Method for identifying stone glacier based on machine learning

Abstract

The invention discloses a method for identifying stone glacier based on machine learning, which comprises the following steps: S1, interpreting different landform features on a stone glacier optical image to obtain an identification result; s2, the reliability of different types of stone glaciers is enhanced according to an InSAR method, and residual stone glacier vector data and complete stone glacier vector data are obtained; s3, image slicing is carried out based on the vector data, a non-stone glacier area is masked, an image only having stone glacier features is obtained, and noise data is cleaned; and S4, importing the InSAR enhanced data set into RF, SVM, LR, DT, KNN and ResNet models for training to obtain the accuracy of stone glacier identification by different machine learning models so as to obtain an optimal machine learning model. According to the method provided by the invention, the problems of low efficiency and strong subjectivity of the existing stone glacier state division can be solved, so that the complete and residual states of the stone glacier can be automatically divided, and meanwhile, the problem that the complete and residual state attributes of a large number of existing stone glacier catalogues are lost can be solved by utilizing the method.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and in particular to a method for identifying rock glaciers based on machine learning. Background Art

[0002] Rock glaciers are a typical periglacial landform found in alpine mountainous areas. They typically take the shape of tongues or lobes, consisting of a core rich in frozen ice and a thick overburden of clastic rocks. As a crucial component of the alpine cryosphere, rock glaciers are often considered the "true value" of local permafrost. Intact rock glaciers are rich in frozen ice and have a plump surface, a sure sign of permafrost. Residual rock glaciers, on the other hand, typically ceased movement hundreds to thousands of years ago, their internal ice having melted, resulting in a lack of visible flow marks on their surfaces. Their collapsed surfaces are often accompanied by vegetation, thus serving as a marker for non-permafrost conditions.

[0003] Existing published data catalogs over 37,000 rock glaciers, yet over 80% remain unrecorded in both intact and residual states. These glaciers are scattered and widespread across various continents, situated in complex terrain. Quantitative estimates of large-scale rock glacier motion are lacking, leading to a relative lack of classification of rock glacier status. Existing catalogs rely primarily on field surveys and visual interpretation of optical remote sensing imagery. This method is highly subjective, time-consuming, and requires high-quality optical imagery. In recent years, global warming has led to dry, hot summers in some regions, causing the gradual loss of ice from intact rock glaciers, leading to their transition to residual rock glaciers. However, there is no simple and timely monitoring method to update the status of these glaciers.

[0004] With the development of space geodesy, the use of InSAR (Infrared SAR) to obtain surface motion information, combined with optical imagery, has become a promising method for quantitatively assessing the status of rock glaciers. However, because most rock glaciers develop at high altitudes, obtaining information on rock glacier motion over large areas using InSAR remains challenging due to factors such as the temporal and spatial decoupling of SAR interferometric phases, atmospheric delay errors, and SAR imaging distortion. Furthermore, combining InSAR observations with manual interpretation methods for interpreting rock glaciers over large areas suffers from high workload and low efficiency, making it generally suitable for identifying and cataloging rock glaciers on a local, small-scale. Clearly, an automated and efficient method for identifying and determining active rock glaciers is urgently needed to catalog active rock glaciers on a large, even global, scale. Summary of the Invention

[0005] The purpose of the present invention is to overcome the shortcomings of the prior art and provide a method for identifying rock glaciers based on machine learning.

[0006] The object of the present invention is achieved through the following technical solution: A method for identifying rock glaciers based on machine learning, comprising the following steps:

[0007] S1: Interpret different geomorphological features on the optical image of the rock glacier and obtain identification results;

[0008] S2: Enhance the reliability of different types of rock glaciers based on the InSAR method, and obtain residual rock glacier vector data and complete rock glacier vector data;

[0009] S3: Slice the image based on the vector data, mask the non-rock glacier area, obtain the image with only rock glacier features, and clean the noise data;

[0010] S4: Import the InSAR enhanced dataset into the RF, SVM, LR, DT, KNN and ResNet models for training, obtain the accuracy of different machine learning models in identifying rock glaciers, and obtain the best machine learning model.

