Glacier information extraction method based on deep learning

Abstract

The invention discloses a glacier information extraction method based on deep learning, and relates to the technical field of information extraction, and the method comprises the following steps: collecting required remote sensing image data according to a research area, constructing a sample library, preprocessing the collected remote sensing image data, and obtaining a pre-processed remote sensing image data; glacier features of the preprocessed remote sensing image data are extracted, abnormal conditions in the extraction process are analyzed, an abnormal feature analysis table is obtained, the remote sensing image data in the sample library are labeled, and glacier boundaries and classification labels are determined. Through optimization and customization of the deep learning model, interference caused by natural factors such as cloud and mist shielding, accumulated snow coverage and shadow change is reduced, the model can learn specific texture, shape and context information of the glacier, the glacier area and the non-glacier area can be accurately distinguished even in a complex and changeable environment, and the method has the advantages of being high in practicability and the like. Therefore, the glacier information extraction accuracy is remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of information extraction, and in particular to a glacier information extraction method based on deep learning. Background Art

[0002] As the largest freshwater resource reservoir on Earth, glaciers have a profound and complex impact on the global water cycle and climate system. Glaciers are not only a reservoir of water resources, but also participate in and influence the global water cycle process through meltwater flowing into rivers, lakes and oceans. At the same time, changes in glaciers, including their area, volume and flow speed, are important indicators of climate change, which can reveal regional and even global climate change trends. In the context of global warming, glaciers are facing unprecedented challenges of retreat and melting. As the temperature rises, the melting rate of glaciers accelerates, resulting in a significant reduction in glacier area and a sharp drop in ice reserves. This not only reduces the freshwater resources available for human use, but may also have a chain reaction on the local and even global climate system, such as changing ocean circulation and affecting precipitation patterns. Researchers can use advanced technologies such as deep learning to monitor glaciers. Deep learning has powerful image recognition and feature extraction capabilities, and can monitor the dynamic changes of glaciers in real time or near real time. With the continuous development and improvement of deep learning technology, its application prospects in glacier monitoring, climate research and disaster warning will be broader, providing strong technical support for global climate change response and sustainable development.

[0003] In existing technologies, the complex terrain, changeable climate environment and unstable lighting conditions unique to glacier areas, such as frequent cloud cover, snow cover, and shadow changes, constitute significant technical challenges, which directly affect the accuracy of deep learning models in extracting glacier information. What is more complicated is that the morphology and coverage of glaciers will change significantly with the change of seasons, which requires deep learning models to have a high degree of flexibility and adaptability to capture and accurately reflect these dynamic changes in real time. In response to the above problems, reducing interference factors and improving model adaptability are important directions for the development of current glacier information extraction technology, and are also the key to achieving accurate glacier monitoring and early warning. Summary of the invention

[0004] The purpose of the present invention is to provide a glacier information extraction method based on deep learning to solve the problems raised in the above background technology.

[0005] In order to solve the above technical problems, the technical solution adopted by the present invention is:

[0006] A glacier information extraction method based on deep learning includes the following steps:

[0007] Step 1: Collect the required remote sensing image data according to the research area and build a sample library, where the remote sensing image data includes real-time data and historical data;

[0008] Step 2, preprocessing the collected remote sensing image data, extracting glacier features from the preprocessed remote sensing image data, and analyzing abnormal conditions in the extraction process to obtain an abnormal feature analysis table;

[0009] Step 3: annotate the remote sensing image data in the sample library, determine the glacier boundaries and classification labels, and build a deep learning model based on the abnormal feature analysis table to obtain the environmental anomaly index and seasonal anomaly index;

[0010] Step 4, obtaining the glacier early warning assessment coefficient based on the environmental anomaly index and the seasonal anomaly index, and determining different early warning levels in combination with the preprocessed historical data, setting early warning assessment thresholds for different early warning levels, and obtaining an early warning assessment sequence table;

[0011] Step 5: Analyze the status of the glacier warning assessment coefficient at different warning levels based on the warning assessment sequence table, obtain an assessment report, and perform analysis and warning based on the extracted glacier information.

