A method for extracting water bodies in mountainous areas based on Sentinel-2 images

By analyzing the reflectivity value of the top layer of the atmosphere in Sentinel-2 images and constructing a triangular water index, the difficulty and leakage of water extraction in mountainous water bodies are solved, and high-precision water body extraction is achieved, reducing the error rate of natural disaster prediction.

CN114399514BActive Publication Date: 2024-12-06HARBIN INST OF TECH SHENZHEN GRADUATE SCHOOL
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202111574212.7
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2021-12-21
Publication Date
2024-12-06
Estimated Expiration
2041-12-21

AI Technical Summary

Technical Problem

The prior art has problems such as difficult, leakage and low extraction accuracy in mountain water extraction, especially in the extraction of small water bodies and mountain water bodies, and is easily confused with ice and snow, clouds and mountain shadows, resulting in errors. High rate.

Method used

The mountain water body extraction method based on Sentinel-2 images was adopted. By selecting multiple water body samples, statistically analyzing the reflectivity value of the top layer of the atmosphere, using the Cartesian coordinate system to divide two types of water bodies, and constructing a triangular water body index to calculate the Sentinel-2 images. Water body index, obtaining the optimal threshold value through histogram to extract water bodies, avoiding the establishment of training samples and improving the extraction speed and accuracy.

Benefits of technology

It realizes high-precision mountain water extraction without establishing training samples, avoids leakage, reduces the error rate of natural disasters such as landslides, and can effectively distinguish mountain shadows and water bodies, improving extraction accuracy.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114399514B_ABST
    Figure CN114399514B_ABST
Patent Text Reader

Abstract

The invention discloses a method for extracting water bodies in mountainous areas based on Sentinel-2 images. First, multiple water samples are selected on Sentinel-2 MSI images, their top atmospheric reflectance values ​​in the images are statistically analyzed, and the top atmospheric reflectance values ​​in different bands are calculated; then, according to the distribution of the top atmospheric reflectance values ​​of pixels of each water sample in different bands under a rectangular coordinate system, the water bodies are divided into two categories according to the top atmospheric reflectance values ​​of pixels of different water bodies in each band; a rectangular coordinate system is established and a triangle is constructed according to the reflectance values ​​and bands of the two types of water bodies; finally, the water body index of the two types of water bodies is calculated using the triangle similarity theorem, and the triangle water body index based on the Sentinel-2 image is summed up to obtain accurate water body information distribution. The invention can effectively distinguish mountain shadows from water bodies, avoid the phenomenon of wrong extraction in the process of extracting water bodies in mountainous areas, has high extraction accuracy, and can also be used for the extraction of lakes, reservoirs, rivers and urban surface water.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention belongs to the technical field of remote sensing image intelligent recognition, and in particular relates to a method for extracting water bodies in mountainous areas based on Sentinel-2 images. Background Art

[0002] With the change of the earth's climate, disasters such as floods and landslides occur frequently, causing great losses to people's lives and property. Therefore, it is of great significance to quickly and accurately obtain surface water information for monitoring disasters such as floods and landslides. Using optical remote sensing images to obtain water information is a fast and effective method. In recent years, the method of extracting water from optical remote sensing images has developed from the initial manual visual interpretation to the current automatic classification method of extracting water. At present, the commonly used automatic classification method of extracting water overcomes the shortcomings of manual visual interpretation that cannot extract water in large areas. This type of method mainly includes the threshold method based on pixel classification and the classification method based on target classification. The threshold method is more used in medium and high resolution images. It is mainly based on the spectral characteristics of the ground objects to establish classification models and water indexes. Its algorithm is simple and effective. It is widely used in the extraction of large-area water bodies such as lakes, but it is not suitable for the extraction of small water bodies and mountain water bodies. The classification method is a method for extracting water bodies based on the spectral, texture and spatial characteristics of images. It is mainly used in high-resolution images. The commonly used methods are mainly SVM method, decision tree classification method and object-oriented method. It has the advantage of extracting more detailed water bodies, but it is difficult to effectively distinguish between shadows and water bodies, and it is easy to cause confusion in the extraction of water bodies in mountainous areas. In addition, the algorithm of this method is complex and requires the establishment of training samples. The extraction speed will be reduced due to sample selection.

