A method for identifying stony glaciers based on machine learning
By combining optical imagery and InSAR data with machine learning-based methods, the state of rock glaciers can be automatically identified, solving the problems of low efficiency and high subjectivity in existing technologies, and achieving efficient and accurate identification and monitoring of rock glacier states.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SOUTHWEST JIAOTONG UNIV
- Filing Date
- 2025-04-25
- Publication Date
- 2026-06-30
AI Technical Summary
Existing technologies are insufficient for efficiently and automatically identifying and determining the state of large-scale stony glaciers, especially intact and residual stony glaciers. They suffer from low efficiency and high subjectivity, and lack simple and timely monitoring methods, particularly in the context of ice melting caused by global warming.
A machine learning-based approach was adopted, combining the interpretation of optical images of rock glaciers and InSAR data augmentation. The optimal machine learning model was obtained by training RF, SVM, LR, DT, KNN and ResNet models to automatically identify the integrity and residual state of rock glaciers. Iterative judgment was made using surface motion information, and the identification accuracy was improved through data cleaning and feature extraction.
It achieves automated, accurate, and efficient identification of the state of rock glaciers, overcomes the problems of low efficiency and strong subjectivity in existing technologies, makes up for the lack of state attributes in existing cataloging, improves the identification accuracy, and is suitable for large-scale rock glacier monitoring.
Smart Images
Figure CN120451653B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of image processing technology, and in particular to a method for identifying rocky glaciers based on machine learning. Background Technology
[0002] Rock glaciers are a typical periglacial landform developed in high-altitude, cold mountainous areas. They are generally tongue-shaped or leaf-shaped, composed of a core rich in frozen ice and a layer of clastic rock overlying a certain thickness. As an important component of the alpine cryosphere, the state of rock glaciers is often considered the "true value" of the existence of local permafrost. Intact rock glaciers are rich in frozen ice inside, and their surfaces are full, which is a definite indicator of the existence of permafrost. Residual rock glaciers, on the other hand, usually ceased movement hundreds to thousands of years ago, and their internal ice has melted, resulting in no obvious flow marks on their surfaces. Their surfaces collapse and are often accompanied by vegetation cover, which can serve as an indicator of non-permafrost.
[0003] Currently, there are over 37,000 published rock glacier catalogs, but more than 80% of these glaciers remain undocumented in terms of their intact or residual states. These glaciers are distributed across various continents, in complex terrains, and are scattered and widespread. Furthermore, quantitative estimates of the extent of large-scale glacier movement are lacking, resulting in a relatively limited classification of their states. Existing catalogs largely rely on field investigations and visual interpretation of optical remote sensing images. This method is highly subjective in determining the state of rock glaciers, time-consuming, labor-intensive, and requires high-quality optical images. In recent years, due to global warming, hot and dry summers in some regions have led to the gradual loss of ice from intact rock glaciers, causing them to transition into residual states. However, there are no simple and timely monitoring methods to update the status of these rock glaciers.
[0004] With the development of space geodesy techniques, using InSAR to acquire surface motion information combined with optical imagery has become a favorable method for quantitatively assessing the state of stony glaciers. However, since most stony glaciers develop in high-altitude areas, they are affected by factors such as spatiotemporal loss of correlation in SAR interferometric phase, atmospheric delay errors, and SAR imaging distortion, making it difficult to obtain information on the motion status of stony glaciers over large areas using InSAR. Furthermore, combining InSAR observations with manual interpretation methods suffers from drawbacks such as high workload and low efficiency in interpreting large-scale stony glaciers, generally only suitable for identifying and cataloging stony glaciers at a local, small scale. Clearly, there is an urgent need for an automated and efficient method for identifying and determining active stony glaciers to catalog active stony glaciers on a large-scale or even global scale. Summary of the Invention
[0005] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method for identifying rock glaciers based on machine learning.