[0012] A further improvement of the technical solution of the present invention is that the process of constructing the sample library is:

[0013] Step 101, clarify the geographical location of the study area, such as the latitude and longitude range, administrative boundaries or specific geographical features, and mark the specific location of the study area on the map to facilitate subsequent data query and download, as well as the time range required for real-time data and historical data;

[0014] Step 102, obtain the latest remote sensing image data through satellite sensors, aerial photography, and ground observation stations. The remote sensing image data includes remote sensing images, terrain data, and climate data, and filter real-time data according to the study area and the time range of the remote sensing image data.

[0015] Step 103, using the data disclosed by the remote sensing data platform to obtain the historical data of the remote sensing image data, and integrating the screened real-time data with the obtained historical data to construct a sample library.

[0016] A further improvement of the technical solution of the present invention is that the process of obtaining the abnormal feature analysis table is as follows:

[0017] Step 201, the collected real-time data and historical data are corrected, data enhanced, and cloud and shadow processing is performed, wherein the correction includes radiation correction to eliminate the influence of sensor characteristics, solar altitude angle, atmospheric conditions and other factors on the data; atmospheric correction to remove the influence of the atmosphere on the image, such as aerosol scattering, water vapor absorption, etc., so that the image data is closer to the actual reflectivity or emissivity of the surface; geometric correction to correct the geometric distortion of the image, such as the pixel position offset caused by satellite attitude changes, earth curvature, terrain undulations, etc.

[0018] Step 202, using GIS processing technology to extract features from real-time data and historical data, respectively for glacier area, glacier texture, such as roughness, directionality, glacier shape, such as area, perimeter, aspect ratio, and spatial distribution,

[0019] Step 203, analyze the abnormal situation according to the extracted features, and record the abnormal type, location, impact range and other information to obtain abnormal interference factors, which include cloud data, shadow data, avalanche data and missing value data. For cloud data and shadow data, record their detection methods, mask processing results and impact on glacier feature extraction; for avalanche data and missing value data, record their causes, impact range and possible remedial measures;

[0020] Step 204, compile an abnormal characteristic analysis table based on the abnormal interference factors, and record detailed information of each abnormal situation, including abnormal type, occurrence time, location, scope, impact degree, and treatment measures.

[0021] A further improvement of the technical solution of the present invention is that the process of obtaining the environmental anomaly index and the seasonal anomaly index is as follows:

[0022] Step 301, combining the glacier area, glacier texture, glacier shape and spatial distribution characteristics, cutting the real-time data and historical data in the sample library to obtain image features, such as glaciers, water bodies, bare land, vegetation, etc., and automatically annotating the extracted image features to clarify the glacier area, abnormal area and specific abnormal type;

[0023] Step 302, integrating the abnormal interference factors and the associated data of the annotated image features to form a unified data set;

[0024] Step 303: Use the abnormal feature analysis table to identify abnormal situations in the data set, encode the abnormal situations, and build a deep learning model.

[0025] Step 304: divide the integrated data set into a training set, a validation set, and a test set, use the training set data to train the deep learning model, and use the validation set and test set data to validate the trained deep learning model to evaluate the performance of the deep learning model;

[0026] Step 305, using the trained model and combining it with the preprocessed real-time data to obtain the environmental anomaly index and the seasonal anomaly index, and analyzing the overall abnormal conditions of the glacier area, comparing the seasonal anomaly indexes of different seasons, identifying the laws and characteristics of seasonal changes, evaluating the differences in the degree of anomalies in the glacier area in different seasons, and outputting relevant analysis reports on the environmental anomaly index and the seasonal anomaly index.

[0027] A further improvement of the technical solution of the present invention is that the calculation formula of the environmental anomaly index is:

[0028]

[0029] Among them, EAI is the environmental anomaly index, A i is the reflectivity anomaly value of the i-th pixel, μ A is the average value of all pixel reflectance anomalies, N is the total number of pixel reflectance anomalies, C i is the cloud coverage of the jth pixel, μ c is the average value of cloud coverage of all pixels, and M is the total number of pixels of cloud coverage;

[0030] The calculation formula of the seasonal anomaly index is:

[0031]

[0032] Where SAI is the seasonal anomaly index, G t is the measured value of glacier coverage at the tth time point, G a is the long-term average glacier coverage, G s is the standard deviation of glacier coverage, ω t is the weight coefficient at the tth time point, r t is the seasonal variation coefficient at the t-th time point, and t is the total number of time points considered.

