A method and system for automatically constructing visual geological models

Abstract

This invention relates to the field of data analysis, specifically to a method and system for automatically constructing a visualized geological model. The method includes: calculating the degree of feature representation based on a 3D geological model to filter target marker point types; identifying candidate control points; generating candidate type combination schemes and determining the optimal combination type based on the uniformity of distribution; determining the candidate control points corresponding to the optimal combination type to form an optimized control point set and evaluating the reliability of its reference value; performing cross-image feature matching on the control points in the optimized control point set and, combined with the reliability of the reference value, removing abnormal control points to obtain a purified control point set; and constructing and visualizing a 3D geological model based on the purified control point set. This invention can improve the accuracy and efficiency of geological model construction.

Description

Technical Field

[0001] This invention relates to the field of data analysis, and specifically to a method and system for automatically constructing a visual geological model. Background Technology

[0002] Automatic geological modeling based on control points significantly improves modeling efficiency and objectivity. This method rapidly processes massive amounts of data through algorithms, reducing subjective biases caused by human intervention and ensuring that the constructed model remains consistent with mathematical constraints and geological laws. Simultaneously, the automated process facilitates parameter adjustment and ambiguity analysis, thus providing a repeatable and verifiable reliable three-dimensional geological framework for resource assessment and engineering decision-making.

[0003] Traditional methods rely on manual selection of control points, which is not only arduous but also prone to human error. Furthermore, traditional methods fail to adequately consider the accuracy differences between different reference 2D images, leading to insufficient evidence and unreasonable selection during the control point selection process. Summary of the Invention

[0004] This invention provides a method and system for automatically constructing visual geological models to solve existing problems.

[0005] The present invention provides an automatic method for constructing a visual geological model, which employs the following technical solution:

[0006] One embodiment of the present invention provides a method for automatically constructing a visual geological model, the method comprising the following steps:

[0007] Obtain the three-dimensional geological model and its corresponding two-dimensional image data from historical data, and obtain the image dataset to be analyzed for the model to be built;

[0008] Based on the three-dimensional geological model, the characteristic performance degree of each type of geological marker is calculated, and the target marker type is selected according to the characteristic performance degree.

[0009] Identify candidate control points belonging to the target marker type from the image dataset to be analyzed;

[0010] Different categories of target marker point types are combined to generate candidate type combination schemes. Based on the uniformity of the distribution of candidate control points in each image of the image dataset to be analyzed, the optimal combination type is determined from the candidate type combination schemes.

[0011] Candidate control points corresponding to the optimal combination type are determined from the image dataset to be analyzed, forming an optimized control point set. The reference value and credibility of each control point in the optimized control point set are evaluated.

[0012] Cross-image feature matching is performed on the control points in the optimized control point set, and abnormal control points are removed from the optimized control point set based on the reference value credibility, thus obtaining the purified control point set.

[0013] A three-dimensional geological model was constructed and visualized based on the set of purification control points.

[0014] Optionally, based on a three-dimensional geological model, the degree of characteristic representation of each type of geological marker is calculated, and the target marker type is selected according to the degree of characteristic representation, specifically including:

[0015] For each type of geological marker, the mean number of markers in all three-dimensional geological models is calculated to obtain the mean frequency of that type of geological marker.

[0016] Obtain the point density of each geological marker in its corresponding three-dimensional geological model, where the point density is the point density of the geological marker within a preset range around its corresponding three-dimensional geological model;

[0017] For any type of geological marker, calculate the arithmetic mean of the point density of all geological markers of that type to obtain the average distribution density. The ratio of the average distribution density to the maximum average distribution density is determined as the distribution complexity index of that type of geological marker. The maximum average distribution density is the maximum value of the average distribution density among all types of geological markers.

[0018] For each type of geological marker, the reciprocal of the mean frequency and the distribution complexity index are multiplied, and the product is normalized to obtain the characteristic performance of the geological marker of that type.

[0019] Geological marker types whose feature expression level is greater than the preset expression level threshold are identified as target marker types.

[0020] Optionally, candidate control points belonging to the target marker type are identified from the image dataset to be analyzed, specifically including:

[0021] A control point type recognition model is trained based on pixels labeled with geological marker types in two-dimensional image data.

[0022] The training process of the control point type recognition model includes:

[0023] Extract the multi-dimensional features of each labeled geological marker type pixel, and train the neural network with the multi-dimensional features and their corresponding type labels;

[0024] The trained control point type recognition model is used to identify the image dataset to be analyzed. The model outputs the pixel positions that belong to the target marker type as candidate control points.

[0025] Optionally, different categories of target marker types can be combined to generate candidate type combination schemes, specifically including:

[0026] Enumerate non-empty subsets for all target marker types, and treat each subset as a candidate type combination scheme. Alternatively, select different types from all target marker types to form combinations according to a preset range of combination numbers, and treat each combination as a candidate type combination scheme.

[0027] Optionally, based on the uniformity of the distribution of candidate control points in each image of the image dataset to be analyzed, according to different candidate type combination schemes, specifically including:

[0028] For any candidate type combination scheme and any image in the image dataset to be analyzed, obtain the candidate control points in the image that belong to the candidate type combination scheme;

[0029] Calculate the nearest neighbor distance between each pair of candidate control points;

[0030] Find the minimum and maximum values ​​of the nearest neighbor distances and calculate their difference;

[0031] Calculate the absolute deviation of each nearest neighbor distance from the arithmetic mean of the nearest neighbor distances, and sum all the absolute deviations;

[0032] Obtain the arithmetic mean of the nearest neighbor distances, and the total number of candidate control points for this candidate type combination scheme in the image;

[0033] Calculate the product of the arithmetic mean of the nearest neighbor distances and the total number of candidate control points, and use it as the comprehensive distance point count.

