavation; determining estimated earthwork data for each divided area based on the three-dimensional point cloud data before foundation excavation, the three-dimensional point cloud data after foundation excavation, and the multiple divided areas of the foundation; determining a cross-sectional video shot for each divided area based on the multiple divided areas of the foundation and the cross-sectional video shot during foundation excavation; generating a simulated three-dimensional foundation image for each divided area and excavation information for each divided area based on the cross-sectional video shot for each divided area; determining an earthwork correction coefficient for each divided area based on the simulated three-dimensional foundation image for each divided area and the excavation information for each divided area; and determining corrected total earthwork result data based on the estimated earthwork data for each divided area and the correction coefficient for each divided area.

[0014] According to a fourth aspect, this embodiment provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the aforementioned foundation data processing method based on three-dimensional point clouds, the method comprising: obtaining three-dimensional point cloud data before foundation excavation, three-dimensional point cloud data after foundation excavation, and a cross-sectional video shot during foundation excavation; determining multiple divided areas of the foundation based on the three-dimensional point cloud data before foundation excavation; determining estimated earthwork data for each divided area based on the three-dimensional point cloud data before foundation excavation, the three-dimensional point cloud data after foundation excavation, and the multiple divided areas of the foundation; determining a cross-sectional video shot for each divided area based on the multiple divided areas of the foundation and the cross-sectional video shot during foundation excavation; generating a simulated three-dimensional foundation image for each divided area and excavation information for each divided area based on the cross-sectional video shot for each divided area; determining an earthwork correction coefficient for each divided area based on the simulated three-dimensional foundation image for each divided area and the excavation information for each divided area; and determining all corrected earthwork result data based on the estimated earthwork data for each divided area and the correction coefficient for each divided area.

[0015] The present invention provides a foundation data processing method and system based on three-dimensional point clouds, which include obtaining three-dimensional point cloud data before foundation excavation, three-dimensional point cloud data after foundation excavation, and cross-sectional video captured during the foundation excavation process; determining multiple divided areas of the foundation based on the three-dimensional point cloud data before foundation excavation; determining estimated earthwork data for each divided area based on the three-dimensional point cloud data before foundation excavation, the three-dimensional point cloud data after foundation excavation, and the multiple divided areas of the foundation; determining a cross-sectional video captured for each divided area based on the multiple divided areas of the foundation and the cross-sectional video captured during the foundation excavation process; generating a simulated three-dimensional foundation image of each divided area and excavation information of each divided area based on the cross-sectional video captured for each divided area; determining an earthwork correction coefficient for each divided area based on the simulated three-dimensional foundation image of each divided area and the excavation information of each divided area; and determining all corrected earthwork result data based on the estimated earthwork data of each divided area and the correction coefficient of each divided area. This method can accurately determine the earthwork data of foundation excavation. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 A schematic flow chart of a ground-based data processing method based on three-dimensional point clouds provided by an embodiment of the present invention;

[0017] Figure 2 A schematic diagram of the 3D point cloud data before denoising provided by an embodiment of the present invention;

[0018] Figure 3A schematic diagram of the denoising effect of three-dimensional point cloud data provided by an embodiment of the present invention;

[0019] Figure 4 A schematic diagram of colorless three-dimensional point cloud data before foundation excavation provided by an embodiment of the present invention;

[0020] Figure 5 A schematic diagram of color 3D point cloud data before foundation excavation provided by an embodiment of the present invention;

[0021] Figure 6 A schematic diagram of a process for determining multiple divided areas of a foundation based on three-dimensional point cloud data before foundation excavation provided by an embodiment of the present invention;

[0022] Figure 7 A schematic diagram of a process for determining multiple division areas of a foundation using a neural network model based on the multiple division schemes provided in an embodiment of the present invention;

[0023] Figure 8 A schematic diagram of a constructed scheme map provided in an embodiment of the present invention;

[0024] Figure 9 A schematic diagram of a ground-based data processing system based on three-dimensional point clouds provided by an embodiment of the present invention; DETAILED DESCRIPTION

[0025] The present invention will be further described in detail below by means of specific embodiments in conjunction with the accompanying drawings. Similar elements in different embodiments are numbered with associated similar elements. In the following embodiments, many detailed descriptions are provided to enable the present invention to be better understood. However, those skilled in the art will readily appreciate that some of the features may be omitted under different circumstances, or may be replaced by other elements, materials, or methods. In some cases, some operations related to the present invention are not shown or described in the specification. This is to avoid the core of the present invention being overwhelmed by excessive descriptions, and for those skilled in the art, it is not necessary to describe these related operations in detail. They can fully understand the related operations based on the description in the specification and the general technical knowledge in the art.

[0026] In an embodiment of the present invention, there is provided Figure 1 A ground-based data processing method based on a three-dimensional point cloud is shown, and the ground-based data processing method based on a three-dimensional point cloud comprises steps S1 to S7:

[0027] Step S1, obtaining three-dimensional point cloud data before foundation excavation, three-dimensional point cloud data after foundation excavation, and cross-sectional video during the foundation excavation process.

[0028] In some embodiments, the foundation excavation is an airstrip straight foundation excavation.

[0029] 3D point cloud data is a dataset of multiple discrete points in three-dimensional space that can be used to describe the surface features of an object or terrain. Each point in the 3D point cloud data contains X, Y, and Z coordinates.

