Wetland classification method based on mutual collaboration of multi-source remote sensing data of GEE platform

Abstract

The invention discloses a wetland classification method based on mutual cooperation of multi-source remote sensing data of a GEE platform. The wetland classification method comprises the following steps: acquiring Landsat 8OLI data, Sentinel-2 data, Sentinel-1 data, SRTM-DEM elevation data, land coverage data and wetland map data of a research area based on the GEE platform; the method comprises the following steps: extracting a plurality of characteristic variables based on Landsat 8OLI data, Sentinel-2 data, Sentinel-1 data and SRTM-DEM elevation data, and obtaining wetland sample data by combining automatic sampling and visual interpretation based on land coverage data and wetland map data; and carrying out feature combination on the plurality of feature variables, and carrying out classification by utilizing different feature combination schemes, combining wetland sample data and based on a random forest classification algorithm of a GEE platform to obtain wetland classification results of different feature combination schemes. According to the method, the classification precision and efficiency are improved by utilizing mutual collaboration of various data and combining various characteristic variables.

Description

Technical Field

[0001] The present invention relates to the technical field of wetland ecological remote sensing classification, and in particular to a wetland classification method based on the mutual collaboration of multi-source remote sensing data on a GEE platform. Background Art

[0002] Wetland ecosystems are one of the most important ecosystems on Earth. They form a transitional zone between terrestrial and aquatic ecosystems and are known as the "Kidneys of the Earth." Wetlands play a key hydrological role in flood control and drought relief, groundwater recharge, and coastline protection. However, in recent years, the global area of wetlands has been rapidly decreasing due to climate change and human activities. Frequent extreme climate events have impacted the ecological stability of wetlands, and rising sea levels have led to the accelerated degradation and even disappearance of coastal wetlands. Wetland ecosystems are facing severe challenges, and the protection and restoration of wetlands has become a key issue in global environmental governance. Therefore, obtaining timely and accurate wetland monitoring information is crucial for protecting biodiversity, rationally utilizing water resources, and supporting sustainable development goals.

[0003] Wetland mapping is mainly carried out through two methods: field measurement and remote sensing. Field measurement collects wetland vegetation, soil and hydrological data through ground surveys, providing a basis for wetland classification and supporting the verification and correction of remote sensing data. Remote sensing methods use satellite imagery, drone imagery, radar or lidar data, combined with a series of spectral features, to identify and classify wetlands. With the development of these methods, more and more wetland mapping datasets have been established to support wetland monitoring and research. To date, many wetland mapping datasets are available, such as single wetland type datasets (such as salt marshes, mangroves, tidal flats, etc.), multiple wetland type datasets, global non-floodplain wetland datasets, China Coastal Wetland Map, etc. However, it is challenging to perform fine wetland classification in large areas such as countries or even continents. It is necessary to develop an effective, robust and accurate wetland classification algorithm. Summary of the Invention

[0004] The purpose of the present invention is to overcome the shortcomings and deficiencies of the existing technology and provide a wetland classification method based on the mutual coordination of multi-source remote sensing data on the GEE platform. Based on the GEE platform, the fusion of Landsat 8OLI data and Sentinel-2 data and the mutual coordination of Sentinel-1 data are utilized to construct a characteristic variable combination scheme to achieve high-precision classification of large-scale wetlands.

[0005] To achieve the above objectives, the present invention provides a technical solution: a wetland classification method based on the mutual collaboration of multi-source remote sensing data on the GEE platform, comprising the following steps:

[0006] 1) Based on the GEE platform, Landsat 8OLI data, Sentinel-2 data, Sentinel-1 data, SRTM-DEM elevation data, land cover data, and wetland map data of the study area were obtained;

[0007] 2) Extract multiple characteristic variables based on Landsat 8OLI data, Sentinel-2 data, Sentinel-1 data, and SRTM-DEM elevation data, including spectral characteristics, spectral index characteristics, texture characteristics, and terrain characteristics. Wetland sample data is obtained based on land cover data and wetland map data combined with automatic sampling and visual interpretation;

[0008] 3) Feature combinations were performed on multiple feature variables. Different feature combination schemes were used, combined with wetland sample data, and classification was performed based on the random forest classification algorithm used for wetland classification in the GEE platform to obtain wetland classification results for different feature combination schemes.