[0006] The objective of this invention is achieved through the following technical solution: a method for identifying stony glaciers based on machine learning, comprising the following steps:
[0007] S1: Interpret different geomorphic features on the optical image of the stone glacier and obtain identification results;
[0008] S2: Enhance the reliability of different types of rock glaciers using the InSAR method to obtain vector data of residual rock glaciers and vector data of intact rock glaciers;
[0009] S3: Based on vector data, image tiling is performed, non-rock glacier areas are masked to obtain images with only rock glacier features, and noisy data is cleaned.
[0010] S4: Import the InSAR-enhanced dataset into RF, SVM, LR, DT, KNN and ResNet models for training, obtain the accuracy of different machine learning models in identifying glaciers, and obtain the best machine learning model.
[0011] Preferably, step S1 further includes the following step:
[0012] S11: Several glacier identification personnel used optical images of Daxueshan Mountain to determine and identify the minimum envelope shape of the glacier;
[0013] S12: Assign a state value to each stone glacier based on its geomorphological characteristics, including two types: complete and residual, to obtain the cataloging results and initial state attributes of stone glaciers in the Daxueshan area;
[0014] S13: Based on the cataloging results, stone glaciers with the same complete residual attributes are included in the stone glacier dataset, while stone glaciers with different complete residual attributes are repeatedly identified and iterated, and the stone glacier state attributes are updated.
[0015] Preferably, step S2 further includes the following step:
[0016] S21: Obtain surface movement information in the Daxueshan area as a basis for enhancing the classification of glacier states;
[0017] S22: Extract the movement characteristics of the rock glacier using surface movement information, and determine the movement area of the rock glacier. Based on the initial integrity and residual state values of the rock glacier obtained in step S12, perform another iteration and delete rock glaciers with poor coherence.
[0018] Preferably, in step S21, a SAR interferogram is generated based on the single-view complex image, and then Goldstein filtering is performed.
[0019]
[0020] in, S(u,v) represents the filtered phase, S(u,v) represents the spectrum, the subscript m represents the smoothing process, and the superscript a represents the filter strength parameter in the Goldstein filter.
[0021] Based on the above results, the tropospheric phase delay error was corrected using differential interferometry short baseline set time series analysis technology to obtain the surface motion rate in the Daxueshan area.
[0022] Preferably, step S3 further includes the following step:
[0023] S31: Perform optical image masking on the raster data of the complete and residual rock glaciers obtained in step S22;
[0024] S32: Slice the optical image according to the resolution size of the optical image, and store the complete type rock glacier data image and the residual type rock glacier data image separately;
[0025] S33: Recheck the dataset and delete image slices that are heavily covered by clouds and fog.
[0026] Preferably, step S4 further includes the following step:
[0027] S41: Define the feature transformation function for the image to extract the geomorphic features of the glacier from the optical imagery in each image.
[0028] Interquartile Rang(IQR) = S 75th -S 25th ;
[0029]
[0030] Among them, S 75th S is the third quartile in the data. 25th Let S be the first quartile in the data, the Interquartile Range (IQR) be the interquartile range, and S be a value in the data sample. median S is the sample median. robust For robust and standardized statistics;
[0031] S42: Divide the dataset in a 4:1 ratio and evaluate the model using the cross-entropy loss function.
[0032]
[0033] Where loss(S,y) is the loss function, and y i For sample i, 1 represents an intact glacier and 0 represents a remnant glacier. i Predict the probability that sample i is a complete type of stone glacier;
[0034] Construct a stochastic gradient descent optimizer and introduce momentum.
[0035] θ t+1 =θ t -γg t ;
[0036] Where, θ t+1 Let θ be the weight at step t+1. t Let θ be the weight at step t, γ be the parameter, γ be the learning rate, and g be the gradient. t This is the momentum term for the current step.