[0011] Preferably, step S1 further includes the following steps:

[0012] S11: Several rock glacier identification personnel use optical images of the Great Snow Mountain to identify the minimum envelope shape of rock glaciers;

[0013] S12: Assign the state of each rock glacier according to its geomorphic characteristics, including two types: complete and residual, to obtain the inventory results and initial state attributes of the rock glaciers in the Daxueshan area;

[0014] S13: According to the cataloging results, rock glaciers with the same complete residual attributes are included in the rock glacier dataset. Rock glaciers with different complete residual attributes are repeatedly identified and iterated, and the rock glacier status attributes are updated.

[0015] Preferably, step S2 further includes the following steps:

[0016] S21: Obtain surface motion information in the Great Snow Mountain region as a basis for enhancing the classification of rock glacier status;

[0017] S22: Using the surface movement information to extract the movement characteristics of the rock glacier, and judging the movement area of the rock glacier, iterate again based on the initial rock glacier complete and residual state values obtained in step S12, and delete the rock glaciers with poor coherence.

[0018] Preferably, in step S21, a SAR interferogram is generated based on the single-vision complex image, and then Goldstein filtering is performed.

[0019]

[0020] in, is the phase after filtering, S(u,v) is the spectrum, subscript m is the smoothing process, and superscript a is the filter intensity parameter in Goldstein filtering;

[0021] Based on the above results, the differential interferometry short baseline set timing analysis technology is used to correct the tropospheric phase delay error and obtain the surface motion rate in the Great Snow Mountain area.

[0022] Preferably, step S3 further includes the following steps:

[0023] S31: performing optical image masking on the complete rock glacier and residual rock glacier grid data obtained in step S22;

[0024] S32: Slicing the optical image according to the resolution of the optical image, and storing the complete type rock glacier data image and the residual type rock glacier data image separately;

[0025] S33: Check the dataset again and delete the image slices with severe cloud cover.

[0026] Preferably, step S4 further includes the following steps:

[0027] S41: Define the feature conversion function of the image to extract the geomorphological features of the rock glacier in each image on the optical image.

[0028] Interquartile Range (IQR) = S 75th -S 25th ;

[0029]

[0030] Among them, S 75th is the third quartile in the data, S 25th is the first quartile in the data, Interquartile Range (IQR) is the interquartile range, S is a value in the data sample, S median is the sample median, S robust is a robust standardized statistic;

[0031] S42: Divide the data set into a ratio of 4:1 and evaluate the model using the cross entropy loss function.

[0032]

[0033] Among them, loss(S,y) is the loss function, y i is the label of sample i, the complete rock glacier is 1, the residual rock glacier is 0, and p i The probability of sample i being predicted as a complete type of rock glacier;

[0034] Construct a stochastic gradient descent optimizer and introduce momentum.

[0035] θ t+1 =θ t -γg t ;

[0036] Among them, θ t+1 is the weight of step t+1, θ t is the weight of the tth step, θ is the parameter, γ is the learning rate, g is the gradient, g t is the momentum term of the current step.

[0037] The present invention has the following advantages: compared with the conventional method of judging the state of rock glaciers based on geomorphological features and the method of judging complete and residual rock glaciers by combining InSAR and optical images, the method proposed by the present invention can overcome the problems of low efficiency and strong subjectivity in the existing rock glacier state division, thereby realizing the automatic division of complete and residual states of rock glaciers. At the same time, this method can make up for the problem of missing complete and residual state attributes in a large number of existing rock glacier catalogs. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 A schematic diagram of the process of identifying rock glaciers based on machine learning;

[0039] Figure 2 This is a schematic diagram of the surface movement rate in the northern part of the Great Snow Mountain;

[0040] Figure 3 This is a schematic diagram of the surface motion characteristics in the southern part of the Great Snow Mountain;

[0041] Figure 4 This is a schematic diagram of the spatial distribution of rock glaciers in the northern part of the Daxue Mountain;

[0042] Figure 5 This is a schematic diagram of the spatial distribution of rock glaciers in the southern part of the Daxue Mountain.

