A method, device, storage medium, and equipment for intelligent classification of underground ice structures.

Abstract

This invention proposes an intelligent classification method, device, storage medium, and equipment for underground ice structures. It acquires parameterized data of each independent ice structure unit in a permafrost sample to be classified, preprocesses the parameterized data to obtain preprocessed parameterized data, constructs derived features for each independent ice structure unit, and selects target data for classification from the derived features and preprocessed parameterized data. The target data for each independent ice structure unit is then input into a pre-trained classification model to obtain the classification results. By sampling and analyzing the permafrost sample to be classified, target data for determining the category of independent ice structure units is obtained. The classification model is then used to identify and analyze the target data of independent ice structure units, thereby obtaining the classification results. This approach eliminates reliance on manual classification and improves classification efficiency while ensuring consistent classification standards.

Description

Technical Field

[0001] This invention relates to the field of geology, and more specifically, to a method, apparatus, storage medium, and device for intelligent classification of underground ice structures. Background Technology

[0002] Underground ice refers to any type of ice in the Earth's crust, often referred to as the underground solid water reservoir of permafrost regions, with enormous reserves. Under the backdrop of climate warming, the melting of underground ice will significantly alter hydrological processes and water resource allocation patterns in permafrost regions, disrupt ecosystem stability, and break carbon balance. Disasters caused by underground ice melting severely impact regional ecological environments, the safety of major engineering projects, and socio-economic development. Its distribution, morphology, and structural characteristics have a significant influence on the stability of permafrost engineering projects, hydrogeological conditions, and climate change response. In cold-region engineering construction (such as the Qinghai-Tibet Railway, the China-Russia oil and gas pipeline, and highway subgrades in permafrost regions), the type and content of underground ice are key geological parameters determining the stability and long-term service performance of the projects. Different types of underground ice (such as layered ice, vein ice, network ice, and pure ice) exhibit significant differences in their formation mechanisms, spatial distribution patterns, and engineering impacts. Accurate identification and classification of underground ice structure types are crucial for site selection, foundation treatment design, and frost damage early warning in permafrost regions.

[0003] Traditional classification of underground ice structures mainly relies on manual visual judgment, that is, geologists classify ice structure types based on personal experience and subjective judgment by observing CT scan images or on-site drilling core photos. This method is inefficient and lacks standardized criteria. Summary of the Invention

[0004] The purpose of this invention is to provide a method, apparatus, storage medium, and device for intelligent classification of underground ice structures to improve the above-mentioned problems.

[0005] To achieve the above objectives, the technical solutions adopted in the embodiments of the present invention are as follows: In a first aspect, embodiments of the present invention provide an intelligent classification method for underground ice structures, the method comprising: Obtain parameterized data of each independent ice structure unit in the frozen soil sample to be classified. The parameterized data includes at least major axis data, thickness data, surface area data, and volume data. The parameterized data is preprocessed to obtain preprocessed parameterized data, wherein the preprocessing includes numerical span reduction processing and dimension elimination processing; Based on the preprocessed volume data, surface area data, major axis data, and thickness data of the independent ice structural units, derived features of the independent ice structural units are constructed, wherein the derived features include volume-to-surface-area ratio data and length-to-thickness ratio data. From the derived features corresponding to the independent ice structural units and the preprocessed parameterized data, target data for classification is selected, and the target data of the independent ice structural units is input into the pre-trained classification model to obtain the classification result of the independent ice structural units.

[0006] Secondly, embodiments of the present invention provide an intelligent classification device for underground ice structures, the device comprising: The first processing unit is used to acquire parameterized data of each independent ice structure unit in the frozen soil sample to be classified. The parameterized data includes at least major axis data, thickness data, surface area data, and volume data. The first processing unit is further configured to preprocess the parameterized data to obtain preprocessed parameterized data, wherein the preprocessing includes numerical span reduction processing and dimension elimination processing; The first processing unit is further configured to construct derived features of the independent ice structure unit based on the preprocessed volume data, surface area data, major axis data, and thickness data of the independent ice structure unit, wherein the derived features include volume-to-surface-area ratio data and length-to-thickness ratio data. The second processing unit is used to filter out target data for classification from the derived features corresponding to the independent ice structure unit and the preprocessed parameterized data, and input the target data of the independent ice structure unit into the pre-trained classification model to obtain the classification result of the independent ice structure unit.

[0007] Thirdly, embodiments of the present invention provide a storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described method.

[0008] Fourthly, embodiments of the present invention provide an electronic device, the electronic device comprising: a processor and a memory, the memory being used to store one or more programs; when the one or more programs are executed by the processor, the above-described method is implemented.

[0009] Compared to existing technologies, the present invention provides an intelligent classification method, apparatus, storage medium, and device for underground ice structures. This method acquires parameterized data of each independent ice structure unit in a frozen soil sample to be classified. The parameterized data is preprocessed to obtain preprocessed parameterized data. Based on the preprocessed volume, surface area, major axis, and thickness data of the independent ice structure units, derived features of each unit are constructed. Target data for classification is selected from the derived features and the preprocessed parameterized data of each independent ice structure unit. This target data is then input into a pre-trained classification model to obtain the classification results for the independent ice structure units. By sampling and analyzing the frozen soil sample to be classified, target data for determining the category of independent ice structure units is obtained. The classification model is then used to identify and analyze this target data, thereby obtaining the classification results for the independent ice structure units. This eliminates the reliance on manual classification and improves classification efficiency while ensuring consistent classification standards.

[0010] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0011] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention.

[0013] Figure 2 This is one of the flowcharts illustrating the intelligent classification method for underground ice structures provided in this embodiment of the invention.

[0014] Figure 3 This is the second flowchart illustrating the intelligent classification method for underground ice structures provided in this embodiment of the invention.

[0015] Figure 4 This is the third flowchart illustrating the intelligent classification method for underground ice structures provided in this embodiment of the invention.

[0016] Figure 5 This is a feature importance ranking diagram for the random forest model provided in an embodiment of the present invention.

[0017] Figure 6 A thermogram showing the correlation between underground ice structure parameters provided in an embodiment of the present invention.

[0018] Figure 7 The intelligent classification confusion matrix diagram of underground ice structure provided in the embodiments of the present invention.

[0019] Figure 8 Pure ice CT scan images provided for embodiments of the present invention.