[0033] A further improvement of the technical solution of the present invention is that the process of obtaining the early warning evaluation sequence table is:

[0034] Step 401, combining the pre-processed historical data, combining the environmental anomaly index and the seasonal anomaly index, and analyzing the correlation between the environmental anomaly index and the seasonal anomaly index;

[0035] Step 402, assigning different weights to the environmental anomaly index and the seasonal anomaly index, performing weighted calculation on the two indexes, and obtaining a glacier early warning assessment coefficient, and assigning different weights to the anomaly index and the seasonal anomaly index according to their importance in affecting glacier changes;

[0036] Step 403, according to the distribution of the environmental anomaly index and the seasonal anomaly index in the historical data, the glacier early warning assessment coefficient is divided into different early warning levels, such as low risk, medium risk, and high risk;

[0037] Step 404, matching the result of the glacier early warning assessment coefficient with each early warning level, and setting a corresponding assessment threshold;

[0038] Step 405, arrange all glacier early warning assessment coefficients in descending order, assign a corresponding early warning level to each value, make an early warning assessment sequence table, and list the corresponding relationship between the environmental anomaly index and the seasonal anomaly index.

[0039] A further improvement of the technical solution of the present invention is that the calculation formula of the glacier early warning assessment coefficient is:

[0040]

[0041] Among them, ω EAI and ω SAI are the weight coefficients of the environmental anomaly index and the seasonal anomaly index, and ω SAI +ω EAI =1, EAI is the environmental anomaly index, SAI is the seasonal anomaly index, β and γ are the adjustment parameters for controlling the speed of index decrease and increase, and μSAI is the average value of the seasonal anomaly index.

[0042] A further improvement of the technical solution of the present invention is that the warning levels and their evaluation threshold ranges are:

[0043] Low risk warning level: GAAC <Q1;

[0044] Medium risk warning level: Q1≤GAAC <Q2;

[0045] High-risk warning level: GAAC>Q2.

[0046] A further improvement of the technical solution of the present invention is that the process of analyzing and warning based on the extracted glacier information is as follows:

[0047] Step 501: classify the glacier warning assessment coefficient according to the warning level, analyze the distribution range of different warning levels under each level standard, determine the frequency and characteristics of each warning level, and obtain the analysis results, and examine the correlation between the GAC value and other environmental parameters, such as temperature, precipitation, glacier thickness change rate, etc., to further understand the physical mechanism behind each level.

[0048] Step 502: Output an assessment report based on the analysis results, and prepare a risk assessment report, in which low risk, medium risk, and high risk are marked, and the influencing factors and possible consequences under different warning levels are analyzed;

[0049] Step 503, formulate response measures based on the assessment report and risk analysis results, continuously monitor the glacier early warning assessment coefficient, and regularly update the early warning assessment sequence table and assessment report.

[0050] Due to the adoption of the above technical solution, the present invention has the following technical advances compared with the prior art:

[0051] 1. The present invention provides a glacier information extraction method based on deep learning. Through the optimization and customization of the deep learning model, the interference caused by natural factors such as cloud obstruction, snow cover and shadow changes is reduced. The model can learn the unique texture, shape and contextual information of the glacier, and can accurately distinguish between glacier areas and non-glacier areas even in complex and changeable environments, thereby significantly improving the accuracy of glacier information extraction and providing reliable data support for glacier monitoring and protection.

[0052] 2. The present invention provides a method for extracting glacier information based on deep learning. The deep learning model adapts to the seasonal changes in the morphology and coverage of glaciers through continuous learning and optimization. By introducing strategies such as transfer learning and online learning, the model can continuously learn new glacier characteristics and quickly adapt to the dynamic changes in environmental conditions. In addition, the use of multi-source data fusion technology, combined with multi-dimensional information such as satellite images and ground observations, further enhances the model's adaptability to complex environments, ensuring that glacier information can be accurately extracted in different seasons and weather conditions.