[0030] In some embodiments, the three-dimensional point cloud data can be scanned and collected by a three-dimensional laser scanner, and denoising data processing can be integrated into the collection process to ultimately obtain usable three-dimensional point cloud data.

[0031] In some embodiments, during the process of collecting 3D point cloud data, denoising processing can be performed on the 3D point cloud data. Figure 2 A schematic diagram of a three-dimensional point cloud data before denoising provided by an embodiment of the present invention. Figure 3 A schematic diagram of the denoising effect of three-dimensional point cloud data provided by an embodiment of the present invention.

[0032] The 3D point cloud data before foundation excavation is collected by scanning the construction area with a 3D laser scanner. De-noising data processing is integrated into the data collection process to obtain the final 3D discrete point data. The 3D point cloud data before foundation excavation consists of a large number of discrete points, each with X, Y, and Z coordinates.

[0033] In some embodiments, the three-dimensional point cloud data before foundation excavation may include colorless three-dimensional point cloud data and color three-dimensional point cloud data. Figure 4 This is a schematic diagram of colorless three-dimensional point cloud data before foundation excavation provided by an embodiment of the present invention. Figure 5 A schematic diagram of color 3D point cloud data before foundation excavation provided by an embodiment of the present invention.

[0034] In addition to the X, Y, and Z coordinates, each point in the colored three-dimensional point cloud data before foundation excavation also contains reflection intensity and color attributes.

[0035] The three-dimensional point cloud data before foundation excavation can accurately record the terrain surface coordinates and terrain surface material reflection characteristics.

[0036] The three-dimensional point cloud data after foundation excavation is obtained by re-scanning and collecting the foundation construction area with a three-dimensional laser scanner after the foundation excavation work is completed. De-noising data processing is integrated during the collection process to finally obtain three-dimensional spatial discrete point data.

[0037] In some embodiments, the three-dimensional point cloud data after foundation excavation includes colorless three-dimensional point cloud data and color three-dimensional point cloud data.

[0038] The cross-sectional video shot during the foundation excavation process is a video obtained by real-time shooting of the cross section of the foundation excavation area using high-resolution video acquisition equipment during the foundation excavation construction process.

[0039] By shooting cross-sectional videos during foundation excavation, dynamic information such as changes in the soil structure of the cross section, progress in excavation depth, operation trajectory of construction equipment, and physical state of the soil can be fully recorded.

[0040] Step S2: determining a plurality of divided areas of the foundation based on the three-dimensional point cloud data before the foundation excavation.

[0041] In some embodiments, Figure 6 A schematic diagram of a process for determining multiple divided areas of a foundation based on the three-dimensional point cloud data before the foundation excavation is provided in an embodiment of the present invention. The process for determining multiple divided areas of the foundation based on the three-dimensional point cloud data before the foundation excavation includes steps S61 to S64:

[0042] Step S61 : determining a plurality of significant point cloud sets based on the three-dimensional point cloud data before foundation excavation.

[0043] In some embodiments, a point cloud analysis model can be used to determine multiple significant point cloud sets based on the three-dimensional point cloud data before foundation excavation. The point cloud analysis model is a convolutional neural network model, the input of the point cloud analysis model is the three-dimensional point cloud data before foundation excavation, and the output of the point cloud analysis model is multiple significant point cloud sets.

[0044] Convolutional neural network models include convolutional neural networks (CNNs), which are feedforward neural networks with deep structures that incorporate convolutional computations. CNNs extract features from input data through convolutional layers and then reduce network parameters using local connections and weight sharing. CNNs are effective at processing data with grid-like topologies.

[0045] Multiple significant point clouds are collections of discrete points with significant terrain or material characteristics, extracted from pre-excavation 3D point cloud data using a point cloud analysis model. Each point in a significant point cloud contains X, Y, and Z coordinates, as well as attributes such as reflection intensity and color. Points in a significant point cloud can be located where the terrain structure changes significantly, such as sudden slope changes or landform inflection points, or where the material reflectance characteristics are unique, such as areas of varying soil moisture or roughness. For example, in a foundation construction area, slope vertices with significant terrain undulations, intersections of different soil types, or points where the reflectance intensity is significantly abnormal due to high soil moisture can all constitute a significant point cloud.

[0046] By extracting multiple representative significant point cloud sets, the differences in physical characteristics of different areas of the foundation can be identified more accurately.

[0047] The local receptive field characteristics of the convolutional layers of a convolutional neural network can be used to extract local geometric features from pre-excavation 3D point cloud data, such as the neighborhood distribution of points and curvature changes. Multi-layer convolution and pooling operations can then filter out points with significant geometric or spatial distribution characteristics from the raw point cloud data. The model then maps these points with significant features from the high-dimensional feature space to 3D space based on a feature similarity threshold, thereby clustering them into multiple significant point cloud sets. Each point cloud set contains discrete points where the terrain structure changes suddenly or where the material reflectance characteristics are unusual, such as slope vertices and soil type boundaries.