[0003] The water index method can realize the extraction of water bodies in mountainous areas. The commonly used water indexes are: ① Normalized Difference Water Index (NDWI), which can suppress vegetation information to the greatest extent and highlight water information, but it is difficult to distinguish buildings and soil from water bodies, and is greatly affected by ice, snow, clouds and mountain shadows. ② Improved Normalized Difference Water Index (MNDWI), which can suppress building and soil information to the greatest extent, highlight water characteristics, and has a good effect in urban water body extraction. ③ Enhanced Water Index (EWI), which can suppress building, soil and vegetation information, highlight water bodies in semi-arid areas, but is easily confused with shadows and snow and difficult to distinguish from river beaches, and is not easy to extract mountain water bodies. ④ Shadow Water Index (SWI), which has the advantage of distinguishing water bodies from shadows, and can also reduce the influence of snow and mountain shadows. However, the above water index methods have the phenomenon of missing extraction in the extraction of mountain water bodies and small water bodies, and there is a deficiency of extracting ice, snow and clouds as water bodies, the extraction accuracy is not high, and there are errors in predicting natural disasters such as landslides. Summary of the invention

[0004] In view of the deficiencies pointed out in the above background technology, the present invention provides a method for extracting water bodies in mountainous areas based on Sentinel-2 images, aiming to solve the problems in the above background technology such as difficulty in extracting water bodies in mountainous areas, omissions, and low extraction accuracy.

[0005] To achieve the above object, the technical solution adopted by the present invention is:

[0006] A method for extracting water bodies in mountainous areas based on Sentinel-2 images is as follows:

[0007] Step 1: Select a variety of water samples of different sizes and properties on the Sentinel-2MSI image, statistically analyze the top atmospheric reflectance values ​​of the various water samples in the Sentinel-2MSI image, and calculate the top atmospheric reflectance value of each water sample in different image bands; preferably, the image bands include B 11 (Shortwave infrared band 1), B 12 (shortwave infrared band 2), B2 (blue light band), B3 (green light band), B4 (red light band), B5 (vegetation red edge band 1), B6 ​​(vegetation red edge band 2), B7 (vegetation red edge band 3), B8 (near infrared band), and B9 (water vapor band).

[0008] Step 2: According to the top atmosphere reflectance value of each water body sample in different image bands, the top atmosphere reflectance value is divided into two distribution trends with a straight line at an angle of 45° to the horizontal axis of the rectangular coordinate system as the boundary. According to the two distribution trends, the water bodies on the Sentinel-2MSI image are divided into two types of water bodies; then the image bands that can reflect the two types of water bodies and the top atmosphere reflectance values ​​corresponding to the image bands are selected, and their reflectance values ​​and bands are expressed in a new rectangular coordinate system, and different triangles are constructed in the new rectangular coordinate system.

[0009] The specific method for constructing the triangle is as follows: the average value of the maximum atmospheric top reflectance value and the minimum atmospheric top reflectance value corresponding to the image band of the water body sample is used as the ordinate, and the image band is used as the abscissa.

[0010] In this step, the image bands contained in the two types of water bodies are:

[0011] Category I water body: B 12 , B3, B4, B5, B6;

[0012] Category II water body: B 12 , B2, B3, B9.

[0013] Step 3: According to the characteristics of the triangles constructed by the two types of water bodies, the initial water body index calculation method is selected, and the water body index of the two types of water bodies is calculated separately by using the triangle similarity theorem, and then the water body index of the two types of water bodies is summed to obtain the water body index for the Sentinel-2 image, that is, the triangle water body index. The triangle water body index is used to calculate the Sentinel-2 image to obtain an enhanced water body map, and then the optimal threshold is obtained through the histogram to extract the water body to obtain the final water body information distribution map.