[0034] Multiply the reciprocal of the difference, the reciprocal of the sum of absolute deviations, and the reciprocal of the total number of distance points to obtain the uniformity of the distribution of the candidate type combination scheme in the image.

[0035] Calculate the average of the distribution uniformity evaluation values ​​of each candidate type combination scheme across all images, and use it as the comprehensive score of the scheme;

[0036] Compare the comprehensive scores of all candidate type combinations, and determine the combination with the highest comprehensive score as the optimal combination type.

[0037] Optionally, the reliability of the reference value of each control point in the optimized control point set is evaluated, specifically including:

[0038] For any control point in the optimized control point set, obtain the two-dimensional image of that control point;

[0039] The source of the two-dimensional image is determined, and the reliability level number corresponding to the two-dimensional image is obtained according to the preset source reliability grading rules. The smaller the reliability level number, the higher the reliability.

[0040] Obtain the spatial resolution of the two-dimensional image;

[0041] Calculate the ratio of the reliability level number to the highest reliability level number among all sources, take the negative value of the ratio and add one to obtain the source reliability factor;

[0042] The reliability factor of the source is multiplied by the reciprocal of the spatial resolution to obtain the reference value credibility of the control point.

[0043] Optionally, cross-image feature matching is performed on the control points in the optimized control point set, specifically including:

[0044] For any control point in the optimized control point set, extract the local image features of that control point in its corresponding two-dimensional image and generate a feature descriptor for that control point.

[0045] Compare the similarity between feature descriptors of different control points;

[0046] Control points whose similarity exceeds a preset matching threshold and belong to different two-dimensional images are paired to establish matching pairs between control points.

[0047] Based on all matching pairs, construct a set of control points representing the projection of the same geological entity in different images, where the control point set for each geological entity includes at least two mutually matching control points from different images.

[0048] Optionally, based on the reliability of the reference value, abnormal control points are removed from the optimized control point set to obtain the purified control point set, which specifically includes:

[0049] Based on the set of control points, for any control point in the optimized set of control points, the sum of the three-dimensional spatial position deviations between the control point and all other matching control points in the set is calculated as the total position deviation of the control point.

[0050] Calculate the ratio of the reference value confidence of the control point to the maximum reference value confidence of all control points in the optimized control point set, and multiply the ratio by the total position deviation of the control point to obtain the anomaly index of the control point.

[0051] The anomaly index is normalized to obtain normalized outlier values;

[0052] The normalized outlier is compared with the preset outlier threshold. If the normalized outlier is greater than the preset outlier threshold, the control point is determined to be an outlier control point.

[0053] All abnormal control points are removed from the optimized control point set, and the remaining control points constitute the purified control point set.

[0054] The three-dimensional spatial position of any control point is obtained by solving the control point and all other matching control points in the set using the bundle adjustment method.

[0055] Optionally, based on the set of purification control points, a three-dimensional geological model is constructed and visualized, specifically including:

[0056] Using the set of purification control points as high-precision control conditions, the exterior orientation elements and the three-dimensional coordinates of all feature points of each image to be analyzed are optimized and solved by the bundle adjustment algorithm.

[0057] Based on the optimized image exterior orientation elements, multi-view image dense matching is performed to generate dense 3D point cloud;

[0058] Filtering and triangulation of dense 3D point clouds are performed to generate a 3D mesh surface model.

[0059] The image texture of the image dataset to be analyzed is mapped to a three-dimensional mesh surface model to generate a three-dimensional geological model;

[0060] Import the 3D geological model into geological software or a 3D visualization platform for display, rendering, and interactive operations.

[0061] This invention proposes an automatic visualization geological model construction system, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is executed by the processor, it implements the steps of the automatic visualization geological model construction method described above.

[0062] The beneficial effects of the technical solution of the present invention are:

[0063] In this embodiment of the invention, algorithms are used to rapidly process massive amounts of data, reducing subjective biases introduced by human interpretation and ensuring that the constructed model maintains inherent consistency in both mathematical criteria and geological laws. Simultaneously, the automated process facilitates parameter adjustment and ambiguity analysis, thereby providing a repeatable and verifiable reliable three-dimensional geological framework for resource assessment and engineering decision-making. Attached Figure Description

[0064] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0065] Figure 1 A flowchart illustrating an automatic method for constructing a visual geological model, as provided in one embodiment of the present invention;

[0066] Figure 2 This is a structural diagram of an automatic visualization geological model construction system provided in one embodiment of the present invention. Detailed Implementation

[0067] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a method for automatically constructing a visual geological model according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0068] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0069] The following description, in conjunction with the accompanying drawings, details the specific scheme of the automatic construction method for visual geological models provided by this invention.

[0070] This invention provides a method and system for automatically constructing visual geological models. Please refer to [link / reference]. Figure 1 The diagram illustrates a flowchart of an automatic method for constructing a visual geological model according to an embodiment of the present invention, the method comprising the following steps:

[0071] S101. Obtain the three-dimensional geological model and its corresponding two-dimensional image data from historical data, and obtain the image dataset to be analyzed for the model to be constructed.

[0072] For example, the acquisition of basic data in this embodiment specifically includes two categories:

[0073] The first category is historical data, used to learn from and extract key patterns. This part of the data covers three-dimensional geological models completed in previous projects, as well as the corresponding two-dimensional image data on which these models were based. This historical data contains information such as the spatial distribution and type characteristics of various geological markers.

[0074] The second category is current data, which is the dataset of images to be analyzed for which a new model needs to be built. This dataset typically consists of multiple two-dimensional images, representing the latest or target survey data for the area to be modeled.

[0075] By simultaneously acquiring both historical and current data, complete input conditions are provided for subsequent analysis of target marker types based on historical experience and intelligent identification of candidate control points in the current image.

