Machine learning based detection of ice content in the ground

Abstract

This invention relates to the field of machine learning technology, and more particularly to a machine learning-based method for detecting underground ice content. The method includes: uniformly dividing the detection area into several sub-regions; determining ice-containing areas based on parameter characterization values; determining spatiotemporal variation labels corresponding to the detection areas or correcting regional parameters based on the proportion of ice-containing areas and feature tendency values; inputting measured sensor parameters into a Clapeyron equation model and an underground permafrost model respectively, and calculating the ice-to-water ratio based on the model output results; and searching a database based on the spatiotemporal variation labels, including: extracting only data from sample datasets and matching them with measured sensor parameters, and calling the sample datasets based on the matching results; or: aggregating data based on the differences in data features of sample datasets and corresponding preprocessed parameters, filtering and analyzing each aggregated dataset, and calling the aggregated datasets. This invention can improve the accuracy of machine learning-based underground ice content detection.

Description

Technical Field

[0001] This invention relates to the field of machine learning technology, and in particular to a method for detecting underground ice content based on machine learning. Background Technology

[0002] As a key hydrological and engineering geological element in permafrost regions, underground ice content directly affects permafrost stability, regional water resource circulation, and ecological environment evolution, holding core research value in areas such as railway construction in cold regions, oil and gas resource extraction, and climate change response. Accurately obtaining underground ice content is a prerequisite for solving problems such as engineering disease prevention and permafrost degradation assessment in cold regions, thus becoming an important research direction in the field of cold region science and engineering technology. The development of technologies such as cold region geophysical exploration and remote sensing monitoring has accumulated a large amount of multi-source data, providing rich training samples for machine learning models. Machine learning can achieve multi-source data fusion analysis, automatically extract deep features related to ice content, avoid interference from single data sources, and improve detection robustness.

[0003] Chinese Patent Publication No. CN119151368A discloses an intelligent monitoring and assessment method and system for terrestrial carbon disturbance in permafrost regions, belonging to the field of carbon cycle research in permafrost regions. The method includes: collecting data on 13 background elements within the permafrost region; calculating carbon disturbance assessment indicators for the permafrost region; conducting a reliability assessment of the carbon disturbance assessment indicators for the permafrost region; monitoring the characteristics and changes in regional background elements and river particulate organic carbon output in the permafrost region, and calculating the rate of change of the "carbon disturbance assessment indicators" based on the monitoring data; using numerical models to identify high-risk areas and periods of carbon disturbance in the permafrost region, and identifying key disturbance factors and their impacts; analyzing the changing trends of the carbon disturbance assessment indicators, and establishing a predictive model using machine learning methods. This method is comprehensive, dynamic, effective, and supports carbon disturbance attribution and predictive analysis. However, the above scheme has the following problems: it fails to combine multiple models and calculation methods to correct the preprocessed data to improve data accuracy, and the classification and summarization of a large number of data samples is insufficient, failing to provide accurate data matching. Summary of the Invention

[0004] To address this, the present invention provides a machine learning-based method for detecting underground ice content, which overcomes the problems in existing technologies that fail to combine multiple models and calculation methods to correct preprocessed data in order to improve data accuracy, and that the classification and summarization of a large number of data samples are insufficient, thus failing to provide accurate data matching.

[0005] To achieve the above objectives, this invention provides a machine learning-based method for detecting underground ice content, comprising:

[0006] The detection area is evenly divided into several sub-regions. Based on the measured sensor parameters and sample dataset, the parameter characterization values ​​corresponding to each sub-region are determined. The ice-containing areas are determined according to the parameter characterization values. The measured sensor parameters are preprocessed to obtain preprocessed parameters, and the remote sensing data samples corresponding to the measured sensor parameters are obtained.

[0007] The spatiotemporal variation label corresponding to the detection area is determined or the regional parameters are corrected based on the proportion of ice-containing areas and the feature tendency value.

[0008] If regional parameter correction is required, the measured sensor parameters are input into the Clapeyron equation model and the underground permafrost model respectively, and the ice-water ratio is calculated based on the model output results.

[0009] The database is searched based on the aforementioned spatiotemporal anomaly tags, including:

[0010] Only the data from the sample dataset is extracted and matched with the measured sensor parameters, and the sample dataset is retrieved based on the matching results;

[0011] Alternatively, based on the differences in data characteristics of the sample dataset and the corresponding preprocessing parameters, several aggregated datasets are obtained. Each aggregated dataset is then filtered and analyzed, and the aggregated dataset is called based on the filtering and analysis results.