[0053] 3. The present invention provides a glacier information extraction method based on deep learning. The efficient computing power of the deep learning model significantly improves the monitoring efficiency and real-time performance. It can quickly process large-scale remote sensing data and realize real-time extraction and updating of glacier information. This not only reduces labor costs, but also shortens the data processing cycle, enabling decision makers to quickly obtain the latest status of the glacier, providing the possibility for timely response measures.

[0054] 4. The present invention provides a glacier information extraction method based on deep learning. The efficient computing power of the deep learning model enables real-time data processing to be faster, and provides timely intelligence through real-time monitoring capabilities, enabling them to take quick action to reduce potential disaster risks and protect human life and property. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0056] Figure 1 is a flow chart of the method of the present invention;

[0057] Figure 2 The present invention is a flowchart for obtaining an environmental anomaly index and a seasonal anomaly index;

[0058] Figure 3 The figure is a flow chart for obtaining the early warning evaluation sequence table of the present invention. DETAILED DESCRIPTION

[0059] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0060] Examples, such as Figure 1-3 As shown, the present invention provides a glacier information extraction method based on deep learning, comprising the following steps:

[0061] Step 1: Collect the required remote sensing image data according to the study area and build a sample library, where the remote sensing image data includes real-time data and historical data. The process of building the sample library is as follows: clarify the geographical location of the study area, such as the latitude and longitude range, administrative boundaries or specific geographical features, mark the specific location of the study area on the map for subsequent data query and download, and the time range required for real-time data and historical data, obtain the latest remote sensing image data through satellite sensors, aerial photography, and ground observation stations. Remote sensing image data includes remote sensing images, terrain data, and climate data, and filter real-time data according to the study area and the time range of remote sensing image data. Use the data disclosed by the remote sensing data platform to obtain the historical data of remote sensing image data, and integrate the filtered real-time data with the acquired historical data to build a sample library;

[0062] Step 2, preprocess the collected remote sensing image data, extract the glacier features of the preprocessed remote sensing image data, and analyze the abnormal conditions in the extraction process to obtain an abnormal feature analysis table; the process of obtaining the abnormal feature analysis table is: correcting the collected real-time data and historical data, data enhancement, and cloud and shadow processing, where the correction includes radiation correction to eliminate the influence of sensor characteristics, solar altitude angle, atmospheric conditions and other factors on the data, atmospheric correction to remove the influence of the atmosphere on the image, such as aerosol scattering, water vapor absorption, etc., to make the image data closer to the actual reflectivity or emissivity of the surface, geometric correction to correct the geometric distortion of the image, such as pixel position offset caused by satellite attitude changes, earth curvature, terrain undulations, etc., and use GIS processing technology to extract the real-time data and historical data. Perform feature extraction, respectively for glacier area, glacier texture, such as roughness, directionality, glacier shape, such as area, perimeter, aspect ratio, and spatial distribution. Perform abnormal situation analysis based on the extracted features, and record information such as abnormal type, location, and impact range to obtain abnormal interference factors. Abnormal interference factors include cloud data, shadow data, avalanche data, and missing value data. For cloud data and shadow data, record their detection methods, mask processing results, and impact on glacier feature extraction. For avalanche data and missing value data, record their causes, impact range, and possible remedial measures. Prepare an abnormal feature analysis table based on the abnormal interference factors, and record detailed information on each abnormal situation, including abnormal type, occurrence time, location, range, impact degree, and treatment measures.

