An Adaptive Octree Three-Dimensional Geological Property Model Compression Method

The 3D geological attribute model is compressed by using an adaptive octree method. The spatial heterogeneity of geological attributes is preserved by using a local attribute complexity function, which solves the problem of insufficient feature preservation in existing methods and achieves efficient data compression and feature fidelity.

CN123118795APending Publication Date: 2026-11-17湖南省地质地理信息所(湖南省地质大数据中心) +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611214862.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-11
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing methods for compressing three-dimensional geological property models are insufficient in preserving the spatial heterogeneity of properties, especially when gradient changes are drastic in local areas, and cannot effectively retain key geological features.

Method used

An adaptive octree method is adopted to preprocess geological attribute data through logarithmic transformation. A local attribute complexity function is constructed by combining indicators such as normalized standard deviation, average gradient strength, information entropy and directional disorder. Bottom-up recursive merging is then performed to establish an adaptive octree compression model, which preserves the geological attribute boundaries and spatial heterogeneity characteristics.

Benefits of technology

While compressing the data size, it effectively preserves the key spatial heterogeneity features and boundaries of geological attributes, avoiding the loss of local features caused by information smoothing in traditional methods, and achieving efficient compression of three-dimensional geological attribute models.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123118795A_ABST
    Figure CN123118795A_ABST
Patent Text Reader

Abstract

This invention relates to the field of geological data processing technology, specifically disclosing an adaptive octree three-dimensional geological attribute model compression method. The method includes: mapping the original voxel data into a three-dimensional discrete array; performing logarithmic transformation preprocessing on geological attributes with strong skewed distribution characteristics; defining indices to quantify the local attribute complexity and numerical error within the three-dimensional grid of the voxel data, providing a basis for octree merging; employing depth-first search to perform bottom-up recursive judgment and merging, realizing voxel aggregation and multi-scale representation, while establishing a quantitative evaluation system oriented towards the preservation of geological attribute features. This invention ensures that the compressed model not only possesses good compression performance but also effectively preserves the spatial heterogeneity of attributes.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application relates to the field of geological data processing technology, and specifically discloses an adaptive octree three-dimensional geological attribute model compression method. Background Technology

[0002] Three-dimensional geological attribute models, as a key representation of the physical properties of underground geological structures, are of great significance for mineral exploration, underground engineering design, and geological environment research. Three-dimensional geological attribute voxel models are widely used to represent the distribution of continuous spatial variables such as porosity and permeability in underground space. However, as model resolution increases, the number of voxels grows exponentially, leading to significant pressure on data storage, transmission, and visualization. Especially in high-precision geological modeling scenarios, single-resolution voxel grids often contain a large amount of redundant information, while simple downsampling (reducing resolution) methods can cause local features of key attribute regions to be ignored, thus affecting model fidelity. How to effectively compress three-dimensional geological attribute models while ensuring numerical accuracy and geological feature fidelity has become a core problem that urgently needs to be solved.

[0003] For data compression, the octree method is adopted. For example, the literature on octree-based three-dimensional model hierarchical merging data compression method is based on the octree method. The octree is divided equally, and the number, level, attributes and orientation attribute values ​​of each node are determined. Then, the real nodes with level (0,0), level (1,1), and level (0,2) are merged in order by determining the common surface. This greatly reduces the number of real nodes, thus achieving better data compression. Combined with linear octree coding and three-dimensional run-length coding, it can greatly reduce the amount of data storage and better achieve compressed data storage.

[0004] However, since geological properties (such as porosity and permeability) usually exhibit significant spatial heterogeneity and show dramatic gradient changes in local areas (such as mineralized enrichment areas or fault-affected zones), existing compression methods have poor balance between accuracy and efficiency for large-scale data, and the spatial heterogeneity characteristics of the properties cannot be preserved during compression. Summary of the Invention

[0005] The purpose of this invention is to solve the problem that existing methods cannot preserve the spatial heterogeneity characteristics of attributes.

[0006] To achieve the above objectives, the basic solution of this invention provides an adaptive octree three-dimensional geological attribute model compression method, comprising the following steps: Step 1, mapping the acquired original voxel data three-dimensional grid into a three-dimensional discrete array, and performing logarithmic transformation preprocessing on the geological attribute fields with strong skewed distribution characteristics in the original voxel data to obtain a three-dimensional attribute model; Step 2, defining an index to quantify the local attribute complexity and numerical error of voxel blocks within the three-dimensional grid of the voxel data. The index is based on normalized standard deviation, normalized average gradient strength, normalized information entropy, and directional disorder to calculate comprehensive local attributes. The complexity function provides the basis for merging the octree; Step 3, the 3D attribute model is used as the root node of the octree, and it is recursively divided into eight child nodes of equal size each time, until the finest granularity of a single voxel is reached to obtain the leaf node. An adaptive octree is constructed based on the local attribute complexity, and a bottom-up recursive judgment is performed by using depth-first search to judge whether the child block meets the merging condition and thus merge it into its parent block, realizing the aggregation of voxels and multi-scale representation. At the same time, a quantitative evaluation system for the preservation of geological attribute features is established, and an adaptive octree attribute compression model is generated.

[0007] The principle and effectiveness of this basic scheme are as follows: Addressing the storage redundancy problem in 3D geological attribute models, a multidimensional attribute complexity function is constructed by integrating attribute change intensity, information entropy, and structural tensor anisotropy. Based on this, a bottom-up octree recursive merging criterion for attribute numerical error constraints is established, and a boundary penalty term is introduced to ensure that key spatial heterogeneity features and geological attribute boundaries are adaptively preserved while significantly compressing the data size. This method overcomes the information smoothing limitations caused by traditional uniform downsampling, achieving efficient compression of 3D geological attribute models.