[0009] Further, the step 1) includes the following steps:

[0010] Image data covering the study area and with an appropriate time span were selected from the acquired Landsat 8OLI and Sentinel-2 data. Cloud masking was performed on the Landsat 8OLI and Sentinel-2 data using the Fmask and S2Cloudless algorithms, respectively, in the GEE platform. The cloud-masked Landsat 8OLI and Sentinel-2 data were resampled to a 10-meter resolution and projected to the same WGS84 coordinate system.

[0011] The Lee filter in GEE was used to denoise the Sentinel-1 data.

[0012] Further, the step 2) includes the following steps:

[0013] Extract spectral features and spectral index features from Landsat 8OLI data and Sentinel-2 data, extract VV polarization features and VH polarization features from Sentinel-1 data, and calculate the ratio of VV polarization features to VH polarization features and texture features;

[0014] The specific steps for fusing Landsat 8OLI data and Sentinel-2 data are as follows: establish a band correspondence table between Landsat 8OLI data and Sentinel-2 data, including the correspondence between the B2 blue band of Landsat 8OLI data and the B2 band of Sentinel-2 data, the B3 green band of Landsat 8OLI data and the B3 band of Sentinel-2 data, the B5 near-infrared band of Landsat 8OLI data and the B8 band of Sentinel-2 data, and the B7 near-infrared band of Landsat 8OLI data and the B8 band of Sentinel-2 data. The B6 and B7 shortwave infrared bands of the Landsat 8OLI data correspond to the B11 and B12 bands of the Sentinel-2 data. The bands of the Landsat 8OLI data and Sentinel-2 data are unified into common names: "Blue", "Green", "Red", "NIR", and "SWIR" through the renaming function to eliminate the differences between the Landsat 8OLI data and Sentinel-2 data. The "Blue", "Green", "Red", "NIR", and "SWIR" bands are normalized to ensure that the numerical ranges are consistent. The spectral index characteristics: NDVI, NDWI, MNDWI, and EVI are calculated based on the normalized bands.

[0015] Texture features are calculated using Sentinel-1 data. GLCM is a two-dimensional matrix where each element represents the number of times two pixels with a spatial relationship appear in the image. Its expression is as follows:

[0016] P(i,j,δ,θ)={[(x,y),(x+dx,y+dy)]|f(x,y)=i,f(x+dx,y+dy)=j}

[0017] Where x is the horizontal coordinate of the image, i.e., the column index; y is the vertical coordinate of the image, i.e., the row index; (x, y) is the position coordinate of a pixel in the image; f(x, y) = i indicates that the grayscale value at (x, y) is i; (x+dx, y+dy) is the deviation point; f(x+dx, y+dy) = j indicates that the grayscale value of (x+dx, y+dy) after deviation is j; i and j are the grayscale values corresponding to (x, y) and (x+dx, y+dy), δ and θ represent the distance and direction between pixels, respectively; P(i, j, δ, θ) is the GLCM matrix; a series of statistics, i.e., texture features, can be calculated through the GLCM matrix. In the GEE platform, VV polarization features and VH polarization features are used as input, and the ee.Image.glcmTexture() function is used to calculate the GLCM matrix. In the process of extracting texture features, the window size and direction parameters must also be set;

[0018] Terrain feature calculation: In the GEE platform, SRTM-DEM elevation data is used to calculate terrain features, including slope and aspect, through the ee.terrain.slope() and ee.Terrain.aspect() terrain calculation functions. The calculation formula is as follows:

[0019]

[0020] In the formula, Z represents the elevation, X and Y represent the horizontal coordinate axes, and π is the circumference of a circle. Represents the rate of change of surface elevation along the east-west direction of the X-axis, Represents the rate of change of surface elevation along the north-south direction of the Y axis;

[0021] Based on land cover data and wetland map data, wetland sample data containing different wetland types are generated. Specifically, water sample data are generated by combining global hydrological data (GSW), mangrove and tidal flat sample data are generated by combining global mangrove data (GMW) and global intertidal zone data, and forest and other types of random sample data are generated by combining GLC_FCS30D data and ESA10 data to enrich wetland sample data. Finally, Google Earth high-definition images are used to conduct detailed visual interpretation and correction of the wetland sample data to obtain the final wetland sample data.

