CN112036264B - Automatic extraction method of superglacial moraine covering type glacier - Google Patents

Abstract

The invention discloses an automatic extraction method of moraine-covered glaciers, comprising: step 1, preprocessing the optical image of the research area to obtain multi-spectral bands and texture features of the research area; step 2, selecting appropriate terrain data to perform topography Analyze and obtain the corresponding slope, plane curvature, and section curvature of the study area; step 3 calculate the flow displacement characteristics of the study area by using the optical offset method; step 4: use the random forest algorithm to classify and obtain the surface moraine glacier profile. The present invention makes full use of the fluidity characteristics of surface moraine glaciers, uses optical offset technology to obtain flow displacement, and then combines random forest algorithm to extract surface moraine glaciers, which has a high degree of automation and high extraction accuracy; random forest algorithm solves The problems of feature threshold, sequence of feature usage, and combination are difficult to determine, and the accuracy of classification has been improved compared with decision tree; the addition of flow rate has greatly improved the accuracy of surface moraine and glacier boundary extraction.

Description

An automatic extraction method of moraine-covered glaciers

technical field

The invention belongs to the field of studying modern glaciology by using optical remote sensing technology, and relates to an automatic extraction method of moraine-covered glaciers.

Background technique

Changes in glaciers are seen as a sensitive indicator of climate change, and the land's most important freshwater resource. In recent years, under the background of global warming, glaciers in the Himalayas, known as the "Asian water tower", are an important part of the global mountain glaciers, which has attracted continuous attention from the glaciology community. For a long time, optical remote sensing technology has played an important role in studying the changes of glaciers on a large scale. Among them, moraine-covered glaciers are an important part of mountain glaciers, and their automatic extraction is of great significance to the study of glacier remote sensing changes. However, due to the spectral similarity between moraine-covered glaciers and the surrounding bare land and its own complexity, the automatic extraction of moraine-covered glaciers has always been a difficult point to be overcome.

In the previous studies on glacier changes, the automatic extraction of threshold classification was only for net glaciers, while for the extraction of surface moraine-covered glaciers, the method of visual interpretation by experts was often used. For the study of large-scale glacier changes, This method is time-consuming and labor-intensive. Later, scholars gradually introduced features such as slope, temperature, surface curvature, texture, and coherence into the automatic extraction process of surface moraines and glaciers, which promoted the development of automatic extraction of surface moraine glaciers.

Haeberli (1986) suggested that sometimes it is also difficult to distinguish the boundary between surface moraine glacier and surrounding bare land by relying on visual interpretation alone. Frank Paul (2003) proposed that it would be difficult to find the boundary between surface moraine glacier and surrounding debris only relying on multispectral information, and gradually introduced topographical factors into the extraction of surface moraine glacier.

Song Bo et al. (2006) used Landsat images and DEM to generate 5 factors of NDSI, NDVI, slope, curvature, and thermal infrared band, and used threshold classification and spatial analysis to semi-automatically extract surface moraine glaciers in Gomba Glacier. However, the method of threshold classification makes this method have geographical limitations, and the choice of threshold is questionable. For example, the author regards the area with a slope greater than 24° as the non-surface moraine glacier area.

Jiang Zongli (2012) judged the boundary of the moraine glacier based on the fluidity of the moraine glacier and its low coherence in the SAR image interferogram. STEFAN LIPPL (2018) added the post-processing process of morphological filtering on this basis, and automatically extracted a single surface moraine glacier. However, the coherence map generated by InSAR technology will be affected by many factors. For example, long time and space baselines, large terrain fluctuations, and atmospheric errors will all cause serious decoherence phenomena, resulting in the extraction results relying on many manual processes. Moreover, due to the difference in imaging principles between SAR images and optical images, registration errors will be introduced as a result.

Wu Miao (2017) used the object-oriented band ratio method combined with the feature partition method to classify the glaciers in Bomi County. According to the classification of remote sensing image features, the research area is divided into net glacier area under light conditions, surface moraine glacier area and shadow glacier area. During this period, multiple features such as band ratio, vegetation index, water body index, slope, and texture were used for threshold classification. Thresholds were determined artificially by trial and error and induction. And pointed out that the classification accuracy of net glaciers is 100%, and that of surface moraine glaciers is 93%. However, according to multiple experiments, the threshold classification method is difficult to achieve 93% accuracy for the classification of surface moraine glaciers in large areas, and the article does not give a comparison chart of the extraction results of surface moraine glaciers, so the effectiveness of this method is open to question.