[0008] Further, in step 1, the three-dimensional grid mapping includes the following sub-steps: Sub-step 1.1, storing the original voxel data used in the form of linear records, each linear record being represented as: ; in, The physical space coordinates of the voxel center. For the geological attribute values ​​of the corresponding location; Sub-step 1.2, establish physical spatial coordinates. With discrete integer index The positive mapping relationship between them is obtained by acquiring the voxel spacing in three spatial directions of the three-dimensional geological attribute model. and the origin of space The conversion from spatial coordinates to discrete integer indices is performed using coordinate index mapping formulas based on rounding. The various coordinate index mapping formulas are as follows: ; ; ; This transformation yields a three-dimensional discrete array for addressing. Sub-step 1.3 expands the dimensions of the three-dimensional discrete array in three dimensions to... In this form, the extended region is filled with NaN and used as a validity mask in subsequent calculations.

[0009] The beneficial effect is that the extended operation ensures that the statistics are calculated based solely on the actual observed data and are not affected by missing values.

[0010] Furthermore, in step 2, the normalized standard deviation is the standard deviation calculated based on the number of voxels, which is then normalized to the [0,1] interval using the same normalization process as the gradient, and is expressed as follows: ; in, Let B be the standard deviation of the number of voxels. This is the theoretical maximum standard deviation estimated based on the global attribute value range. For the introduction of the smallest positive number; The normalized average gradient intensity is obtained by normalizing the average gradient intensity within the voxel block and mapping it to the [0,1] interval, expressed as: ; in, The average gradient intensity within the voxel block. The global maximum gradient intensity is calculated based on the global property range and the minimum voxel spacing. For the introduction of the smallest positive number; The normalized information entropy is a histogram distribution of attribute values ​​within the voxel block. Based on the definition of basic information entropy using the natural logarithm, and normalized to obtain, it is expressed as: ; in, Based on basic information entropy, K represents the maximum information entropy when the voxel block attributes are uniformly distributed across all bins, where K is the number of bins. The directional disorder is obtained by inverting the anisotropy index, and is expressed as: ; The anisotropy index is expressed as: , , These are the maximum and minimum eigenvalues ​​of the gradient-based structure tensor, respectively. This is the smallest positive number introduced.

[0011] The beneficial effect is that by calculating multiple indicators, the spatial heterogeneity of voxel blocks can be characterized from multiple perspectives, such as the intensity of attribute changes, local volatility, and distribution complexity.

[0012] Furthermore, in step 2, the calculation process for the average gradient intensity within the voxel block is as follows: first, define the geological body attribute field. The attribute gradient is: ; The average gradient within the voxel block is then calculated based on the attribute gradient: ; in, Let be the modulus of the p-th individual block.

[0013] The beneficial effect is that by defining the attribute gradient, the spatial variation characteristics of geological body attributes can be characterized.

[0014] Furthermore, in step 2, the gradient-based structure tensor is obtained by de-meaning the gradient components. The structure tensor is then represented as: ; Calculate the eigenvalues ​​of the structure tensor and sort them in descending order. .

[0015] The beneficial effects are: by setting the structural tensor, the spatial orientation structure of geological properties in a local area can be accurately characterized, and the interference of local constant gradient (i.e. linear trend) on the statistics of directional variability can be eliminated.

[0016] Furthermore, in step 2, the comprehensive local attribute complexity function is expressed as: ; in, To normalize the average gradient intensity, To normalize the standard deviation, To normalize information entropy, For directional disorder, , , , The weighting coefficients satisfy .

[0017] The beneficial effect is that by integrating the local attribute complexity function, the spatial heterogeneity of voxel blocks can be characterized from multiple perspectives, such as change intensity, local fluctuation, attribute complexity, and directional structural features.

[0018] Furthermore, in step 2, a boundary-aware penalty is added to the comprehensive local attribute complexity function, expressed as: ; in, This is the penalty coefficient, with a value range of [0,1]. This represents the proportion of valid voxels within a block, i.e., the ratio of the number of non-missing voxels to the total number.

[0019] The beneficial effects are: In actual geological data processing, since boundary regions and missing values ​​(NaN) are common, the addition of boundary-aware penalty can reduce the impact of invalid regions on statistical results, dynamically amplify the local complexity of boundary regions, retain higher grid resolution at the outer edge of the geological attribute model, and thus preserve irregular geological attribute boundaries, avoiding block artifacts caused by excessive merging of edge regions.

[0020] Furthermore, in step 3, the process of recursively determining whether a sub-block meets the merging condition and thus merging it into its parent block includes the following sub-steps: Sub-step 3.1: Initialize leaf nodes. When the recursion depth reaches the maximum and the voxel block size corresponding to the current node is 1, it is regarded as a leaf node. At the same time, the attribute values ​​of the valid voxels are returned to the parent, and the information that merging is allowed is returned. The missing value NaN only exists as a mask and the information that merging is refused is returned directly to the parent. Sub-step 3.2: Collect child node states. During the backtracking phase, the parent node must wait for all 8 child nodes within it to complete traversal and return their states. If any of the 8 child nodes refuses to merge, the parent node will immediately terminate the aggregation operation of the current branch. At the same time, the 8 child nodes will each be directly pushed into the final result set as independent child blocks. Sub-step 3.3: Calculate local attribute features. When all 8 child nodes agree to merge, start calculating the internal features of the parent block. The missing NaN positions in the parent block are only temporarily filled during the gradient calculation stage in step 2. The filling value is calculated from the mean of the effective voxels in the block. Sub-step 3.4, Model Boundary Protection: Before performing the merging decision, first check the number of effective voxels in the parent block. If the number of effective voxels in the block is less than the preset minimum voxel count threshold, it is considered an extremely sparse region and the merging failure is returned directly. For voxel blocks that meet the minimum count requirement, the boundary penalty coefficient is calculated based on the effective voxel ratio to dynamically amplify the complexity index of the edge region. Sub-step 3.5, merge determination. Before determination, calculate the comprehensive local attribute complexity C(B) of the parent block. Sub-block merging must meet the error constraint. Merging is only performed when the comprehensive local attribute complexity C(B) meets the preset constraint merging condition. Sub-step 3.6, complexity consistency check: when the constraint merging condition is met, the attribute complexity between parent and child blocks is checked for consistency according to the consistency check condition. In sub-step 3.7, node merging and coordinate mapping, if all merging conditions in sub-steps 3.5 and 3.6 are met, the node is merged. The average value of the effective child voxel attributes within the parent block is used as the attribute value of the merged parent voxel, and the allowed merging state is passed up the chain. Otherwise, the child node structure is preserved and the subsequent merging of this branch is terminated. Finally, a set of voxel blocks of different scales is output, resulting in an adaptive octree.