[0014] Preferably, the initial water index calculation formula selected when calculating the water index of the above-class I water body is as follows:

[0015] TWI1=(B3-B 12 ) / (B3+B 12 ).

[0016] The calculation formula of the initial water index selected when calculating the water index of Class II water bodies is as follows:

[0017] TWI2 = (B9 - B8) / (B8 + B9).

[0018] The above-mentioned mountain water extraction method can also be used for extracting lakes, reservoirs, rivers or urban surface water.

[0019] Compared with the shortcomings and deficiencies of the prior art, the present invention has the following beneficial effects:

[0020] The present invention does not need to establish training samples, the extraction speed is not affected by sample selection, and can effectively distinguish mountain shadows from water bodies. The present invention avoids the phenomenon of missing extraction during the extraction of water bodies in mountainous areas, has high extraction accuracy, and can effectively reduce the error rate of predicting natural disasters such as landslides. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 These are 11 water body samples and Sentinel-2MSI image maps selected from Sentinel-2MSI images.

[0022] Figure 2 It is the statistical distribution map of the top atmosphere reflectance values ​​of water samples in Sentinel-2MSI images and the statistical distribution map of the top atmosphere reflectance values ​​of each water sample in different image bands.

[0023] Figure 3 It is a triangle diagram constructed in the rectangular coordinate system using the top atmospheric reflectance values ​​and image bands of two types of water bodies.

[0024] Figure 4These are the Sentinel-2 MSI images of different areas of the Brahmaputra River in Example 1 and the results of extracting water bodies using the TWI index method.

[0025] Figure 5 This is a detailed location map of the study area of ​​Example 2 and Example 3.

[0026] Figure 6 This is a graph showing the extraction results of different water body indices for the water bodies in the local study area of ​​Qinghai Province in Example 2.

[0027] Figure 7 This is a result diagram of different water body indices in the local study area of ​​Tibet in Example 3. DETAILED DESCRIPTION

[0028] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0029] Sentinel-2 images are used as remote sensing images for analysis and explanation of the present invention. Eleven water samples of different sizes and properties are selected from Sentinel-2 MSI images, including rivers, lakes, reservoirs and small rivers. Figure 1 As shown in the figure, the atmospheric top reflectance values ​​of 11 water samples in Sentinel-2MSI images were statistically analyzed. The statistical results are as follows Figure 2 As shown in (k), it can be seen that the reflectivity of water bodies with different properties in each band is similar but also exhibits different characteristics. Figure 2 The reflectivity of water bodies in (k) shows two trends. The reflectivity of one type of water body is in band B. 11 From band B1 to band B2, it shows a decreasing trend, and from band B8 to band B9, it shows a significant increasing trend. 11 From band B8 to band B9, there is no obvious increasing trend and the change is stable. 11 (Shortwave infrared band 1), B 12 The pixel top atmosphere reflectance values ​​of B (shortwave infrared band 2), B2 (blue light band), B3 (green light band), B4 (red light band), B5 (vegetation red edge band 1), B6 ​​(vegetation red edge band 2), B7 (vegetation red edge band 3), B8 (near infrared band), and B9 (water vapor band) are as follows: Figure 2 As shown in (a)-(j), Figure 2(a), (b), (c), (d), (e), (f), (g), (h), (i), and (j) are the reflectance values ​​of different water samples on band 2, band 3, band 4, band 5, band 6, band 7, band 8, band 9, band 11, and band 12 respectively. Figure 2 As can be seen from (a)-(j), the reflectance values ​​of water samples also show two concentrated distribution trends, both of which divide the water samples into two categories based on the reflectance intensity along the direction from the lower right point 0 to the upper left of the rectangular coordinate system with 45° as the boundary.