[0020] Figure 9 This is a schematic diagram of the three-dimensional morphological reconstruction of pure ice provided in an embodiment of the present invention.

[0021] Figure 10 Example 1: SHAP feature importance analysis diagram of pure ice structure provided for embodiments of the present invention.

[0022] Figure 11 The image provided is a layered ice CT scan image for an embodiment of the present invention.

[0023] Figure 12 This is a schematic diagram of the three-dimensional morphological reconstruction of layered ice provided in an embodiment of the present invention.

[0024] Figure 13 This is a diagram illustrating the importance of SHAP features in layered ice structures provided in an embodiment of the present invention.

[0025] Figure 14 The image provided in this embodiment of the invention is a mesh ice CT scan image.

[0026] Figure 15 This is a schematic diagram of the three-dimensional morphological reconstruction of reticulated ice provided in an embodiment of the present invention.

[0027] Figure 16 This is a feature importance analysis diagram of the SHAP network ice structure provided in an embodiment of the present invention.

[0028] Figure 17 This is a schematic diagram of a unit of the intelligent classification device for underground ice structures provided in an embodiment of the present invention.

[0029] In the diagram: 10-Processor; 11-Memory; 12-Bus; 13-Communication interface; 501-First processing unit; 502-Second processing unit. Detailed Implementation

[0030] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0031] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0032] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this invention, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0033] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0034] In the description of this invention, it should be noted that the terms "upper," "lower," "inner," "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, or the orientation or positional relationship in which the product of this invention is usually placed when in use. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limiting this invention.

[0035] In the description of this invention, it should also be noted that, unless otherwise explicitly specified and limited, the terms "set" and "connection" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0036] The following detailed description of some embodiments of the present invention is provided in conjunction with the accompanying drawings. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0037] Traditional classification of underground ice structures mainly relies on manual visual judgment, which has the following prominent problems: (1) High subjectivity: Different researchers often have inconsistent classification results for the same ice structure sample, and the classification criteria vary from person to person, making it difficult to establish a unified classification system. Related studies have shown that the consistency rate of classification of the same sample by different observers is only 60% to 75%, which seriously restricts the comparability of classification results and the engineering application value.

[0038] (2) Inefficient: Manual classification requires observing and judging each ice structure unit one by one, with a processing speed of about 500 samples / day, while a single CT scan can generate tens of thousands or even hundreds of thousands of ice structure units. Manual classification cannot meet the needs of large-scale data processing. (3) Unable to quantify: Manual classification mainly relies on qualitative descriptions (such as morphological terms like layered and vein-like), lacking quantitative morphological parameters to support it, making it difficult to establish a quantitative relationship between classification results and the physical and mechanical properties of ice structures.

[0039] (4) Lack of model interpretability. Although some studies have attempted to use machine learning methods such as support vector machines (SVM) and convolutional neural networks (CNN) for ice structure classification, most of these methods are black box models and cannot provide a physical basis for classification decisions. Geologists find it difficult to understand why the model classifies a certain sample as a specific type of ice, and the classification results lack physical interpretability, which limits their acceptance and application in engineering practice.

[0040] This invention provides an electronic device, which may be a server device, a computer device, a mobile phone device, etc. Please refer to... Figure 1 This is a schematic diagram of the structure of an electronic device. The electronic device includes a processor 10, a memory 11, and a bus 12. The processor 10 and the memory 11 are connected via the bus 12. The processor 10 is used to execute executable modules, such as computer programs, stored in the memory 11.

[0041] Processor 10 can be an integrated circuit chip with signal processing capabilities. During implementation, each step of the intelligent classification method for underground ice structures can be completed through integrated logic circuits in the hardware or software instructions within processor 10. The aforementioned processor 10 can be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), etc.; it can also be a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field-Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0042] The memory 11 may include high-speed random access memory (RAM) and may also include non-volatile memory, such as at least one disk storage.

[0043] Bus 12 can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus, or an EISA (Extended Industry Standard Architecture) bus, etc. Figure 1 The symbol is represented by a single double-headed arrow, but this does not mean that there is only one bus 12 or one type of bus 12.

[0044] The memory 11 is used to store programs, such as the program corresponding to the intelligent classification device for underground ice structures. The intelligent classification device for underground ice structures includes at least one software functional module that can be stored in the memory 11 in the form of software or firmware or embedded in the operating system (OS) of the electronic device. After receiving an execution instruction, the processor 10 executes the program to implement the intelligent classification method for underground ice structures.

[0045] The electronic device provided in this embodiment of the invention may further include a communication interface 13. The communication interface 13 is connected to the processor 10 via a bus.

[0046] It should be understood that, Figure 1 The structure shown is only a partial schematic diagram of the electronic device; the electronic device may also include components that are larger than... Figure 1 The more or fewer components shown, or having the same Figure 1 The different configurations shown. Figure 1 The components shown can be implemented using hardware, software, or a combination thereof.

[0047] The intelligent classification method for underground ice structures provided in this invention can be applied, but is not limited to, to... Figure 1 For the specific process of the electronic devices shown, please refer to [link / reference]. Figure 2 The intelligent classification method for underground ice structures includes S110, S120, S130 and S140, which are described in detail below.

[0048] S110, Obtain parameterized data of each independent ice structure unit in the frozen soil sample to be classified. The parameterized data shall include at least the major axis data, thickness data, surface area data, and volume data.

[0049] Optionally, the parameterized data includes major axis data (Length_3d, i.e., the maximum projected distance of the ice structure element along the principal axis in three-dimensional space), thickness data (Thickness_3d, the minimum projected distance along the direction perpendicular to the principal axis), minor axis data (Breadth_3d, the intermediate projected distance along the direction perpendicular to both the principal axis and the thickness direction), surface area data (Area_3d, the closed surface area of ​​the ice structure element), volume data (Volume_3d, the spatial volume occupied by the ice structure element), flatness data (defined as the ratio of the minimum eigenvalue to the intermediate eigenvalue of the covariance matrix of the connected components, reflecting the flatness of the ice structure element), aspect ratio data (Length-Breadth ratio, i.e., Length_3d / Breadth_3d), and orientation angle data (Orientation Phi and Orientation). Theta (describing the spatial orientation of ice structural units in three-dimensional space), shape factor data (Shape_3d, a dimensionless parameter comprehensively reflecting the shape complexity of ice structural units), and centroid coordinate data (BaryCenterX, BaryCenterY, BaryCenterZ, the coordinates of the geometric center of the ice structural unit). These parameters quantitatively characterize the geometric morphology, spatial orientation, and shape characteristics of underground ice structures.