[0037] The present invention has the following advantages: Compared with conventional methods for judging the state of rock glaciers based on geomorphic features and methods for judging complete and residual rock glaciers by combining InSAR and optical imagery, the method proposed in this invention can overcome the problems of low efficiency and strong subjectivity in existing rock glacier state classification, thereby realizing the automatic classification of complete and residual states of rock glaciers. At the same time, this method can make up for the lack of complete and residual state attributes in a large number of existing rock glacier catalogs. Attached Figure Description
[0038] Figure 1 This is a schematic diagram of the process of identifying rock glaciers based on machine learning.
[0039] Figure 2 A schematic diagram illustrating the surface movement rate in the northern region of the Daxueshan Mountains;
[0040] Figure 3 A schematic diagram illustrating the surface movement characteristics of the southern region of the Daxueshan Mountains;
[0041] Figure 4 A schematic diagram showing the spatial distribution of the stone glaciers in the northern part of the Daxue Mountain;
[0042] Figure 5 A schematic diagram showing the spatial distribution of the stone glaciers in the southern part of the Daxue Mountain;
[0043] Figure 6 A schematic diagram of the surface motion rate obtained from InSAR processing in the Gangdise Mountains region;
[0044] Figure 7 A schematic diagram of the surface movement rate of a complete type of stony glacier;
[0045] Figure 8 A schematic diagram of the surface movement rate of residual type stony glaciers;
[0046] Figure 9 A schematic diagram showing the results of identifying stone glaciers in the Himalayas based entirely on geomorphological features;
[0047] Figure 10This is a schematic diagram of the results of the judgment of the stone glaciers in the Himalayas region based on the present invention. Detailed Implementation
[0048] 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.
[0049] 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.
[0050] It should be noted that, unless otherwise specified, the embodiments and features described in this invention can be combined with each other.
[0051] It should be noted that similar labels 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.
[0052] In the description of this invention, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "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 commonly used when the product of this invention is in use, or the orientation or positional relationship commonly understood by those skilled in the art. They are only used 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, and therefore should not be construed as a limitation of this invention. In addition, the terms "first," "second," etc., are only used to distinguish descriptions and should not be construed as indicating or implying relative importance.
[0053] In the description of this invention, it should also be noted that, unless otherwise explicitly specified and limited, the terms "set," "install," "connect," and "link" 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.
[0054] In this embodiment, as Figure 1 As shown, a method for identifying rocky glaciers based on machine learning includes the following steps:
[0055] S1: Interpret the different geomorphic features on the optical image of the glacier to obtain identification results. Specifically, step S1 mainly utilizes the different geomorphic features of intact and remnant glaciers on high-precision optical images to obtain accurate manual identification results, providing a foundation for the reliability of subsequent InSAR augmented data. Furthermore, step S1 also includes the following steps:
[0056] S11: Several glacier identification personnel used optical images of Daxueshan Mountain to determine and identify the minimum envelope shape of the glacier;
[0057] S12: Assign a state value to each stone glacier based on its geomorphological characteristics, including two types: complete and residual, to obtain the cataloging results and initial state attributes of stone glaciers in the Daxueshan area;
[0058] S13: Based on the cataloging results, stone glaciers with the same complete residual attributes are included in the stone glacier dataset, while stone glaciers with different complete residual attributes are repeatedly identified and iterated, and the stone glacier state attributes are updated.
[0059] S2: Enhance the reliability of different types of rock glaciers using the InSAR method to obtain residual rock glacier vector data and complete rock glacier vector data; specifically, the main function of step S2 is to obtain the basic rock glacier spatial information data for model training, thereby overcoming the subjective problem of the existence of rock glacier state in manual interpretation.
[0060] S3: Image tiling is performed based on vector data. Non-rock glacier areas are masked to obtain images with only rock glacier features. Noisy data is cleaned to control the quality of training data and improve the reliability of the dataset.