[0030] Therefore, according to the distribution of the pixel atmospheric top reflectance values ​​of each water body sample in different image bands in the rectangular coordinate system, the water bodies in the Sentinel-2MSI image are divided into two types of water bodies along the direction from the 0 point in the lower right corner of the rectangular coordinate system to the upper left corner, with a straight line at an angle of 45° to the horizontal axis of the rectangular coordinate system. The reflectance values ​​of the two types of water bodies and the corresponding image band divisions are shown in Table 1:

[0031] Table 1 Water bodies in Sentinel-2A images

[0032]

[0033] The characteristic of Class I water bodies is that they are light green in color on Sentinel-2MSI images, as shown in sample water1-7; the characteristic of Class II water bodies is that they are dark in color on Sentinel-2MSI images, as shown in sample water8-11. Figure 2 The water reflectance values ​​in (k) are highest in bands B3 and B5. 12 and B6 have the lowest values; the Class II water body has the lowest values ​​at Figure 2 The water reflectance value in (k) is in band B 12 The reflection is strong in band B9 and band B2 and B3, and the absorption is strong in band B3. According to the above characteristics, the triangles of the corresponding bands and the reflectivity values ​​of the top layer of the atmosphere of the two types of water bodies in the rectangular coordinate system are established respectively, as follows: Figure 3 shown. Figure 3 (a) is the triangle constructed by the first type of water body, and (b) is the triangle constructed by the second type of water body. Figure 3 B5 and B3 of (a) are the average values ​​of the maximum and minimum values ​​of the first type of water samples; Figure 3 B9 in (b) is the average of the maximum and minimum values ​​of the second type of water samples. In this way, it can be converted into a mathematical problem and the water index can be established using mathematical characteristics.

[0034] According to the triangular characteristics of the two types of water bodies, the initial water index calculation method for the two types of water bodies is selected, and the water index of the two types of water bodies is calculated separately using the triangle similarity theorem, and then the water index of the two types of water bodies is summed up to obtain the water index for Sentinel-2 images, namely the triangle water index (TWI). The mountain water body on the Sentinel-2 image is determined based on the obtained triangle water index. The detailed calculation process is as follows:

[0035] The spectral characteristics of the first type of water bodies conform to the improved normalized difference water index (MNDWI) research method, which suppresses the building and soil information to the greatest extent, highlights the water characteristics, and has a good effect in urban water body extraction. The MNDWI calculation formula in Sentinel-2 images is as follows:

[0036] MNDWI=(B3-B 12 ) / (B3+B 12 ) (1)

[0037] Therefore, the present invention is based on the improved normalized difference water index (MNDWI) to construct the (G-SIR) using triangle ( Figure 3 (a)) principle is replaced to calculate the water index of Class I water bodies. Therefore, the initial water index calculation formula selected when calculating the water index of Class I water bodies is as follows:

[0038] TWI1=(B3-B 12 ) / (B3+B 12 ) (2)

[0039] Figure 3 In (a), according to the triangle similarity theorem, we know that ΔACO≌ΔDEO, ΔBEO≌ΔHFO. Therefore, according to the similarity theorem of similar triangles: the corresponding sides of similar triangles are proportional, we can get that in ΔACO and ΔDEO:

[0040]

[0041] From formula (3), we can get:

[0042]

[0043] In ΔBEO and ΔHFO:

[0044]

[0045] From formula (5), we can get:

[0046]

[0047] Substituting formula (4) into formula (6) yields:

[0048]

[0049] because Figure 3 (a) contains:

[0050] B6-B5=(B 12 -B5)-(B 12 -B6) (8)

[0051] Substituting formula (4) and formula (7) into formula (8) yields:

[0052]

[0053] Combining formula (9) we get:

[0054]

[0055] From formula (10), we can get:

[0056]

[0057] Therefore, TWI1 = (B3-B 12 ) / (B3+B 12 ) combined with formula (11) can be rewritten into the following formula (12), which is the water body index of Class I water body:

[0058]

