Search

CN-121999164-A - Three-dimensional geological topography modeling method and system based on multi-eye digital image comparison and identification

CN121999164ACN 121999164 ACN121999164 ACN 121999164ACN-121999164-A

Abstract

The invention relates to the technical field of geological topography modeling, in particular to a three-dimensional geological topography modeling method and system based on multi-objective digital image comparison and identification, which are used for constructing an initial three-dimensional geological model based on multi-source exploration data, acquiring a face image through a single-camera multi-objective system calibrated on site, automatically extracting quantitative structural features with observation uncertainty through three-dimensional reconstruction and deep learning fusion analysis, establishing a mapping relation between the structural features and a geological potential field, coupling the observation uncertainty soft constraint with the initial probability model based on Bayesian inference to form a posterior joint probability model, and finally realizing dynamic iteration correction and optimization of the model along with construction progress through an incremental Bayesian updating engine, and synchronously generating a spatial confidence three-dimensional map reflecting the reliability of the model. The invention realizes the fundamental transition of geological modeling from static, subjective, deterministic to dynamic, objective and probabilistic, and remarkably improves the precision, reliability and engineering guiding value of the model.

Inventors

  • LIAO CHANGHAO

Assignees

  • 西南大学

Dates

Publication Date
20260508
Application Date
20260128

Claims (10)