[0063] Step 3: Annotate the remote sensing image data in the sample library, determine the glacier boundaries and classification labels, and build a deep learning model based on the abnormal feature analysis table to obtain the environmental anomaly index and seasonal anomaly index. The process of obtaining the environmental anomaly index and seasonal anomaly index is as follows: combine the glacier area, glacier texture, glacier shape and spatial distribution characteristics, crop the real-time data and historical data in the sample library to obtain image features, such as glaciers, water bodies, bare land, vegetation, etc., and automatically annotate the extracted image features to clarify the glacier area, abnormal area and specific abnormal type, integrate the abnormal interference factors and the associated data of the annotated image features to form a unified data set, use the abnormal feature analysis table to identify the abnormal conditions in the data set, and encode the abnormal conditions. At the same time, build a deep learning model, and divide the integrated data set into training set, validation set and test set. Use the training set data to analyze the deep learning model. The deep learning model is trained and the trained deep learning model is verified using the validation set and test set data to evaluate the performance of the deep learning model. The trained model is used in combination with the preprocessed real-time data to obtain the environmental anomaly index and seasonal anomaly index, and the overall anomaly status of the glacier area is analyzed. The seasonal anomaly index of different seasons is compared to identify the laws and characteristics of seasonal changes, and the differences in the degree of anomaly of the glacier area in different seasons are evaluated. The relevant analysis reports on the environmental anomaly index and seasonal anomaly index are output. Through the optimization and customization of the deep learning model, the interference caused by natural factors such as cloud cover, snow cover and shadow changes is reduced. The model can learn the unique texture, shape and contextual information of the glacier, and can accurately distinguish between glacier areas and non-glacier areas even in complex and changing environments, thereby significantly improving the accuracy of glacier information extraction and providing reliable data support for glacier monitoring and protection.

[0064] The calculation formula of the environmental anomaly index is:

[0065]

[0066] Among them, EAI is the environmental anomaly index, A i is the reflectivity anomaly value of the i-th pixel, μ A is the average value of all pixel reflectance anomalies, N is the total number of pixel reflectance anomalies, C i is the cloud coverage of the jth pixel, μ c is the average value of cloud coverage of all pixels, and M is the total number of pixels of cloud coverage;

[0067] The calculation formula of seasonal anomaly index is:

[0068]

[0069] Where SAI is the seasonal anomaly index, G tis the measured value of glacier coverage at the tth time point, G a is the long-term average glacier coverage, G s is the standard deviation of glacier coverage, ω t is the weight coefficient at the tth time point, r t is the seasonal variation coefficient at the t-th time point, and t is the total number of time points considered;

[0070] Step 4, based on the environmental anomaly index and the seasonal anomaly index, obtain the glacier early warning assessment coefficient, and determine different early warning levels in combination with the preprocessed historical data, set early warning assessment thresholds for different early warning levels, and obtain the early warning assessment sequence table. The early warning assessment sequence table is obtained by combining the environmental anomaly index and the seasonal anomaly index in combination with the preprocessed historical data, analyzing the degree of correlation between the environmental anomaly index and the seasonal anomaly index, assigning different weights to the environmental anomaly index and the seasonal anomaly index, performing weighted calculation on the two indexes, and obtaining the glacier early warning assessment coefficient. According to the importance of the anomaly index and the seasonal anomaly index on glacier changes, different weights are assigned to the two. According to the distribution of the environmental anomaly index and the seasonal anomaly index in the historical data, the glacier early warning assessment coefficient is divided into different early warning levels, such as low risk, medium risk, and high risk. The result of the glacier early warning assessment coefficient is matched with each early warning level, and the corresponding assessment threshold is set. All glacier early warning assessment coefficients are arranged in order from high to low, and the corresponding warning level is assigned to each value. A early warning assessment sequence table is made, and the corresponding relationship between the environmental anomaly index and the seasonal anomaly index is listed;

[0071] The calculation formula of the glacier early warning assessment coefficient is:

[0072]

[0073] Among them, ω EAI and ω SAI are the weight coefficients of the environmental anomaly index and the seasonal anomaly index, and ω SAI +ω EAI =1, EAI is the environmental anomaly index, SAI is the seasonal anomaly index, β and γ are the adjustment parameters for controlling the speed of index decrease and increase, and μSAI is the average value of the seasonal anomaly index;

[0074] The warning levels and their assessment threshold ranges are as follows:

[0075] Low risk warning level: GAAC <Q1;

[0076] Medium risk warning level: Q1≤GAAC <Q2;

[0077] High-risk warning level: GAAC>Q2;