[0059] The distribution of spectral reflectance of Class II water bodies was analyzed, and according to the construction principle of vegetation index, Class II water body index (TWI2) was established based on Sentinel-2 images using B8 (near infrared band) and B9 (water vapor band). The calculation formula for the initial water body index of Class II water bodies is as follows:

[0060] TWI2=(B9-B8) / (B8+B9) (13)

[0061] Since band B9 is a water vapor band in Sentinel-2 images, it is not stable and practical, so B9 is replaced by other bands using the triangle similarity principle. Figure 3 From ΔBGH in (b), we know that: (B9-B8)=(B9-B2)-(B8-B2), so we can change formula (13) to:

[0062]

[0063] Figure 3 In (b), ΔOHB≌ΔODC and ΔAEG≌ΔBHG. According to the triangle similarity theorem, we know that:

[0064]

[0065] By deducing from formula (15), we can get:

[0066]

[0067]

[0068] From formula (15) and formula (17), we can know that:

[0069]

[0070] Transforming formula (18) yields:

[0071]

[0072] Let |AE|=a, |BH|=c, formula (19) can be rewritten as:

[0073]

[0074] Therefore, we can get B9 as:

[0075]

[0076] Substituting formula (19) and (21) into formula (14), we can get the water index of Class II water body:

[0077]

[0078] According to the above calculation process, the water index calculation formulas of the two types of water bodies were obtained respectively. Therefore, the total triangular water index is the sum of the two types of water body indices:

[0079]

[0080] |DE|, |BE|, |HF|, |AC| all represent magnitudes. For ease of use, the coefficients are rounded off. Among them, |DE|≈0.021, |BE|≈0.041, |HF|≈0.031, |AC|≈0.032, a≈0.055, c≈0.199. Based on the reflectance of the water sample being less than 1, Taking the reciprocal becomes The triangular water index TWI based on Sentinel-2 images is defined as:

[0081]

[0082] In the Sentinel-2 image, B 12The TWI water index eliminates non-water pixels and can extract water bodies in high-altitude mountainous areas, suppress non-water pixels such as mountain shadows and snow, and can also suppress dark building surfaces in urban background areas and extract urban water bodies.

[0083] Example 1

[0084] Sentinel-2A image data of the Brahmaputra River in India on July 28, 2021 was selected as the experimental point. The northern area of ​​the Brahmaputra River has a high altitude and many mountains, with small lakes and puddles between the mountains. The specific location of the Brahmaputra River area is as follows: Figure 4 (a) and (d) are shown in the GEE platform using the TWI index method. Figure 4 The enhanced water bodies in (a) and (d) are as follows: Figure 4 As shown in (b) and (e), this study uses the OTSU algorithm to automatically extract the threshold of the TWI index, and obtains the final water body information distribution as shown in Figures (c) and (f). Comparing the figures before and after water body extraction, it can be seen that the distribution position and area of ​​water body information extracted by the TWI water body index method of the present invention are almost consistent with the actual water body information distribution displayed by the Sentinel-2MSI image. Therefore, the TWI water body index method has a good resolution effect, can well extract the water body between the mountains, effectively reduce the influence of noise, make the water body-non-water body boundary smoother, and obtain a more complete distribution map of rivers and lakes.

[0085] Example 2

[0086] In order to qualitatively evaluate the effectiveness of the method of the present invention and reflect the advantages of the TWI index of the present invention in extracting water body information, the current six classic water body indices are selected as comparisons for water body extraction analysis. The six classic water body indices are: wetness index (LSWI), improved normalized difference water index (MNDWI), normalized difference water index (NDWI), automatic extraction water index (AWEI), shadow water index (SWI) and enhanced water index (EWI). The Sentinel-2A image data of the Tibet region on October 13, 2021 were selected. The water bodies in the Tibet region include rivers, lakes and other conventional water bodies, as well as special water bodies containing sediments, salt and ice, and there are a lot of ice and snow. These conditions are the best conditions for testing the effectiveness of the TWI index. The detailed location of the study area is as follows: Figure 5 As shown, the study area in Tibet is located between 89.88 and 91.76 east longitude and 31.06 and 32.37 north latitude.