[0012] The filtering and analysis includes matching the data in the aggregated dataset with the measured sensor parameters.

[0013] Furthermore, the process of determining the ice-containing region based on the parameter characterization values ​​includes,

[0014] Extract the sample sensor parameters from the sample dataset, including temperature, dielectric constant, wave velocity, and resistivity;

[0015] Several factors affecting the differences between the sample sensor parameters and the measured sensor parameters were calculated, including the average temperature difference ratio, the average dielectric constant difference ratio, the average wave velocity difference ratio, and the average resistivity difference ratio.

[0016] The parameter values ​​are obtained by weighted summation of the various differential factors.

[0017] Furthermore, the step of determining the ice-containing region based on parameter characterization values, wherein,

[0018] If the parameter characterization value is greater than or equal to the preset parameter characterization value, then the sub-region corresponding to the parameter characterization value is determined to be an ice-containing region.

[0019] If the parameter representation value is less than the preset parameter representation value, then the sub-region corresponding to the parameter representation value is determined to be a region to be determined.

[0020] Furthermore, if the proportion of ice-containing areas is greater than or equal to the preset proportion of ice-containing areas or the feature tendency value is less than or equal to the preset feature tendency value, then the spatiotemporal variation label corresponding to the detection area is determined.

[0021] If the proportion of ice-containing areas is less than the preset proportion of ice-containing areas and the feature tendency value is greater than the preset feature tendency value, then it is determined that the region parameters should be corrected.

[0022] Furthermore, the measured sensor parameters are input into the Clapeyron equation model and the underground permafrost model, respectively, and the results are weighted to calculate the ice-water ratio.

[0023] Furthermore, a search is performed in the database based on the aforementioned spatiotemporal anomaly tags, including:

[0024] If the sample dataset has a weakly variable label, then only the data in the sample dataset is extracted and matched with the measured sensor parameters, and the sample dataset is called based on the matching result;

[0025] If the sample dataset has strongly variable labels, then the data features of the sample dataset and the corresponding preprocessing parameters are aggregated to obtain several aggregate datasets. Each aggregate dataset is then filtered and analyzed, and the aggregate dataset is called based on the filtering and analysis results.

[0026] Furthermore, the process of matching the extracted sample dataset data with the measured sensor parameters includes,

[0027] Extract model data from the sample dataset;

[0028] Calculate the similarity between the model data and the remote sensing data samples.

[0029] Furthermore, the step of calling the sample dataset based on the matching results, wherein,

[0030] If the similarity of the sample dataset is greater than or equal to the preset similarity, then it is determined that the sample dataset needs to be called.

[0031] Furthermore, the differences in data features based on the sample dataset and corresponding preprocessing parameters are aggregated to obtain several aggregated datasets, wherein...

[0032] Several sample datasets were aggregated according to the aggregation criteria.

[0033] The aggregation condition is that the difference factor between any dataset in the aggregation set is less than the preset aggregation factor value.

[0034] Furthermore, the process of calling the aggregated dataset based on the filtered parsing results includes,

[0035] In the aggregated dataset, any data is retrieved, and the retrieved data is compared with remote sensing data samples to calculate the similarity.

[0036] If the similarity is greater than the preset similarity, it is determined that the aggregated dataset needs to be invoked.

[0037] Compared with the prior art, the beneficial effects of the present invention are as follows: The present invention divides the detection area into several sub-regions, extracts and analyzes the data features of the sample dataset and the corresponding versions of the dataset, extracts the difference factors between the data features, determines the parameter characterization value corresponding to each sub-region based on the measured sensor parameters and the sample dataset, and determines the ice-containing area according to the parameter characterization value. Based on the proportion of the ice-containing area and the feature tendency value, the present invention determines the spatiotemporal variation label corresponding to the detection area or performs regional parameter correction. Subsequently, the spatiotemporal variation label is used to search the database to match the measured sensor parameters, thereby improving the analysis efficiency when analyzing a large amount of data, accelerating the data matching efficiency, and improving the accuracy of underground ice content detection based on machine learning.