[0061] S4: The InSAR-enhanced dataset is imported into RF (Random Forest), SVM (Support Vector Machine), LR (Logistic Regression), DT (Decision Tree), KNN (K-Nearest Neighbor), and ResNet (Residual Neural Network) models for training. The accuracy of different machine learning models in identifying rock glaciers is obtained, and the optimal machine learning model is determined. Compared with conventional methods based on geomorphological features to determine the state of rock glaciers and methods that combine InSAR and optical imagery to determine complete and residual rock glaciers, the method proposed in this invention can overcome the problems of low efficiency and high subjectivity in existing rock glacier state classification, thereby realizing the automatic classification of complete and residual states of rock glaciers. At the same time, this method can compensate for the lack of complete and residual state attributes in many existing rock glacier catalogs.
[0062] Furthermore, such as Figures 2-5 As shown, step S2 also includes the following steps:
[0063] S21: Obtain surface motion information in the Daxueshan area as a basis for enhancing the classification of glacier status; further, in step S21, generate SAR interferograms based on single-view complex images, including image registration, differential phase interferometry, phase filtering, unwrapping, and geocoding, and then perform Goldstein filtering.
[0064]
[0065] in, S(u,v) represents the filtered phase, S(u,v) represents the spectrum, the subscript m represents the smoothing process, and the superscript a represents the filter strength parameter in the Goldstein filter.
[0066] Based on the above results, the tropospheric phase delay error was corrected using differential interferometry short baseline set time series analysis technology to obtain the surface motion rate in the Daxueshan area;
[0067] S22: Extract the movement characteristics of the rock glacier using surface movement information, and determine the movement area of the rock glacier. Based on the initial complete and residual state values of the rock glacier obtained in step S12, perform another iteration and delete rock glaciers with poor coherence, thereby ensuring the high reliability of the complete and residual state dataset of the rock glacier.
[0068] In this embodiment, step S3 further includes the following step:
[0069] S31: Perform optical image masking on the complete stony glacier and residual stony glacier raster data obtained in step S22 to avoid noise interference from non-stony glacier areas on model identification;
[0070] S32: Slice the optical image according to the resolution size of the optical image, and store the complete type rock glacier data image and the residual type rock glacier data image separately;
[0071] S33: The dataset is checked again, and image slices with severe cloud cover are deleted to ensure the reliability of the constructed glacier state dataset.
[0072] Furthermore, step S4 also includes the following steps:
[0073] S41: Define the feature transformation function for the image to extract the geomorphic features of the glacier from the optical imagery in each image.
[0074] Interquartile Range (IQR) = S 75th -S 25th ;
[0075]
[0076] Among them, S 75th S is the third quartile in the data. 25th Let S be the first quartile in the data, the Interquartile Range (IQR) be the interquartile range, and S be a value in the data sample. median S is the sample median. robust For robust and standardized statistics;
[0077] S42: Divide the dataset in a 4:1 ratio and evaluate the model using the cross-entropy loss function.
[0078]
[0079] Where loss(S,y) is the loss function, and y i For sample i, 1 represents an intact glacier and 0 represents a remnant glacier. i Predict the probability that sample i is a complete type of stone glacier;
[0080] Construct a stochastic gradient descent optimizer and introduce momentum.
[0081] θ t+1 =θ t -γg t ;
[0082] Where, θ t+1 Let θ be the weight at step t+1. t Let θ be the weight at step t, γ be the parameter, γ be the learning rate, and g be the gradient. t This is the momentum term for the current step.
[0083] In this embodiment, to verify the effectiveness of the method disclosed in this invention, the method was applied to glaciers in the Gangdise Mountains and Himalayas for experiments. The results of this method were compared with the results of glacier identification based on conventional geomorphological features and surface deformation information obtained by InSAR. Figures 6-10 As shown:
[0084] Based on the assessment results in the Gangdise Mountains region, the predicted results include 7 complete types of rock glaciers and 4 residual types of rock glaciers. Further comparison of the model assessment results with InSAR processed rate image maps and 3D images showed that the assessment results of 9 of the rock glaciers were correct, with an accuracy rate of 81.81%.