[0050] S120, preprocess the parameterized data to obtain preprocessed parameterized data, wherein the preprocessing includes numerical span reduction processing and dimension elimination processing.

[0051] S130. Based on the preprocessed volume data, surface area data, major axis data, and thickness data of the independent ice structure unit, the derived features of the independent ice structure unit are constructed. The derived features include volume-to-surface-area ratio data and length-to-thickness ratio data. The formula for the volume-area-ratio is:

[0052] in, This represents the volume-to-surface area ratio. Represents volume data, Represents surface area data; The volume-area-ratio reflects the ratio of the volume to the surface area of ​​an ice structural unit. Physically, it represents the equivalent characteristic scale of the ice structural unit. A larger value indicates that the ice structure is more spherical or blocky, while a smaller value indicates that the ice structure is more plate-like or needle-like.

[0053] The formula for the aspect ratio is:

[0054] in, This represents the length-to-thickness ratio data. Represents thickness data. Represents the major axis data; The aspect ratio directly reflects the extent of the ice structural unit's extension in one dimension and is a key morphological indicator for distinguishing between layered ice (large aspect ratio) and pure ice (small aspect ratio).

[0055] The reason for introducing the above two derived features is that although the original parameters each carry morphological information of the ice structure, a single parameter is difficult to fully characterize the three-dimensional morphological features of the ice structure. The above two ratio-type derived features can comprehensively reflect the overall morphological tendency of the ice structure in a dimensionless form, effectively improving the discrimination ability of the classification model.

[0056] S140: Target data for classification is selected from the derived features corresponding to independent ice structural units and the preprocessed parameterized data. The target data of independent ice structural units is then input into the pre-trained classification model to obtain the classification results of independent ice structural units.

[0057] Target data can be important data determined based on correlations in historical training data and factors influencing classification results, or data determined based on correlation analysis results of parametric data of individual ice structural units in the permafrost sample to be classified. Target data must include at least derived features. The purpose of screening is to reduce feature dimensionality, eliminate the adverse effects of multicollinearity on model performance, reduce computational overhead, and improve the model's generalization ability. Classification results include layered ice, suspended ice, vein-like ice, porous ice, crust-like ice, network-like ice, and pure ice.

[0058] In one optional implementation, the classification model outputs the classification result of an independent ice structure unit along with the corresponding confidence level. When the confidence level of the classification result is lower than the confidence threshold T (e.g., T=0.8), a manual review can be requested for the independent ice structure unit, and a permafrost expert can make the final judgment. This review mechanism effectively reduces the risk of misclassification in low-confidence areas and improves the overall reliability of the classification system.

[0059] In the intelligent classification method for underground ice structures provided in this embodiment of the invention, target data for determining the category of independent ice structure units is obtained by sampling and analyzing the frozen soil samples to be classified, and the target data of independent ice structure units is identified and analyzed by the classification model, thereby obtaining the classification results of independent ice structure units. This eliminates the reliance on manual classification and improves classification efficiency while ensuring the uniformity of classification standards.

[0060] Optionally, in step S110, parameterized data of each independent ice structure unit in the frozen soil sample to be classified are obtained, including S111, S112 and S113, which are described in detail below.

[0061] S111, scan the frozen soil sample to be classified to obtain the corresponding raw image data.

[0062] The scanning method can be, but is not limited to, industrial CT scanning. The parameters for industrial CT scanning can be: spatial resolution less than 80μm, tube voltage 50~300kV, tube current 70~300μA, scanning angle range of 0°~360°, and projection frame count of not less than 720 frames.

[0063] S112, the original image data is reconstructed into three dimensions using the Filtered Back Projection (FBP) algorithm to obtain three-dimensional volume data.

[0064] S113 utilizes machine learning segmentation algorithms to segment three-dimensional volume data and extract parameterized data of each independent ice structure unit in the frozen soil sample to be classified.

[0065] Optionally, a machine learning segmentation algorithm is used to segment the three-dimensional volume data and extract the parameterized data of each independent ice structure unit in the frozen soil sample to be classified. This includes: using a machine learning segmentation algorithm to separate the ice phase from the soil matrix phase and pores in the three-dimensional volume data to obtain a segmented binary image of the ice phase; performing three-dimensional connected component labeling (using the 26-adjacent connectivity criterion) on the segmented binary image of the ice phase, and taking the closed-loop ice structure formed by the same three-dimensional connected component label as an independent ice structure unit; and extracting the parameterized data of each independent ice structure unit.

[0066] Optionally, in S120, the parameterized data is preprocessed to obtain preprocessed parameterized data, including S121 and S122, which are described in detail below.

[0067] S121, Perform a logarithmic transformation on the first type of parameters in the parameterized data to reduce the numerical span and obtain the first type of parameters after logarithmic transformation; The first type of parameters includes major axis data, thickness data, minor axis data, surface area data, and volume data. These parameters exhibit significant dimensional differences and a wide distribution range. The purpose of applying a large logarithmic transformation is to compress the dynamic range of these parameters, making their distribution closer to a normal distribution and avoiding interference from extreme values ​​in subsequent standardization and model training.

[0068] Alternatively, the formula for the logarithmic transformation is:

[0069] in, This represents the j-th first-class parameter. This represents the first type of parameter after logarithmic transformation.

[0070] S122, perform dimensionless standardization on the second type of parameters and the first type of parameters after logarithmic transformation to obtain standardized parameterized data (i.e., preprocessed parameterized data), wherein the second type of parameters are the parameters in the parameterized data other than the first type of parameters.

[0071] Optionally, the formula for standardization is:

[0072] in, This represents the i-th parameterized data. This represents the i-th parameterized data after standardization. This represents the arithmetic mean of the i-th parameterized data point. This represents the standard deviation corresponding to the i-th parameterized data. and It can be calculated based on the i-th parameterized data of each independent ice structural unit in all training permafrost samples, or it can be calculated based on the i-th parameterized data of each independent ice structural unit in the permafrost sample to be classified. Standardization unifies all parameters to a scale with a mean of 0 and a standard deviation of 1, eliminating the influence of dimensional differences on distance metrics and model weights.