[0021] The beneficial effect is that by recursively merging, fine-grained sub-blocks with gradual attribute changes are merged into higher-level coarse-grained parent blocks, thereby reducing data redundancy while preserving local key attribute features, and ultimately generating a non-uniformly distributed multi-scale voxel set.

[0022] Furthermore, in sub-step 3.5, the constraint merging condition is that, for any voxel block B, the comprehensive local attribute complexity function after adding the boundary-aware penalty satisfies: ;in, The attribute complexity threshold is used; in sub-step 3.6, the consistency check condition is: ; in, This is the set of complexity metrics for the 8 child nodes of the current parent block. This is the tolerance parameter.

[0023] The beneficial effects are: such a constraint mechanism can effectively prevent the attribute mutation region from being over-smoothed, thus preserving the geological attribute characteristics. It can also effectively detect the abnormal increase in the complexity of the parent block, avoid the unreasonable phenomenon that the attribute complexity of the parent block is higher than that of the child block, and effectively preserve the local attribute change characteristics during the compression process.

[0024] Furthermore, in sub-step 3.7, for the three-dimensional attribute model with size S and starting index... The voxel block, whose lower left vertex coordinates are... The inverse mapping formula is: ; in, The lower left vertex of the voxel. The original voxel spacing.

[0025] The beneficial effects are: by establishing a precise correspondence between voxel blocks and spatial positions, it is easier to perform 3D visualization of the model in the future, and to ensure spatial consistency under the octree multi-resolution representation. Attached Figure Description

[0026] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0027] Figure 1 A flowchart of an adaptive octree three-dimensional geological attribute model compression method proposed in an embodiment of this application is shown.

[0028] Figure 2 The flowchart of the adaptive octree voxel block recursive merging algorithm in the adaptive octree three-dimensional geological attribute model compression method proposed in the embodiments of this application is shown.

[0029] Figure 3 A schematic block diagram of the octree in an adaptive octree three-dimensional geological attribute model compression method proposed in an embodiment of this application is shown.

[0030] Figure 4 This paper presents a visualization of a three-dimensional geological attribute voxel model compressed using an adaptive octree method proposed in an embodiment of this application. Detailed Implementation

[0031] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0032] An adaptive octree 3D geological attribute model compression method, implementing, for example... Figure 1 As shown, it includes the following steps: Step 1: Map the acquired raw voxel data into a 3D discrete array using a 3D grid. Perform logarithmic transformation preprocessing on the geological attribute fields with strong skewed distribution characteristics in the raw voxel data. v is an attribute field. Using tolerance parameters, a three-dimensional attribute model is obtained. The model consists of voxels, each with corresponding coordinate positions and attribute fields. The three-dimensional attribute model is... Dimensions, number of rows, number of columns, and number of levels are all 1 L represents the partitioning level of the octree.

[0033] The mapping of the original voxel data into a 3D grid includes the following sub-steps: Sub-step 1.1: Store the original voxel data as linear records, each represented as follows: ; in, The physical space coordinates of the voxel center. This refers to the geological attribute value for the corresponding location, such as porosity or permeability.

[0034] Sub-step 1.2: Establish physical space coordinates With discrete integer index The positive mapping relationship between them is obtained by acquiring the voxel spacing in three spatial directions of the three-dimensional geological attribute model. and the origin of space , , representing the voxel spacing of the 3D geological attribute model in three spatial dimensions, and , representing the spatial distance between adjacent voxel centers. A coordinate index mapping formula based on rounding is used to convert spatial coordinates to discrete integer indices. The coordinate index mapping formulas are as follows: ; ; ; This transformation yields a three-dimensional discrete array for computer addressing. That is, through coordinate index mapping, linear record data is reorganized into a size of A three-dimensional structured array.

[0035] Sub-step 1.3 expands the dimensions of the three-dimensional discrete array in three dimensions to To adapt to the requirements of the hierarchical structure in the recursive partitioning of octrees, the extended region of the three-dimensional discrete array is padded with NaN and used as a validity mask in subsequent calculations. This ensures that the statistics are calculated based solely on real observation data and are unaffected by missing values. This algorithm independently introduces three dimensions of independent grid spacing. Therefore, this framework can support the anisotropic voxel structure commonly used in geological modeling, namely cuboid voxel units with varying lengths, widths, and heights, thus improving the algorithm's adaptability and versatility to complex geological data.

[0036] Step 2 defines indices for the local attribute complexity and numerical error of voxel blocks within the 3D grid of quantized voxel data. These indices are calculated based on a comprehensive local attribute complexity function derived from normalized standard deviation, normalized average gradient strength, normalized information entropy, and directional disorder, providing a basis for octree merging. By constructing a comprehensive spatial variation complexity function based on gradient information, the spatial heterogeneity of voxel blocks can be characterized from multiple perspectives, including attribute variation intensity, local volatility, and distribution complexity.

[0037] (1) The normalized standard deviation is the standard deviation calculated based on the number of voxels. The standard deviation is normalized in the same way as the gradient and mapped to the [0,1] interval, as follows: ; in, Let B be the standard deviation of the number of voxels. This is the theoretical maximum standard deviation estimated based on the global attribute value range. This is the smallest positive number introduced.

[0038] The mean squared error (MSE) calculated for a voxel block is a numerical error measure, and the average error is expressed as: ; in, The effective voxel attribute values ​​within the voxel block. This represents the average effective voxel value within the voxel block. This represents the number of valid voxels, i.e., the number of voxels after removing missing attribute values. Since voxels include valid voxels and invalid voxels, invalid voxels are those with the missing value NaN. Defined as the number of effective voxels, and eliminating ineffective voxels.

[0039] According to the theory of bounded variance upper bounds, the standard deviation of the data distribution of any bounded data should theoretically not exceed its range. Since it is half of the global theoretical maximum standard deviation, the global theoretical maximum standard deviation is used as the normalization benchmark. The mean squared error is normalized, and the normalized mean squared error is expressed as: ; in, , The smallest positive number is introduced to ensure the numerical stability of the denominator.