[0043] Figure 6 Schematic diagram of the surface motion rate obtained by InSAR processing in the Gangdise Mountains area;

[0044] Figure 7 Schematic diagram of the surface movement rate of a complete type of rock glacier;

[0045] Figure 8 Schematic diagram of the surface movement rate of residual rock glaciers;

[0046] Figure 9 This is a schematic diagram of the rock glaciers in the Himalayas based solely on geomorphological features.

[0047] Figure 10This is a schematic diagram of the judgment results of the present invention on the stone glaciers in the Himalayas. DETAILED DESCRIPTION

[0048] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions of the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings herein can be arranged and designed in various different configurations.

[0049] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention as claimed, but rather merely represents selected embodiments of the present invention. All other embodiments derived by persons of ordinary skill in the art based on the embodiments of the present invention without creative effort are also within the scope of protection of the present invention.

[0050] It should be noted that, in the absence of conflict, the embodiments and features of the embodiments of the present invention can be combined with each other.

[0051] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, it does not need to be further defined or explained in subsequent drawings.

[0052] In the description of the present invention, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," "outer," etc., indicating orientations or positional relationships, are based on the orientations or positional relationships shown in the accompanying drawings, or are the orientations or positional relationships in which the inventive product is typically placed when in use, or are the orientations or positional relationships commonly understood by those skilled in the art. These terms are intended only to facilitate the description of the present invention and to simplify the description, and are not intended to indicate or imply that the device or element referred to must have a specific orientation, be constructed, or operate in a specific orientation. Therefore, they should not be construed as limiting the present invention. Furthermore, the terms "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0053] In the description of the present invention, it should also be noted that, unless otherwise expressly specified or limited, the terms "disposed," "installed," "connected," and "connected" should be understood in a broad sense. For example, they may refer to fixed connections, detachable connections, or integral connections; they may refer to mechanical connections or electrical connections; they may refer to direct connections or indirect connections through an intermediate medium; and they may refer to internal communication between two components. Those skilled in the art will understand the specific meanings of the above terms in the present invention based on the specific circumstances.

[0054] In this embodiment, if Figure 1 As shown, a method for identifying rock glaciers based on machine learning includes the following steps:

[0055] S1: Interpret the different geomorphic features on the rock glacier optical image to obtain identification results. Specifically, in step S1, the different geomorphic features of the complete rock glacier and the residual rock glacier on the high-precision optical image are mainly used to interpret the different geomorphic features to obtain accurate manual identification results, providing a basis for subsequent InSAR data enhancement. Furthermore, step S1 also includes the following steps:

[0056] S11: Several rock glacier identification personnel use optical images of the Great Snow Mountain to identify the minimum envelope shape of rock glaciers;

[0057] S12: Assign the state of each rock glacier according to its geomorphic characteristics, including two types: complete and residual, to obtain the inventory results and initial state attributes of the rock glaciers in the Daxueshan area;

[0058] S13: According to the cataloging results, rock glaciers with the same complete residual attributes are included in the rock glacier dataset. Rock glaciers with different complete residual attributes are repeatedly identified and iterated, and the rock glacier status attributes are updated.

[0059] S2: Enhance the reliability of different types of rock glaciers according to the InSAR method, and 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 manual interpretation of rock glacier status.

[0060] S3: Slice the image based on the vector data, mask the non-rock glacier area, obtain images with only rock glacier features, and clean the noise data to control the quality of the training data and improve the reliability of the dataset;

[0061] S4: The InSAR-enhanced dataset was trained using RF (Random Forest), SVM (Support Vector Machine), LR (Logistic Regression), DT (Decision Tree), KNN (K-Nearest Neighbor), and ResNet (Residual Neural Network) models. The accuracy of different machine learning models in identifying rock glaciers was determined, and the optimal machine learning model was obtained. Compared with conventional methods for determining rock glacier status based on geomorphological features and combining InSAR and optical imagery to determine intact and residual rock glaciers, the proposed method overcomes the low efficiency and high subjectivity of existing rock glacier status classification, thereby automatically classifying rock glaciers into intact and residual states. Furthermore, this method can address the problem of missing attributes for the intact and residual states of a large number of existing rock glacier catalogs.