[0076] The acquisition of the first type of data can include the following specific approaches: by systematically sorting and extracting internal digital asset libraries, published literature and software project files, original data packages or standard vector files containing complete spatial coordinates and map projection information are obtained, thereby completely compiling the geological three-dimensional models and their corresponding two-dimensional image data that have been built in historical projects.

[0077] S102. Based on the three-dimensional geological model, calculate the degree of characteristic expression of each type of geological marker and select the target marker type according to the degree of characteristic expression.

[0078] In this embodiment, based on a three-dimensional geological model, the characteristic representation degree of each type of geological marker is calculated, and the target marker type is selected according to the characteristic representation degree, specifically including:

[0079] For each type of geological marker, the mean number of markers in all three-dimensional geological models is calculated to obtain the mean frequency of that type of geological marker.

[0080] Obtain the point density of each geological marker in its corresponding three-dimensional geological model, where the point density is the point density of the geological marker within a preset range around its corresponding three-dimensional geological model;

[0081] For any type of geological marker, calculate the arithmetic mean of the point density of all geological markers of that type to obtain the average distribution density. The ratio of the average distribution density to the maximum average distribution density is determined as the distribution complexity index of that type of geological marker. The maximum average distribution density is the maximum value of the average distribution density among all types of geological markers.

[0082] For each type of geological marker, the reciprocal of the mean frequency and the distribution complexity index are multiplied, and the product is normalized to obtain the characteristic performance of the geological marker of that type.

[0083] Geological marker types whose feature expression level is greater than the preset expression level threshold are identified as target marker types.

[0084] For example, various geological markers are identified and categorized in a historical 3D geological model. Their statistical characteristics and spatial distribution patterns are quantitatively analyzed to assess their suitability as control points. The specific process is as follows:

[0085] For each type of geological marker 'a', calculate its average number of occurrences across all historical models. This is the frequency mean. This value reflects the prevalence of this type of point; the lower the value, the stronger the uniqueness.

[0086] For each geological marker point p (belonging to type a), obtain the number of other geological points in its corresponding 3D geological model, centered on that point and with a preset range R as its radius, and calculate the point density of that point. For type a, calculate the point density of all its geological markers. The arithmetic mean of the values ​​yields the average distribution density of this type. This value reflects the average complexity of the geological structure surrounding this type of point.

[0087] Find the maximum value of the average distribution density among all types. That is, the maximum average distribution density. For type a, its average distribution density... With the maximum average distribution density ratio This ratio is used as an indicator of the distribution complexity of this type of geological marker. The higher the ratio, the more complex and distinctive the geological structure surrounding the marker.

[0088] For type a, its average frequency reciprocal With distribution complexity index Multiply to obtain the initial fitness of this type as a control point. The addition of 0.01 is to prevent the denominator from being zero. Subsequently, for all types... The values ​​are normalized and mapped to the [0,1] interval to obtain the feature representation level of this type. .

[0089] Optionally, 0.01 in this embodiment and all subsequent embodiments is a minimum value set to prevent the denominator from being 0, and can also be other values, such as 0.001, etc.

[0090] Set a preset threshold for feature representation (e.g., 0.6). This will affect the feature representation level. Geological marker types exceeding this threshold are identified as target marker types.

[0091] This method transforms the qualitative experience of geological experts on "key points" into calculable indicators by establishing a quantitative model, and automatically selects landmark point types that are both unique and structurally representative, serving as a reliable source of control points for the subsequent construction of high-precision three-dimensional geological models.

[0092] Optionally, the preset performance threshold can be set and adjusted according to actual modeling accuracy requirements, data quality, or specific application scenarios. In the above embodiments, it is exemplarily set to 0.6 for illustrative purposes only and is not intended to limit the invention. In practical applications, this threshold can be dynamically determined through empirical setting, cross-validation, or optimization algorithms.

[0093] Similarly, the preset parameters involved in other steps of this invention—such as the preset range R for calculating point density, the preset matching threshold for feature matching, and the preset anomaly threshold for identifying abnormal control points—are all configurable parameters. Their specific values ​​can be set and adjusted according to the actual situation of the project, data characteristics, and model accuracy requirements to achieve the best modeling effect. This design enhances the flexibility and applicability of the method, enabling it to adapt to different types and levels of geological modeling tasks.

[0094] S103. Identify candidate control points belonging to the target marker type from the image dataset to be analyzed.

[0095] In this embodiment, identifying candidate control points belonging to the target marker type from the image dataset to be analyzed specifically includes:

[0096] A control point type recognition model is trained based on pixels labeled with geological marker types in two-dimensional image data.

[0097] The training process of the control point type recognition model includes:

[0098] Extract the multi-dimensional features of each labeled geological marker type pixel, and train the neural network with the multi-dimensional features and their corresponding type labels;

[0099] The trained control point type recognition model is used to identify the image dataset to be analyzed. The model outputs the pixel positions that belong to the target marker type as candidate control points.

[0100] For example, for pixels belonging to the same target marker type in historical two-dimensional images, multi-dimensional features are extracted. These features include the specific primitives associated with the point, its spatial relationship with surrounding lines, local curvature patterns, and relevant geological attribute information. Based on these extracted multi-dimensional feature samples and their corresponding target marker type labels, a machine learning model (e.g., a graph-based neural network) is trained under supervision. During training, the model learns the mapping relationship from multi-dimensional features to geological marker types, thereby mastering the comprehensive performance pattern of this type of marker in the image, ultimately forming a control point type recognition model.

[0101] A trained control point type recognition model is used to perform full-map scanning and recognition of the image dataset to be analyzed for geological model construction. The model receives image data as input, automatically analyzes and outputs the pixel positions belonging to each target marker type in the image, thereby obtaining the spatial distribution of candidate control points.