To sum up, through the efforts of a large number of scholars, the research on the automatic extraction of moraine-covered glaciers has made great progress. However, as far as the results are concerned, there are still problems such as low accuracy of extraction results, difficulty in determining the threshold value of each feature and the sequence of feature use, which makes these methods unable to be extended or unable to complete the automatic extraction of large-scale surface moraine glaciers.

Contents of the invention

In view of the deficiencies in the prior art, the object of the present invention is to provide an automatic extraction method for moraine-covered glaciers, which solves the problems in the prior art that the accuracy of the extraction results is not high, the degree of automation is not high, and the threshold is difficult to determine It is difficult to apply to the extraction of surface moraines and glaciers in large areas.

In order to solve the above technical problems, the present invention adopts the following technical solutions to achieve:

An automatic extraction method for surface moraine-covered glaciers, comprising the following steps:

Step 1: Preprocess the optical image of the study area to obtain multispectral bands; perform principal component analysis on the multispectral bands to obtain the first principal component with the most information, and use this as the input band for texture calculation to obtain synergy bands As the texture feature of the study area; the optical image of the study area contains thermal infrared bands;

Step 2: Select appropriate terrain data for terrain analysis: project the terrain data into the same projection band as the multispectral band of the image, and then obtain the corresponding slope, plane curvature, and section curvature of the study area through 3D surface analysis tools;

Step 3: Calculate the flow displacement characteristics of the study area using the optical offset method:

In step 3.1, the optical images with a difference of 1-2 years in the study area are regarded as the front-phase and back-phase images respectively as the input data for the calculation of the optical offset, and the search window size, step size, signal-to-noise ratio threshold and robustness are set. The iterative value of the rod is calculated to obtain the preliminary EW and NS horizontal displacement images of the study area;

Step 3.2: Perform error correction on the preliminary EW and NS horizontal displacement images obtained in step 3.1 to obtain accurate EW and NS horizontal displacement maps, and synthesize the horizontal flow displacement map through the band calculation tool to obtain the flow displacement characteristics of the study area ;

Step 4: Use the random forest algorithm to classify and obtain the surface moraine glacier profile:

Step 4.1: Manually select samples on the images of the research area. The samples include bare land, clean glaciers, surface moraine glaciers, and glacial lakes. The samples are evenly distributed in the images of the research area;

Step 4.2, combining the multi-spectral bands, thermal infrared bands, texture features, slope, plane curvature, profile curvature and flow displacement features obtained in steps 1, 2 and 3 into a multi-band image;

In step 4.3, input the samples in step 4.1 and the multi-band images in step 4.2 into the random forest classification algorithm for learning and modeling; then input the unchecked part of the image of the study area into the random forest algorithm for Automatically classify, and then output the classification results;

In step 4.4, the classification results are subjected to morphological classification and post-processing to obtain accurate contours of surface moraine glaciers.

The present invention also includes following technical characteristics:

Specifically, the preprocessing in step 1 includes atmospheric correction and image stitching;

The texture calculation in step 1 adopts the texture calculation method based on second-order probability and statistics filtering.

Specifically, the error correction in step 3.2 includes long-wavelength track errors and fringe artifact errors;

The long-wavelength track error is corrected by a polynomial curve fitting method; the fringe artifact error is corrected by a mean value subtraction principle.

Specifically, step 4.4 performs morphological classification post-processing on the classification results, including the following steps:

Step 4.4.1, through two or more Majority/Minority analysis algorithms to remove classification loopholes and delete isolated points in the surface moraine glacier area;

Step 4.4.2, using the Clump Classis algorithm for cluster analysis;

Step 4.4.3, vectorizing and smoothing the surface moraine glacier to obtain the preliminary surface moraine glacier contour vector;

In step 4.4.4, the accurate surface moraine glacier profile is obtained through a certain area gradient screening.

Compared with the prior art, the present invention has the following technical effects:

The present invention makes full use of the fluidity characteristics of surface moraine glaciers, uses optical offset technology to obtain flow displacement, and then combines random forest algorithm to extract surface moraine glaciers; in the whole process, except for manually selecting samples, the degree of automation is high , the extraction accuracy is high; the random forest algorithm solves the problem of difficult determination of feature threshold, feature use sequence, and combination, and improves the classification accuracy compared with decision tree. The addition of flow displacement greatly improves the accuracy of surface moraine glacier boundary extraction.