[0038] Furthermore, this invention determines ice-containing areas by calculating several difference factors of sample sensor parameters and measured sensor parameters respectively, based on parameter characterization values. In practice, the data in the underground permafrost model undergoes multiple rounds of processing to meet training requirements. Due to differences in processing methods, the differences between the data in these underground permafrost models and the original data in terms of data feature dimensions are relatively discrete. If the differences are small, the original data tends to meet the same training requirements. Therefore, several difference factors are identified, and parameter characterization values ​​are calculated. These difference factors are the basic features of the data. Basic features can be obtained quickly and can characterize the differences between data. Therefore, calculating parameter characterization values ​​provides data support for setting spatiotemporal variation labels for the detection area, facilitating the adaptive retrieval of data in the underground permafrost model. This improves the accuracy of underground ice content detection based on machine learning while enhancing matching robustness.

[0039] Furthermore, this invention generates the ice-water ratio by weighting the results of the measured sensor parameters input into the Clapeyron equation model and the underground permafrost model, or by searching a database based on spatiotemporal variation labels. By classifying the labels of the sample dataset into weak and strong variation labels, different analyses and matching are performed on sample data with different degrees of variation. Through weighted fusion of the dual-model results and classification adaptation to spatiotemporal variation data, the reliability and generalization ability of machine learning for underground ice content detection are significantly improved in terms of both data processing accuracy and scene adaptability. This is combined with the Clapeyron equation model. The physical mechanism advantages of the equation model and the scene adaptability of the underground permafrost model are combined through weighted calculation to achieve complementarity between the two types of results. This avoids the problems of single models being affected by parameter interference or scene limitations, making the ice-water ratio calculation more consistent with actual geological conditions and providing more accurate basic data for ice content inversion. The sample data is classified and analyzed according to weak and strong anomalies, and a differentiated matching strategy is adopted for different degrees of anomalies. This reduces the impact of data noise and external interference, improves the model's adaptability to complex geological conditions, and further improves the accuracy of underground ice content detection based on machine learning.

[0040] Furthermore, this invention aggregates similar data from a sample dataset to form a pooled dataset. Then, it compares remote sensing data samples with any data in the pooled dataset, matches them based on the comparison results, and calls up pooled datasets with high similarity. This precise matching of similar sample aggregation with remote sensing data enhances the representativeness of data features and reduces noise interference. This improves the accuracy and stability of machine learning detection in terms of both data quality and matching efficiency. The dataset formed by similar data aggregation eliminates redundant information and focuses on core features, allowing the model to more efficiently capture key patterns strongly correlated with ice content. It avoids interference from irrelevant information in scattered samples, making model learning more targeted and reducing feature extraction bias. Comparing remote sensing data with the high-similarity pooled dataset quickly filters out reference data suitable for the current detection scenario, reducing interference from irrelevant factors such as soil texture and terrain differences. Similar data aggregation reduces redundant calculations, and high-similarity matching shortens data retrieval and adaptation time, improving overall detection efficiency and further enhancing the accuracy of underground ice content detection based on machine learning. Attached Figure Description

[0041] Figure 1 This is an overall flowchart of the underground ice content detection method based on machine learning, according to an embodiment of the present invention.

[0042] Figure 2 This is a flowchart illustrating the process of determining ice-containing regions based on parameter characterization values ​​according to an embodiment of the present invention;

[0043] Figure 3This is a logic block diagram for determining ice-containing regions based on parameter characterization values ​​in an embodiment of the present invention;

[0044] Figure 4 This is a block diagram illustrating the logic of calling the sample dataset based on the matching results in an embodiment of the present invention. Detailed Implementation

[0045] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0046] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0047] It should be noted that in the description of this invention, the terms "upper", "lower", "left", "right", "inner", "outer", etc., which indicate directions or positional relationships, are based on the directions or positional relationships shown in the accompanying drawings. This is only for the convenience of description and is not intended to indicate or imply that the device or element must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, it should not be construed as a limitation of this invention.

[0048] Please see Figure 1 , Figure 2 , Figure 3 and Figure 4 The diagrams shown are, respectively, an overall flowchart of the machine learning-based underground ice content detection method according to an embodiment of the present invention; a flowchart of determining ice-containing areas based on parameter characterization values ​​according to an embodiment of the present invention; a logic block diagram of determining ice-containing areas based on parameter characterization values ​​according to an embodiment of the present invention; and a logic block diagram of calling the sample dataset based on matching results according to an embodiment of the present invention. This embodiment of the present invention provides a machine learning-based underground ice content detection method, including:

[0049] Step S1: Divide the detection area into several sub-regions evenly, determine the parameter characterization value corresponding to each sub-region based on the measured sensor parameters and sample dataset, determine the ice-containing area based on the parameter characterization value, preprocess the measured sensor parameters to obtain preprocessed parameters, and obtain the remote sensing data sample corresponding to the measured sensor parameters.

