CN-122023689-A - Mine three-dimensional dynamic mapping system based on multi-source heterogeneous point cloud fusion and self-evolution time sequence modeling

CN122023689ACN 122023689 ACN122023689 ACN 122023689ACN-122023689-A

Abstract

The invention relates to the technical field of mining detection, in particular to a three-dimensional dynamic mapping system of a mine based on multi-source heterogeneous point cloud fusion and self-evolution time sequence modeling, which comprises a data acquisition layer, a data storage layer and a data storage layer, wherein the data acquisition layer is used for integrating sensing units for acquiring data, a time sequence point cloud data set is formed when each sensing unit synchronously acquires, and each point cloud is uniformly mapped to a system coordinate system after time synchronization and space registration; the method comprises a calculation layer, a time sequence modeling layer and a self-feedback learning layer, wherein the calculation layer carries out space registration and confidence weighting superposition on multi-source heterogeneous point clouds, adopts a collaborative entropy minimization fusion algorithm to process and output fusion points Yun Juzhen, the time sequence modeling layer carries out residual modeling on continuous time sequence points through a self-evolution time sequence topology model and outputs a time sequence topology structure, and the self-feedback learning layer comprises a deformation driving self-labeling and feedback correction module.

Inventors

JIANG HE
WANG YUELIN
YUAN JIYING
AN NA
SHU HAI
TIAN LINA
LI JUAN
WANG GUIYAN
BAO YIN
SUN YANFEI

Assignees

内蒙古自治区国土空间规划院

Dates

Publication Date: 20260512
Application Date: 20251226

Claims (10)

1. A mine three-dimensional dynamic mapping system based on multi-source heterogeneous point cloud fusion and self-evolution time sequence modeling is characterized by comprising the following components: the data acquisition layer is used for integrating sensing units for acquiring data, forming a time sequence point cloud data set when the sensing units synchronously acquire, and uniformly mapping each point cloud to a system coordinate system after time synchronization and spatial registration; the fusion calculation layer is used for carrying out space registration and confidence weighting superposition on the multi-source heterogeneous point cloud, adopting a collaborative entropy minimization fusion algorithm for processing, and outputting a fusion point Yun Juzhen; The time sequence modeling layer carries out residual modeling on the continuous time sequence points through a self-evolution time sequence topology model and outputs a time sequence topology structure; the self-feedback learning layer comprises a deformation driving self-labeling and feedback correction module, realizes automatic labeling of a deformation region, self-supervision training and parameter dynamic calibration by analyzing a residual error matrix and a prediction error, and establishes a self-feedback correction function to reversely act on a collaborative entropy minimization fusion algorithm and a self-evolution time sequence topology model.
2. The mine three-dimensional dynamic mapping system based on multi-source heterogeneous point cloud fusion and self-evolution time sequence modeling as claimed in claim 1, wherein in the data acquisition layer, a single sensing source is arranged at the moment The point cloud expression of (2) is as follows: ; wherein the method comprises the steps of Is the first Three-dimensional coordinate vectors of the individual points, The number of source points.
3. The mine three-dimensional dynamic mapping system based on multi-source heterogeneous point cloud fusion and self-evolution time sequence modeling according to claim 2, wherein in the fusion calculation layer, the expression of the output fusion point cloud matrix is as follows: ; wherein the method comprises the steps of In order to fuse the weight coefficients, For the total number of data sources, satisfy 。
4. The mine three-dimensional dynamic mapping system based on multi-source heterogeneous point cloud fusion and self-evolution time sequence modeling according to claim 3, wherein in the time sequence modeling layer, the geometric change of the earth surface between two periods is described by a time sequence residual error matrix, and the expression of the time sequence residual error matrix is as follows: ; The expression of the local rate of change is as follows: ; The expression of the output timing topology is as follows: ; wherein the method comprises the steps of A set of nodes is represented and, Representing the deformed association edges between the nodes, The side weight matrix represents the coupling variation intensity between the areas.
5. The mine three-dimensional dynamic mapping system based on multi-source heterogeneous point cloud fusion and self-evolution time sequence modeling according to claim 4, wherein the expression of the self-feedback correction function in the self-feedback learning layer is as follows: ; Wherein the method comprises the steps of As the error coupling coefficient, the coupling coefficient, In order to predict the error of the signal, Is a fusion error.
6. The mine three-dimensional dynamic mapping system based on multi-source heterogeneous point cloud fusion and self-evolution time sequence modeling as claimed in claim 1, wherein the system is in a longitudinal layered and transverse coupling mode: The data acquisition layer is positioned at the bottom and is responsible for multi-source information input; The fusion calculation layer is positioned in the middle layer, and heterogeneous point cloud space consistency is executed; The time sequence modeling layer is arranged at the upper layer to realize topology evolution and morphological prediction; the uppermost self-feedback learning layer forms a closed loop path, and error information is fed back to a preceding algorithm to realize system self-correction.
7. The mine three-dimensional dynamic mapping system based on multi-source heterogeneous point cloud fusion and self-evolution time sequence modeling according to claim 2, wherein the data acquisition layer, the fusion calculation layer, the time sequence modeling layer and the self-feedback learning layer keep data and parameter transmission consistency through unified state vectors, and the expression of the unified state is as follows: ; wherein the method comprises the steps of In order to fuse the error function, In order for the information to be a difference in entropy, In order to fuse the weight coefficients, For the noise variance estimation, Is a resolution scaling factor.
8. The mine three-dimensional dynamic mapping system based on multi-source heterogeneous point cloud fusion and self-evolution time sequence modeling according to claim 7, wherein a fusion point cloud matrix obtained based on the collaborative entropy minimization fusion algorithm updates a state vector, and the updated state vector expression is as follows: ; wherein the method comprises the steps of Representing the mean square error between the fused point cloud and the reference model as a fusion error function; Is the information entropy difference; is a fusion weight; estimating for noise variance; Is a resolution scaling factor.
9. The mine three-dimensional dynamic mapping system based on multi-source heterogeneous point cloud fusion and self-evolution time sequence modeling according to claim 4, wherein a topology evolution function is introduced into the time sequence modeling layer to realize self-learning and updating of a time sequence structure, a predictive modeling process is executed after the topology structure is updated, and the expression of a predictive model is as follows: ; wherein the method comprises the steps of For the next point cloud form predicted by the system, the function Representing a deformation propagation mapping function based on the current topology.
10. The mine three-dimensional dynamic mapping system based on multi-source heterogeneous point cloud fusion and self-evolution time sequence modeling according to claim 5, wherein in the self-feedback learning layer, when a feedback error exceeds a set threshold value, the system triggers a parameter self-correction process, and fusion and learning parameters are adjusted according to the following updating rules: ; wherein the method comprises the steps of Is the first The fusion weight of the sensor source is determined, Is the learning step length; 、 Respectively update rate control parameters.