Description of drawings

Fig. 1 is a flowchart of the present invention;

Fig. 2 is that the left and right sides of step 1 test area (Landsat4, 3, 2 combination) of the present invention are typical surface moraine glaciers;

Fig. 3 is the first principal component band and texture feature of step 1 of the present invention;

Fig. 4 is step 2 terrain features of the present invention, and A is DEM, B is slope, C is section curvature, and D is plane curvature;

Fig. 5 is the step 3 glacier flow displacement strip processing process of the present invention; A image before detrending error; B detrending factor; C image after detrending; D strip image after removal;

Figure 6 is the post-processing result of step 4.2 classification, the red area is the classification result, and the green line area is the manually demarcated surface moraine glacier area;

Fig. 7 is the surface moraine glacier extraction result after step 4.3 area gradient screening and vectorization;

Fig. 8 (a) is the outline figure that traditional classification method obtains; (b) is the outline figure that the random forest classification of the present application obtains;

Figure 9(a) is the result graph without displacement feature participation; (b) is the result graph with displacement feature participation of the present application;

Figure 10 is a comparison of the extracted surface moraine glacier area and the manually delineated area.

The present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments.

Detailed ways

The flow velocity of mountain glaciers in the Himalayas is 2-50m/y, and the flow velocity of surface moraine glaciers is about 0.5-10m/y, which is a relatively large displacement. After multiple verifications, the accuracy of optical offset technology is generally about 1/20 of a pixel, and the accuracy of measuring glacier flow velocity with landsat images is only about 0.75-1.5m. Therefore, using optical offset technology to measure surface moraine glacier flow velocity is feasible. Mobility is an important factor for a glacier to distinguish its surrounding bare land, so the addition of the velocity factor (ie, flow displacement) can theoretically improve the extraction accuracy of surface moraine glaciers.

After a lot of research, it has been shown that the random forest algorithm has the characteristics of strong model generalization ability, low requirement for the number of samples, and automatic determination of the importance of features. It happened to solve the problem that the threshold value was difficult to determine, and the order and importance of features were difficult to determine in the previous classification methods. Less sample requirements and strong generalization ability enhance the universality of the classification method, allowing the automatic extraction of surface moraine glaciers to adapt to a wider area.

The present invention provides a kind of automatic extraction method of moraine-covered glacier, and this method comprises the following steps:

Step 1: Preprocess the optical image of the study area, including atmospheric correction and image stitching; the optical image of the study area contains thermal infrared bands; the preprocessed image contains multispectral bands; the multispectral band of the preprocessed image Perform principal component analysis on spectral bands to obtain the first principal component component PC1 with the largest amount of information, and use this as the input band for texture calculation to obtain the synergy band as the texture feature of the research area; the texture calculation in step 1 is based on the second-order probability Texture Computation Methods in Statistical Filtering.

Specifically, this embodiment performs preprocessing on the Landsat image of the study area in 2018. The Landsat image provided by the USGS is L2 level. The image has been orthorectified and geometrically corrected, so the preprocessing only includes atmospheric correction and image stitching; and then Perform principal component analysis on multi-spectral bands to obtain the first principal component component PC1 with the largest amount of information, and perform texture analysis on it, and select the texture calculation method based on second-order probability statistical filtering; select the synergistic bands for comparison and analysis. Texture analysis, because it can best distinguish surface moraine glaciers; from this, three groups of multispectral 6 bands, thermal infrared bands, and texture bands are jointly obtained; the texture calculation is completed in the corresponding Co-occurrencemeasures tool in ENVI5.3. As shown in Figure 3, Figure 3(a) is the first principal component component image after principal component analysis of Landsat image multispectral bands; Figure 3(b) is the texture feature image calculated by the first principal component component .