[0050] Step S2: Determine the spatiotemporal variation label corresponding to the detection area or correct the regional parameters based on the proportion of ice-containing areas and the feature tendency value;

[0051] Step S3: If regional parameter correction is required, input the measured sensor parameters into the Clapeyron equation model and the underground permafrost model respectively, and calculate the ice-water ratio based on the model output results.

[0052] Step S4: Search the database based on the spatiotemporal anomaly tags, including...

[0053] Only the data from the sample dataset is extracted and matched with the measured sensor parameters, and the sample dataset is retrieved based on the matching results;

[0054] Step S5, or, based on the differences in data features of the sample dataset and the corresponding preprocessing parameters, several aggregated datasets are obtained, and each aggregated dataset is filtered and analyzed, and the aggregated dataset is called based on the filtering and analysis results;

[0055] The filtering and analysis process includes matching the data in the aggregated dataset with the measured sensor parameters.

[0056] Understandably, the process of preprocessing the measured sensor parameters to obtain the preprocessed parameters involves removing high-frequency noise related to temperature and resistivity using Kalman filtering (filtering window 5s); and performing temperature correction on the ultrasonic wave velocity using the formula: Vp_corr=Vp_meas×(1+0.002×(T+20)), where Vp_corr is the ultrasonic wave velocity correction value in m / s; Vp_meas is the measured ultrasonic wave velocity value in m / s; and T is the temperature in °C.

[0057] Specifically, this invention divides the detection area into several sub-regions, extracts and analyzes the data features of the sample dataset and the corresponding versions of the dataset, extracts the difference factors between the data features, determines the parameter characterization values ​​corresponding to each sub-region based on the measured sensor parameters and the sample dataset, and determines the ice-containing area based on the parameter characterization values. Based on the proportion of the ice-containing area and the feature tendency value, it determines the spatiotemporal variation label corresponding to the detection area or performs regional parameter correction. Subsequently, it searches the database based on the spatiotemporal variation label to match the measured sensor parameters, thereby improving the analysis efficiency when analyzing large amounts of data, accelerating the data matching efficiency, and improving the accuracy of underground ice content detection based on machine learning.

[0058] Specifically, in step S1, the process of determining the ice-containing region based on the parameter characterization values ​​includes,

[0059] Step S101: Extract the sample sensor parameters from the sample dataset, including temperature, dielectric constant, wave velocity, and resistivity;

[0060] Step S102: Calculate several difference factors between the sample sensor parameters and the measured sensor parameters, including the average temperature difference ratio, the average dielectric constant difference ratio, the average wave velocity difference ratio, and the average resistivity difference ratio.

[0061] Step S103: Weighted summation of each difference factor to obtain parameter characterization value.

[0062] It is understandable that the difference ratio is the ratio of the difference between two values ​​to the mean of the two values.

[0063] It is understandable that the measured sensor parameters need to be compared with the parameters of multiple sample sensors. The temperature difference ratio, dielectric constant difference ratio, wave velocity difference ratio and resistivity difference ratio can be solved one by one, and then the average temperature difference ratio, average dielectric constant difference ratio, average wave velocity difference ratio and average resistivity difference ratio can be solved. This will not be elaborated here.

[0064] In practice, since the importance of the measured sensor parameters to the comparison results is relatively small, the weights of the average temperature difference ratio, average dielectric constant difference ratio, average wave velocity difference ratio, and average resistivity difference ratio are 0.25, 0.25, 0.25, and 0.25, respectively.

[0065] Specifically, in step S1, the ice-containing region is determined based on the parameter characterization values, wherein,

[0066] If the parameter representation value is greater than or equal to the preset parameter representation value, then the sub-region corresponding to the parameter representation value is determined to be an ice-containing region.

[0067] If the parameter representation value is less than the preset parameter representation value, then the sub-region corresponding to the parameter representation value is determined to be a region to be determined.

[0068] In one specific embodiment, a preset parameter characterization value is set to 0.9. If the parameter characterization value is 0.98, which is greater than the preset parameter characterization value, then the sub-region corresponding to the parameter characterization value is determined to be an ice-containing region.