[0048] In some embodiments, the point cloud analysis model includes a feature initial screening layer, a significant point quantization layer, and a point cloud generation layer. The input of the feature initial screening layer is the three-dimensional point cloud data before the foundation excavation, the output of the feature initial screening layer is a set of terrain mutation points and a set of color and reflection intensity abnormal points, the input of the significant point quantization layer is a set of terrain mutation points and a set of color and reflection intensity abnormal points, the output of the significant point quantization layer is a plurality of pre-selected significant points, the feature difference between each point and its adjacent points, the neighborhood point cloud density of each point, the position importance, and the surrounding humidity gradient change, the input of the point cloud generation layer is a plurality of pre-selected significant points, the feature difference between each point and its adjacent points, the neighborhood point cloud density of each point, the position importance, and the surrounding humidity gradient change, and the output of the point cloud generation layer is a plurality of pre-selected significant points.

[0049] A terrain mutation point set is a collection of points extracted from 3D point cloud data that show significant changes in terrain structure. These points include slope abruptness points and elevation anomalies, such as steep inflection points, rock protrusions, or sunken pits. This set of terrain mutation points can reflect the surface undulations and vertical anomalies of the foundation.

[0050] Color and reflection intensity outlier points are collections of points in 3D point cloud data where the color attributes or laser reflection intensity significantly deviate from the surrounding area. Color outliers correspond to areas of sudden material color changes in the foundation, such as wet areas. Reflection intensity outliers correspond to areas of abnormal soil moisture, such as areas with high humidity where the reflection intensity is low.

[0051] The multiple preliminarily selected significant points are a set of candidate significant points that are preliminarily screened.

[0052] The feature difference between each point and its adjacent points is the degree of difference between each preliminary significant point and its adjacent preliminary significant points in terrain features and reflection properties.

[0053] The neighborhood point cloud density of each point is the number of point clouds containing other preliminary significant points per unit area in the local space centered on a single point.

[0054] Position importance can indicate the representativeness of the primary significant point to the characteristics of the surrounding area.

[0055] The surrounding humidity gradient is the rate of change of humidity around the primary significant point. It reflects the speed of humidity transition within a unit distance and can be used to identify humidity transition zones or abrupt boundaries.

[0056] Different layers are responsible for different levels of information abstraction and processing. The feature screening layer is responsible for extracting basic feature points such as sudden changes in terrain and anomalies in color and reflection intensity from 3D point cloud data. The significant point quantification layer can calculate the initial selection of significant points based on these basic feature points, as well as quantitative indicators such as the feature difference, point cloud density, location importance, and humidity gradient change of these points. The point cloud set generation layer is responsible for determining the final significant point cloud set based on these quantitative indicators. Through this layered processing, large and complex point cloud data can be processed more efficiently. By building multiple layers, the modularity of the system can be increased, and the accuracy and efficiency of feature recognition of 3D point cloud data can be improved.

[0057] Step S62 : determining a plurality of high-saliency dividing points, a plurality of medium-saliency dividing points, and a plurality of low-saliency dividing points based on the plurality of saliency point clouds.

[0058] In some embodiments, a segmentation model can be used based on the multiple salient point cloud sets to determine multiple high-salient segmentation points, multiple medium-salient segmentation points, and multiple low-salient segmentation points. The segmentation model is a Transformer model, the input of the segmentation model is the multiple salient point cloud sets, and the output of the segmentation model is multiple high-salient segmentation points, multiple medium-salient segmentation points, and multiple low-salient segmentation points.

[0059] The Transformer model is a deep learning model based on the self-attention mechanism. The Transformer model can capture long-distance semantic associations by calculating the global dependencies of each element in the input sequence, and can realize feature extraction and semantic understanding of sequence data through multi-head attention mechanism and feedforward neural network.

[0060] Multiple highly significant partition points are a set of discrete points extracted from multiple significant point clouds using a partitioning model that have a decisive influence on the differences in soil moisture characteristics on the foundation surface. The reflectance intensity attributes of the areas where highly significant partition points are located can show significant anomalies due to extremely high or low soil moisture.

[0061] For example, in the runway foundation construction area, points with locally saturated soil due to water infiltration, or points with extremely low humidity and abnormally low reflection intensity due to solar evaporation, are all highly significant demarcation points.

[0062] Highly significant dividing points serve as humidity mutation points and can be used to accurately define the boundaries of different humidity areas.

[0063] Moderately significant points are discrete points that have a moderate impact on the differences in soil moisture characteristics on the foundation surface, as determined by the partitioning model. Changes in the reflectance intensity attributes of these points reflect the gradient transition characteristics of soil moisture. Moderately significant points are located in areas of gradual moisture transition between highly significant points, such as the transition zone from moist to semi-dry soil areas, or the continuous change in soil moisture from 60% to 80%.

[0064] For example, in the runway foundation construction area, the regional points where the humidity increases uniformly due to the difference in soil particle distribution have a regular increase in reflection intensity, so they can be classified as medium-significant dividing points.

[0065] Multiple low-significance partition points are discrete points whose partitioning model has a low degree of influence on the differences in soil moisture characteristics on the foundation surface. Low-significance partition points are distributed in areas with relatively uniform soil moisture. These points have similar reflection intensity attribute values, indicating a high consistency in soil moisture in the area. Examples include homogeneous clay layers with a humidity maintained at 50% ± 5% and concentrated sandy areas with humidity fluctuations of less than 3%.