[0022] Furthermore, in step 3), feature combination is performed based on the obtained characteristic variables, and the random forest classification algorithm for wetland classification is selected in the GEE platform and relevant parameters are determined. Different feature combination schemes and wetland sample data are used for classification to obtain wetland classification results;

[0023] Four feature combination schemes were designed based on characteristic variables to compare their impact on wetland classification accuracy. The spectral features and spectral index features in each scheme were obtained by matching and mapping the original bands of Landsat 8OLI data and Sentinel-2 data, and then normalizing the bands and calculating the spectral index. The texture features were calculated based on Sentinel-1 data. The terrain features included slope and aspect calculated from SRTM-DEM elevation data. The four feature combination schemes include:

[0024] Solution 1: spectral features + spectral index features;

[0025] Option 2: spectral characteristics + spectral index characteristics + terrain characteristics;

[0026] Solution 3: spectral features + spectral index features + texture features;

[0027] Option 4: spectral features + spectral index features + terrain features + texture features.

[0028] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0029] 1. The present invention utilizes the fusion of Landsat 8OLI data and Sentinel-2 data and the synergy of Sentinel-1 data to construct a feature combination scheme with greater discriminative capabilities. Landsat 8OLI data has good temporal continuity and spectral stability, while Sentinel-2 data has higher spatial resolution and richer spectral bands, making it suitable for fine-grained identification of land feature types. Sentinel-1 data can stably obtain information on surface moisture status under conditions of cloud cover or low light, making it particularly suitable for tropical and subtropical wetland scenes. With the synergistic effect of Landsat 8OLI data, Sentinel-2 data, and Sentinel-1 data, the complementary advantages of various data types can be fully utilized, effectively improving the accuracy of wetland boundary extraction and wetland type identification. This data fusion strategy significantly improves the classification robustness under conditions of high cloud cover and complex terrain, providing reliable data support for large-scale wetland ecological mapping.

[0030] 2. Wetland classification tasks are often affected by the spectral similarity, spatial heterogeneity, and seasonal dynamics of land features. A single feature cannot fully describe the characteristics of the target land feature. This invention enhances the distinguishing ability of wetland classification algorithms from different dimensions by constructing a multi-level, multi-dimensional feature combination system that integrates spectral, topographic, and spatial information. Feature combination can improve the recognition ability of various types of land features in complex wetland systems, such as water bodies, tidal flats, mangroves, herbaceous swamps, woody swamps, etc., providing a more comprehensive and high-precision classification scheme for wetland classification.

[0031] 3. The present invention relies on the GEE platform and combines the random forest classification algorithm to construct a classification method suitable for large-scale wetlands. This method utilizes the global multi-source remote sensing data set and distributed computing resources built into the GEE platform. It does not require local data downloads and high-performance computing equipment to achieve full-process automation from image preprocessing, feature extraction, model training to classification result generation. Compared with traditional classification methods that rely on local computing environments, the present invention greatly improves the classification processing efficiency and spatial coverage capabilities, and supports batch processing of long-term sequence data and cross-regional wetland information extraction. While ensuring classification accuracy, it significantly reduces computing costs and time overhead, has good visualization capabilities and result sharing mechanisms, and is suitable for wetland resource surveys and dynamic monitoring tasks at regional, national and even intercontinental scales, and has broad prospects for promotion and application. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 It is a schematic diagram of the wetland classification process provided by the present invention.

[0033] Figure 2 This is an overview map of the study area provided by the present invention.

[0034] Figure 3 This is the spatial distribution map of wetland sample points provided by the present invention.

[0035] Figure 4 This is the wetland classification result diagram of solution 1 provided by the present invention.

[0036] Figure 5 This is the wetland classification result diagram of Scheme 2 provided by the present invention.

[0037] Figure 6 This is the wetland classification result diagram of Scheme 3 provided by the present invention.

[0038] Figure 7 This is the wetland classification result diagram of Scheme 4 provided by the present invention. DETAILED DESCRIPTION

[0039] The present invention will be further described below with reference to specific embodiments and drawings, but the embodiments of the present invention are not limited thereto.