[0073] Please refer to Figure 3 In one optional implementation, after obtaining the classification results of independent ice structure units, the intelligent classification method for underground ice structures further includes: S150, which is described in detail below.

[0074] S150: Obtain the marginal contribution value of each target data when the classification model obtains independent ice structural units, and generate the corresponding contribution report.

[0075] The contribution report demonstrates the attribution of characteristics to the classification results. This interpretability analysis ensures that the classification results are no longer a black box output, but can be traced back to specific physical parameters, making it easier for geologists to understand and verify the rationality of the classification results.

[0076] Alternatively, the SHAP (Shapley Additive explanations) method can be used to obtain the marginal contribution value of each target data when the classification model obtains independent ice structural units.

[0077] Building upon the preceding text, this embodiment of the invention also provides an optional implementation method for the training process of the classification model. Please refer to [link / reference needed]. Figure 4 The training process of the classification model includes steps S210 to S280, which are described in detail below.

[0078] S210, obtain suspected relevant parameters of multiple training ice structural units, wherein the training ice structural units are each independent ice structural unit in the training permafrost sample (which carries artificial classification labels), and the suspected relevant parameters include at least the major axis data, thickness data, surface area data, and volume data.

[0079] S220, preprocess the suspected correlation parameters to obtain preprocessed suspected correlation parameters.

[0080] The preprocessing includes numerical span reduction and dimensionless elimination.

[0081] S230. Based on the preprocessed volume data, surface area data, major axis data, and thickness data of the training ice structure units, the derived features of the training ice structure units are constructed. Among them, the derived features include volume-to-surface-area ratio data and length-to-thickness ratio data.

[0082] S240, combining the suspected target data corresponding to the training ice structure units and the manual classification labels, the random forest algorithm is used to train the classification model for one stage. The suspected target data includes the derived features corresponding to the training ice structure units and the preprocessed suspected relevance parameters.

[0083] S250 determines the importance assessment value of each suspected target data for accurate classification based on the results of the first stage of training.

[0084] Optionally, the classification model is a subsurface ice classification model built based on the random forest algorithm. The hyperparameters of the random forest model are set as follows: number of decision trees n_estimators=100, maximum tree depth max_depth=20, minimum number of split samples min_samples_split=5, and random seed random_state=42. Please refer to [reference needed]. Figure 5 , Figure 5 This is a feature importance ranking diagram for the random forest model provided in an embodiment of the present invention.

[0085] Figure 5 The contribution of each feature to the classification results is shown in the form of a horizontal bar chart, arranged from highest to lowest importance. Volume-to-area ratio (19.0%) and thickness (12.6%) rank first and second, indicating that the density and thickness of ice structures are key indicators for distinguishing different ice types. The derived feature, volume-to-area ratio, has the highest importance, verifying the effectiveness of the feature engineering design of this invention.

[0086] S260, performs correlation analysis on multiple suspected target data from various training ice structure units to determine the comprehensive correlation coefficient of each suspected target data with other suspected target data. For example, it obtains the individual correlation coefficients of the i-th suspected target data with the 1st to (i-1th)th and i+1th to Kth suspected target data, respectively, and performs summation and averaging on all individual correlation coefficients of the i-th suspected target data to obtain its corresponding comprehensive correlation coefficient.

[0087] Please refer to Figure 6 , Figure 6 A thermogram showing the correlation between underground ice structure parameters provided in an embodiment of the present invention.

[0088] The correlation strength between different CT scan data is visualized using a Pearson correlation coefficient matrix, with darker colors indicating stronger correlations. The figure clearly shows the high correlation between the three basic size parameters Length_3d and Thickness_3d, Thickness_3d and Breadth_3d (all correlation coefficients are greater than 0.97), as well as the strong correlation between Area_3d and Volume_3d, providing a basis for subsequent feature selection and dimensionality reduction.

[0089] S270: Target data is selected based on the importance assessment values ​​and comprehensive correlation coefficients of various suspected target data.

[0090] S280, combining the target data corresponding to the training ice structural units and the manual classification labels, uses the random forest algorithm to perform two-stage training on the classification model to obtain a classification model for classifying each independent ice structural unit in the permafrost sample to be classified.

[0091] Optionally, target data can be selected based on the importance assessment values ​​and comprehensive correlation coefficients of various suspected target data, including: excluding suspected target data whose importance assessment values ​​are lower than the first threshold and whose comprehensive correlation coefficients are higher than the second threshold, and using the remaining types of suspected target data as target data.

[0092] Please refer to Figure 7 , Figure 7 This is a confusion matrix diagram for intelligent classification of underground ice structures provided in an embodiment of the present invention. Specifically, it is a confusion matrix diagram obtained based on the intelligent classification method for underground ice structures provided in an embodiment of the present invention.

[0093] Figure 7 The heatmap illustrates the correspondence between the classification model's predictions and the true labels on the test set, with the diagonal elements representing the proportion of correctly classified samples. The graph shows that the classification accuracy for most categories exceeds 85%, validating the effectiveness of the proposed method. The darker color of the diagonal in the confusion matrix indicates the model's strong ability to distinguish between categories.

[0094] To further demonstrate the improvements brought about by the intelligent classification method for underground ice structures provided in the embodiments of the present invention, three examples are given below for further explanation.

[0095] Example 1: Classification of Pure Ice Structures (1) Sample information and CT scan Sample source: G109 National Highway, Zhiduo County, sampling depth 3.2 meters. The sample, located at 3.5m, is part of the permafrost layer. It was a cylindrical, undisturbed permafrost core, approximately 100mm in diameter and 300mm in height. Immediately after sampling, it was transported in a -20°C insulated box and stored at -20°C in the laboratory for analysis. The sample appeared as a uniform, translucent ice mass with a few visible microbubbles, but no obvious stratification or fissures. CT scanning equipment: A Bruker SKYSCAN1273 industrial microfocus CT system (Phoenix, Germany) was used. Scanning parameters: tube voltage 130kV, tube current 300μA, spatial resolution 30μm (voxel size 30×30×30μm), scanning angle range 0-360°, 1923 frames, total scan time approximately 11 minutes. A copper filter (0.5mm thick) was used to reduce X-ray hardening artifacts. The obtained CT images are as follows: Figure 8 As shown. Figure 8 Pure ice CT scan images provided for embodiments of the present invention.