[0078] Step 5: Analyze the status of the glacier early warning assessment coefficient at different warning levels based on the early warning assessment sequence table to obtain an assessment report, and perform analysis and early warning based on the extracted glacier information. The process of performing analysis and early warning based on the extracted glacier information is as follows: classify the glacier early warning assessment coefficient according to the warning level, analyze the distribution range at different warning levels under each level standard, determine the frequency and characteristics of each warning level, and obtain the analysis results; examine the correlation between the GAC value and other environmental parameters, such as temperature, precipitation, and glacier thickness change rate, to further understand the physical mechanism behind each level; output an assessment report based on the analysis results, and prepare a risk assessment report, marking low risk, medium risk, and high risk in the report, and analyzing the influencing factors and possible consequences at different warning levels; formulate response measures based on the assessment report and risk analysis results, and continuously monitor the glacier early warning assessment coefficient, and regularly update the early warning assessment sequence table and assessment report; formulate response measures based on the assessment report and risk analysis results. This may include: enhanced monitoring: increasing the frequency of monitoring of high-risk areas; disaster prevention: formulating and implementing disaster prevention and mitigation plans; public warning: issuing warning information to the public and relevant departments; policy adjustment: adjusting relevant policies and plans based on risk assessment. The efficient computing power of deep learning models has significantly improved monitoring efficiency and real-time performance. It can quickly process large-scale remote sensing data and realize real-time extraction and updating of glacier information. This not only reduces labor costs, but also shortens the data processing cycle, enabling decision makers to quickly obtain the latest status of glaciers, making it possible to take timely response measures.

[0079] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A glacier information extraction method based on deep learning, characterized by: The following steps are involved: Step 1: Collect the required remote sensing image data according to the research area and build a sample library, where the remote sensing image data includes real-time data and historical data; Step 2, preprocessing the collected remote sensing image data, extracting glacier features from the preprocessed remote sensing image data, and analyzing abnormal conditions in the extraction process to obtain an abnormal feature analysis table; Step 3: annotate the remote sensing image data in the sample library, determine the glacier boundaries and classification labels, and build a deep learning model based on the abnormal feature analysis table to obtain the environmental anomaly index and seasonal anomaly index; Step 4, obtaining the glacier early warning assessment coefficient based on the environmental anomaly index and the seasonal anomaly index, and determining different early warning levels in combination with the preprocessed historical data, setting early warning assessment thresholds for different early warning levels, and obtaining an early warning assessment sequence table; Step 5: Analyze the status of the glacier warning assessment coefficient at different warning levels based on the warning assessment sequence table, obtain an assessment report, and perform analysis and warning based on the extracted glacier information.

2. The method for extracting glacier information based on deep learning according to claim 1, characterized in that: The process of building the sample library is as follows: Step 101, clarify the geographical location of the study area and the time range required for real-time data and historical data; Step 102, obtaining the latest remote sensing image data through satellite sensors, aerial photography, and ground observation stations, and filtering real-time data according to the study area and the time range of the remote sensing image data; Step 103, using the data disclosed by the remote sensing data platform to obtain the historical data of the remote sensing image data, and integrating the screened real-time data with the obtained historical data to construct a sample library.

3. The method for extracting glacier information based on deep learning according to claim 2, characterized in that: The process of obtaining the abnormal feature analysis table is as follows: Step 201, correcting, enhancing, and processing clouds and shadows on the collected real-time data and historical data, wherein the correction includes radiation correction, atmospheric correction, and geometric correction; Step 202, using GIS processing technology to extract features from real-time data and historical data, and extracting features for glacier area, glacier texture, glacier shape and spatial distribution respectively; Step 203, performing abnormal situation analysis according to the extracted features to obtain abnormal interference factors, wherein the abnormal interference factors include cloud data, shadow data, avalanche data, and missing value data; Step 204, compile an abnormal feature analysis table based on the abnormal interference factors, and record detailed information of each abnormal situation.