[0085] Based on the results from the Himalayan region, the identification method disclosed in this invention performs satisfactorily compared to the results of judging the integrity and remnant type of stony glaciers entirely based on extracting geomorphic features from optical images. Specifically, the model's judgment accuracy rate is 79.03%, successfully predicting the activity status of 48 stony glaciers. It effectively replaces manual methods for predicting the integrity and remnant status of stony glaciers, automating the classification of the integrity and remnant status of stony glaciers. Furthermore, this method can compensate for the lack of integrity and remnant status in many existing stony glacier cataloging results.
[0086] Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A method for identifying stony glaciers based on machine learning, characterized in that: Includes the following steps: S1: Interpret different geomorphic features on the optical image of the stone glacier and obtain identification results; S2: Enhance the reliability of different types of rock glaciers using the InSAR method to obtain vector data of residual rock glaciers and vector data of intact rock glaciers; S3: Based on vector data, image tiling is performed, non-rock glacier areas are masked to obtain images with only rock glacier features, and noisy data is cleaned. S4: Import the InSAR-enhanced dataset into RF, SVM, LR, DT, KNN and ResNet models for training, obtain the accuracy of different machine learning models in identifying glaciers, and obtain the best machine learning model. Step S2 further includes the following steps: S21: Obtain surface movement information in the Daxueshan area as a basis for enhancing the classification of glacier states; S22: Extract the movement characteristics of the rock glacier using surface movement information, and determine the movement area of the rock glacier. Based on the initial integrity and residual state values of the rock glacier obtained in step S12, perform another iteration and delete rock glaciers with poor coherence. In step S21, a SAR interferogram is generated based on the single-view complex image, and then Goldstein filtering is performed. ； in, The filtered phase, For the spectrum, subscript For smoothing purposes, superscript This refers to the filter strength parameter in the Goldstein filter. Based on the filtered phase, the tropospheric phase delay error was corrected using differential interferometry short baseline set time series analysis technology to obtain the surface motion rate in the Daxueshan area; Step S3 further includes the following steps: S31: Perform optical image masking on the complete and residual rock glacier raster data obtained in step S22; S32: Slice the optical image according to the resolution size of the optical image, and store the complete type rock glacier data image and the residual type rock glacier data image separately; S33: Recheck the dataset and delete image slices that are heavily covered by clouds and fog; Step S4 also includes the following steps: S41: Define the feature transformation function for the image to extract the geomorphic features of the glacier from the optical imagery in each image. ； ； in, It is the third quartile in the data. The first quartile in the data. Interquartile range, For a certain value in the data sample, The median of the sample. For robust and standardized statistics; S42: Divide the dataset in a 4:1 ratio and evaluate the model using the cross-entropy loss function. ； in, For loss function, For the sample The label indicates the type of glacier: 1 for intact glaciers and 0 for remnant glaciers. For the sample The probability of it being a complete type of stony glacier; Construct a stochastic gradient descent optimizer and introduce momentum. ； in, For the first Step weights For the first Step weights For parameters, For learning rate, For gradient, This is the momentum term for the current step.
2. The method for identifying rock glaciers based on machine learning according to claim 1, characterized in that: Step S1 further includes the following steps: S11: Several glacier identification personnel used optical images of Daxueshan Mountain to determine and identify the minimum envelope shape of the glacier; S12: Assign a state value to each stone glacier based on its geomorphological characteristics, including two types: complete and residual, to obtain the cataloging results and initial state attributes of stone glaciers in the Daxueshan area; S13: Based on the cataloging results, stone glaciers with the same complete residual attributes are included in the stone glacier dataset, while stone glaciers with different complete residual attributes are repeatedly identified and iterated, and the stone glacier state attributes are updated.