[0096] (2) Image reconstruction and parameter extraction Using the CTVOX reconstruction software provided by Bruker, filtered back projection (FBP) reconstruction was performed, resulting in a 3D grayscale volumetric data with dimensions of 2968×2952×1923 voxels. The grayscale values ​​ranged from 0 to 65535 (16-bit unsigned integers). A machine learning segmentation algorithm based on random forest was used to separate the ice facies from the porosity facies. For the segmented binary images of the ice facies, a 26-adjacency connectivity criterion was used to perform 3D connected component labeling, identifying a total of 12847 independent ice structural units. The Label-Analysis module of AVIZO software was used to extract the structural parameterization data for each ice structural unit. Please refer to [reference needed]. Figure 9 , Figure 9 This is a schematic diagram of the three-dimensional morphological reconstruction of pure ice provided in an embodiment of the present invention, showing a three-dimensional image reconstruction of a layered ice structure.

[0097] (3) Data preprocessing and feature selection Logarithmic transformation is performed on the five parameters Length_3d, Thickness_3d, Breadth_3d, Area_3d, and Volume_3d; then Z-score standardization is performed on all parameters.

[0098] After constructing the derived features Volume_Area_Ratio and Aspect_Ratio, Pearson correlation coefficient analysis was performed on all 15 features (13 original parameters and 2 derived features). Optionally, the results showed that the correlation coefficient between Length_3d and Volume_3d was 0.76, and the correlation coefficient between Area_3d and Volume_3d was 0.97, both belonging to strongly correlated parameter pairs. Based on feature importance comparison, Volume_3d was retained, while Length_3d and Area_3d were removed, ultimately resulting in 14 key classification features.

[0099] (4) Classification results The 14 selected features were input into a trained random forest classification model (n_estimators=100, max_depth=20) to classify 12,847 ice structure units. The classification results are as follows: 11,532 pure ice units (89.7%), 486 layered ice units (3.8%), 312 suspended ice units (2.4%), 298 porous ice units (2.3%), and 219 other types of units (1.7%). The overall classification accuracy was 96.2%. The classification results for this sample, which is predominantly pure ice, are consistent with the visual interpretation by geological experts, validating the effectiveness of this method in classifying high-purity ice samples.

[0100] (5) SHAP analysis results SHAP analysis results showed that the top three features contributing most to the classification decision of pure ice samples were, in descending order: Volume_Area_Ratio (average SHAP absolute value 0.35), Flatness (0.29), and Aspect_Ratio (0.19). Pure ice units typically have a large Volume_Area_Ratio (>500 μm) and a small Flatness absolute value (<0.15), indicating that pure ice tends to be equiaxed in three-dimensional space. This physical characteristic is consistent with the formation mechanism of pure ice (primitive frozen ice, formed by the in-situ freezing and expansion of water in pores). Please refer to [reference needed]. Figure 10 , Figure 10 Example 1: SHAP feature importance analysis diagram of pure ice structure provided for embodiments of the present invention.

[0101] Figure 10The importance ranking of SHAP features for pure ice structures in Example 1 is shown. As clearly seen in the figure, Volume_Area_Ratio, with an average absolute SHAP value of 0.35, ranks first in importance and is the most crucial feature for distinguishing pure ice from other ice types. Flatness (0.29) and Aspect_Ratio (0.19) rank second and third, respectively, together constituting the core discrimination criteria for pure ice classification. These three derived features collectively contribute approximately 78% of the classification decision weight, fully demonstrating the important role of the derived features constructed in pure ice identification.

[0102] Example 2: Classification of Layered Ice Structures (1) Sample information and CT scan Sample source: Section of G214 National Highway from Gonghe to Yushu (Huashixia area), Qinghai Province, sampling depth 2.0~2.6m, located at the bottom of the active permafrost layer. The sample is a cylindrical undisturbed permafrost core with a diameter of 100mm and a height of 390mm. The sample exhibits clear horizontal stratification, with alternating ice and soil layers. The ice layer is approximately 1~5mm thick, and the soil layer is silty clay, grayish-brown in color. This layered ice belongs to the "horizontal layered ice" category of periglacial cryogenic structures, formed by the migration and segregation of water towards the freezing front during multi-year freeze-thaw cycles.

[0103] The CT scanning equipment was the same as in Example 1. Scanning parameters: tube voltage 130kV, tube current 300μA, spatial resolution 40μm (voxel size 40×40×40μm), scanning angle range 0-360°, approximately 4509 projection frames, and total scanning time approximately 10 minutes. A copper filter (0.5mm thick) was used to reduce X-ray hardening artifacts. The obtained CT images are as follows... Figure 11 As shown. Figure 11 The image provided is a layered ice CT scan image for an embodiment of the present invention.

[0104] (2) Image reconstruction and parameter extraction The image reconstruction parameters are the same as in Example 1. Due to the significant grayscale difference between the ice and soil phases in the layered ice sample, a machine learning segmentation algorithm based on random forest was used to separate the ice phase from other phases. For the segmented binary ice phase images, a 26-adjacency connectivity criterion was used to perform three-dimensional connected component labeling, identifying a total of 8,926 independent ice structural units. The structural parameterization data for each ice structural unit was extracted using the Label-Analysis module of AVIZO software. Please refer to [reference needed]. Figure 12 , Figure 12 This is a schematic diagram of the three-dimensional morphological reconstruction of layered ice provided in an embodiment of the present invention.

[0105] (3) Classification results The classification results are as follows: 5,847 lamellar ice units (65.5%), 1,203 suspended ice units (13.5%), 987 porous ice units (11.1%), 412 pure ice units (4.6%), and 477 other types of units (5.3%). The overall classification accuracy was 94.8%. Typical parameters of lamellar ice units are: Length_3d > 1000 μm (average 2,350 μm), Flatness absolute value < 0.3 (average 0.18), aspect ratio between 1.5 and 2.0 (average 1.72), and Aspect_Ratio > 2.0 (average 3.45). These parameters reflect the morphological characteristics of lamellar ice, which is significantly extended in one direction and flattened in the vertical direction, consistent with the formation mechanism of lamellar ice (water migration and segregation along the horizontal direction of the freezing front).

[0106] It is noteworthy that in the boundary region of the layered ice units, the CT grayscale values ​​exhibit a distinct gradient transition zone, approximately 100–300 μm wide. This transition zone corresponds to the unfrozen water film layer at the ice-soil interface. This phenomenon was confirmed by gradient analysis of the CT grayscale profile, indicating that the CT parametric data upon which this method is based can effectively capture the microscopic boundary features of ice structural units.