[0040] Normalized mean square error It can reflect the degree of dispersion of the attribute distribution within the current voxel block. When When the value is small, it indicates that the region's attributes change gradually and can be approximated by a single representative value, thus possessing high compression potential.

[0041] Introducing the standard deviation of the attribute value to characterize the local variability within a local region. , represented as .

[0042] Standard deviation of normalized local volatility As an important component of the overall complexity function, it is used to measure the local fluctuation range of the properties inside the voxel block.

[0043] (2) The normalized average gradient intensity is obtained by normalizing the average gradient intensity within the voxel block and mapping it to the [0,1] interval, expressed as: ; in, The average gradient intensity within the voxel block. The global maximum gradient intensity is calculated based on the global property range and the minimum voxel spacing. This is the smallest positive number introduced.

[0044] Gradient intensity is used to describe attribute change characteristics. The calculation process for the average gradient intensity within a voxel block is as follows: First, define the geological body attribute field. The attribute gradient is: ; The gradient is calculated using finite differences based on voxel spacing, and its magnitude is... The gradient reflects the rate of change of attributes in space and is used to characterize the degree of spatial heterogeneity of the attribute field. A large gradient indicates a significant abrupt change in attributes in the region, which may correspond to a transitional region between different geological units. For example, it often manifests as abrupt changes in porosity or permeability near stratigraphic interfaces or faults. In the numerical implementation, only locally missing values ​​are temporarily filled using the intra-block mean to avoid the propagation of NaN during difference calculations. The final index is calculated only based on effective voxels.

[0045] The average gradient within the voxel block is then calculated based on the attribute gradient: ; in, Let be the modulus of the p-th voxel block. Normalize the average gradient to the interval [0,1] to obtain the normalized average gradient intensity, so as to avoid excessive interference from local extreme values ​​on the comprehensive complexity evaluation.

[0046] (3) The normalized information entropy is the histogram distribution of attribute values ​​within the voxel block. , The physical meaning is: Given a randomly selected valid voxel from a 3D voxel grid, and all valid voxel attribute values ​​(e.g., permeability) within the data block are binned into equal-width bins according to their numerical ranges. The probability that the attribute value falls into the i-th bin is given. Based on the natural logarithm, the fundamental information entropy is defined as: ; The result, obtained by normalizing the basic information entropy, is expressed as: ; in, Based on basic information entropy, K represents the maximum information entropy when the voxel block attributes are uniformly distributed across all bins, where K is the number of bins.

[0047] The gradient-based structure tensor underwent mean-reduction processing on the gradient components to eliminate the interference of local constant gradients (i.e., linear trends) on the statistics of directional variability, thus obtaining the gradient components. By defining the structure tensor, we can characterize the spatial directional structure of geological properties within a local region. The structure tensor is represented as: ; Calculate the eigenvalues ​​of the structure tensor and sort them in descending order. The calculation of eigenvalues ​​follows existing methods and will not be elaborated here. Each eigenvalue corresponds to the gradient energy distribution of the local region in the three orthogonal principal directions. When the regional structure is isotropic, the gradient is uniformly distributed in all directions, resulting in the three eigenvalues ​​being close. When there are obvious directional geological structures (such as bedding, faults, or banded structures), the gradient energy is concentrated in the principal directions, resulting in significant separation of the eigenvalues.

[0048] (4) The directional disorder is obtained by inverting the anisotropy index, and is expressed as: ; The anisotropy index is expressed as: , , These are the maximum and minimum eigenvalues ​​of the gradient-based structure tensor, respectively. The anisotropy index characterizes the directional strength by the difference between the maximum and minimum eigenvalues, and normalization eliminates the influence of the overall gradient magnitude, thus achieving a scale-invariant measure of the directionality of geological structures. Directional disorder. Essentially, it is the normalized discreteness of the eigenvalue spectrum of the structure tensor, used to measure the degree of concentration of local gradient energy in a direction. It focuses on quantifying the disorder of spatial directions. When it tends to 1, it means that the local gradient is uniformly diverged in all directions of space and lacks a single dominant structural surface.

[0049] (5) The complexity function of the combined local attributes is expressed as: ; in, The normalized average gradient strength is used to describe the rate of change of an attribute. The normalized standard deviation is used to characterize local volatility. Normalized information entropy is used to quantify the disorder of attribute numerical distribution. directional disorder is used to quantify the disorder of spatial directions. , , , The weighting coefficients satisfy The weighting coefficients are used to balance the contribution of each index to local complexity. This comprehensive local attribute complexity function can characterize the spatial heterogeneity of voxel blocks from multiple perspectives, such as the intensity of change, local volatility, attribute complexity, and directional structural features.

[0050] (6) Add a boundary-aware penalty to the comprehensive local attribute complexity function, expressed as: ; in, This is the penalty coefficient, with a value range of [0,1]. This represents the proportion of valid voxels within a block, i.e., the ratio of the number of non-missing voxels to the total number.

[0051] In actual geological data processing, since boundary regions and missing values ​​(NaN) are common, the addition of boundary-aware penalty can reduce the impact of invalid regions on statistical results.

[0052] To ensure both the preservation of geological boundary regions and compression efficiency, when the number of voxels within a block is 8, merging is directly prohibited if the number of effective sub-voxels within the block is less than a preset minimum voxel count threshold. Furthermore, the formula dynamically amplifies the local complexity of the boundary regions by introducing a continuous penalty term. The higher the proportion of missing effective voxels within a voxel block, the larger the penalty coefficient, making it more difficult for the block to meet the merging threshold condition. This penalty-forced algorithm retains a higher grid resolution at the outer edges of the geological attribute model, thereby preserving irregular geological attribute boundaries and avoiding block artifacts caused by excessive merging of edge regions.