Description

Mine three-dimensional dynamic mapping system based on multi-source heterogeneous point cloud fusion and self-evolution time sequence modeling Technical Field The application relates to the technical field of mining detection, and particularly discloses a three-dimensional dynamic mapping system for a mine based on multi-source heterogeneous point cloud fusion and self-evolution time sequence modeling. Background Mine mapping technology is changing from traditional manual mapping to high-precision, automatic and dynamic directions along with the expansion of mine exploitation scale and the improvement of intelligent level. The novel mapping means such as three-dimensional laser scanning, unmanned aerial vehicle remote sensing and GNSS high-precision positioning are applied to the field of mining, and a data base is provided for monitoring the terrains of the strip mine side slopes, storage yards and stopes. However, the existing mapping system mostly uses a single sensor as a core, the acquired data types and the acquired resolution are large in difference, and high-precision unified modeling and dynamic updating are difficult to realize under complex terrain and multi-time-phase environments. With the advancement of digital transformation of mining areas, three-dimensional point cloud data has become an important information carrier for surface modeling and deformation analysis. How to realize effective fusion and joint modeling among multi-source heterogeneous point clouds becomes a key research direction of the current mine mapping technology. The traditional method depends on static registration and manual parameter setting, and fusion precision and model stability are difficult to guarantee when the dimensional change of a measurement area, observation noise disturbance and time sequence deformation are faced. In recent years, some methods have been attempted to monitor terrain changes by using unmanned aerial vehicle point clouds, multi-temporal data contrast, laser scanning, or other means. For example, the chinese patent of the disclosure CN113593017B proposes a method for constructing a three-dimensional model of the surface of an open pit, which performs a two-stage fusion process on a local point cloud in time and space, and then updates and fuses the local point cloud with a historical three-dimensional model to form a dynamic three-dimensional model. The method can process the point cloud superposition and shielding filling problems of different time phases to a certain extent, but the fusion strategy depends on experience or fixed parameter setting, and the fusion suitability and the self-adaptive adjustment capability of the multi-source heterogeneous sensor are insufficient. The Chinese patent publication No. CN112945196B proposes a strip mine step line extraction and slope monitoring method based on point cloud data, step lines are extracted through progressive morphological filtering, curvature evaluation and clustering algorithm, and slope deformation is judged based on difference of the two steps. The method is effective under a flat step structure, but has low sensitivity to abnormal deformation in complex terrains or active sliding areas, and lacks multi-source point cloud fusion and integral three-dimensional change prediction capability. In the aspect of academic research, scholars propose to utilize a multi-temporal unmanned aerial vehicle point cloud and point cloud differential algorithm to carry out three-dimensional dynamic monitoring on surface subsidence of strip mines. The method can reveal the capability of the unmanned aerial vehicle point cloud in complex terrain settlement monitoring, but still mainly relies on single-platform data, and the capability of fusion, error correction, self-adaptive weight mechanism and time sequence prediction on the foundation laser scanning data is still insufficient. In addition, researches are presented to provide an automatic three-dimensional point cloud processing flow, a coarse-fine registration strategy, a robust change detection and a volume estimation technology are adopted, the limitation of manual parameter tuning is emphasized, self-tuning of key parameters according to point cloud statistical characteristics is tried, and the volume error of an actual measurement result is kept at a lower level under a noise condition. However, the method is not yet deeply expanded to the self-evolution capability of the topology timing learning, self-feedback correction or fusion mechanism. In summary, the prior art still has the following main problems in the aspects of three-dimensional mapping and earth surface deformation monitoring of mines, such as low fusion precision of multi-source point clouds, lack of a unified coordinate and weight self-adaptive mechanism, insufficient time sequence modeling capability, difficulty in realizing continuous dynamic evolution of earth surface deformation, lack of a self-learning and feedback correction