[0107] (4) SHAP analysis results of layered ice structure To gain a deeper understanding of the decision-making mechanism for layered ice classification, this embodiment conducted a detailed SHAP interpretability analysis on 5847 layered ice units. The analysis results are as follows: Summary table of SHAP analysis results for layered ice structures

[0108] SHAP analysis reveals key decision factors for the classification of layered ice: Flatness is the most important discriminative feature: its average absolute SHAP value reaches 0.45, ranking first in importance. The Flatness value of lamellar ice is usually between 0.15 and 0.30, significantly greater than that of pure ice (<0.15), but less than that of suspended ice (>0.40). A larger Flatness value indicates that lamellar ice is compressed in the thickness direction, exhibiting a flattened morphology, which is highly consistent with the formation mechanism of horizontal segregation of lamellar ice along the freezing front. Aspect_Ratio is the second key feature: its average absolute SHAP value is 0.32, with a positive contribution rate as high as 78.5%, indicating that a larger Aspect_Ratio value is a strong indicator for classifying lamellar ice. The typical Aspect_Ratio value of lamellar ice is between 2.5 and 4.5, reflecting its significant extension in one direction. Volume_Area_Ratio has a negative contribution: 64.8% of the samples show a negative SHAP value, indicating that a smaller Volume_Area_Ratio is beneficial for the classification of lamellar ice. This aligns with the sheet-like morphology of layered ice—sheet structures have a large surface area to volume ratio. Length_3d and Length-Breadth ratio work synergistically: these two features contribute 81.6% and 75.8% positively, respectively, jointly indicating the extensibility of layered ice along its principal axis. When Length_3d > 1500 μm and Length-Breadthratio > 1.5, the probability of the sample being identified as layered ice increases significantly. Please refer to [reference needed]. Figure 13 , Figure 13 This is a diagram illustrating the importance of SHAP features in layered ice structures provided in an embodiment of the present invention.

[0109] Figure 13 This study visually demonstrates the importance ranking of SHAP features for classifying layered ice. Compared to pure ice, the SHAP distribution of layered ice exhibits significantly different characteristics: Flatness jumps to become the primary discriminant factor, while the importance of Volume_Area_Ratio drops to third place. This difference reflects the fundamental morphological distinction between the two ice structures—pure ice is characterized by equiaxed, blocky shapes, while layered ice is characterized by extended, flattened shapes. SHAP analysis provides a quantitative physical framework for understanding this morphological difference.

[0110] Example 3: Classification of Reticulated Ice Structures (1) Sample information and CT scan Sample Source: Section of G214 National Highway from Gonghe to Yushu (Huashixia area), Qinghai Province, sampling depth 4.5-4.8m, belonging to permafrost layer. The sample is a cylindrical undisturbed permafrost core with a diameter of 100mm and a height of 170mm. The sample appears as an irregular network of ice veins interwoven within a grayish-brown silty clay matrix. The ice veins are approximately 0.5-3mm wide, crisscrossing in multiple directions, dividing the soil into irregular polyhedral blocks. This network of ice is a secondary structure of "ice vein network" or "ice wedge," formed by tensile cracks created during repeated freeze-thaw cycles in the permafrost, which are then filled with water and re-frozen. CT Scan Parameters: Tube voltage 130kV, tube current 300μA, spatial resolution 50μm, projection frames 2563, total scan time approximately 7 minutes. Due to the small width and complex distribution of the ice veins in the network of ice samples, a higher tube voltage and higher resolution were used to ensure clear imaging of the ice veins. The obtained CT images are as follows: Figure 14 As shown. Figure 14 The image provided in this embodiment of the invention is a mesh ice CT scan image.

[0111] (2) Image reconstruction and parameter extraction After image reconstruction, the grayscale difference between ice veins and the soil matrix in the 3D grayscale data of the mesh ice sample was not as significant as in the previous two examples (ice vein grayscale approximately 22,000~30,000, soil matrix grayscale approximately 12,000~18,000). A random forest algorithm was used for machine learning segmentation. Since the grayscale values ​​of some small ice veins and the soil matrix were similar, morphological opening operations (structural element radius of 2 voxels) were used to remove noise points after segmentation, and then closing operations (structural element radius of 3 voxels) were used to fill the tiny voids inside the ice veins. After the above processing, a total of 15,632 independent ice structural units were identified.

[0112] Three-dimensional image reconstruction of layered ice structure, such as Figure 15 As shown, Figure 15 This is a schematic diagram of the three-dimensional morphological reconstruction of reticulated ice provided in an embodiment of the present invention.

[0113] (3) Classification results The classification results are as follows: 8214 reticulate ice units (52.6%), 3456 vein-like ice units (22.1%), 1897 porous ice units (12.1%), 1203 layered ice units (7.7%), and 862 other types of units (5.5%). The overall classification accuracy was 93.5%, the weighted F1 score was 0.929, and the Kappa coefficient was 0.912.

[0114] (4) SHAP analysis results of layered ice structure To gain a deeper understanding of the decision-making mechanism for reticulated ice classification, this embodiment conducted a detailed SHAP interpretability analysis on 8214 reticulated ice units. The analysis results are as follows: Summary table of SHAP analysis results for network ice structures

[0115] SHAP analysis revealed the key decision factors for classifying reticulated ice: Volume_Area_Ratio is the most important discriminative feature: with a mean absolute SAP value of 0.34, it ranks first in importance. 77.5% of the samples show negative SAP values, indicating that a very small Volume_Area_Ratio (<150μm) is a strong indicator for classifying reticulated ice. This is highly consistent with the long, needle-like morphological characteristics of reticulated ice. Shape_3d is a hallmark feature of reticulated ice: with a mean absolute SAP value of 0.32 and a positive contribution rate as high as 88.2%, it indicates that a high Shape_3d value (>0.60) is the core criterion for distinguishing reticulated ice from other ice types. Shape_3d effectively quantifies the complex, interwoven, and tortuous geometric characteristics of reticulated ice. Orientation features (Orientation Phi / Theta) contribute significantly: the SAP values ​​of the two orientation features are 0.22 and 0.12, respectively, with a nearly balanced positive / negative contribution ratio, indicating that the absolute value of the orientation angle (i.e., the degree of deviation from the horizontal plane) rather than the orientation angle itself is an important basis for classifying reticulated ice. High dispersion of azimuth angles (standard deviation > 35°) is a typical characteristic of reticulated ice. Aspect_Ratio contributes positively: 58.5% of the samples show a positive SHAP value, indicating that a moderately high Aspect_Ratio (1.8~3.2) is beneficial for the classification of reticulated ice, reflecting its moderate extension in multiple directions. Please refer to [reference needed]. Figure 16 , Figure 16 This is a feature importance analysis diagram of the SHAP network ice structure provided in an embodiment of the present invention.