[0053] Compared to a single error threshold, which often fails to distinguish between transitional regions with drastic attribute changes and relatively homogeneous blocks, easily leading to the smoothing or loss of complex local geological structures, this step proposes a local attribute change metric based on spatial complexity constraints, providing a basis for adaptive grid division. This step defines local attribute change from two dimensions: the local numerical error of quantified voxel block attributes and the spatial complexity of attribute variation, comprehensively reflecting the heterogeneous characteristics of geological attributes.

[0054] Step 3: Using the 3D attribute model as the root node of an octree, it is logically recursively divided into eight child nodes of equal size each time, until the finest granularity of a single voxel is reached to obtain a leaf node. An adaptive octree is constructed based on local attribute complexity. A depth-first search is used to perform bottom-up recursive judgment, sequentially judging whether the child block meets the merging condition and merging it into its parent block, realizing voxel aggregation and multi-scale representation. At the same time, a quantitative evaluation system for preserving geological attribute features is established, generating an adaptive octree attribute compression model. Figure 3The diagram shows a four-level octree structure, where the root node is located at Level 0, and the tree divides into three levels of child nodes. All nodes in the diagram use a uniform solid color to illustrate the node hierarchy and relationships. In reality, whether a node continues to be divided depends on its local property complexity; some nodes terminate the division at the middle level, ultimately forming leaf nodes to represent the smallest voxel unit.

[0055] like Figure 2 As shown, the 3D attribute model is used as the root node of an octree. Logically, it is recursively divided into 8 child nodes of equal size each time, until the finest granularity of a single voxel (i.e., a leaf node) is reached. The process of determining whether a child block meets the merging condition and merging it into its parent block includes the following sub-steps: Sub-step 3.1: Initialize leaf nodes. When the recursion depth reaches the maximum and the voxel block size corresponding to the current node is 1, it is regarded as a leaf node. At the same time, the attribute values ​​of the valid voxels are returned to the parent, and the information that merging is allowed is returned. The missing value NaN exists only as a mask and the information that merging is refused is returned directly to the parent.

[0056] Sub-step 3.2: Collect child node states. During the backtracking phase, the parent node must wait for all 8 child nodes within it to complete traversal and return their states. If any of the 8 child nodes refuses to merge, the parent node will immediately terminate the aggregation operation of the current branch. At the same time, the 8 child nodes will each be pushed directly into the final result set as independent child blocks.

[0057] Sub-step 3.3: Calculate local attribute features. When all 8 child nodes agree to merge, start calculating the internal features of the parent block. The NaN positions with missing attributes in the parent block are only temporarily filled in the gradient calculation stage in step 2 to ensure the stability of the gradient calculation. The filling value is calculated from the mean of the effective voxels in the block. This filling can avoid gradient anomalies caused by missing local voxel attribute values.

[0058] Sub-step 3.4, Model Boundary Protection: Before performing the merge determination, first check the number of valid voxels in the parent block. If the number of valid voxels in the block is less than the preset minimum voxel count threshold, it is considered an extremely sparse region and the merge fails directly. The preset minimum voxel count threshold is set according to actual needs. For voxel blocks that meet the minimum count requirement, the boundary penalty coefficient is calculated based on the proportion of valid voxels, and the complexity index of the edge region is dynamically amplified. This boundary protection is used to prevent excessive merging of irregular boundary regions.

[0059] Sub-step 3.5, Merging Decision: Before making the decision, calculate the comprehensive local attribute complexity C(B) of the parent block. To ensure that sub-block merging satisfies error constraints, merging is only performed when the comprehensive local attribute complexity C(B) meets the preset constraint merging condition, thus simultaneously guaranteeing numerical accuracy and preserving attribute space characteristics. The constraint merging condition is: for any voxel block B, the comprehensive local attribute complexity function after adding the boundary-aware penalty satisfies: ;in, This threshold represents the attribute complexity, controlling the degree to which local attribute variations are preserved. Merging operations are only permitted when spatial attribute variations within a region are gradual. This constraint mechanism effectively prevents over-smoothing in areas of abrupt attribute changes, thus preserving the fidelity of geological attribute characteristics. It ensures that merging is performed only in regions with gradual attribute variations and controllable errors, thereby achieving a balance between compression efficiency and attribute structure preservation.

[0060] Sub-step 3.6, complexity consistency check: When the constraint merging condition is met, perform a consistency check on the attribute complexity between parent and child blocks according to the consistency check conditions. The consistency check conditions are: ;in, Right now Figure 2 The parent complexity in the middle, This is the set of complexity metrics for the 8 child nodes of the current parent block, i.e. Figure 2 The sub-complexity in the context, i.e., the consistency check condition, is expressed as: , This is a very small positive number. Consistency constraints prevent the introduction of abnormal attribute changes when forcibly merging regions with abrupt attribute mutations, thus preserving true attribute characteristics. Through consistency checks, the spatial continuity of attribute voxel blocks at different scales is guaranteed, avoiding abnormal and unrealistic attribute change patterns after merging parent blocks. After consistency checks, voxel blocks that meet the merging conditions are selected, providing a data foundation for subsequent node merging.

[0061] Because geological models contain regions of abrupt attribute changes, if adjacent sub-blocks have significantly different attribute characteristics, forced merging may result in the attribute values ​​of the sub-blocks being averaged and assigned to the parent block, thus masking the true local spatial variations in attributes. Consistency checks effectively detect abnormal increases in parent block complexity, preventing unreasonable situations where the parent block's attribute complexity is higher than that of the sub-blocks, and effectively preserving local attribute variation characteristics during compression. Specifically, since leaf layer voxels do not possess internal spatial variation complexity, the algorithm explicitly skips consistency checks at this level when aggregating from the bottom layer to the parent unit.

[0062] Sub-step 3.7, node merging and coordinate mapping: When all merging conditions in sub-steps 3.5 and 3.6 are satisfied, the node is merged. The average value of the effective child voxel attributes within the parent block is used as the merged parent voxel attribute value, and the merging status continues to be passed upwards. Otherwise, the child node structure is preserved, and the subsequent merging of this branch is terminated. Finally, a set of voxel blocks of unequal scales is output, resulting in an adaptive octree. After processing by the adaptive octree, the 3D attribute model is represented as a set of voxel blocks of unequal scales. Each block contains its starting index, scale information, and attribute mean. For a 3D attribute model with size S and starting index S... The voxel block, whose lower left vertex coordinates are... The inverse mapping formula is: ; in, The lower left vertex of the voxel. This represents the original voxel spacing. Mapping facilitates subsequent 3D visualization of the model and ensures spatial consistency across octree multi-resolution representations. Furthermore, for skewed geological attributes with logarithmic preprocessing enabled, their values ​​need to be inversely mapped to the true attribute space using an exponential transformation with offset.