[0066] Multiple significant point clouds contain key points where terrain structure changes suddenly, such as slope inflection points and landform intersections, as well as points where material reflectance characteristics change abnormally, such as soil moisture differences. Their coordinates and attribute data can accurately reflect the differences in physical characteristics across different areas of the foundation. For example, a point cloud in an area with dramatic terrain fluctuations can characterize the complexity of the geological structure, while a point cloud with abnormal reflectance intensity corresponds to a sudden change in soil moisture. Using multiple significant point clouds, key feature points that play a decisive role in regional division can be directly located, and high, medium, and low-significance division points can be selected based on the degree of feature significance.

[0067] The Transformer model uses a multi-head attention mechanism to calculate the weight of each point in the global point cloud. The statistical correlation between reflection intensity and soil moisture is converted into feature weights. For example, anomalies in reflection intensity at a sudden change in moisture receive a high weight in the Transformer model's self-attention calculation, thus being identified as highly significant points. Points in areas of gradual moisture changes have a more gradual change in reflection intensity, so their weights are lower than those of highly significant points, resulting in them being classified as moderately significant points. Points in areas of relatively uniform moisture have the lowest weights and are therefore classified as lowly significant points.

[0068] Step S63 : determining multiple sets of division schemes based on the multiple high-saliency division points, the multiple medium-saliency division points, and the multiple low-saliency division points.

[0069] In some embodiments, a scheme output model can be used to determine multiple sets of division schemes based on the multiple highly significant division points, the multiple moderately significant division points, and the multiple low-significance division points. The scheme output model is a deep neural network model. The input of the scheme output model is the multiple highly significant division points, the multiple moderately significant division points, and the multiple low-significance division points. The output of the scheme output model is multiple sets of division schemes.

[0070] Deep neural network models include deep neural networks, which are multi-layered neural network models. Deep neural networks combine a large number of neurons to form complex computational networks, enabling them to automatically learn high-level abstract features from input data. Deep neural networks consist of input layers, hidden layers, and output layers. Hidden layers can include multiple nonlinear transformation layers, such as fully connected layers and convolutional layers, and network parameters can be optimized using backpropagation algorithms.

[0071] Multiple partitioning schemes are generated by the partitioning model and are used to define the boundaries of regions with different soil moisture characteristics before foundation excavation. Each partitioning scheme passes through all highly significant partitioning points, some moderately significant partitioning points, and some low-significance partitioning points. The differences between the partitioning schemes are reflected in the granularity of the moisture partitions and the direction of their boundaries.

[0072] For example, one division scheme uses all highly significant division points as the boundary core, selects moderately significant division points where soil moisture increases from 70% to 80% and lowly significant division points where moisture is uniform, and ultimately divides the area before foundation excavation into a high-humidity zone (humidity ≥ 80%) and a moderate-humidity zone (humidity < 80%). Another division scheme also includes all highly significant division points, but selects moderately significant division points with moisture levels of 60%, 70%, and 80%, and lowly significant division points in zones with uniform moisture, further dividing the area before foundation excavation into saturated, high-humidity, moderate-humidity, and dry zones.

[0073] Highly significant points serve as boundary cores, enabling rapid identification of large-scale humidity zones and avoiding unnecessary subdivisions. In complex construction areas like runways, highly significant points can be used to mark structural abrupt changes, such as steep slope inflection points and landform intersections, providing a foundational framework for subsequent delineation.

[0074] In the gradient area between high-significance division points, medium-significance division points can refine the boundary direction of the moisture partition, making the division more consistent with the actual moisture distribution. By introducing medium-significance points, while ensuring the accuracy of the division, regional redundancy caused by over-division can be avoided. Low-significance division points are located in areas with relatively uniform soil moisture (such as homogeneous clay layers or concentrated sand areas), and their reflection intensity attribute values are similar. In areas with uniform moisture, low-significance points can act as "noise filters" to avoid misjudging uniform areas as areas of moisture variation. The core purpose of dividing multiple high-significance division points, medium-significance division points, and low-significance division points is to achieve refined division and efficient modeling of the foundation area by extracting key feature points on the foundation surface and combining the degree of influence of points with different significance levels on soil moisture characteristics.

[0075] Deep neural networks can automatically learn how the spatial distribution patterns of multiple highly significant, moderately significant, and lowly significant demarcation points are associated with soil moisture characteristics through multi-layer nonlinear transformations. Deep neural networks take the three-dimensional coordinates and reflectance intensity of these demarcation points as input. Using a combination of neurons in the hidden layers, they extract the geometric relationships and moisture feature similarities between the demarcation points. This allows for semantic fusion of the moisture-sustaining characteristics of highly significant demarcation points with the moisture-gradient characteristics of moderately and lowly significant demarcation points. For example, the model can learn that highly significant demarcation points must serve as the core anchor points of moisture region boundaries, while the connection between moderately and lowly significant demarcation points can form different boundary orientations. Ultimately, the deep neural network can determine multiple demarcation schemes through different combinations of hidden layer weights.

[0076] Step S64: Determine multiple division areas of the foundation using a neural network model based on the multiple division schemes.