4. The method for extracting glacier information based on deep learning according to claim 3, characterized in that: The process of obtaining the environmental anomaly index and the seasonal anomaly index is as follows: Step 301, combining the glacier area, glacier texture, glacier shape and spatial distribution characteristics, cutting the real-time data and historical data in the sample library to obtain image features, and automatically annotating the extracted image features; Step 302, integrating the abnormal interference factors and the associated data of the annotated image features to form a unified data set; Step 303, using the abnormal feature analysis table to identify abnormal situations in the data set, encode the abnormal situations, and build a deep learning model; Step 304: divide the integrated data set into a training set, a validation set, and a test set, use the training set data to train the deep learning model, and use the validation set and test set data to validate the trained deep learning model to evaluate the performance of the deep learning model; Step 305, using the trained model and combining it with the preprocessed real-time data to obtain the environmental anomaly index and the seasonal anomaly index, and analyzing the overall abnormal conditions of the glacier area, comparing the seasonal anomaly indexes of different seasons, identifying the laws and characteristics of seasonal changes, evaluating the differences in the degree of anomalies in the glacier area in different seasons, and outputting relevant analysis reports on the environmental anomaly index and the seasonal anomaly index.

5. The method for extracting glacier information based on deep learning according to claim 4, characterized in that: The calculation formula of the environmental anomaly index is: Among them, EAI is the environmental anomaly index, A i is the reflectivity anomaly value of the i-th pixel, μ A is the average value of all pixel reflectance anomalies, N is the total number of pixel reflectance anomalies, C i is the cloud coverage of the jth pixel, μ c is the average value of cloud coverage of all pixels, and M is the total number of pixels of cloud coverage; The calculation formula of the seasonal anomaly index is: Where SAI is the seasonal anomaly index, G t is the measured value of glacier coverage at the tth time point, G a is the long-term average glacier coverage, G s is the standard deviation of glacier coverage, ω t is the weight coefficient at the tth time point, r t is the seasonal variation coefficient at the t-th time point, and t is the total number of time points considered.

6. The method for extracting glacier information based on deep learning according to claim 5, characterized in that: The acquisition process of the early warning evaluation sequence table is as follows: Step 401, combining the pre-processed historical data, combining the environmental anomaly index and the seasonal anomaly index, and analyzing the correlation between the environmental anomaly index and the seasonal anomaly index; Step 402, assigning different weights to the environmental anomaly index and the seasonal anomaly index, performing weighted calculation on the two indexes, and obtaining a glacier early warning assessment coefficient; Step 403, classifying the glacier early warning assessment coefficient into different early warning levels according to the distribution of the environmental anomaly index and the seasonal anomaly index in the historical data; Step 404, matching the result of the glacier early warning assessment coefficient with each early warning level, and setting a corresponding assessment threshold; Step 405, arrange all glacier early warning assessment coefficients in descending order, assign a corresponding early warning level to each value, make an early warning assessment sequence table, and list the corresponding relationship between the environmental anomaly index and the seasonal anomaly index.

7. The method for extracting glacier information based on deep learning according to claim 6, characterized in that: The calculation formula of the glacier early warning assessment coefficient is: Among them, ω EAI and ω SAI are the weight coefficients of the environmental anomaly index and the seasonal anomaly index, and ω SAI +ω EAI =1, EAI is the environmental anomaly index, SAI is the seasonal anomaly index, β and γ are the adjustment parameters for controlling the speed of index decrease and increase, and μSAI is the average value of the seasonal anomaly index.

8. The method for extracting glacier information based on deep learning according to claim 7, characterized in that: The warning levels and their assessment threshold ranges are as follows: Low risk warning level: GAAC <Q1; Medium risk warning level: Q1≤GAAC <Q2; High-risk warning level: GAAC>Q2.

9. The method for extracting glacier information based on deep learning according to claim 8, characterized in that: The process of analyzing and warning based on the extracted glacier information is as follows: Step 501, classify the glacier warning assessment coefficients according to the warning levels, analyze the distribution ranges under different warning levels under each level standard, determine the frequency and characteristics of each warning level, and obtain the analysis results; Step 502: Output an assessment report based on the analysis results, and prepare a risk assessment report, in which low risk, medium risk, and high risk are marked, and the influencing factors and possible consequences under different warning levels are analyzed; Step 503, formulate response measures based on the assessment report and risk analysis results, continuously monitor the glacier early warning assessment coefficient, and regularly update the early warning assessment sequence table and assessment report.

Images