Step 2: Select the appropriate terrain data for terrain analysis: project the terrain data into the same projection band as the multispectral band of the image, and then obtain the slope map, plane curvature, and section curvature corresponding to the study area through 3D surface analysis tools;

Specifically, in order to obtain more realistic terrain data, ALOS-World-3DDSM data is used; this data is a high-precision global digital surface model data provided free of charge by Japan Aerospace Exploration Agency (JAXA) in May 2015, with horizontal resolution The rate is 30 meters, and the elevation accuracy is 5 meters, which is one of the most accurate data in the world. In arc-gis, the dsm is projected to the same projection band as the multispectral band WGS84-45N. Use the 3D surface analysis tool in arc-gis to obtain the corresponding slope map, plane curvature, and section curvature. As shown in Figure 4, A: DEM of the study area; B: slope image generated by DEM; C: plane curvature image generated by DEM; D: profile curvature image generated by DEM.

Step 3: Calculate the flow displacement in the study area using the optical offset method;

In step 3.1, the optical images with a difference of 1-2 years in the study area are regarded as the front-phase and back-phase images respectively as the input data for the calculation of the optical offset, and the search window size, step size, signal-to-noise ratio threshold and robustness are set. The iterative value of the rod is calculated to obtain the preliminary EW and NS horizontal displacement images of the study area;

In step 3.1, set the search window size and step size according to the resolution of the optical image; for example, the sentinel-2 image resolution used in this method verification experiment is 10m. After many experiments, the search window is selected as 64×64, and the step size is selected as 4 For the Landsat series images, choose 32×32, and the step size is 4 for the best effect; in addition, the signal-to-noise ratio threshold value is 0.9, and the robust iteration value is 2; thus, the preliminary EW and NS horizontal displacement images of the study area can be obtained;

Step 3.2: Perform error correction on the preliminary EW and NS horizontal displacement images obtained in step 3.1 to obtain accurate EW and NS horizontal displacement maps, and synthesize the horizontal flow displacement map through the band calculation tool to obtain the flow displacement characteristics of the study area .

Due to satellite orbit errors and imaging errors, the horizontal displacement image obtained in the previous step contains many errors such as long-wavelength orbit errors and fringe artifacts with a large impact area, so it is necessary to correct the errors in the horizontal displacement images in two directions. Error correction includes long-wavelength orbit error and fringe artifact error; long-wavelength orbit error is corrected by polynomial curve fitting method; fringe artifact error is corrected by mean value subtraction principle. Among them, the long-wavelength orbital error can be well corrected by the polynomial curve fitting method; specifically, it can be realized by the image detrending tool provided in the COSI-CORR software; for the stripe artifact error, the Destripe image tool provided in the COSI-CORR software has a large Therefore, this method adopts the principle of mean value subtraction proposed by Feng Guangcai of Central South University, and implements it through matalab language programming. The result has a good streak removal effect. After the error processing of the above two steps, a relatively accurate horizontal displacement map in the EW and NS directions can be obtained, and finally the horizontal flow displacement map is synthesized through the band calculation tool in ENVI, where the formula is: Sqrt(b1^2+b2^2 ).

Specifically, using optical offset technology to calculate the flow displacement of moraine-covered glaciers in the study area from 2016 to 2018; after comparison, the COSI-CORR software with better processing results was selected for processing; the software was developed by the California Institute of Technology based on the IDL language Developed and embedded in ENVI5.3; for the banding problem in the offset calculation results, we use the mean value subtraction method and realize it through matlab programming; finally obtain a more accurate flow displacement map. As shown in Figure 5, A: the horizontal displacement image in the NS direction obtained on the right side of the study area; B: the image detrending tool removes the error factor of the long-wave band orbit error; C: the horizontal displacement image in the NS direction after the long-wave band orbit error is removed; D: NS horizontal displacement image after removing streak artifacts by mean subtraction method.

From Figure A, it can be seen that there are serious stripes in the horizontal displacement image without error processing, while in Figure D, it can be seen that the stripes have been well corrected.

Step 4: Use the random forest algorithm to classify and obtain the surface moraine glacier profile:

Step 4.1: Manually select samples on the images of the research area. The samples include bare land, clean glaciers, surface moraine glaciers, and glacial lakes. The samples are evenly distributed in the images of the research area;

Step 4.2, combining the multi-spectral bands, thermal infrared bands, texture features, slope, plane curvature, profile curvature and flow displacement features obtained in steps 1, 2 and 3 into a multi-band image;

In step 4.3, input the samples in step 4.1 and the multi-band images in step 4.2 into the random forest classification algorithm for learning and modeling; then input the unchecked part of the image of the study area into the random forest algorithm for Automatically classify, and then output the classification results;

The random forest algorithm belongs to the supervised classification algorithm in the field of machine learning. It needs to select samples in advance and provide them to the algorithm for learning and modeling. According to the characteristics of the ground objects in the area where the glacier is located, in order to improve the accuracy of the classification algorithm, the sample selection is divided into four categories: bare land, net glaciers, surface moraine glaciers, and glacial lakes. The samples are evenly distributed in the study area, and the sample area only accounts for the total area. 0.1% or so.