[0077] In some embodiments, Figure 7 A schematic diagram of a process for determining multiple division areas of a foundation using a neural network model based on the multiple division schemes provided in an embodiment of the present invention, wherein the process for determining multiple division areas of a foundation using a neural network model based on the multiple division schemes includes steps S71 to S73:

[0078] Step S71: construct a scheme map based on the multiple sets of division schemes. The scheme map includes multiple division scheme nodes and edges between nodes. The node features of the division scheme nodes include three-dimensional point cloud data of each division area corresponding to a division scheme. The edges between the division scheme nodes are the differences of the division schemes.

[0079] The scheme graph is a data structure composed of multiple partitioning scheme nodes and the edges between them. A partitioning scheme node contains node features, which are the 3D point cloud data for each partitioned area corresponding to a partitioning scheme. Node features describe the spatial distribution and soil moisture characteristics of the partitioning scheme. Edges connect different partitioning scheme nodes and can describe the relationship between two partitioning schemes. Edges represent the degree of difference between the partitioning schemes. Figure 8 A schematic diagram of a scheme map constructed according to an embodiment of the present invention is provided. Figure 8 As shown, Figure 8 It includes partitioning scheme node A, partitioning scheme node B, partitioning scheme node C, and partitioning scheme node D. The edge between two partitioning scheme nodes is the difference between the partitioning schemes contained in these two partitioning scheme nodes.

[0080] The 3D point cloud data for each partitioned area corresponding to the partitioning scheme is the 3D point cloud data corresponding to each humidity partitioned area in the partitioning scheme. The 3D point cloud data for each partitioned area corresponding to a partitioning scheme is extracted from the original 3D point cloud data before foundation excavation. Then, based on the area boundary coordinates of the partitioning scheme, the 3D point cloud data corresponding to each partitioned area is filtered through spatial matching.

[0081] The difference between the partitioning schemes can be used to quantify the degree of difference between different partitioning schemes in terms of moisture partition boundaries, soil moisture feature coverage, etc. In some embodiments, the difference between the partitioning schemes can be calculated using a deep neural network.

[0082] Step S72: Process the solution map based on the graph neural network to obtain the target partitioning solution.

[0083] A graph neural network (GNN) is a deep learning model that can be used to process graph data. GNNs can learn to represent nodes in graph data by transferring information between nodes in the graph through a message passing mechanism.

[0084] The target partitioning scheme is an optimal pre-excavation area partitioning scheme output by processing the solution map using a graph neural network. This target partitioning scheme can keep humidity differences within each partition within a small fluctuation range while avoiding redundant partitions caused by over-segmentation.

[0085] The scheme map clearly illustrates the relationship between each partitioning scheme and its neighbors. This association information has a direct impact on selecting the optimal foundation partitioning scheme. Using the 3D point cloud data of each partitioned area corresponding to a partitioning scheme node as node features, and the differences between the partitioning schemes as edge features, we can fully utilize the potential association information between the schemes. This helps the graph neural network better understand the interactive effects of different schemes on the continuity of humidity zone boundaries and the uniformity of individual zones, thereby improving the accuracy of selecting the target partitioning scheme.

[0086] The graph neural network can use the message passing mechanism to enable each partitioning scheme node to combine the three-dimensional point cloud data features and difference information of adjacent nodes, and then use graph convolution operations to analyze the optimization effect of different combinations of medium and low-significance partitioning points on the partition boundary fitting the actual humidity distribution, such as whether a continuous transition humidity gradient area is formed, and the ability to control the humidity uniformity within the partition. Finally, the target partitioning scheme is screened out that can form a reasonable humidity partition boundary to avoid the number of partitioned areas, and can control the soil moisture difference in each partitioned area within an acceptable small fluctuation range.

[0087] Step S73: Divide the foundation based on the target division scheme to obtain multiple division areas of the foundation.

[0088] The multiple divided areas of the foundation are multiple divided area units obtained by segmenting the area before foundation excavation based on a target partitioning scheme, and the soil in each divided area has similar moisture characteristics.

[0089] Step S3 : determining estimated earthwork data for each divided area based on the three-dimensional point cloud data before foundation excavation, the three-dimensional point cloud data after foundation excavation, and the multiple divided areas of the foundation.

[0090] In some embodiments, an earthwork analysis model can be used to determine estimated earthwork data for each divided area based on the three-dimensional point cloud data before foundation excavation, the three-dimensional point cloud data after foundation excavation, and the multiple divided areas of the foundation. The earthwork analysis model is a Transformer model. The input of the earthwork analysis model is the three-dimensional point cloud data before foundation excavation, the three-dimensional point cloud data after foundation excavation, and the multiple divided areas of the foundation. The output of the earthwork analysis model is the estimated earthwork data for each divided area.

[0091] The estimated earthwork data for each zone is output by the earthwork analysis model, representing the projected amount of excavated earthwork in each zone. The estimated earthwork data is measured in cubic meters and represents the volume of earth removed from the corresponding zone after foundation excavation.