[0040] like Figures 1 to 7 As shown, this embodiment discloses a wetland classification method based on the mutual coordination of multi-source remote sensing data from the GEE platform. Taking Kenya in eastern Africa as the research area, this method achieves high-precision classification of Kenyan wetlands by fusion of Landsat 8OLI data and Sentinel-2 data and mutual coordination of Sentinel-1 data. This wetland classification method extracts spectral features and spectral indices from Landsat 8OLI data and Sentinel-2 data, and texture features and topographic features from Sentinel-1 data, constructs a feature combination, combines it with wetland sample data, and uses a random forest algorithm to perform wetland classification to obtain wetland classification results.

[0041] Figure 1 The wetland classification process of the present invention is presented, and the specific implementation includes the following steps:

[0042] Step 1: Based on the GEE platform, obtain Landsat 8OLI data, Sentinel-2 data, Sentinel-1 data, SRTM-DEM elevation data, land cover data and wetland map data of the study area. The overview of the study area is as follows: Figure 2 shown.

[0043] Image data covering the study area and with an appropriate time span were selected from the acquired Landsat 8OLI and Sentinel-2 data. Cloud masking was performed on the Landsat 8OLI and Sentinel-2 data using the Fmask and S2Cloudless algorithms, respectively, in the GEE platform. The cloud-masked Landsat 8OLI and Sentinel-2 data were resampled to a 10-meter resolution and projected to the same WGS84 coordinate system.

[0044] The Lee filter in GEE was used to denoise the Sentinel-1 data.

[0045] Step 2: Extract multiple characteristic variables based on Landsat 8OLI data, Sentinel-2 data, Sentinel-1 data, and SRTM-DEM elevation data, including spectral characteristics, spectral index characteristics, texture characteristics, and terrain characteristics. Wetland sample data is obtained based on land cover data and wetland map data combined with automatic sampling and visual interpretation. The specific steps are as follows:

[0046] Establish a band correspondence table between Landsat 8OLI data and Sentinel-2 data, including the correspondence between the B2 blue band of Landsat 8OLI data and the B2 band of Sentinel-2 data, the B3 green band of Landsat 8OLI data and the B3 band of Sentinel-2 data, the B5 near-infrared band of Landsat 8OLI data and the B8 band of Sentinel-2 data, and the B6 and B7 short-wave infrared bands of Landsat 8OLI data and the B11 and B12 bands of Sentinel-2 data. Use the renaming function to unify the bands of Landsat 8OLI data and Sentinel-2 data into common names: "Blue", "Green", "Red", "NIR", and "SWIR", eliminating the Landsat 8OLI bands. 8. To account for the differences between OLI data and Sentinel-2 data, the "Blue", "Green", "Red", "NIR", and "SWIR" bands are normalized to ensure that the numerical ranges are consistent.

[0047] Spectral index features are calculated based on the normalized bands: NDVI, NDWI, MNDWI, and EVI.

[0048] Texture features are calculated using Sentinel-1 data. GLCM is a two-dimensional matrix where each element represents the number of times two pixels with a certain spatial relationship appear in the image. Its expression is as follows:

[0049] P(i,j,δ,θ)={[(x,y),(x+dx,y+dy)]|f(x,y)=i,f(x+dx,y+dy=j)}

[0050] Where x is the image's horizontal coordinate, or column index; y is the image's vertical coordinate, or row index; (x, y) is the position coordinate of a pixel in the image; f(x, y) = i indicates that the grayscale value at (x, y) is i; (x + dx, y + dy) is the deviation point; f(x + dx, y + dy) = j indicates that the grayscale value at (x + dx, y + dy) after deviation is j; i and j are the grayscale values corresponding to (x, y) and (x + dx, y + dy), respectively; δ and θ represent the distance and direction between pixels, respectively; and P(i, j, δ, θ) is the GLCM matrix. The GLCM matrix can be used to calculate a series of commonly used statistics, namely texture features. In the GEE platform, VV polarization features and VH polarization features are used as input, and the GLCM matrix is calculated using the ee.Image.glcmTexture() function. The window size and direction parameters must also be set during texture feature extraction.