[0069] If the parameter representation value is 0.76, which is less than the preset parameter representation value, then the sub-region corresponding to the parameter representation value is determined to be a region to be determined.

[0070] Understandably, in practice, when the differences between the sensor parameters and those of existing ice-containing samples are small, the sub-regions corresponding to the measured sensor parameters can be determined to have underground ice, and the preset parameter characterization values ​​range from 0.85 to 0.95.

[0071] Specifically, in step S2, if the proportion of ice-containing areas is greater than or equal to the preset proportion of ice-containing areas or the feature tendency value is less than or equal to the preset feature tendency value, then the spatiotemporal variation label corresponding to the detection area is determined.

[0072] If the proportion of ice-containing areas is less than the preset proportion of ice-containing areas and the feature tendency value is greater than the preset feature tendency value, then it is determined that the region parameters should be corrected.

[0073] Specifically, the characteristic tendency values ​​include temperature characteristic tendency value, dielectric characteristic tendency value, wave velocity reciprocal tendency value, and resistivity reciprocal tendency value. If any characteristic tendency value is less than or equal to the corresponding preset characteristic tendency value, then the spatiotemporal variation tag corresponding to the detection area is determined.

[0074] Understandably, the preset temperature characteristic tendency value, preset dielectric characteristic tendency value, preset wave velocity reciprocal tendency characteristic value, and preset resistivity reciprocal tendency characteristic value need to be determined based on the specific environment of the region. Generally, the preset temperature characteristic tendency value is 0, the preset dielectric characteristic tendency value is 10, the preset wave velocity reciprocal tendency characteristic value is 1 / 2500, and the preset resistivity reciprocal tendency characteristic value is 1 / 10. 4 .

[0075] It is understandable that the larger the proportion of the number of sub-regions identified as ice-containing areas to the total number of sub-regions in the detection area, the greater the degree of ice content in the detection area. Therefore, the preset range for the proportion of ice-containing areas is 0.6 to 0.8.

[0076] Specifically, this invention determines ice-containing areas by calculating several difference factors of sample sensor parameters and measured sensor parameters respectively, based on parameter characterization values. In practice, the data in the underground permafrost model undergoes multiple rounds of processing to meet training requirements. Due to differences in processing methods, the differences between the data in these underground permafrost models and the original data in terms of data feature dimensions are relatively discrete. If the differences are small, the original data tends to meet the same training requirements. Therefore, several difference factors are identified, and parameter characterization values ​​are calculated. These difference factors are the basic features of the data. Basic features can be obtained quickly and can characterize the differences between data. Therefore, calculating parameter characterization values ​​provides data support for setting spatiotemporal variation labels for the detection area, facilitating subsequent adaptive retrieval of data in the underground permafrost model. While improving matching robustness, this further improves the accuracy of underground ice content detection based on machine learning.

[0077] Specifically, in step S3, the measured sensor parameters are input into the Clapeyron equation model and the underground permafrost model respectively, and the results are weighted to calculate the ice-water ratio.

[0078] Specifically, in practice, the weights of the results output by the Clapeyron equation model and the results output by the underground permafrost model are 0.3 and 0.7, respectively.

[0079] Specifically, in step S4, a search is performed in the database based on the spatiotemporal anomaly tags, including:

[0080] If the sample dataset has a weakly variable label, then only the data in the sample dataset is extracted and matched with the measured sensor parameters, and the sample dataset is called based on the matching results;

[0081] If the sample dataset has strongly variable labels, then the data features of the sample dataset and the corresponding preprocessing parameters are aggregated to obtain several aggregate datasets. Each aggregate dataset is then filtered and analyzed, and the aggregate dataset is called based on the filtering and analysis results.

[0082] Specifically, in step S4, the process of matching the extracted sample dataset data with the measured sensor parameters includes,

[0083] Extract model data from the sample dataset;

[0084] Calculate the similarity between the model data and the remote sensing data samples.

[0085] Specifically, the model data in the sample dataset is regional remote sensing data. The similarity is calculated by solving the similarity between regional remote sensing data. The purpose is to characterize the similarity between remote sensing data. For example, the goodness of fit can be obtained by calculating the ratio of the data differences between the regional remote sensing data and the remote sensing data samples. Of course, other methods can also be used, which will not be elaborated here.

[0086] Specifically, in step S4, the sample dataset is retrieved based on the matching results, wherein...