[0116] Figure 16 This study visually demonstrates the importance ranking of SHAP features in the classification of reticulated ice. Compared to pure ice and laminar ice, the SHAP distribution of reticulated ice exhibits unique characteristics: Volume_Area_Ratio and Shape_3d together constitute the top two discriminant factors, and the importance of orientation spectral features (Orientation Phi / Theta) is significantly increased. This distribution pattern reflects the core logic of reticulated ice classification: its extremely small volume-to-surface-area ratio indicates its elongated shape, its high shape complexity captures its interwoven structure, and its dispersed orientation spectral features characterize its multi-directional extension. SHAP analysis provides a clear quantitative framework for understanding the complex morphological characteristics of reticulated ice.

[0117] Through comparative experiments with existing technical methods, it can be concluded that the intelligent classification method for underground ice structures provided in this embodiment of the invention has the following advantages.

[0118] First, verify the classification accuracy.

[0119] To verify the technical advantages of this invention, sample datasets from Examples 1 to 3 (a total of 37,405 ice structural units) were used to conduct classification and comparison experiments using the following methods: (1) Manual visual classification: Three geologists with more than five years of experience in permafrost research independently interpreted the CT images, and the majority vote was used as the final classification label. The pairwise agreement rates of the three observers were 68.3%, 71.5%, and 66.8%, respectively, and the three-person agreement rate was only 58.2%, indicating that there were significant subjective differences in manual classification. Using the three-person majority vote as the "benchmark label", the overall accuracy of manual classification was 75.2%.

[0120] (2) Support Vector Machine (SVM) classification: Radial basis function (RBF) was used, and the optimal hyperparameters (C=10, γ=0.01) were determined by grid search. The classification accuracy was 88.5%, and the weighted F1 score was 0.876. The SVM method performed well in identifying pure ice and layered ice, but had a high confusion rate with network ice and vein ice.

[0121] (3) Decision tree classification: The CART algorithm was used, with max_depth=10. The classification accuracy was 85.3% and the weighted F1 score was 0.841. A single decision tree is prone to overfitting. The accuracy on the training set (98.7%) is much higher than that on the test set, indicating insufficient generalization ability.

[0122] (4) K-Nearest Neighbors (KNN) classification: The optimal K=7 was determined through cross-validation, with a classification accuracy of 82.1% and a weighted F1 score of 0.808. The KNN method is sensitive to feature scale and suffers from the "curse of dimensionality" in high-dimensional feature spaces.

[0123] (5) The method of this invention (random forest + feature engineering + SHAP interpretability analysis): classification accuracy was 94.8%, weighted F1 score was 0.943, and Kappa coefficient was 0.928. The comparison of each method is shown in the table below: Performance comparison of various classification methods

[0124] This demonstrates that the method of this invention outperforms existing methods in three dimensions: classification accuracy, processing efficiency, and interpretability. Compared to manual classification, the accuracy of the method of this invention is improved by 19.6 percentage points, and the processing speed is improved by approximately 200 times. Compared to traditional machine learning methods such as SVM, decision trees, and KNN, the accuracy of the method of this invention is improved by 6.3, 9.5, and 12.7 percentage points, respectively. Furthermore, the method of this invention, through SHAP interpretability analysis, can provide feature-level attribution explanations for each classification result, an advantage not possessed by existing machine learning methods.

[0125] Please see Figure 17 , Figure 17 An intelligent classification device for underground ice structures is provided as an embodiment of the present invention. Optionally, the intelligent classification device for underground ice structures is applied to the electronic equipment described above.

[0126] The intelligent classification device for underground ice structures includes: a first processing unit 501 and a second processing unit 502.

[0127] The first processing unit 501 is used to acquire parameterized data of each independent ice structure unit in the frozen soil sample to be classified. The parameterized data includes at least major axis data, thickness data, surface area data and volume data. The first processing unit 501 is also used to preprocess the parameterized data to obtain preprocessed parameterized data, wherein the preprocessing includes numerical span reduction processing and dimension elimination processing. The first processing unit 501 is also used to construct derived features of the independent ice structure unit based on the preprocessed volume data, surface area data, major axis data and thickness data of the independent ice structure unit, wherein the derived features include volume-to-surface-area ratio data and length-to-thickness ratio data. The second processing unit 502 is used to filter out target data for classification from the derived features corresponding to the independent ice structure units and the preprocessed parameterized data, and input the target data of the independent ice structure units into the pre-trained classification model to obtain the classification results of the independent ice structure units.

[0128] The second processing unit 502 can execute S140 and S150 as described above, and the first processing unit 501 can execute other steps in the above method embodiments.

[0129] It should be noted that the intelligent classification device for underground ice structures provided in this embodiment can execute the method flow shown in the above-described method flow embodiment to achieve the corresponding technical effects. For the sake of brevity, any parts not mentioned in this embodiment can be referred to the corresponding content in the above-described embodiments.

[0130] This invention also provides a storage medium storing computer instructions and programs, which, when read and executed, perform the intelligent classification method for underground ice structures described in the above embodiments. The storage medium may include memory, flash memory, registers, or a combination thereof.

[0131] The following provides an electronic device, which may be a server device, a computer device, a mobile phone device, etc. This electronic device is as follows: Figure 1 As shown, the above-described intelligent classification method for underground ice structures can be implemented. Specifically, the electronic device includes: a processor 10, a memory 11, and a bus 12. The processor 10 may be a CPU. The memory 11 is used to store one or more programs, which, when executed by the processor 10, execute the intelligent classification method for underground ice structures described in the above embodiment.