[0087] The water bodies in the Tibet study area were extracted using LSWI, MNDWI, NDWI, AWEI, SWI, EWI and the TWI method of the present invention on the GEE platform. The results of the water body index extraction are shown in Figure 2. Figure 6 shown.

[0088] Figure 6 (a) is the Google map of the study area. Figure 6 (b) Figure 6 (a) shows the Sentinel-2A image data. It can be seen that the ground objects covered by the image data include snow, lakes, mountains, shadows and ice. The water enhancement maps obtained by EWI, AWEI, SWI, LSWI, MNWI, MNDWI and TWI water indexes are as follows: Figure 6 As shown in (c), (d), (e), (f), (g), (h), and (i). Figure 6 (d), (e) and (f) show that the separation effect between water and background is not obvious. Figure 6 (c), (g) and (h) have significantly better water enhancement effects than Figure 6 (d), (e) and (f) water enhancement performance. It can be seen from the enhanced water performance that LSWI, MNDWI, NDWI, AWEI, SWI and EWI are difficult to identify or even miss some frozen water bodies (square areas in the figure), and small rivers are not fully identified or even not identified as water bodies (as shown in the elliptical area in the figure). TWI solves these shortcomings and can better extract ice water bodies, effectively distinguish water bodies from snow, and enhance small rivers. It can be seen that the TWI of the present invention can effectively realize the extraction of mountain lakes, frozen water and rivers, and can distinguish snow, clouds and mountain shadows from water bodies.

[0089] Example 3

[0090] Sentinel-2A image data of a local area of ​​Qinghai Province on October 13, 2021 was selected. The detailed location of the study area is as follows: Figure 5 As shown in the figure, the research area of ​​Qinghai Province is the area of ​​Oring Lake, Xingxinghai and Longricuo Lake at 97.46-98.38 east longitude and 34.54-35.16 north latitude.

[0091] On the GEE platform, LSWI, MNDWI, NDWI, AWEI, SWI, EWI and the TWI method of this invention were used to extract the water bodies in the local study area of ​​Qinghai Province. The extraction results of each water body index are shown in Figure 7 shown. Figure 7 (a) and (b) are the Sentinel-2MSI image and the image of area A on the Sentinel-2MSI image, respectively. Figure 7(d), (e), (h), (i), (l), (m) and (p) are the extraction results of EWI, AWEI, SWI, LSWI, NDWI, MNDWI and TWI of Sentinel-2MSI images, respectively. Figure 7 (c), (f), (g), (j), (k), (n) and (o) are the extraction results of EWI, AWEI, SWI, LSWI, NDWI, MNDWI and TWI of area A on Sentinel-2MSI image, respectively.

[0092] Depend on Figure 7 It can be seen that all types of water indexes can enhance the water body, but there are obvious differences and different effects. In the black circle in the figure, it can be seen that SWI (Figure (h)) and LSWI (Figure (i)) have poor water-to-background distinction, followed by AWEI (Figure (e)). Compared with AWEI, SWI and LSWI, EWI (Figure (d)), MNDWI (Figure (m)) and NDWI (Figure (l)) have obvious water enhancement and can separate the lake water from the background, but some of the lake area has not been enhanced. TWI (Figure (p)) can effectively separate the water body features from the background and can almost completely enhance the lake area. Figure (b) shows the distribution of small rivers. A comprehensive comparison shows that the water body is not enhanced. Figure 7 From (c), (f), (g), (j), (k), (n) and (o), we can see that EWI (Figure (c)), SWI (Figure (g)), LSWI (Figure (j)), MNDWI (Figure (n)), NDWI (Figure (k)) and AWEI (Figure (f)) cannot effectively enhance small rivers in mountainous areas, while TWI (Figure (o)) can highlight mountainous rivers and enhance mountainous rivers and background.