[0087] If the similarity of the sample dataset is greater than or equal to the preset similarity, it is determined that the sample dataset needs to be called.

[0088] If the similarity of the sample dataset is less than the preset similarity, then the sample dataset will not be used.

[0089] Specifically, the preset similarity is calculated in advance, the dataset used in the training process is recorded, the mean similarity between the data in the dataset is determined, the training mean of the mean similarity in several training processes is calculated, and the preset similarity is set as the product of the training mean and the accuracy coefficient, with the accuracy coefficient ranging from 0.9 to 0.95.

[0090] Specifically, this invention generates the ice-water ratio by weighting the results of the Clapeyron equation model and the underground permafrost model, respectively, using measured sensor parameters as inputs. Alternatively, it retrieves data from a database based on spatiotemporal variation labels. By categorizing the labels of the sample dataset into weak and strong variation labels, different analyses and matching are performed on sample data with varying degrees of variation. By weighted fusion of the dual-model results and classification adaptation to spatiotemporal variation data, the reliability and generalization ability of machine learning for detecting underground ice content are significantly improved in terms of both data processing accuracy and scene adaptability. This is combined with the Clapeyron equation model. The physical mechanism advantages of the equation model and the scene adaptability of the underground permafrost model are combined through weighted calculation to achieve complementarity between the two types of results. This avoids the problems of single models being affected by parameter interference or scene limitations, making the ice-water ratio calculation more consistent with actual geological conditions and providing more accurate basic data for ice content inversion. The sample data is classified and analyzed according to weak and strong anomalies, and a differentiated matching strategy is adopted for different degrees of anomalies. This reduces the impact of data noise and external interference, improves the model's adaptability to complex geological conditions, and further improves the accuracy of underground ice content detection based on machine learning.

[0091] Specifically, in step S5, the differences in data features based on the sample dataset and corresponding preprocessing parameters are aggregated to obtain several aggregated datasets, among which...

[0092] Several sample datasets were aggregated according to the aggregation criteria.

[0093] The aggregation condition is that the difference factor between any dataset in the aggregation set is less than the preset aggregation factor value.

[0094] It is understandable that datasets with small differences are aggregated, with the preset aggregate factor value range being 0.9 to 0.95.

[0095] Specifically, in step S5, the process of calling the aggregated dataset based on the filtered parsing results includes:

[0096] In the aggregated dataset, any data is retrieved, and the retrieved data is compared with remote sensing data samples to calculate the similarity.

[0097] If the similarity is greater than the preset similarity, it is determined that the aggregated dataset needs to be called.

[0098] It is understandable that the higher the similarity, the more accurate the data matching. The preset similarity is selected within the range [0.85, 0.95].

[0099] Specifically, this invention aggregates similar data from a sample dataset to form a pooled dataset. Then, remote sensing data samples are compared with any data in the pooled dataset, and matching is performed based on the comparison results. The pooled dataset with high similarity is called. By accurately matching and calling the pooled dataset with remote sensing data, the representativeness of data features is enhanced, and noise interference is reduced. This improves the accuracy and stability of machine learning detection in terms of both data quality and matching efficiency. The dataset formed by the aggregation of similar data eliminates redundant information and focuses on core features, allowing the model to more efficiently capture key patterns strongly correlated with ice content. It avoids interference from irrelevant information in scattered samples, making model learning more targeted and reducing feature extraction bias. Comparing and matching remote sensing data with the high-similarity pooled dataset can quickly filter out reference data suitable for the current detection scenario, reducing interference from irrelevant factors such as soil texture and terrain differences. The aggregation of similar data reduces redundant calculations, and high-similarity matching shortens data retrieval and adaptation time, improving overall detection efficiency and further enhancing the accuracy of underground ice content detection based on machine learning.