[0063] The resulting adaptive octree attribute compression model achieves orders-of-magnitude compression in terms of data scale while preserving areas of significant geological attribute variation (such as high-gradient anomalies or attribute abrupt change zones) at the geological semantic level. When geological interpretation conditions are available, these areas may correspond to geological structures such as stratigraphic interfaces or faults. This model can be directly used for instantiated rendering, multi-level detail representation, and pixel-stream-based remote interactive applications in 3D visualization systems.

[0064] Dataset: This embodiment uses a three-dimensional permeability attribute model as the test dataset. The entire model contains 1,668,940 attribute voxels, as shown in Table 1. This real geological attribute data, with a scale of millions and strong spatial skewness (permeability spanning multiple orders of magnitude), can test the spatial compression potential of the algorithm of this invention under massive voxel counts, as well as its ability to preserve the spatial heterogeneity of geological attributes (which may correspond to high permeability channels and fault abrupt changes).

[0065] Table 1. Example Data Table

[0066] Evaluation metrics: To comprehensively evaluate the performance of the method in this embodiment in terms of numerical accuracy and property structure preservation, an evaluation metric system including data compression efficiency, numerical reconstruction error and spatial structure consistency is adopted.

[0067] ① Data compression efficiency: The compression efficiency of the model is evaluated using the compression ratio (CR) and space saving ratio (SSR), defined as follows: ; ; in, This represents the number of valid voxels in the original regular grid. This represents the total number of voxel blocks generated after adaptive octree merging. CR intuitively quantifies the data size reduction factor, while SSR reflects the storage space saving as a percentage. These two metrics are used to measure the degree of data size compression by the algorithm and are important criteria for evaluating the engineering performance of the algorithm.

[0068] ② Numerical Reconstruction Error: Regarding numerical fidelity, scientific data compression not only needs to measure the overall average deviation but also requires strict control over local distortions in extreme cases. Therefore, the root mean square error (RMSE) is selected to evaluate the reconstruction results. The formula for this index is: ; in, and These represent the original attribute values ​​and the reconstructed attribute values ​​after transformation to the true attribute space, respectively. RMSE is used to evaluate the average deviation of the overall numerical reconstruction; the lower the RMSE value, the closer the reconstruction result is to the true geological attribute field at the numerical level.

[0069] ③ Spatial Structure Fidelity: Relying solely on numerical errors is insufficient to characterize the preservation of key spatial structure features in geological models. Therefore, the Global Gradient Correlation Ratio (GCR) is further introduced as a macroscopic structure evaluation index, defined as: ; in, and Let represent the three-dimensional gradient vector fields of the original attribute field and the reconstructed attribute field, respectively. Describes the covariance operator. The standard deviation is represented by . In actual calculations, a gradient filtering strategy based on effective masking is adopted, calculating the Pearson correlation coefficient only at locations where the original voxel data is valid and uncontaminated by boundary NaN, thereby avoiding statistical bias caused by gradient distortion. The value range of this index is [-1, 1]. The closer it is to 1, the better the reconstructed model preserves the overall structural trend of the original geological attribute field, reflecting the spatial structure fidelity of the compressed model.

[0070] For key attribute abrupt change regions in geological models (which may correspond to stratigraphic interfaces, fault zones, etc.), a High Gradient Characterization Rate (HGCR) index is designed to evaluate the algorithm's ability to retain spatially heterogeneous attribute characteristics. The HGCR is defined as follows: ; During the calculation, the gradient magnitude of the original attribute field is first obtained, i.e., the gradient. norm , representing the intensity of spatial variation, and extracting the voxels of the most significant gradient change regions (i.e., those greater than the 90th quantile) based on the global gradient magnitude, denoted as . The number of voxels adaptively assigned to the highest and second-highest resolution grid levels by the algorithm in these regions with abrupt changes in spatial properties was counted. HGCR characterizes the recall capability of compressed grids in regions of abrupt attribute mutations. It can quantify and verify whether the model retains sufficient high-density grid resolution in critical transitional zones with strong heterogeneity while pursuing the ultimate compression ratio.

[0071] Data compression efficiency (CR / SSR), numerical reconstruction accuracy (RMSE), and spatial structure fidelity (GCR / HGCR) together constitute a complete evaluation system that can comprehensively evaluate the adaptive octree compression algorithm.

[0072] Model compression metric evaluation: The proposed method's compression efficiency (CR / SSR), numerical reconstruction accuracy (RMSE), and spatial structure fidelity (GCR / HGCR) are analyzed under different compression intensities. This is achieved by adjusting the complexity threshold. This allows for the control of the model's balance between high accuracy and maximum compression efficiency. To systematically evaluate the performance of different compression intensities, this study selected three representative parameter configurations from the grid search experiment, as shown in Table 2, corresponding to low compression mode, balanced mode, and extreme compression mode, respectively.

[0073] Table 2. Quantitative evaluation indicators for compression under different combinations of constraint parameters

[0074] Experimental results show that, with With gradual relaxation of the complexity threshold, the model compression intensity significantly improved, with the compression ratio increasing from 1.90 to 17.69 and the space saving rate increasing from 47.44% to 94.35%. This trend demonstrates that the proposed method can achieve continuously adjustable multi-level compression by adjusting the complexity threshold and the adaptive octree structure. (The last sentence appears to be incomplete and possibly refers to a separate point about balancing parameters.) When the value is 0.40, the model achieves a good trade-off between compression efficiency and numerical accuracy, with RMSE controlled at 57.83.