[0051] Terrain feature calculation: Based on the SRTM-DEM elevation data in the GEE platform, terrain features are calculated using the ee.terrain.slope() and ee.Terrain.aspect() terrain calculation functions, including slope and aspect. The calculation formula is as follows:

[0052]

[0053] In the formula, Z represents the elevation, X and Y represent the horizontal coordinate axes, and π is the circumference of a circle. Represents the rate of change of surface elevation along the east-west direction of the X-axis, Represents the rate of change of surface elevation along the north-south direction of the Y axis.

[0054] Based on land cover data and wetland map data, wetland sample data containing different wetland types are generated, such as Figure 3 As shown in the figure, it specifically includes combining global hydrological data GSW to generate water sample data, combining global mangrove data GMW and global intertidal zone data to generate mangrove and tidal flat sample data, combining GLC_FCS30D data and ESA10 data to generate forest and other types of random sample data to enrich wetland sample data, and finally using Google Earth high-definition images to conduct detailed visual interpretation and correction of the wetland sample data to obtain the final wetland sample data.

[0055] Step 3: Perform feature combination on multiple feature variables, use different feature combination schemes, combine with wetland sample data, select the random forest classification algorithm for wetland classification in the GEE platform and determine the relevant parameters (i.e., determine the wetland random forest classifier parameters), use different feature combination schemes and wetland sample data for classification, and obtain the wetland classification results.

[0056] Specifically, four feature combination schemes were designed based on the characteristic variables to compare their effects on wetland classification accuracy. The spectral features and spectral index features in each scheme were the original band matching and mapping of Landsat 8OLI data and Sentinel-2 data, followed by normalized bands and calculated spectral indices. The texture features were mainly calculated based on Sentinel-1 data. The terrain features included slope and aspect calculated based on SRTM-DEM elevation data. The classification results of the four feature combination schemes are shown in Figure 2. Figure 4 、 5 As shown in , 6, and 7, the four feature combination schemes are designed as follows:

[0057]

[0058] The above embodiments are preferred implementation modes of the present invention, but the implementation modes of the present invention are not limited to the above embodiments. Any other changes, modifications, substitutions, combinations, and simplifications that do not deviate from the spirit and principles of the present invention should be considered as equivalent replacement methods and are included in the scope of protection of the present invention.

Claims

1. A wetland classification method based on the mutual collaboration of multi-source remote sensing data on the GEE platform is characterized by: The following steps are involved: 1) Based on the GEE platform, Landsat 8OLI data, Sentinel-2 data, Sentinel-1 data, SRTM-DEM elevation data, land cover data, and wetland map data of the study area were obtained; 2) Extract multiple characteristic variables based on Landsat 8OLI data, Sentinel-2 data, Sentinel-1 data, and SRTM-DEM elevation data, including spectral characteristics, spectral index characteristics, texture characteristics, and terrain characteristics. Wetland sample data is obtained based on land cover data and wetland map data combined with automatic sampling and visual interpretation; 3) Feature combinations were performed on multiple feature variables. Different feature combination schemes were used, combined with wetland sample data, and classification was performed based on the random forest classification algorithm used for wetland classification in the GEE platform to obtain wetland classification results for different feature combination schemes.

2. The wetland classification method based on the mutual collaboration of multi-source remote sensing data on the GEE platform according to claim 1 is characterized in that: The step 1) comprises the following steps: Image data covering the study area and with an appropriate time span were selected from the acquired Landsat 8OLI and Sentinel-2 data. Cloud masking was performed on the Landsat 8OLI and Sentinel-2 data using the Fmask and S2Cloudless algorithms, respectively, in the GEE platform. The cloud-masked Landsat 8OLI and Sentinel-2 data were resampled to a 10-meter resolution and projected to the same WGS84 coordinate system. The Lee filter in GEE was used to denoise the Sentinel-1 data.