[0100] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

Claims

1. A method for detecting underground ice content based on machine learning, characterized in that, include: The detection area is evenly divided into several sub-regions. Based on the measured sensor parameters and sample dataset, the parameter characterization values ​​corresponding to each sub-region are determined. The ice-containing areas are determined according to the parameter characterization values. The measured sensor parameters are preprocessed to obtain preprocessed parameters, and the remote sensing data samples corresponding to the measured sensor parameters are obtained. The spatiotemporal variation label corresponding to the detection area is determined or the regional parameters are corrected based on the proportion of ice-containing areas and the feature tendency value. If regional parameter correction is required, the measured sensor parameters are input into the Clapeyron equation model and the underground permafrost model respectively, and the ice-water ratio is calculated based on the model output results. The database is searched based on the aforementioned spatiotemporal anomaly tags, including: Only the data from the sample dataset is extracted and matched with the measured sensor parameters, and the sample dataset is retrieved based on the matching results; Alternatively, based on the differences in data characteristics of the sample dataset and the corresponding preprocessing parameters, several aggregated datasets are obtained. Each aggregated dataset is then filtered and analyzed, and the aggregated dataset is called based on the filtering and analysis results. The filtering and analysis includes matching the data in the aggregated dataset with the measured sensor parameters. If the proportion of ice-containing areas is greater than or equal to the preset proportion of ice-containing areas or the feature tendency value is less than or equal to the preset feature tendency value, then the spatiotemporal variation label corresponding to the detection area is determined. If the proportion of ice-containing areas is less than the preset proportion of ice-containing areas and the feature tendency value is greater than the preset feature tendency value, then it is determined that the region parameters should be corrected. The characteristic tendency values ​​include temperature characteristic tendency values, dielectric characteristic tendency values, wave velocity reciprocal tendency value, and resistivity reciprocal tendency value. If any characteristic tendency value is less than or equal to the corresponding preset characteristic tendency value, then the spatiotemporal variation tag corresponding to the detection area is determined. The database is searched based on the aforementioned spatiotemporal anomaly tags, including: If the sample dataset has a weakly variable label, then only the data in the sample dataset is extracted and matched with the measured sensor parameters, and the sample dataset is called based on the matching result; If the sample dataset has strongly variable labels, then the data features of the sample dataset and the corresponding preprocessing parameters are aggregated to obtain several aggregate datasets. Each aggregate dataset is then filtered and analyzed, and the aggregate dataset is called based on the filtering and analysis results. The remote sensing data sample is compared with any data in the aggregated dataset, and the aggregated dataset is called up based on the comparison results.

2. The method for detecting underground ice content based on machine learning according to claim 1, characterized in that, The process of determining the ice-containing region based on parameter characterization values ​​includes, Extract the sample sensor parameters from the sample dataset, including temperature, dielectric constant, wave velocity, and resistivity; Several factors affecting the differences between the sample sensor parameters and the measured sensor parameters were calculated, including the average temperature difference ratio, the average dielectric constant difference ratio, the average wave velocity difference ratio, and the average resistivity difference ratio. The parameter values ​​are obtained by weighted summation of the various differential factors.

3. The method for detecting underground ice content based on machine learning according to claim 2, characterized in that, The process of determining the ice-containing region based on parameter characterization values, wherein, If the parameter characterization value is greater than or equal to the preset parameter characterization value, then the sub-region corresponding to the parameter characterization value is determined to be an ice-containing region. If the parameter representation value is less than the preset parameter representation value, then the sub-region corresponding to the parameter representation value is determined to be a region to be determined.

4. The method for detecting underground ice content based on machine learning according to claim 3, characterized in that, The measured sensor parameters are input into the Clapeyron equation model and the underground permafrost model, respectively, and the results are weighted to calculate the ice-water ratio.

5. The method for detecting underground ice content based on machine learning according to claim 4, characterized in that, The process of matching the extracted sample dataset with the measured sensor parameters includes: Extract model data from the sample dataset; Calculate the similarity between the model data and the remote sensing data samples.

6. The method for detecting underground ice content based on machine learning according to claim 5, characterized in that, The sample dataset is called based on the matching results, whereby... If the similarity of the sample dataset is greater than or equal to the preset similarity, then it is determined that the sample dataset needs to be called.

7. The method for detecting underground ice content based on machine learning according to claim 6, characterized in that, The differences in data features based on the sample dataset and corresponding preprocessing parameters are aggregated to obtain several aggregated datasets, among which... Several sample datasets were aggregated according to the aggregation criteria. The aggregation condition is that the difference factor between any dataset in the aggregation set is less than the preset aggregation factor value.

8. The method for detecting underground ice content based on machine learning according to claim 7, characterized in that, The process of calling the aggregated dataset based on the filtered and parsed results includes: In the aggregated dataset, any data is retrieved, and the retrieved data is compared with remote sensing data samples to calculate the similarity. If the similarity is greater than the preset similarity, it is determined that the aggregated dataset needs to be invoked.

Images