[0075] Regarding spatial structure fidelity, the proposed method maintains good attribute feature preservation capabilities under different compression intensities. The Global Gradient Relevance (GCR) reaches 0.76 in the balanced mode, indicating that the compressed model maintains high consistency with the original model in terms of overall attribute space variation trends. The High Gradient Retention Rate (HGCR) further validates the algorithm's attribute structure preservation capability, demonstrating that attribute variation trends can be effectively preserved, ensuring spatial continuity across multi-scale representations. In the balanced mode, the HGCR is 83.66%, while it increases to 99.30% in the conservative mode. This change indicates that as the compression intensity increases, the algorithm prioritizes preserving high-gradient key structures while adaptively merging low-gradient redundant regions, thus achieving an attribute structure-driven compression strategy.

[0076] To clearly demonstrate the details of the merged voxel nodes in the octree, this study conducted a visualization comparison of different profiles between the original model and 3D geological property models under different compression levels, such as... Figure 4 As shown, from left to right, the models are the original model, the low compression ratio model, the medium compression ratio model, and the high compression ratio model: (a) global view of the model; (b) X=643000 profile; (c) Y=4148000 profile; (d) Z=-2500 profile. This method allows for a direct representation of the degree to which the spatial distribution characteristics of permeability attributes are preserved. Key attribute structures with significant permeability variations are forcibly terminated due to excessive local complexity and numerical errors, thus being visualized in three-dimensional space as high-resolution, fine-grained voxels. Meanwhile, low-gradient, relatively homogeneous regions far from these key features are smoothly aggregated into larger-scale voxel units. This method not only effectively avoids the interface blurring problem common in traditional uniform sampling methods but also achieves a good balance between geometric resolution and attribute expression accuracy, realistically restoring the spatial heterogeneity and anisotropic distribution characteristics of geological attributes.

[0077] Taking the 3D geological attribute model generated in the low-compression mode as an example for analysis, the hierarchical distribution of the compressed model exhibits obvious non-uniform refinement characteristics, as shown in Table 3. Combined with statistical data, it was found that fine-scale voxels (Level 1 and Level 2) dominate in number, accounting for a total of 95.42%, but covering only 40.46% of the total spatial volume. The distribution pattern of fine-grained voxels with a small proportion of total volume proves that the algorithm accurately allocates fine-grained voxels to key abrupt change areas with complex geological attribute spatial structures and large attribute spans. Conversely, large-scale nodes (Level 8 and Level 16), although accounting for a negligible proportion in number (only about 0.29%), efficiently occupy more than 22.04% of the volume space. This phenomenon, where the number of voxels is inversely proportional to the volume distribution, strongly illustrates that the algorithm effectively reduces geometric redundancy and computational resource waste caused by traditional single-resolution grids in flat areas while ensuring that attribute abrupt change features are not lost.

[0078] Table 3. Statistical table of voxel distribution at each level of the adaptive octree compression model.

[0079] This embodiment uses a 3D geological attribute model as the object. By introducing attribute complexity constraints, it constructs a multi-index evaluation system capable of characterizing the heterogeneity of the attribute field. Based on this, it achieves recursive merging and multi-scale expression of voxel blocks, thereby significantly reducing the data size while ensuring reconstruction accuracy. Through multi-dimensional indices such as attribute change intensity, local volatility, and distribution complexity, it achieves a comprehensive evaluation of the spatial variation characteristics of geological attributes, enabling the octree partitioning process to adaptively focus on regions with high attribute variation. Furthermore, by combining complexity consistency constraints and boundary awareness mechanisms, it effectively avoids the problem of attribute spatial structure distortion during cross-scale merging. This invention achieves a high compression ratio while maintaining good global structural consistency and local high gradient features, balancing numerical accuracy and geological interpretability, providing an effective solution for the efficient storage and rendering of large-scale 3D geological attribute voxel models.

[0080] Compared to existing methods, this embodiment addresses the issue that some geological attributes (such as permeability) often exhibit significant spatial skewness, with their numerical ranges potentially spanning multiple orders of magnitude. Directly performing multi-scale statistics and merging can easily lead to local features being masked by large-scale values. This method introduces a logarithmic transformation into the effective attribute values ​​during data preprocessing, significantly improving statistical stability. During octree compression, local gradient calculations and statistical analyses are performed in its logarithmic space. Finally, in the error assessment and 3D visualization stages, an exponential transformation is used to restore the original attribute space. This approach effectively alleviates the problem of excessively large attribute distribution ranges and reduces their masking effect on local anomalies.

[0081] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any indirect modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. An adaptive octree 3D geologic attribute model compression method, characterized in that, Includes the following steps: Step 1: Map the acquired original voxel data into a three-dimensional discrete array using a three-dimensional grid. Perform logarithmic transformation preprocessing on the geological attribute fields with strong skewed distribution characteristics in the original voxel data to obtain a three-dimensional attribute model. Step 2: Define the indicators of local attribute complexity and numerical error of voxel blocks within the 3D grid of quantized voxel data. The indicators are based on the normalized standard deviation, normalized average gradient strength, normalized information entropy, and directional disorder to calculate a comprehensive local attribute complexity function, providing a basis for octree merging. Step 3: Using the 3D attribute model as the root node of the octree, it is recursively divided into eight child nodes of equal size each time, until the finest granularity of a single voxel is reached to obtain the leaf node. An adaptive octree is constructed based on the local attribute complexity constraint. A depth-first search is used to perform bottom-up recursive judgment, and the child blocks are judged one by one to determine whether they meet the merging conditions and thus merged into their parent blocks, realizing voxel aggregation and multi-scale representation. At the same time, a quantitative evaluation system for the preservation of geological attribute features is established, and an adaptive octree attribute compression model is generated.

2. The adaptive octree 3D geological attribute model compression method according to claim 1, characterized in that, In step 1, the three-dimensional grid mapping includes the following sub-steps: Sub-step 1.1: Store the original voxel data as linear records, each represented as follows: ; wherein, is a physical space coordinate of a voxel center, is a geological property value of the corresponding location; Sub-step 1.2: Establish physical space coordinates With discrete integer index The positive mapping relationship between them is obtained by acquiring the voxel spacing in three spatial directions of the three-dimensional geological attribute model. and the origin of space The conversion from spatial coordinates to discrete integer indices is performed using coordinate index mapping formulas based on rounding. The various coordinate index mapping formulas are as follows: ; ; ; The conversion thus results in a three-dimensional discrete array for addressing ; Sub-step 1.3 expands the dimensions of the three-dimensional discrete array in three dimensions to In this form, the extended region is filled with NaN and used as a validity mask in subsequent calculations.