The entire classification process is carried out in ENVI5.3. The samples are selected in the form of roi, and finally the random forest algorithm provided by the ENVI extension tool is used for classification. The decision tree is 80-100 trees, and the rest of the parameters can be defaulted. The processing time is 15 minutes. about.

In step 4.4, the classification results are subjected to morphological classification and post-processing to obtain accurate contours of surface moraine glaciers. As shown in Figure 7, the results of vectorization, smoothing, and area gradient screening are carried out on the classification results; wherein the red line (line 1) is the visual interpretation outline of the surface moraine glacier, and the yellow line (line 2) is the extraction of the surface moraine glacier by the method of the present invention contour.

Specifically, samples were selected in the study area. The samples were divided into four categories: bare land, net glaciers, surface moraine glaciers, and glacial lakes. The samples were evenly distributed in the study area, and the sample area only accounted for about 0.1% of the total area. Use the ENVI extension tool random forest algorithm classification to classify, and the processing process takes about 15 minutes. Then, the post-processing operation of morphological classification was performed on the classification results to fill in holes and delete isolated points and clusters, and then vectorized, smoothed, and area gradient was deleted to obtain the surface moraine glacier outline.

Specific embodiments of the present invention are provided below, and it should be noted that the present invention is not limited to the following specific embodiments, and all equivalent transformations done on the basis of the technical solutions of the present application all fall within the scope of protection of the present invention.

Example 1:

In this example, using the landsat-8 image Sentinel-2 and ALOS-World-3ddsm data, the above-mentioned method is used to extract the surface moraine glaciers in the Shishapangma Peak and the Cuolangma Glacier Glacier Lake Group in the middle of the Himalayas. Download the October 2016 and October 2018 images of the region from the usgs website. However, due to cloud cover and image inventory problems in the 2016 images of this area, the Sentinel-2 images of December 2016 and October 2018 were used to obtain the surface moraine glacier displacement from 2016 to 2018. The time difference of two years is mainly because surface moraine glaciers flow more slowly than net glaciers, and the two-year flow rate can highlight the flow characteristics and help classification. But pay special attention, if the time span is too long, there will be loss of correlation.

In order to reflect the advantages of this method, the decision tree threshold classification was carried out using the same features at the same time, and each threshold was confirmed many times through trial and error, and it has basically reached the optimum. The classification result is shown in Figure 8(a); At the same time, in order to reflect the importance of displacement features, random forest classification without displacement features was carried out under the same conditions, and the classification results are shown in Figure 9(a).

Figure 8: a is the threshold classification result of the decision tree (the red line is the line 3 is the visually interpreted contour of the moraine glacier, and the green line is the line 4 is the contour obtained by the decision tree threshold classification), b is the classification result of the random forest (the red line is the line 5 is the visually interpreted contour of the surface moraine glacier, and the yellow line is line 6 is the contour obtained by random forest classification); it can be seen from the figure that the classification result of the traditional decision tree threshold classification is still relatively poor even with the help of displacement features. The random forest classification has obtained better classification results; it reflects the advantage of high accuracy of classification results of this method.

Figure 9: a is the preliminary classification result of the random forest without the participation of the flow displacement feature; b is the preliminary classification result of the random forest with the participation of the flow displacement feature (the method of the present invention).

It can be seen from the comparison that the random forest classification results without the participation of flow displacement features have improved accuracy compared with the threshold classification results in Figure 8, but there are many "loopholes" and large-scale misclassification phenomena. However, in the results of the classification method involving flow displacement, the "holes" and misclassification phenomena have been greatly improved, resulting in a further improvement in the accuracy of the classification results.

From the comparison in Figure 9, it is found that the automatic extraction accuracy of the random forest has been greatly improved compared with the threshold classification results. The addition of the displacement feature has greatly improved the unfavorable conditions such as large and numerous gaps and inaccurate edge extraction in surface moraine glacier extraction. The main reason is that the displacement has a certain continuity in the plane relative to other factors. It handles the problem of many gaps in the extraction results very well. Furthermore, mobility, as an important feature of surface moraine glaciers, is significantly different from the surrounding stable bare land. Therefore, from the perspective of other regions, the addition of displacement features eliminates the wrong extraction and omission of some edge regions. Therefore, the extraction accuracy is further improved.