3. The wetland classification method based on the mutual collaboration of multi-source remote sensing data on the GEE platform according to claim 2 is characterized in that: The step 2) comprises the following steps: Extract spectral features and spectral index features from Landsat 8OLI data and Sentinel-2 data, extract VV polarization features and VH polarization features from Sentinel-1 data, and calculate the ratio of VV polarization features to VH polarization features and texture features; The specific steps for fusing Landsat 8OLI data and Sentinel-2 data are as follows: establish a band correspondence table between Landsat 8OLI data and Sentinel-2 data, including the B2 blue band of Landsat 8OLI data corresponding to the B2 band of Sentinel-2 data, the B3 green band of Landsat 8OLI data corresponding to the B3 band of Sentinel-2 data, the B5 near infrared band of Landsat 8OLI data corresponding to the B8 band of Sentinel-2 data, the B6 and B7 short wave infrared bands of Landsat 8OLI data corresponding to the B11 and B12 bands of Sentinel-2 data, and use the rename function to convert the Landsat 8OLI data into Sentinel-2 data. The bands of Landsat 8OLI data and Sentinel-2 data are unified into common names: "Blue", "Green", "Red", "NIR", and "SWIR". The differences between Landsat 8OLI data and Sentinel-2 data are eliminated. The "Blue", "Green", "Red", "NIR", and "SWIR" bands are normalized to ensure the consistency of the numerical range. The spectral index characteristics: NDVI, NDWI, MNDWI, and EVI are calculated based on the normalized bands. Texture features are calculated using Sentinel-1 data. GLCM is a two-dimensional matrix where each element represents the number of times two pixels with a spatial relationship appear in the image. Its expression is as follows: P(i,j,δ,θ)={[(x,y),(x+dx,y+dy)]|f(x,y)=i,f(x+dx,y+dy)=j} Where x is the horizontal coordinate of the image, i.e., the column index; y is the vertical coordinate of the image, i.e., the row index; (x, y) is the position coordinate of a pixel in the image; f(x, y) = i indicates that the grayscale value at (x, y) is i; (x+dx, y+dy) is the deviation point; f(x+dx, y+dy) = j indicates that the grayscale value of (x+dx, y+dy) after deviation is j; i and j are the grayscale values corresponding to (x, y) and (x+dx, y+dy), δ and θ represent the distance and direction between pixels, respectively; P(i, j, δ, θ) is the GLCM matrix; a series of statistics, i.e., texture features, can be calculated through the GLCM matrix. In the GEE platform, VV polarization features and VH polarization features are used as input, and the ee.Image.glcmTexture() function is used to calculate the GLCM matrix. In the process of extracting texture features, the window size and direction parameters must also be set; Terrain feature calculation: In the GEE platform, SRTM-DEM elevation data is used to calculate terrain features, including slope and aspect, through the ee.terrain.slope() and ee.Terrain.aspect() terrain calculation functions. The calculation formula is as follows: In the formula, Z represents the elevation, X and Y represent the horizontal coordinate axes, and π is the circumference of a circle. Represents the rate of change of surface elevation along the east-west direction of the X-axis, Represents the rate of change of surface elevation along the north-south direction of the Y axis; Based on land cover data and wetland map data, wetland sample data containing different wetland types are generated. Specifically, water sample data are generated by combining global hydrological data (GSW), mangrove and tidal flat sample data are generated by combining global mangrove data (GMW) and global intertidal zone data, and forest and other types of random sample data are generated by combining GLC_FCS30D data and ESA10 data to enrich wetland sample data. Finally, Google Earth high-definition images are used to conduct detailed visual interpretation and correction of the wetland sample data to obtain the final wetland sample data.

4. The wetland classification method based on the mutual collaboration of multi-source remote sensing data on the GEE platform according to claim 3 is characterized in that: In step 3), feature combination is performed based on the obtained characteristic variables, and the random forest classification algorithm for wetland classification is selected in the GEE platform and relevant parameters are determined. Different feature combination schemes and wetland sample data are used for classification to obtain the wetland classification results; Four feature combination schemes were designed based on characteristic variables to compare their impact on wetland classification accuracy. The spectral features and spectral index features in each scheme were obtained by matching and mapping the original bands of Landsat 8OLI data and Sentinel-2 data, and then normalizing the bands and calculating the spectral index. The texture features were calculated based on Sentinel-1 data. The terrain features included slope and aspect calculated from SRTM-DEM elevation data. The four feature combination schemes include: Solution 1: spectral features + spectral index features; Option 2: spectral characteristics + spectral index characteristics + terrain characteristics; Solution 3: spectral features + spectral index features + texture features; Option 4: spectral features + spectral index features + terrain features + texture features.

Images