3. The adaptive octree 3D geologic attribute model compression method of claim 2, wherein, In step 2, the normalized standard deviation is the standard deviation calculated based on the number of voxels. This standard deviation is then normalized to the [0,1] interval using a process consistent with the gradient, and is expressed as follows: ; in, Let B be the standard deviation of the number of voxels. This is the theoretical maximum standard deviation estimated based on the global attribute value range. For the introduction of the smallest positive number; The normalized average gradient intensity is obtained by normalizing the average gradient intensity within the voxel block and mapping it to the [0,1] interval, expressed as: ; in, The average gradient intensity within the voxel block. The global maximum gradient intensity is calculated based on the global property range and the minimum voxel spacing. For the introduction of the smallest positive number; The normalized information entropy is a histogram distribution of attribute values ​​within the voxel block. Based on the definition of basic information entropy using the natural logarithm, and normalized to obtain, it is expressed as: ; in, Based on basic information entropy, K represents the maximum information entropy when the voxel block attributes are uniformly distributed across all bins, where K is the number of bins. The directional disorder is obtained by inverting the anisotropy index, and is expressed as: ; The anisotropy index is expressed as: , , These are the maximum and minimum eigenvalues ​​of the gradient-based structure tensor, respectively. This is the smallest positive number introduced.

4. The adaptive octree three-dimensional geological attribute model compression method according to claim 3, characterized in that, In step 2, the calculation process for the average gradient intensity within the voxel block is as follows: first, define the geological body attribute field. The attribute gradient is: ; The average gradient within the voxel block is then calculated based on the attribute gradient: ; in, Let be the modulus of the p-th individual block.

5. The adaptive octree three-dimensional geological attribute model compression method according to claim 3, characterized in that, In step 2, the gradient-based structural tensor is obtained by demeaning the gradient components. The structure tensor is then represented as: ; Calculate the eigenvalues ​​of the structure tensor and sort them in descending order. .

6. The adaptive octree three-dimensional geological attribute model compression method according to claim 3, characterized in that, In step 2, the comprehensive local attribute complexity function is expressed as: ; in, To normalize the average gradient intensity, To normalize the standard deviation, To normalize information entropy, For directional disorder, , , , The weighting coefficients satisfy .

7. The adaptive octree three-dimensional geological attribute model compression method according to claim 6, characterized in that, In step 2, a boundary-aware penalty is added to the comprehensive local attribute complexity function, expressed as: ; in, This is the penalty coefficient, with a value range of [0,1]. This represents the proportion of valid voxels within a block, i.e., the ratio of the number of non-missing voxels to the total number.

8. The adaptive octree three-dimensional geological attribute model compression method according to claim 7, characterized in that, In step 3, the recursive determination process of whether a sub-block meets the merging condition and thus merges it into its parent block includes the following sub-steps: Sub-step 3.1: Initialize leaf nodes. When the recursion depth reaches the maximum and the voxel block size corresponding to the current node is 1, it is regarded as a leaf node. At the same time, the attribute values ​​of the valid voxels are returned to the parent, and the information that merging is allowed is returned. The missing value NaN only exists as a mask and the information that merging is refused is returned directly to the parent. Sub-step 3.2: Collect child node states. During the backtracking phase, the parent node must wait for all 8 child nodes within it to complete traversal and return their states. If any of the 8 child nodes refuses to merge, the parent node will immediately terminate the aggregation operation of the current branch. At the same time, the 8 child nodes will each be directly pushed into the final result set as independent child blocks. Sub-step 3.3: Calculate local attribute features. When all 8 child nodes agree to merge, start calculating the internal features of the parent block. The missing NaN positions in the parent block are only temporarily filled during the gradient calculation stage in step 2. The filling value is calculated from the mean of the effective voxels in the block. Sub-step 3.4, Model Boundary Protection: Before performing the merging decision, first check the number of effective voxels in the parent block. If the number of effective voxels in the block is less than the preset minimum voxel count threshold, it is considered an extremely sparse region and the merging failure is returned directly. For voxel blocks that meet the minimum count requirement, the boundary penalty coefficient is calculated based on the effective voxel ratio to dynamically amplify the complexity index of the edge region. Sub-step 3.5, merge determination. Before determination, calculate the comprehensive local attribute complexity C(B) of the parent block. Sub-block merging must meet the error constraint. Merging is only performed when the attribute complexity C(B) meets the preset constraint merging conditions. Sub-step 3.6, complexity consistency check: when the constraint merging condition is met, the attribute complexity between parent and child blocks is checked for consistency according to the consistency check condition. In sub-step 3.7, node merging and coordinate mapping, if all merging conditions in sub-steps 3.5 and 3.6 are met, the node is merged. The average value of the effective child voxel attributes within the parent block is used as the attribute value of the merged parent voxel, and the allowed merging state is passed up the chain. Otherwise, the child node structure is preserved and the subsequent merging of this branch is terminated. Finally, a set of voxel blocks of different scales is output, resulting in an adaptive octree.

9. The adaptive octree three-dimensional geological attribute model compression method according to claim 8, characterized in that, In sub-step 3.5, the constraint merging condition is that, for any voxel block B, the comprehensive local attribute complexity function after adding the boundary-aware penalty satisfies: ; in, This represents the attribute complexity threshold. In sub-step 3.6, the consistency check condition is: ; in, This is the set of complexity metrics for the 8 child nodes of the current parent block. This is the tolerance parameter.

10. The adaptive octree three-dimensional geological attribute model compression method according to claim 8, characterized in that, In sub-step 3.7, for a three-dimensional attribute model with size S and starting index... The voxel block, whose lower left vertex coordinates are... The inverse mapping formula is: ; in, The lower left vertex of the voxel. The original voxel spacing.