Figure 10 to verify the extraction accuracy. Combined with the cataloged data, the surface moraine and glaciers in this area were manually delineated as the standard result. The 15 large glaciers were selected for comparison and found that the average area accuracy was 89.7%. A large number of small surface moraine glaciers were found in the classification results. After careful manual interpretation and extraction of corresponding flow velocities, it can be seen that these areas do have the characteristics of spectrum and mobility of surface moraine glaciers, so these small areas are indeed surface moraine glaciers. It proves that artificial delineation of surface moraine glaciers has the risk of missing small glaciers, but the method proposed in this paper can avoid most of the missing problems. Therefore, this method better solves the problems of difficulty in determining the threshold value, poor feature combination, poor applicability in large areas, and low degree of automation in the automatic extraction of surface moraines and glaciers.

Claims (3)

1. An automatic extraction method of superglacial moraine covering type glacier is characterized by comprising the following steps:

step 1: preprocessing an optical image of a research area to obtain a multispectral waveband; performing principal component analysis on the multispectral wave bands to obtain a first principal component with the most information content, and performing texture calculation by taking the first principal component as an input wave band to obtain a collaborative wave band serving as a texture feature of a research area; the optical image of the research area comprises a thermal infrared band;

and 2, step: selecting appropriate terrain data for terrain analysis: projecting the terrain data into a projection zone which is the same as the multispectral wave band of the image, and further obtaining the corresponding slope, plane curvature and section curvature of the research area through a 3D surface analysis tool;

and step 3: calculating the flow displacement characteristics of the study area by using an optical offset method:

step 3.1, respectively taking optical images with a difference of 1-2 years between the front time phase and the rear time phase of the research region as input data for calculating the optical offset, setting the size of a search window, the step length, the signal-to-noise ratio threshold value and the robust iteration value, and calculating to obtain preliminary EW and NS horizontal displacement images of the research region;

step 3.2, respectively carrying out error correction on the preliminary EW and NS direction horizontal displacement images obtained in the step 3.1 to obtain accurate EW and NS direction horizontal displacement images, and synthesizing the horizontal flow displacement images by a wave band calculation tool to obtain the flow displacement characteristics of the research area;

and 4, step 4: classifying by using a random forest algorithm and obtaining a superglacial moraine profile:

step 4.1, manually selecting samples on the image of the research area, wherein the samples comprise nude land, clean glacier, superglacial moraine and ice lake four types, and the samples are uniformly distributed in the image of the research area;

step 4.2, synthesizing the multispectral wave bands, the thermal infrared wave bands, the textural features, the gradients, the plane curvatures, the section curvatures and the flow displacement features obtained in the step 1, the step 2 and the step 3 into a multiband image;

4.3, inputting the sample in the step 4.1 and the multiband image in the step 4.2 into a random forest classification algorithm for learning and modeling; inputting the part which is not selected in the image of the research area into a random forest algorithm for automatic classification, and then outputting a classification result;

step 4.4, performing morphological classification post-treatment on the classification result to obtain an accurate superglacial moraine contour;

step 4.4 morphological classification post-processing of the classification results comprises the following steps:

4.4.1, removing classification holes and deleting isolated points in the superglacial moraine region by using a Majority/least analysis algorithm for two times or more;

step 4.4.2, performing cluster analysis by using a column Classis algorithm;

4.4.3, carrying out vectorization and smoothing operation on the superglacial moraine to obtain a preliminary superglacial moraine contour vector;

and 4.4.4, screening through a certain area gradient to obtain the accurate superglacial moraine profile.

2. The automated superglacial moraine covering-type glacier extraction method according to claim 1, wherein the pretreatment in the step 1 comprises atmospheric correction and image stitching;

the texture calculation in the step 1 adopts a texture calculation method based on second-order probability statistical filtering.

3. The automated extraction method of tillite covered glacier according to claim 1, wherein the step 3.2 of error correction comprises long wavelength orbit errors and streak artifact errors;

correcting the long wavelength orbit error by a polynomial curve fitting method; the streak artifact error is corrected by the mean subtraction principle.

Images