[0092] The 3D point cloud data before and after foundation excavation contain the original topography of the construction area before and after foundation excavation, respectively, as well as physical features such as color and reflection intensity, thus providing a spatial data foundation for representing terrain changes. The multiple subdivisions of the foundation are defined by spatial units with different soil moisture characteristics through 3D spatial coordinate ranges, allowing the model to process differentiated soil properties at a regional granularity. The Transformer model can capture the spatial correlation between the coordinate distribution and the boundaries of the subdivisions in the 3D point cloud data before and after foundation excavation through a self-attention mechanism, and automatically learn the mapping rules between terrain change characteristics, moisture change characteristics, and earthwork data. The Transformer model also uses a feedforward neural network to extract the implicit correlation between features such as point cloud reflection intensity and soil moisture, and can then output estimated earthwork data for each subdivision.

[0093] Step S4 : determining a cross-sectional video of each divided area based on the multiple divided areas of the foundation and the cross-sectional video during the foundation excavation process.

[0094] In some embodiments, a gated loop unit can be used to determine a cross-sectional video of each divided area based on the multiple divided areas of the foundation and the cross-sectional video captured during the foundation excavation process. The input of the gated loop unit is the multiple divided areas of the foundation and the cross-sectional video captured during the foundation excavation process, and the output of the gated loop unit is the cross-sectional video captured for each divided area.

[0095] The Gated Recurrent Unit (GRU) can be used to process sequence data and time series information. The GRU consists of three components: a memory unit, an update gate, and a reset gate.

[0096] The cross-sectional video of each divided area is obtained through a gated loop unit according to the multiple divided areas of the foundation, and then the exclusive cross-sectional video corresponding to each divided area is segmented from the cross-sectional video during the foundation excavation process.

[0097] The gated recurrent unit models the temporal features of the video frame sequence through a gating mechanism, and then uses an update gate to remember the dynamic changes of the boundaries of the divided regions in consecutive frames. At the same time, it performs spatial matching on each frame in combination with the region coordinate range, so that the global video can be accurately divided into cross-sectional videos corresponding to each divided region according to the spatial region.

[0098] Step S5: generating a simulated three-dimensional foundation image of each divided area and excavation information of each divided area based on the cross-sectional shooting video of each divided area.

[0099] In some embodiments, a generative adversarial network can be used to generate a simulated three-dimensional foundation image of each divided area and excavation information of each divided area based on the cross-sectional video of each divided area. The input of the generative adversarial network is the cross-sectional video of each divided area, and the output of the generative adversarial network is a simulated three-dimensional foundation image of each divided area and excavation information of each divided area.

[0100] A generative adversarial network (GAN) is a deep learning model consisting of a generator and a discriminator. GANs generate data through an adversarial training mechanism. The generator attempts to generate samples from random noise that resemble the distribution of real data, while the discriminator distinguishes between generated samples and real data. Through a game of optimization, the generator can ultimately produce realistic samples that are difficult for the discriminator to distinguish.

[0101] A simulated 3D ground image of a divided area is generated using a generative adversarial network (GAN) based on a video of a cross-section of the area. It represents a 3D visualization of the soil's internal structure. This image contains physical features such as soil stratification, internal soil moisture distribution, and the location of geological anomalies (such as cavities and faults). For example, a simulated 3D ground image of a high-wet soil area clearly shows the depth range of the saturated soil layer, the moisture gradient, and its relationship to the surrounding area. The color value of each pixel in the image corresponds to a specific soil moisture value.

[0102] The mining information of each partitioned area is the quantitative information related to the physical properties of the foundation soil and the mining process in each partitioned area output by the generative adversarial network. The mining information includes soil structure information, soil physical properties, and material content information.

[0103] Soil structure information includes soil void area, void distribution density, and void connectivity within the divided area.

[0104] Soil physical properties include soil looseness, soil moisture gradient, and soil internal friction angle.

[0105] Material content information includes the percentage of stone content, the percentage of clay content, and the organic matter content.

[0106] Cross-sectional video footage of the divided areas records continuous dynamic information, including soil structure and moisture content changes, throughout the foundation excavation process. By analyzing changes in soil color and texture in the video, the internal moisture gradient and material content distribution can be determined. The model uses stereo matching between consecutive frames to construct the 3D geometry of each divided area, including the location, size, and connectivity of voids. This data accurately supports the geometric modeling and physical property assignment of simulated 3D foundation images for each divided area, as well as the dynamic quantification of excavation information for each divided area.

[0107] Generative Adversarial Networks (GANs), through their unique adversarial training mechanism, can efficiently convert 2D temporal information from cross-sectional videos into 3D spatial data and physical parameters. The generator learns visual features such as soil texture, color, and geometric relationships from video frames to generate physically accurate 3D foundation structures, such as the layered interface between clay and sand. The discriminator continuously compares the generated data with the actual geological structure to ensure the accuracy and reliability of the output simulated 3D foundation images. Furthermore, the GAN maps visual features from the cross-sectional videos with physical parameters such as soil void area and looseness to output excavation information for each demarcated area.

[0108] Step S6: determining an earthwork correction coefficient for each divided area based on the simulated three-dimensional foundation image of each divided area and the excavation information of each divided area.

[0109] In some embodiments, a correction model can be used to determine the earthwork correction coefficient for each divided area based on the simulated three-dimensional foundation image of each divided area and the excavation information of each divided area. The correction model is a convolutional neural network, the input of the correction model is the simulated three-dimensional foundation image of each divided area and the excavation information of each divided area, and the output of the correction model is the earthwork correction coefficient for each divided area.

[0110] The earthwork correction coefficient for each divided area is a value output by the correction model to accurately correct the compacted earthwork volume under ideal conditions.