[0102] Through the above steps, the process of identifying and extracting all candidate control points belonging to the target marker type from the image dataset to be analyzed is completed, providing a complete set of alternative points for subsequent optimization and screening.

[0103] S104. Combine different types of target marker points to generate candidate type combination schemes. Based on the uniformity of the distribution of candidate control points in each image of the image dataset to be analyzed, determine the optimal combination type from the candidate type combination schemes.

[0104] In this embodiment, different categories of target marker types are combined to generate candidate type combination schemes, specifically including:

[0105] Enumerate non-empty subsets for all target marker types, and treat each subset as a candidate type combination scheme. Alternatively, select different types from all target marker types to form combinations according to a preset range of combination numbers, and treat each combination as a candidate type combination scheme.

[0106] The uniformity of the distribution of candidate control points in each image of the image dataset to be analyzed is determined based on different combinations of candidate types, specifically including:

[0107] For any candidate type combination scheme and any image in the image dataset to be analyzed, obtain the candidate control points in the image that belong to the candidate type combination scheme;

[0108] Calculate the nearest neighbor distance between each pair of candidate control points;

[0109] Find the minimum and maximum values ​​of the nearest neighbor distances and calculate their difference;

[0110] Calculate the absolute deviation of each nearest neighbor distance from the arithmetic mean of the nearest neighbor distances, and sum all the absolute deviations;

[0111] Obtain the arithmetic mean of the nearest neighbor distances, and the total number of candidate control points for this candidate type combination scheme in the image;

[0112] Calculate the product of the arithmetic mean of the nearest neighbor distances and the total number of candidate control points, and use it as the comprehensive distance point count.

[0113] Multiply the reciprocal of the difference, the reciprocal of the sum of absolute deviations, and the reciprocal of the total number of distance points to obtain the uniformity of the distribution of the candidate type combination scheme in the image.

[0114] Calculate the average of the distribution uniformity evaluation values ​​of each candidate type combination scheme across all images, and use it as the comprehensive score of the scheme;

[0115] Compare the comprehensive scores of all candidate type combinations, and determine the combination with the highest comprehensive score as the optimal combination type.

[0116] For example, a specific implementation method for generating candidate type combination schemes is proposed to achieve optimized combination of target marker point types.

[0117] In one implementation, an exhaustive enumeration method can be used: a complete set is constructed from all the selected target marker types, and all non-empty subsets of this set are systematically enumerated. Each generated subset represents a possible type combination strategy and is defined as a candidate type combination scheme. This method ensures coverage of all possible type combinations, providing a complete search space for subsequent optimization.

[0118] In another implementation, a limited combination method can be used: based on actual modeling needs or prior knowledge, a range for the number of combinations is set (e.g., only combinations consisting of 2 to 4 types are considered). Under this constraint, different types that meet the quantity requirement are selected from all target marker types to form multiple combinations, and each such combination is also considered as a candidate type combination scheme. This method can significantly reduce computational complexity and improve screening efficiency while taking into account the diversity of combinations.

[0119] The above two methods can be used individually or in combination. For example, the scope can be narrowed down first by limiting the combination method, and then enumeration can be performed within the subspace. Finally, the generated candidate type combination schemes will be used as input for subsequent steps to conduct comprehensive evaluation and selection based on criteria such as distribution uniformity, thereby determining the optimal combination type for actual modeling.

[0120] Determining the optimal combination type requires a quantitative evaluation of the distribution uniformity of different candidate combination schemes, and the determination is based on this parameter. For each scheme's candidate control point set in each image, its spatial distribution quality is calculated through multi-index fusion. The specific steps are as follows:

[0121] First, for any candidate type combination scheme and any image in the image dataset to be analyzed, extract all candidate control points belonging to the scheme in the image to form the point set to be evaluated.

[0122] Next, calculate the spatial distribution statistics within this point set:

[0123] Calculate the distance between each point in the point set and its nearest neighbor to obtain a set of nearest neighbor distances.

[0124] Find the minimum value from this set of distances. With the maximum value And calculate the difference. This difference reflects the extreme fluctuation range of the point set distribution; the smaller the difference, the more compact and uniform the distribution.

[0125] Calculate the arithmetic mean of the distances to all nearest neighbors. At the same time, calculate each distance value. Compared with the average absolute deviation And sum them to obtain the total absolute deviation. This value reflects the degree of dispersion of the distance within a set of points relative to the average level; the smaller the value, the more consistent the spacing between points.

[0126] Record the total number N of candidate control points in the point set.

[0127] Furthermore, the above statistics are transformed into evaluation factors with clear geometric meaning:

[0128] Distance range factor: is the range reciprocal The smaller the range, the larger the factor value, indicating a more uniform distribution.

[0129] Distance dispersion factor: the total absolute deviation reciprocal The smaller the dispersion, the larger the value of this factor, indicating better consistency of the point spacing.

[0130] Density distance factor: the product of average distance and number of points. reciprocal This factor comprehensively reflects the average sparsity of points and the total number of points. The larger the value, the more reasonable the total number of points is while maintaining appropriate proximity.

[0131] Multiplying the above three evaluation factors, we obtain the evaluation value W for the uniformity of distribution of this candidate type combination scheme in a single image:

[0132] ;

[0133] The larger the value W, the more uniform, compact, and appropriately sized the control points corresponding to this scheme are distributed in this image.

[0134] The distribution uniformity evaluation value is normalized by using the maximum-minimum normalization method to obtain the normalized distribution uniformity evaluation value, whose value range is [0,1].

[0135] Then, the above calculation is repeated for each image in the dataset to be analyzed, resulting in a set of distribution uniformity evaluation values. The arithmetic mean of these values ​​is calculated as the comprehensive score for this candidate combination scheme. The comprehensive score objectively reflects the average distribution quality of this combination across all images.