  1. 1. The modeling method for identifying the three-dimensional geological landform based on the comparison of the multi-view digital images is characterized by comprising the following steps of: Constructing a scalar geological potential field based on the pre-exploration data, and generating an initial three-dimensional geological probability model containing the optimal estimation position and uncertainty of each geological interface space by adopting a probabilistic interpolation method; Acquiring a three-dimensional image pair of the face by using a calibrated single-camera multi-view system, and reconstructing to generate a three-dimensional point cloud model and a real coordinate texture image of the face; associating each construction example in the face construction characteristic data set with a corresponding geological potential field to establish a characteristic mapping relation, and performing Bayesian coupling on potential field soft constraint generated based on the observation uncertainty and the initial three-dimensional geological probability model to form a posterior joint probability model; and taking the posterior joint probability model as a new priori, and carrying out iterative correction and optimization on the posterior joint probability model based on the newly added face construction feature data set, extracting an updated three-dimensional geological structure model and generating an associated spatial confidence three-dimensional map.
  2. 2. The method for modeling three-dimensional geologic features based on multi-view digital image contrast recognition according to claim 1, wherein the initial three-dimensional geologic potential field probability model construction step specifically comprises: Acquiring multisource early exploration data, and uniformly characterizing the multisource early exploration data as exploration constraint triples comprising position coordinates, interface identifiers, direction vectors and constraint intensity weights; defining a scalar potential field function for each geological interface, wherein the geological interface corresponds to a zero-valued surface of the potential field function; And taking the exploration constraint triplets as condition data, solving each geological potential field as a Gaussian random process by adopting a collaborative kriging interpolation method fused with gradient direction constraints, and obtaining the conditional probability distribution of potential values at any point in space, wherein the mean value of the conditional probability distribution forms the optimal estimated position of an interface, and the variance of the conditional probability distribution forms the spatial uncertainty measure, so that the initial three-dimensional geological probability model is generated.
  3. 3. The method for modeling three-dimensional geologic features based on multi-dimensional digital image contrast recognition according to claim 1, wherein the steps of recognizing and extracting linear structures through deep learning and generating a face structure feature dataset comprising three-dimensional geometric parameters and observation uncertainties comprise: calculating a point cloud curvature map based on the three-dimensional point cloud model of the face; the RGB channel of the real coordinate texture image and the point cloud curvature map are input into a multi-branch feature fusion neural network together, and pixel-level linear construction semantic segmentation and instance segmentation are carried out; For each partitioned construction example, a construction point cluster is extracted from the corresponding three-dimensional point cloud, three-dimensional geometric parameters are calculated through geometric fitting, meanwhile, network identification confidence and geometric fitting residual errors are evaluated, and a measurement error propagation model is combined to generate the face construction characteristic data set.
  4. 4. The method for modeling three-dimensional geologic features based on multi-view digital image contrast recognition of claim 1, wherein associating each construction instance in the face construction feature dataset with a corresponding geologic potential field, creating a feature map, comprises: for the planar construction example in the face construction characteristic data set, an exposure point constraint equation and a potential field gradient direction constraint equation taking the normal vector direction as a reference are established, a mapping operator of the current example to the target geological potential field is formed together, and a characteristic mapping relation is established.
  5. 5. The method of modeling three-dimensional geologic features based on multi-view digital image contrast recognition of claim 1, wherein generating potential field soft constraints based on the observed uncertainties comprises: The overall observation uncertainty is decomposed into geometric measurement errors, identification interpretation errors and representative errors, and the uncertainty of the construction characteristic parameters is converted into soft constraint uncertainty of a potential field constraint equation through error propagation of the characteristic mapping relation.
  6. 6. The method for modeling three-dimensional geologic features based on multi-dimensional digital image contrast recognition of claim 1, wherein bayesian coupling the potential field soft constraints to the initial three-dimensional geologic probability model comprises: And regarding the potential field soft constraint as observation with noise for the prior of the Gaussian process, carrying out regression updating calculation through the Gaussian process, combining the prior distribution of the initial three-dimensional geological probability model with an observation likelihood function, and calculating a posterior mean value field and a posterior covariance field of the posterior joint probability model.
  7. 7. The method for modeling three-dimensional geologic features based on multi-objective digital image contrast recognition as defined in claim 1, wherein iteratively modifying and optimizing the posterior joint probability model based on the newly added face construction feature dataset comprises: continuously inputting a new construction feature data set corresponding to the new tunnel face disclosed by the subsequent tunnel excavation into an incremental Bayesian updating engine; the current posterior joint probability model is used as a new priori, updating is carried out only based on the newly added data and covariance between the newly added data and the model, the optimal estimated position of the geological interface is dynamically refreshed, uncertainty of a set area is reduced, and progressive optimization of the model along with the construction progress is realized.
  8. 8. The method for modeling three-dimensional geologic features based on multi-view digital image contrast recognition as defined in claim 1, The spatial confidence three-dimensional map is generated by extracting posterior variances of the geological interface positions in the posterior joint probability model and normalizing the posterior variances into spatial confidence indexes, and is used for visualizing the credibility of each region of the identification model.
  9. 9. The method for modeling three-dimensional geologic features based on multi-view digital image contrast recognition as defined in claim 1, The on-site geometric calibration of the single-camera multi-view system adopts a self-adaptive calibration method based on a portable coding target range, and the relative pose of a virtual stereo camera is calibrated in a construction site by referencing the camera and is determined through accurate baseline movement so as to ensure that the absolute precision of three-dimensional reconstruction is matched with the on-site conditions.
  10. 10. A three-dimensional geological topography modeling system based on multi-eye digital image comparison and identification is applied to the three-dimensional geological topography modeling method based on multi-eye digital image comparison and identification as claimed in claim 1, The three-dimensional geological topography modeling system based on the multi-view digital image comparison and identification comprises an initial model construction module, a three-dimensional reconstruction module, an uncertainty association module and a model correction module; The initial model construction module is used for constructing a scalar geological potential field based on the previous exploration data and generating an initial three-dimensional geological probability model containing the optimal estimation position of each geological interface space and uncertainty of each geological interface space by adopting a probabilistic interpolation method; The three-dimensional reconstruction module is used for acquiring a face stereopair by using the calibrated single-camera multi-view system, reconstructing and generating a face three-dimensional point cloud model and a real coordinate texture image; The uncertainty association module is used for associating each construction example in the face construction characteristic data set with a corresponding geological potential field to establish a characteristic mapping relation, and performing Bayesian coupling on potential field soft constraint generated based on the observation uncertainty and the initial three-dimensional geological probability model to form a posterior joint probability model; the model correction module is used for taking the posterior joint probability model as a new priori, carrying out iterative correction and optimization on the posterior joint probability model based on the newly added face construction feature data set, extracting an updated three-dimensional geological structure model and generating an associated spatial confidence three-dimensional map.