[0132] In summary, the intelligent classification method, apparatus, storage medium, and device for underground ice structures provided by this invention acquires parameterized data of each independent ice structure unit in a frozen soil sample to be classified. The parameterized data is preprocessed to obtain preprocessed parameterized data. Based on the preprocessed volume, surface area, major axis, and thickness data of the independent ice structure units, derived features of the independent ice structure units are constructed. From the derived features corresponding to the independent ice structure units and the preprocessed parameterized data, target data for classification is selected. The target data of the independent ice structure units is input into a pre-trained classification model to obtain the classification results of the independent ice structure units. By sampling and analyzing the frozen soil sample to be classified, target data for determining the category of independent ice structure units is obtained. The classification model is then used to identify and analyze the target data of the independent ice structure units, thereby obtaining the classification results of the independent ice structure units. This eliminates the reliance on manual classification and improves classification efficiency while ensuring uniform classification standards.

[0133] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

[0134] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A method for intelligent classification of underground ice structures, characterized in that, The method includes: Obtain parameterized data of each independent ice structure unit in the frozen soil sample to be classified. The parameterized data includes at least major axis data, thickness data, surface area data, and volume data. The parameterized data is preprocessed to obtain preprocessed parameterized data, wherein the preprocessing includes numerical span reduction processing and dimension elimination processing; Based on the preprocessed volume data, surface area data, major axis data, and thickness data of the independent ice structural units, derived features of the independent ice structural units are constructed, wherein the derived features include volume-to-surface-area ratio data and length-to-thickness ratio data. From the derived features corresponding to the independent ice structural units and the preprocessed parameterized data, target data for classification is selected, and the target data of the independent ice structural units is input into the pre-trained classification model to obtain the classification result of the independent ice structural units.

2. The intelligent classification method for underground ice structures as described in claim 1, characterized in that, The process of obtaining parameterized data for each independent ice structural unit in the permafrost sample to be classified includes: The frozen soil samples to be classified were scanned to obtain the corresponding raw image data; A filtered back projection algorithm is used to reconstruct the three-dimensional image data to obtain three-dimensional volume data. The three-dimensional volume data is segmented using a machine learning segmentation algorithm to extract parameterized data of each independent ice structure unit in the frozen soil sample to be classified.

3. The intelligent classification method for underground ice structures as described in claim 2, characterized in that, The process of segmenting the three-dimensional volume data using a machine learning segmentation algorithm to extract parameterized data for each independent ice structure unit in the permafrost sample to be classified includes: The ice phase, soil matrix phase, and pores in the three-dimensional volume data are separated using a machine learning segmentation algorithm to obtain a segmented binary image of the ice phase. Three-dimensional connected component labeling is performed on the segmented binary ice phase image, and closed-loop ice structures formed by the same three-dimensional connected component labeling are used as independent ice structure units; Extract the parameterized data of each independent ice structure unit.

4. The intelligent classification method for underground ice structures as described in claim 1, characterized in that, The preprocessing of the parameterized data to obtain preprocessed parameterized data includes: Logarithmic transformation is performed on the first type of parameters in the parameterized data to reduce the numerical span and obtain the first type of parameters after logarithmic transformation; The first type of parameters includes major axis data, thickness data, minor axis data, surface area data, and volume data. The second type of parameters and the first type of parameters after logarithmic transformation are standardized by eliminating dimensions to obtain standardized parameterized data, wherein the second type of parameters are the parameters in the parameterized data other than the first type of parameters.

5. The intelligent classification method for underground ice structures as described in claim 1, characterized in that, After obtaining the classification results of the independent ice structural units, the method further includes: When the classification model obtains independent ice structural units, it obtains the marginal contribution value of each target data and generates a corresponding contribution report.

6. The intelligent classification method for underground ice structures as described in claim 1, characterized in that, The training process of the classification model includes: Acquire potential related parameters for multiple training ice structure units, wherein the training ice structure unit is each independent ice structure unit in the training permafrost sample, and the potential related parameters include at least major axis data, thickness data, surface area data, and volume data. The suspected relevant parameters are preprocessed to obtain preprocessed suspected relevant parameters; Based on the preprocessed volume data, surface area data, major axis data, and thickness data of the training ice structure units, derived features of the training ice structure units are constructed, wherein the derived features include volume-to-surface-area ratio data and length-to-thickness ratio data. By combining suspected target data corresponding to the training ice structure units and manual classification labels, a random forest algorithm is used to train the classification model for one stage. The suspected target data includes derived features corresponding to the training ice structure units and preprocessed suspected relevance parameters. Based on the results of the first phase of training, determine the importance assessment value of each suspected target data for accurate classification; Correlation analysis was performed on multiple suspected target data from various training ice structural units to determine the comprehensive correlation coefficient of each suspected target data with other suspected target data. Target data was selected based on the importance assessment values ​​and comprehensive correlation coefficients of various suspected target data. By combining the target data corresponding to the training ice structural units and the manual classification labels, the random forest algorithm is used to train the classification model in two stages to obtain a classification model for classifying each independent ice structural unit in the frozen soil sample to be classified.

7. The intelligent classification method for underground ice structures as described in claim 6, characterized in that, The process involves filtering out target data based on the importance assessment values ​​and comprehensive correlation coefficients of various suspected target data, including: Suspected target data with importance assessment values ​​below the first threshold and overall correlation coefficients above the second threshold are excluded, and the remaining types of suspected target data are used as target data.

8. An intelligent classification device for underground ice structures, characterized in that, The device includes: The first processing unit is used to acquire parameterized data of each independent ice structure unit in the frozen soil sample to be classified. The parameterized data includes at least major axis data, thickness data, surface area data, and volume data. The first processing unit is further configured to preprocess the parameterized data to obtain preprocessed parameterized data, wherein the preprocessing includes numerical span reduction processing and dimension elimination processing; The first processing unit is further configured to construct derived features of the independent ice structure unit based on the preprocessed volume data, surface area data, major axis data, and thickness data of the independent ice structure unit, wherein the derived features include volume-to-surface-area ratio data and length-to-thickness ratio data. The second processing unit is used to filter out target data for classification from the derived features corresponding to the independent ice structure unit and the preprocessed parameterized data, and input the target data of the independent ice structure unit into the pre-trained classification model to obtain the classification result of the independent ice structure unit.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the method as described in any one of claims 1-7.

10. An electronic device, characterized in that, include: Processor and memory, the memory being used to store one or more programs; When the one or more programs are executed by the processor, the method as described in any one of claims 1-7 is implemented.

Images