[0136] Finally, the comprehensive scores of all candidate type combinations are compared, and the scheme with the highest score is determined as the optimal combination type. This scheme represents a set of globally optimal target marker point type combinations in spatial distribution, which can provide a geometrically constrained and highly reliable control point foundation for subsequent 3D modeling.

[0137] S105. Determine the candidate control points corresponding to the optimal combination type from the image dataset to be analyzed, form an optimized control point set, and evaluate the reference value and credibility of each control point in the optimized control point set.

[0138] In this embodiment, evaluating the reliability of the reference value of each control point in the optimized control point set specifically includes:

[0139] For any control point in the optimized control point set, obtain the two-dimensional image of that control point;

[0140] The source of the two-dimensional image is determined, and the reliability level number corresponding to the two-dimensional image is obtained according to the preset source reliability grading rules. The smaller the reliability level number, the higher the reliability.

[0141] Obtain the spatial resolution of the two-dimensional image;

[0142] Calculate the ratio of the reliability level number to the highest reliability level number among all sources, take the negative value of the ratio and add one to obtain the source reliability factor;

[0143] The reliability factor of the source is multiplied by the reciprocal of the spatial resolution to obtain the reference value credibility of the control point.

[0144] For example, based on the determined optimal combination type, all candidate control points belonging to any type within that combination are extracted from each image in the image dataset to be analyzed. These points from all images belonging to the optimal type combination are then aggregated to form the optimized control point set. This set, after being filtered through two layers of "type suitability" and "distribution uniformity," theoretically possesses good structural representativeness and spatial layout rationality, serving as input for subsequent refined processing.

[0145] For each control point in the optimized control point set, its reference value and reliability need to be evaluated. This evaluation integrates the authority of the data source and the accuracy of the data itself, specifically based on two dimensions:

[0146] Two-dimensional image data from different sources have different levels of reliability. For example, results from national-level geological surveying and mapping institutions are generally the most reliable, followed by authoritative academic publications, while general engineering reports or historical exploration sketches should be treated with caution. Therefore, it is necessary to establish a set of source reliability grading rules in advance, assigning a reliability level number x to each common data acquisition source (generally, the smaller the number, the more reliable). During evaluation, the corresponding level number x can be retrieved based on the source attribute of the image where the control point is located.

[0147] The precision of an image directly affects the accuracy of the locations marked on it. The most direct indicator of precision is the spatial resolution s (unit: meters per pixel). The smaller the value, the stronger the ability to resolve ground details, and theoretically, the higher the precision of the location.

[0148] Combining the above two dimensions, a quantitative model is used to calculate the reference value confidence level E of each control point. The calculation method is as follows:

[0149] ;

[0150] in, It can represent the source reliability factor. This represents the highest reliability level index among all sources, which is also the least reliable source index. It can be calculated from the index, which is then converted into a positive coefficient between 0 and 1. Thus, the closer the source reliability factor is to 1 (the more authoritative the source) and the smaller the resolution s value (the clearer the image), the higher the calculated credibility E value.

[0151] This step assigns a quantified reliability score to each point in the optimized control point set. This score not only identifies the inherent quality differences between points, but more importantly, it provides a crucial basis for identifying and eliminating abnormal control points caused by source errors, insufficient accuracy, or manual annotation errors in subsequent steps.

[0152] S106. Perform cross-image feature matching on the control points in the optimized control point set, and combine the reference value credibility to remove abnormal control points from the optimized control point set to obtain the purified control point set.

[0153] In this embodiment, cross-image feature matching is performed on the control points in the optimized control point set, specifically including:

[0154] For any control point in the optimized control point set, extract the local image features of that control point in its corresponding two-dimensional image and generate a feature descriptor for that control point.

[0155] Compare the similarity between feature descriptors of different control points;

[0156] Control points whose similarity exceeds a preset matching threshold and belong to different two-dimensional images are paired to establish matching pairs between control points.

[0157] Based on all matching pairs, construct a set of control points representing the projection of the same geological entity in different images, where the control point set for each geological entity includes at least two mutually matching control points from different images.

[0158] Based on the reliability of the reference value, abnormal control points are removed from the optimized control point set to obtain the purified control point set, which specifically includes:

[0159] Based on the set of control points, for any control point in the optimized set of control points, the sum of the three-dimensional spatial position deviations between the control point and all other matching control points in the set is calculated as the total position deviation of the control point.

[0160] Calculate the ratio of the reference value confidence of the control point to the maximum reference value confidence of all control points in the optimized control point set, and multiply the ratio by the total position deviation of the control point to obtain the anomaly index of the control point.

[0161] The anomaly index is normalized to obtain normalized outlier values;

[0162] The normalized outlier is compared with the preset outlier threshold. If the normalized outlier is greater than the preset outlier threshold, the control point is determined to be an outlier control point.

[0163] All abnormal control points are removed from the optimized control point set, and the remaining control points constitute the purified control point set.

[0164] The three-dimensional spatial position of any control point is obtained by solving the control point and all other matching control points in the set using the bundle adjustment method.

[0165] For example, the purpose of this embodiment is to optimize the discovery and removal of inconsistent or erroneous points within the control point set. Its core idea is "cross-validation, using high-confidence points as a benchmark to discover and remove outliers." The entire process is divided into two closely connected stages: feature matching to establish geometric relationships, and confidence-guided anomaly detection.

[0166] For each control point in the optimized control point set, a local image region (image patch) is extracted centered on its pixel location in the corresponding 2D image. This image patch is then processed using feature extraction algorithms (such as SIFT, ORB, or features based on convolutional neural networks) to generate a high-dimensional, discriminative feature descriptor. This descriptor encodes the unique geological texture, shape, and pattern information surrounding the point.