[0093] Error description:

[0094] Tibet and Qinghai Province were selected as the verification test areas. Since Google Earth images have the advantage of sub-meter resolution, Google Earth images were used as reference images to verify the extraction accuracy. A certain number of sampling points were randomly generated on the Google Earth images. The water body distribution obtained by Sentinel-2MSI images based on 7 indexes was evaluated for accuracy. A confusion matrix was established, and quantitative evaluation indicators such as the error and omission rate, overall accuracy, and Kappa coefficient were calculated, as shown in Tables 2 and 3:

[0095] Table 2 Water body index evaluation indicators

[0096]

[0097] Table 3 Water body index evaluation indicators

[0098]

[0099]

[0100] It can be seen from Tables 2 and 3 that, taking Tibet and Qinghai as the study areas, compared with several other water body indices, the overall accuracy and Kappa coefficient of TWI water body identification results are better, with an overall accuracy of 98%, and low false identification and missed identification rates of 2.5% and 2.3% respectively, which has a good effect on the extraction and identification of water bodies in mountainous areas.

[0101] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for extracting water bodies in mountainous areas based on Sentinel-2 images. It is characterized in that The steps include: Step 1: Select a variety of water samples of different sizes and properties on the Sentinel-2 MSI image, statistically analyze the top atmospheric reflectance values ​​of the various water samples, and calculate the top atmospheric reflectance value of each water sample in different image bands; the image bands include short-wave infrared band 1 B 11 , shortwave infrared band 2 B 12 , Blue light band B 2 、Green band B 3 、Red light band B 4 , vegetation red edge band 1 Band5 B 5 , vegetation red edge band 2 B 6 , vegetation red edge band 3 B 7 , near infrared band B 8 , Water vapor band B 9 ; Step 2: Based on the top atmosphere reflectance values ​​of each water body sample in different image bands, the top atmosphere reflectance values ​​are divided into two distribution trends with a straight line at an angle of 45° to the horizontal axis of the rectangular coordinate system as the boundary. The water bodies on the Sentinel-2MSI image are divided into two types of water bodies according to the two distribution trends; then the image bands reflecting the two types of water bodies and the top atmosphere reflectance values ​​corresponding to the image bands are selected to represent them in a new rectangular coordinate system, and different triangles are constructed in the new rectangular coordinate system; The image bands contained in the two types of water bodies are: Category I water body: B 12 , B 3 , B 4 , B 5 , B 6 ; Category II water body: B 12 , B 2 , B 3 , B 9 ; Step 3: According to the characteristics of the triangles constructed by the two types of water bodies, the initial water body index calculation method is selected, and the water body indexes of the two types of water bodies are calculated respectively by using the triangle similarity theorem, and then the water body indexes of the two types of water bodies are summed up to obtain the water body index for the Sentinel-2MSI image, that is, the triangle water body index. The triangle water body index is used to calculate the Sentinel-2MSI image to obtain an enhanced water body information distribution map, and then the optimal threshold is obtained through the histogram to extract the water body, and the final water body information distribution map is obtained; The calculation formula for the initial water index selected when calculating the water index for the Class I water body is as follows: TWI1=(B 3 -B 12 ) / (B 3 +B 12 ); The initial water index calculation formula selected when calculating the water index of Class II water bodies is as follows: TWI2=(B 9 -B 8 ) / (B 8 +B 9 )。 2. The method for extracting water bodies in mountainous areas based on Sentinel-2 images as claimed in claim 1, It is characterized in that In step 2, the method of constructing different triangles is: in the new rectangular coordinate system, the average value of the maximum atmospheric top reflectance value and the minimum atmospheric top reflectance value corresponding to the image band of the water body sample is used as the ordinate, and the image band is used as the abscissa.

3. A method for extracting water bodies in mountainous areas based on Sentinel-2 images as claimed in claim 1 or 2, It is characterized in that The method is used for extracting lakes, reservoirs, rivers or urban surface water.