[0062] Further, such as Figures 2 to 5 As shown, step S2 further includes the following steps:

[0063] S21: Obtaining surface motion information in the Great Snow Mountain area as the basis for enhanced classification of rock glacier status; further, in step S21, generating SAR interferograms based on single-vision complex images, including image registration, differential phase interferometry, phase filtering, unwrapping, and geocoding, and then performing Goldstein filtering,

[0064]

[0065] in, is the phase after filtering, S(u,v) is the spectrum, subscript m is the smoothing process, and superscript a is the filter intensity parameter in Goldstein filtering;

[0066] Based on the above results, the short baseline set timing analysis technique of differential interferometry is used to correct the tropospheric phase delay error and obtain the surface motion velocity in the Great Snow Mountain area.

[0067] S22: Using the surface movement information to extract the movement characteristics of the rock glacier, and determine the movement area of the rock glacier, iterate again based on the initial rock glacier complete and residual state values obtained in step S12, and delete the rock glaciers with poor coherence, thereby ensuring the high reliability of the rock glacier complete residual state data set.

[0068] In this embodiment, step S3 further includes the following steps:

[0069] S31: performing optical image masking on the complete rock glacier and residual rock glacier grid data obtained in step S22, so as to avoid noise interference of non-rock glacier areas on model identification;

[0070] S32: Slicing the optical image according to the resolution of the optical image, and storing the complete type rock glacier data image and the residual type rock glacier data image separately;

[0071] S33: Check the dataset again and delete the image slices with severe cloud cover to ensure the reliability of the constructed rock glacier status dataset.

[0072] Furthermore, step S4 further includes the following steps:

[0073] S41: Define the feature conversion function of the image to extract the geomorphological features of the rock glacier in each image on the optical image.

[0074] Interquartile Range (IQR) = S 75th -S 25th ;

[0075]

[0076] Among them, S 75th is the third quartile in the data, S 25th is the first quartile in the data, Interquartile Range (IQR) is the interquartile range, S is a value in the data sample, S median is the sample median, S robust is a robust standardized statistic;

[0077] S42: Divide the data set into a ratio of 4:1 and evaluate the model using the cross entropy loss function.

[0078]

[0079] Among them, loss(S,y) is the loss function, y i is the label of sample i, the complete rock glacier is 1, the residual rock glacier is 0, and p i The probability of sample i being predicted as a complete type of rock glacier;

[0080] Construct a stochastic gradient descent optimizer and introduce momentum.

[0081] θ t+1 =θ t -γg t ;

[0082] Among them, θ t+1 is the weight of step t+1, θ t is the weight of the tth step, θ is the parameter, γ is the learning rate, g is the gradient, g t is the momentum term of the current step.

[0083] In this example, in order to verify the effect of the method disclosed in the present invention, the method was applied to the rock glaciers in the Gangdise Mountains and the Himalayas for experiments, and the results of the method were compared with the results of the rock glaciers based on conventional geomorphological features and the surface deformation information obtained by InSAR. Figures 6 to 10 As shown:

[0084] The Gangdise Mountains region predicted seven complete rock glaciers and four residual rock glaciers. Further comparison of the model prediction results with InSAR processing rate images and three-dimensional images showed that nine of the rock glaciers were correctly identified, with an accuracy rate of 81.81%.

[0085] Judging from the judgment results in the Himalayas, the overall performance of the identification method proposed in the present invention is satisfactory compared with the results of judging the complete residual type of stone glaciers based entirely on optical image extraction of geomorphological features, that is, the accuracy of the model judgment results is 79.03%, and the activity status of 48 stone glaciers has been successfully predicted. It is a good alternative to manual prediction of the complete and residual status of stone glaciers, and realizes the automation of the division of complete and residual states of stone glaciers. At the same time, this method can make up for the lack of complete and residual states in a large number of existing stone glacier cataloging results.