[0167] Calculate the similarity (e.g., cosine similarity, reciprocal of Euclidean distance) between each pair of feature descriptors of all control points. When the similarity between two control points exceeds a preset matching threshold, and the two points originate from different 2D images, they are considered to have a high probability of corresponding to the same real-world geological point. Record such point pairs to form a matching pair.

[0168] All matching pairs are analyzed and clustered. Points connected by matching relationships are grouped together to form a control point set. Ideally, such a set contains points from multiple different images, all of which are projections of the same geological entity onto the images. For example, a marker layer feature point exposed on the surface may be identified and marked on aerial image A, satellite image B, and geological profile C; these three points constitute a set.

[0169] Thus, the optimized control point set has been organized from a loose list of points into a network consisting of multiple control point sets, with each set having a strong geometric association assumption that the points are "the same point".

[0170] For a set of control points, suppose it contains k points. These k points are derived from k images. Using bundle adjustment, an optimal estimation algorithm in photogrammetry, the corresponding image observations (pixel coordinates) are combined with possible camera parameters (obtainable from image metadata or coarse estimates) for overall optimization adjustment. One result of the adjustment is calculating the most probable 3D spatial coordinates of these k points. If all point observations are perfect and the model is correct, these coordinates should coincide. However, in reality, due to various errors, they will be scattered over a small area. For any point in the set... Calculate its solved three-dimensional coordinates The sum of the Euclidean distances to all other points in the set is used as the total positional deviation of that point. This value is somewhat quantified. The degree of deviation relative to the overall deviation of all its "partners".

[0171] Not all points with large deviations are necessarily erroneous. The reliability of an observation's reference value depends on its accuracy. If the accuracy is very low (e.g., from a low-precision sketch), then a large deviation is likely understandable measurement noise. Conversely, if a point claims to be highly reliable, the credibility of the observation reference value is higher. If a point is very high, but its location is significantly different from its counterparts, then it is likely the source of the error. Therefore, it is necessary to combine reliability and geometric deviation to determine the anomaly. Specifically, the approach is as follows: First, find the maximum reliability value of the reference value for all points in the optimized control point set. Then, for point... Calculate its relative credibility ratio , This represents the reliability of the reference value of the i-th point. Finally, this ratio is compared with its total positional deviation. Multiply them to obtain an anomaly index. The core logic of this indicator lies in amplifying the deviation weight of high-credibility points. The more "credible" a point is, the greater its suspicion of being abnormal if it deviates from its partners.

[0172] Since the absolute deviation scales of different control point sets may differ, an anomaly index for all points will be used for a unified assessment. Perform normalization (e.g., max-min normalization) to transform the value to the [0,1] interval, and obtain the normalized outliers. Set a preset anomaly threshold T (e.g., 0.8). If the point... of If a point is identified as an anomalous control point, it is determined to be an anomalous control point. All points are iterated over, and all points deemed anomalous are removed from the optimization control point set.

[0173] After the above two stages of processing, points in the original optimized control point set that could not form consistent geometric constraints across multiple perspectives, or points with high credibility but seriously contradicting multi-perspective evidence, were effectively removed. The points that were ultimately retained constituted the purified control point set. This set not only optimized the type combination and spatial distribution, but also had highly consistent geometric evidence across multiple images. Furthermore, the reliability of the data source for each point underwent quantitative evaluation and cross-validation, laying the most solid data foundation for the final construction of a high-precision, high-reliability 3D geological model.

[0174] S107. Based on the set of purification control points, construct and visualize a three-dimensional geological model.

[0175] In this embodiment, a three-dimensional geological model is constructed and visualized based on the set of purification control points, specifically including:

[0176] Using the set of purification control points as high-precision control conditions, the exterior orientation elements and the three-dimensional coordinates of all feature points of each image to be analyzed are optimized and solved by the bundle adjustment algorithm.

[0177] Based on the optimized image exterior orientation elements, multi-view image dense matching is performed to generate dense 3D point cloud;

[0178] Filtering and triangulation of dense 3D point clouds are performed to generate a 3D mesh surface model.

[0179] The image texture of the image dataset to be analyzed is mapped to a three-dimensional mesh surface model to generate a three-dimensional geological model;

[0180] Import the 3D geological model into geological software or a 3D visualization platform for display, rendering, and interactive operations.

[0181] For example, the purpose of this embodiment is to use a high-quality, highly consistent set of purified control points obtained through the aforementioned multi-step rigorous screening and purification process to drive and constrain the entire 3D reconstruction process, ultimately generating a realistic and high-precision 3D geological model suitable for geological analysis. This process integrates professional techniques from photogrammetry, computer vision, and geological modeling, and is implemented as follows:

[0182] The control conditions, with the precise image coordinates of the cleaned control point set as the core, are input into the bundle adjustment algorithm. This algorithm is a large-scale global optimization process. Its goal is to accurately solve for the six exterior orientation elements (i.e., the camera's three-dimensional position and three attitude angles in space at the time of capture) of each image in the dataset to be analyzed, and simultaneously, to accurately calculate the three-dimensional coordinates of a large number of feature points (not limited to control points) corresponding to all matching points in the real world. The cleaned control point set plays a crucial "anchor" role here, greatly constraining the convergence direction and accuracy of the adjustment solution, ensuring that the recovered three-dimensional geometric framework not only conforms to multi-view geometric constraints but also closely matches the actual ground control information, thus laying the foundation for the high geometric accuracy of the model.

[0183] After obtaining the precise exterior orientation elements of each image, pixel-level dense matching is performed between image pairs or image sequences using stereo vision or multi-view stereo matching techniques. This process no longer involves finding sparse feature points, but rather searching for corresponding points in other images for a large number (typically millions or even hundreds of millions) of pixels in one image. Using the forward intersection principle, the three-dimensional coordinates of each successfully matched pixel are calculated, thereby generating an extremely dense 3D point cloud. This point cloud meticulously depicts the undulations and fine structures of the earth's surface or geological bodies with extremely high point density.