[0111] For example, in one zone, the soil is high-moisture clay. This excessive moisture content creates a "rubbery" soil, where water forms a film between soil particles, preventing them from packing tightly together. This results in a volume after compaction exceeding the expected volume under normal compaction conditions. The correction model analyzes and determines a volume correction factor of 0.95 for this zone, meaning the volume after compaction is 95% of the estimated volume for the zone.

[0112] For another divided area, the soil composition is sandy soil with a large number of cavities. The compacted volume of the excavated soil in this area is 82% of the excavated amount, and the soil correction coefficient is correspondingly set to 0.82.

[0113] The convolutional neural network can slide the convolution kernel of the convolution layer on the simulated three-dimensional foundation image, and then automatically extract spatial features such as soil layer structure and cavity distribution. The pooling layer reduces the data dimension while retaining key information. For information such as soil structure indicators, soil physical properties, and material content indicators in the mined information, the convolutional neural network can establish an association mapping between them and the three-dimensional foundation image features through the fully connected layer. Through this multi-level feature extraction and fusion mechanism, the convolutional neural network can capture the nonlinear relationship between the physical properties of the foundation and the earthwork correction coefficient. For example, it can identify high humidity areas with larger correction coefficients and areas with many cavities with smaller correction coefficients, and ultimately output an earthwork correction coefficient for each divided area that fits the actual situation.

[0114] Step S7: determining corrected total earthwork result data based on the estimated earthwork data of each divided area and the correction coefficient of each divided area.

[0115] The corrected total earthwork result data refers to the volume of soil removed after foundation excavation is completed and compacted as a whole, calculated by combining the estimated earthwork data for each divided area with the earthwork correction factor for that area.

[0116] The revised total earthwork result data is calculated by multiplying the estimated earthwork data of each divided area with the correction coefficient of the corresponding divided area to obtain the compacted earthwork volume of each divided area, and then accumulating the compacted earthwork volumes of all divided areas to finally obtain the compacted earthwork volume covering the entire foundation excavation construction area, which is the revised total earthwork result data.

[0117] Based on the same inventive concept, Figure 9 A schematic diagram of a ground-based data processing system based on a three-dimensional point cloud provided by an embodiment of the present invention, wherein the ground-based data processing system based on a three-dimensional point cloud includes:

[0118] A data acquisition module 91 is used to acquire three-dimensional point cloud data before foundation excavation, three-dimensional point cloud data after foundation excavation, and cross-sectional video during foundation excavation;

[0119] A region division module 92 is configured to determine a plurality of divided regions of the foundation based on the three-dimensional point cloud data before the foundation excavation;

[0120] an earthwork estimation module 93 for determining estimated earthwork data for each divided area based on the three-dimensional point cloud data before foundation excavation, the three-dimensional point cloud data after foundation excavation, and the multiple divided areas of the foundation;

[0121] A video processing module 94 is configured to determine a cross-sectional video of each divided area based on the plurality of divided areas of the foundation and the cross-sectional video captured during the foundation excavation process;

[0122] An image generation module 95 is configured to generate a simulated three-dimensional foundation image of each divided area and excavation information of each divided area based on the cross-sectional video of each divided area;

[0123] A correction coefficient determination module 96 is configured to determine an earthwork correction coefficient for each divided area based on the simulated three-dimensional foundation image of each divided area and the excavation information of each divided area;

[0124] The earthwork correction module 97 is configured to determine corrected total earthwork result data based on the estimated earthwork data of each divided area and a correction coefficient of each divided area.

[0125] In addition, unless expressly stated in the claims, the order of the processing elements and sequences, the use of alphanumeric characters, or the use of other names described in this specification are not intended to limit the order of the processes and methods of this specification. Although the above disclosure discusses some of the invention embodiments currently considered useful through various examples, it should be understood that such details are for illustrative purposes only, and the appended claims are not limited to the disclosed embodiments. On the contrary, the claims are intended to cover all modifications and equivalent combinations that are consistent with the spirit and scope of the embodiments of this specification. For example, although the system components described above can be implemented by hardware devices, they can also be implemented only by software solutions, such as installing the described system on an existing server or mobile device.

[0126] Similarly, it should be noted that, in order to simplify the presentation of this specification and thus facilitate understanding of one or more embodiments of the invention, the foregoing descriptions of the embodiments of this specification sometimes combine multiple features into a single embodiment, figure, or description thereof. However, this disclosure method does not imply that the subject matter of this specification requires more features than those recited in the claims. In fact, an embodiment may have fewer features than all of the features of a single disclosed embodiment.

[0127] Finally, it should be understood that the embodiments described in this specification are intended only to illustrate the principles of the embodiments of this specification. Other variations may also fall within the scope of this specification. Therefore, by way of example and not limitation, alternative configurations of the embodiments of this specification may be considered consistent with the teachings of this specification. Accordingly, the embodiments of this specification are not limited to the embodiments explicitly described and illustrated in this specification.