Description

Three-dimensional geological topography modeling method and system based on multi-eye digital image comparison and identification Technical Field The invention relates to the technical field of geological topography modeling, in particular to a three-dimensional geological topography modeling method and system based on multi-order digital image comparison and identification. Background In the field of underground engineering, the precise three-dimensional cognition of complex geological and geomorphic structures and structures of engineering crossing areas is a foundation stone for guaranteeing engineering safety, optimizing design and construction decision. The target area usually undergoes multi-stage structural movement to form a complex geological structure pattern such as tight folds, multi-stage faults, joint dense bands and the like, and the surface topography with remarkable fluctuation and the underground structure with changeable hidden and changeable are molded under the internal and external dynamic geological action. Traditional geologic modeling methods rely heavily on pre-limited surface mapping, sparse drilling, and geophysical prospecting profiles. Not only is the spatial coverage limited, but there is a great deal of inferences and uncertainty in the characterization of deep complex constructs such as the spatial morphology of faults, spatial distribution and connectivity of joint groups. The modeling process is highly dependent on the experience of a geological engineer, and discrete data are connected into a continuous two-dimensional section through manual interpretation, so that a three-dimensional geological model is constructed. The mode of manual sketching and two-dimensional pushing three-dimensional is not only strong in subjectivity and poor in consistency when facing to hidden structures under complex landforms (such as deep-cut valleys and steep mountains), but also static and disposable in built models, and mass high-precision face information which is disclosed in the construction process and can directly observe internal details of the geological structures cannot be effectively fused. Therefore, the traditional method is difficult to realize the precise three-dimensional modeling which dynamically evolves and quantitatively describes the uncertainty of the geological interface space along with the engineering disclosure, so that the model precision and reliability are insufficient, and the urgent requirements of engineering safety early warning and dynamic design under the condition of complex geological landforms cannot be met. Disclosure of Invention The invention aims to provide a three-dimensional geological landform modeling method and system based on multi-view digital image comparison and identification, which are used for fundamentally converting static, subjective, deterministic into dynamic, objective and probabilistic, and remarkably improving the precision, reliability and engineering guiding value of a model. To achieve the above object, in a first aspect, the present invention provides a modeling method for identifying three-dimensional geological features based on multi-view digital image contrast, comprising the steps of: Constructing a scalar geological potential field based on the pre-exploration data, and generating an initial three-dimensional geological probability model containing the optimal estimation position and uncertainty of each geological interface space by adopting a probabilistic interpolation method; Acquiring a three-dimensional image pair of the face by using a calibrated single-camera multi-view system, and reconstructing to generate a three-dimensional point cloud model and a real coordinate texture image of the face; associating each construction example in the face construction characteristic data set with a corresponding geological potential field to establish a characteristic mapping relation, and performing Bayesian coupling on potential field soft constraint generated based on the observation uncertainty and the initial three-dimensional geological probability model to form a posterior joint probability model; and taking the posterior joint probability model as a new priori, and carrying out iterative correction and optimization on the posterior joint probability model based on the newly added face construction feature data set, extracting an updated three-dimensional geological structure model and generating an associated spatial confidence three-dimensional map. The initial three-dimensional geological potential field probability model construction step specifically comprises the following steps: Acquiring multisource early exploration data, and uniformly characterizing the multisource early exploration data as exploration constraint triples comprising position coordinates, interface identifiers, direction vectors and constraint intensity weights; defining a scalar potential field function for each geological interface, wherein t