[0184] The original dense 3D point cloud is a discrete dataset. To construct a continuous surface model that can be used for analysis and visualization, the following steps are required:

[0185] Filtering and denoising: Remove outlier noise points and non-terrain points (such as vegetation, buildings, etc.) from the point cloud. These can be retained or removed depending on the geological modeling requirements.

[0186] Triangulation construction: The filtered point cloud is connected using triangulation algorithms (such as Delaunay triangulation or Poisson reconstruction) to construct a continuous 3D mesh surface model. This model consists of a large number of interconnected triangular facets, clearly defining the geometric boundaries of the geological body.

[0187] Subsequently, to give the model a realistic visual appearance, texture mapping is performed: the high-resolution images in the dataset to be analyzed are used to solve for the precise exterior orientation elements, which are then used as textures to "wrap" or "fit" onto the corresponding triangular faces of the 3D mesh surface model. The resulting model not only has an accurate 3D shape but also possesses realistic image colors and texture details, becoming a realistic 3D geological model.

[0188] Finally, the generated realistic 3D geological model (usually output in common formats such as OBJ, OSGB, and LAS) is imported into professional geological software (such as GOCAD, Surpac, and Petrel) or a general 3D visualization platform. On these platforms, the model can be displayed and rendered in high fidelity: utilizing rendering techniques such as lighting, shadows, and materials to realistically showcase geological features.

[0189] Interactive operations: Enable model rotation, scaling, translation, cutting (generating arbitrary cross-sections), and transparency adjustment.

[0190] Geological interpretation and attribute assignment: Geological engineers can directly perform professional analysis work on the model, such as drawing geological boundaries, interpreting faults, assigning rock strata attributes, and estimating reserves, transforming the geometric model into a decision support tool containing geological knowledge.

[0191] In summary, this invention employs a four-layer progressive intelligent screening mechanism—based on historical data learning for type selection, globally balanced combination distribution optimization, credibility assessment of data source quality, and multi-view geometric consistency cross-validation—to achieve fully automated processing of control points from "massive candidates" to "refined and reliable" selection. This method significantly overcomes the drawbacks of traditional manual point selection, such as high subjectivity, low efficiency, and neglect of data quality differences. It can automatically, efficiently, and objectively construct three-dimensional geological models with high geometric accuracy, good geological rationality, and realistic visualization, providing a repeatable and verifiable digital decision-making basis for resource assessment, engineering planning, and geological scientific research.

[0192] This invention also proposes an automatic system for constructing visual geological models; please refer to [link / reference]. Figure 2 The diagram shows a structural diagram of an automatic visualization geological model construction system provided by an embodiment of the present invention. The system includes: a data acquisition module 101, a data processing module 102, and a model construction module 103.

[0193] The data acquisition module 101 is used to acquire the three-dimensional geological model and its corresponding two-dimensional image data from historical data, and to acquire the image dataset to be analyzed for the model to be constructed.

[0194] The data processing module 102 is used to calculate the characteristic performance degree of each type of geological marker based on the three-dimensional geological model, and to select the target marker type according to the characteristic performance degree.

[0195] Identify candidate control points belonging to the target marker type from the image dataset to be analyzed;

[0196] Different categories of target marker point types are combined to generate candidate type combination schemes. Based on the uniformity of the distribution of candidate control points in each image of the image dataset to be analyzed, the optimal combination type is determined from the candidate type combination schemes.

[0197] Candidate control points corresponding to the optimal combination type are determined from the image dataset to be analyzed, forming an optimized control point set. The reference value and credibility of each control point in the optimized control point set are evaluated.

[0198] Cross-image feature matching is performed on the control points in the optimized control point set, and abnormal control points are removed from the optimized control point set based on the reference value credibility, thus obtaining the purified control point set.

[0199] Model building module 103 is used to build and visualize a three-dimensional geological model based on the set of purification control points.

[0200] It should be noted that the system provided in the above embodiments is only an example of the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the computer device can be divided into different functional modules to complete all or part of the functions described above. In addition, the visualization geological model automatic construction system and the visualization geological model automatic construction method embodiment provided in the above embodiments belong to the same concept, and their specific implementation process can be found in the method embodiment, which will not be repeated here.

[0201] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0202] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0203] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for automatically constructing a visual geological model, characterized in that, include: Obtain the three-dimensional geological model and its corresponding two-dimensional image data from historical data, and obtain the image dataset to be analyzed for the model to be built; For each type of geological marker, the mean number of markers in all three-dimensional geological models is calculated to obtain the mean frequency of that type of geological marker. Obtain the point density of each geological marker in its corresponding three-dimensional geological model, where the point density is the point density of the geological marker within a preset range around its corresponding three-dimensional geological model; For any type of geological marker, calculate the arithmetic mean of the point density of all geological markers of that type to obtain the average distribution density. The ratio of the average distribution density to the maximum average distribution density is determined as the distribution complexity index of that type of geological marker. The maximum average distribution density is the maximum value of the average distribution density among all types of geological markers. For each type of geological marker, the reciprocal of the mean frequency and the distribution complexity index are multiplied, and the product is normalized to obtain the characteristic performance of the geological marker of that type. Geological marker types with a feature expression level greater than a preset expression level threshold are identified as target marker types; Identify candidate control points belonging to the target marker type from the image dataset to be analyzed; Different categories of target marker point types are combined to generate candidate type combination schemes. Based on the uniformity of the distribution of candidate control points in each image of the image dataset to be analyzed, the optimal combination type is determined from the candidate type combination schemes. Candidate control points corresponding to the optimal combination type are determined from the image dataset to be analyzed, forming an optimized control point set, and the reference value and credibility of each control point in the optimized control point set are evaluated. Cross-image feature matching is performed on the control points in the optimized control point set, and abnormal control points are removed from the optimized control point set based on the reference value credibility, thus obtaining the purified control point set. A three-dimensional geological model was constructed and visualized based on the set of purification control points.