[0086] Although the present invention has been described in detail with reference to the aforementioned embodiments, it is still possible for those skilled in the art to modify the technical solutions described in the aforementioned embodiments, or to make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for identifying rock glaciers based on machine learning, characterized by: The following steps are involved: S1: Interpret different geomorphological features on the optical image of the rock glacier and obtain identification results; S2: Enhance the reliability of different types of rock glaciers based on the InSAR method, and obtain residual rock glacier vector data and complete rock glacier vector data; S3: Slice the image based on the vector data, mask the non-rock glacier area, obtain the image with only rock glacier features, and clean the noise data; S4: Import the InSAR enhanced dataset into the RF, SVM, LR, DT, KNN and ResNet models for training, obtain the accuracy of different machine learning models in identifying rock glaciers, and obtain the best machine learning model.

2. The method for identifying rock glaciers based on machine learning according to claim 1, characterized in that: The step S1 further includes the following steps: S11: Several rock glacier identification personnel use optical images of the Great Snow Mountain to identify the minimum envelope shape of rock glaciers; S12: Assign the state of each rock glacier according to its geomorphic characteristics, including two types: complete and residual, to obtain the inventory results and initial state attributes of the rock glaciers in the Daxueshan area; S13: According to the cataloging results, rock glaciers with the same complete residual attributes are included in the rock glacier dataset. Rock glaciers with different complete residual attributes are repeatedly identified and iterated, and the rock glacier status attributes are updated.

3. The method for identifying rock glaciers based on machine learning according to claim 2, characterized in that: The step S2 further includes the following steps: S21: Obtain surface motion information in the Great Snow Mountain region as a basis for enhancing the classification of rock glacier status; S22: Using the surface movement information to extract the movement characteristics of the rock glacier, and judging the movement area of the rock glacier, iterate again based on the initial rock glacier complete and residual state values obtained in step S12, and delete the rock glaciers with poor coherence.

4. The method for identifying rock glaciers based on machine learning according to claim 3, characterized in that: In step S21, a SAR interferogram is generated based on the single-vision complex image, and then Goldstein filtering is performed. in, is the phase after filtering, S(u,v) is the spectrum, subscript m is the smoothing process, and superscript a is the filter intensity parameter in Goldstein filtering; Based on the above results, the differential interferometry short baseline set timing analysis technology is used to correct the tropospheric phase delay error and obtain the surface motion rate in the Great Snow Mountain area.

5. The method for identifying rock glaciers based on machine learning according to claim 4, characterized in that: The step S3 further includes the following steps: S31: performing optical image masking on the complete rock glacier and residual rock glacier grid data obtained in step S22; S32: Slicing the optical image according to the resolution of the optical image, and storing the complete type rock glacier data image and the residual type rock glacier data image separately; S33: Check the dataset again and delete the image slices with severe cloud cover.

6. The method for identifying rock glaciers based on machine learning according to claim 5, characterized in that: The step S4 further includes the following steps: S41: Define the feature conversion function of the image to extract the geomorphological features of the rock glacier in each image on the optical image. Interquartile Range(IQR)=S 75th -S 25th ; Among them, S 75th is the third quartile in the data, S 25th is the first quartile in the data, Interquartile Range (IQR) is the interquartile range, S is a value in the data sample, S median is the sample median, S robust is a robust standardized statistic; S42: Divide the data set into a ratio of 4:1 and evaluate the model using the cross entropy loss function. Among them, loss(S,y) is the loss function, y i is the label of sample i, the complete rock glacier is 1, the residual rock glacier is 0, and p i The probability of sample i being predicted as a complete type of rock glacier; Construct a stochastic gradient descent optimizer and introduce momentum. i t+1 =θ t -γg t ; Among them, θ t+1 is the weight of step t+1, θ t is the weight of the tth step, θ is the parameter, γ is the learning rate, g is the gradient, g t is the momentum term of the current step.

Images