Claims

1. A ground data processing method based on three-dimensional point cloud, characterized in that: include: Obtaining 3D point cloud data before and after foundation excavation, and cross-sectional video during foundation excavation; determining a plurality of divided areas of the foundation based on the three-dimensional point cloud data before the foundation excavation; Determining estimated earthwork data for each divided area based on the three-dimensional point cloud data before foundation excavation, the three-dimensional point cloud data after foundation excavation, and the multiple divided areas of the foundation; determining a cross-sectional captured video of each divided area based on the plurality of divided areas of the foundation and the cross-sectional captured video during the excavation of the foundation; generating a simulated three-dimensional foundation image of each divided area and excavation information of each divided area based on the cross-sectional shot video of each divided area; determining an earthwork correction coefficient for each divided area based on the simulated three-dimensional foundation image of each divided area and the excavation information of each divided area; Corrected total earthwork result data is determined based on the estimated earthwork data of each divided area and the correction coefficient of each divided area.

2. The ground-based data processing method based on three-dimensional point cloud according to claim 1, characterized in that: The determining of a plurality of divided areas of the foundation based on the three-dimensional point cloud data before the foundation excavation comprises: determining a plurality of significant point cloud sets based on the three-dimensional point cloud data before the foundation excavation; Determining a plurality of high-saliency dividing points, a plurality of medium-saliency dividing points, and a plurality of low-saliency dividing points based on the plurality of saliency point clouds; Determining multiple sets of partitioning schemes based on the multiple high-saliency partitioning points, the multiple medium-saliency partitioning points, and the multiple low-saliency partitioning points; A neural network model is used to determine multiple divided areas of the foundation based on the multiple sets of division schemes.

3. The ground-based data processing method based on three-dimensional point cloud according to claim 1, characterized in that: The foundation excavation is the straight foundation excavation of the runway.

4. The ground-based data processing method based on three-dimensional point cloud according to claim 2, characterized in that: The method of determining the multiple division areas of the foundation using a neural network model based on the multiple division schemes includes: Constructing a scheme graph based on the multiple sets of partitioning schemes, the scheme graph includes multiple partitioning scheme nodes and edges between the nodes, the node features of the partitioning scheme nodes include three-dimensional point cloud data of each partitioned area corresponding to a partitioning scheme, and the edges between the partitioning scheme nodes represent the differences between the partitioning schemes; The target partitioning scheme is obtained by processing the scheme graph based on the graph neural network; The foundation is divided based on the target division scheme to obtain multiple division areas of the foundation.

5. A ground-based data processing system based on three-dimensional point clouds, for implementing the ground-based data processing method based on three-dimensional point clouds according to any one of claims 1 to 4, characterized in that: include: A data acquisition module is used to acquire three-dimensional point cloud data before foundation excavation, three-dimensional point cloud data after foundation excavation, and cross-sectional video during the foundation excavation process; A region division module, configured to determine a plurality of divided regions of the foundation based on the three-dimensional point cloud data before the foundation excavation; an earthwork estimation module, configured to determine estimated earthwork data for each divided area based on the three-dimensional point cloud data before foundation excavation, the three-dimensional point cloud data after foundation excavation, and the multiple divided areas of the foundation; A video processing module, configured to determine a cross-sectional video of each divided area based on the plurality of divided areas of the foundation and the cross-sectional video during the foundation excavation process; An image generation module is used to generate a simulated three-dimensional foundation image of each divided area and excavation information of each divided area based on the cross-sectional video of each divided area; a correction coefficient determination module, configured to determine an earthwork correction coefficient for each divided area based on the simulated three-dimensional foundation image of each divided area and the excavation information of each divided area; The earthwork correction module is used to determine the corrected total earthwork result data based on the estimated earthwork data of each divided area and the correction coefficient of each divided area.

6. The ground-based data processing system based on three-dimensional point cloud according to claim 5, characterized in that: The area division module is also used for: determining a plurality of significant point cloud sets based on the three-dimensional point cloud data before the foundation excavation; Determining a plurality of high-saliency dividing points, a plurality of medium-saliency dividing points, and a plurality of low-saliency dividing points based on the plurality of saliency point clouds; Determining multiple sets of partitioning schemes based on the multiple high-saliency partitioning points, the multiple medium-saliency partitioning points, and the multiple low-saliency partitioning points; A neural network model is used to determine multiple divided areas of the foundation based on the multiple sets of division schemes.

7. The ground-based data processing system based on three-dimensional point cloud according to claim 5, characterized in that: The foundation excavation is the straight foundation excavation of the runway.

8. The ground-based data processing system based on three-dimensional point cloud according to claim 6, characterized in that: The area division module is also used for: Constructing a scheme graph based on the multiple sets of partitioning schemes, the scheme graph includes multiple partitioning scheme nodes and edges between the nodes, the node features of the partitioning scheme nodes include three-dimensional point cloud data of each partitioned area corresponding to a partitioning scheme, and the edges between the partitioning scheme nodes represent the differences between the partitioning schemes; The target partitioning scheme is obtained by processing the scheme graph based on the graph neural network; The foundation is divided based on the target division scheme to obtain multiple division areas of the foundation.

9. An electronic device, characterized in that: include: processor; Memory; And a computer program; wherein, the computer program is stored in the memory and is configured to be executed by the processor to implement the ground-based data processing method based on three-dimensional point cloud according to any one of claims 1 to 4.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the ground-based data processing method based on three-dimensional point cloud as described in any one of claims 1 to 4 is implemented.

Images