2. The method for automatically constructing a visual geological model according to claim 1, characterized in that, The step of identifying candidate control points belonging to the target marker type from the image dataset to be analyzed specifically includes: A control point type recognition model is trained based on pixels labeled with geological marker types in two-dimensional image data. The training process of the control point type recognition model includes: Extract the multi-dimensional features of each labeled geological marker type pixel, and train the neural network with the multi-dimensional features and their corresponding type labels; The trained control point type recognition model is used to identify the image dataset to be analyzed. The model outputs the pixel positions that belong to the target marker type as candidate control points.

3. The method for automatically constructing a visual geological model according to claim 1, characterized in that, The step of combining different categories of target marker types to generate candidate type combination schemes specifically includes: Enumerate non-empty subsets for all target marker types, and treat each subset as a candidate type combination scheme. Alternatively, select different types from all target marker types to form combinations according to a preset range of combination numbers, and treat each combination as a candidate type combination scheme.

4. The method for automatically constructing a visual geological model according to claim 1, characterized in that, The step of determining the optimal combination type from the candidate control points of different candidate type combination schemes in each image of the image dataset to be analyzed, based on the uniformity of distribution of these control points, specifically includes: For any candidate type combination scheme and any image in the image dataset to be analyzed, obtain the candidate control points in the image that belong to the candidate type combination scheme; Calculate the nearest neighbor distance between each pair of candidate control points; Find the minimum and maximum values ​​of the nearest neighbor distances and calculate their difference; Calculate the absolute deviation of each nearest neighbor distance from the arithmetic mean of the nearest neighbor distances, and sum all the absolute deviations; Obtain the arithmetic mean of the nearest neighbor distances, and the total number of candidate control points for this candidate type combination scheme in the image; Calculate the product of the arithmetic mean of the nearest neighbor distances and the total number of candidate control points, and use it as the comprehensive distance point count. Multiply the reciprocal of the difference, the reciprocal of the sum of absolute deviations, and the reciprocal of the total number of distance points to obtain the uniformity of the distribution of the candidate type combination scheme in the image. Calculate the average of the distribution uniformity evaluation values ​​of each candidate type combination scheme across all images, and use it as the comprehensive score of the scheme; Compare the comprehensive scores of all candidate type combinations, and determine the combination with the highest comprehensive score as the optimal combination type.

5. The method for automatically constructing a visual geological model according to claim 1, characterized in that, The evaluation of the reliability of the reference value of each control point in the optimized control point set specifically includes: For any control point in the optimized control point set, obtain the two-dimensional image of that control point; The source of the two-dimensional image is determined, and the reliability level number corresponding to the two-dimensional image is obtained according to the preset source reliability grading rules. The smaller the reliability level number, the higher the reliability. Obtain the spatial resolution of the two-dimensional image; Calculate the ratio of the reliability level number to the highest reliability level number among all sources, take the negative value of the ratio and add one to obtain the source reliability factor; The reliability factor of the source is multiplied by the reciprocal of the spatial resolution to obtain the reference value credibility of the control point.

6. The method for automatically constructing a visual geological model according to claim 1, characterized in that, The cross-image feature matching of control points in the optimized control point set specifically includes: For any control point in the optimized control point set, extract the local image features of that control point in its corresponding two-dimensional image and generate a feature descriptor for that control point. Compare the similarity between feature descriptors of different control points; Control points whose similarity exceeds a preset matching threshold and belong to different two-dimensional images are paired to establish matching pairs between control points. Based on all matching pairs, construct a set of control points representing the projection of the same geological entity in different images, where the control point set for each geological entity includes at least two mutually matching control points from different images.

7. The method for automatically constructing a visual geological model according to claim 6, characterized in that, The process of combining the reliability of reference value to remove abnormal control points from the optimized control point set, resulting in a purified control point set, specifically includes: Based on the set of control points, for any control point in the optimized set of control points, the sum of the three-dimensional spatial position deviations between the control point and all other matching control points in the set is calculated as the total position deviation of the control point. Calculate the ratio of the reference value confidence of the control point to the maximum reference value confidence of all control points in the optimized control point set, and multiply the ratio by the total position deviation of the control point to obtain the anomaly index of the control point. The anomaly index is normalized to obtain normalized outlier values; The normalized outlier is compared with the preset outlier threshold. If the normalized outlier is greater than the preset outlier threshold, the control point is determined to be an outlier control point. All abnormal control points are removed from the optimized control point set, and the remaining control points constitute the purified control point set. The three-dimensional spatial position of any control point is obtained by solving the control point and all other matching control points in the set using the bundle adjustment method.

8. The method for automatically constructing a visual geological model according to claim 1, characterized in that, The construction and visualization of a three-dimensional geological model based on a set of purification control points specifically includes: Using the set of purification control points as high-precision control conditions, the exterior orientation elements and the three-dimensional coordinates of all feature points of each image to be analyzed are optimized and solved by the bundle adjustment algorithm. Based on the optimized image exterior orientation elements, multi-view image dense matching is performed to generate dense 3D point cloud; Filtering and triangulation of dense 3D point clouds are performed to generate a 3D mesh surface model. The image texture of the image dataset to be analyzed is mapped onto a three-dimensional mesh surface model to generate a three-dimensional geological model; Import the 3D geological model into geological software or a 3D visualization platform for display, rendering, and interactive operations.

9. An automatic visualization geological model construction system, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the computer program is executed by the processor, it implements the steps of the method for automatically constructing a visual geological model as described in any one of claims 1-8.

Images