A large-scale remote sensing intelligent extraction method for tidal flats

Abstract

The present invention discloses a method for intelligent extraction of large-scale remote sensing of tidal flats, including: extraction of sample points; extraction of shoreline data; determination of the extraction range of tidal flats based on the shoreline data; construction of a remote sensing feature data set, calculating different features based on remote sensing data and synthesizing them into a new data set; construction of a classification decision tree, using the distribution frequency of sample points in the remote sensing feature data set to determine appropriate segmentation thresholds and construct a decision tree; synthesis of the lowest tide level data, using the proportion of each pixel identified as a tidal flat in three years of remote sensing data, and synthesizing the lowest tide level data using the top 5% and 5%-25% of the images; extraction of the tidal flat spatial range, applying the classification decision tree to the synthesized lowest tide level data, obtaining products and performing data post-processing and statistical mapping. The present invention realizes the extraction and synthesis of large-scale lowest tide level data and tidal flat mapping through remote sensing big data, while overcoming the limitations of existing methods and improving the robustness and application scope of the method.

Description

Technical Field

[0001] The present invention relates to the field of remote sensing big data for feature recognition, and particularly to a method for large-scale remote sensing intelligent extraction of tidal flats. Background Technique

[0002] Coastal ecosystems are one of the most dynamic ecosystems globally, characterized by high spatio-temporal dynamics, intertwined land-sea influences, diverse ecosystems, and large external disturbances. As the most dynamically changing natural geographical unit in the coastal zone, tidal flats are affected by multiple factors such as tidal dynamic coverage, vegetation invasion, and coastal reclamation. Tidal flat wetlands provide ecological service values for coastal protection and bird habitats. Understanding their high spatio-temporal dynamic characteristics and spatial extent provides scientific data support for coastal zone protection, management, restoration, and policy formulation.

[0003] The intertidal zone poses challenges for accurate tidal flat mapping due to diverse terrain conditions, random tidal coverage, and low accessibility. Traditional on-site sampling surveys are difficult to conduct large-scale surveys in tidal flat areas due to factors such as cost and accessibility. Thanks to the development of remote sensing technology and the growth of satellite data, large-scale tidal flat monitoring based on remote sensing big data has been carried out. Existing tidal flat mapping either adopts a land cover classification strategy, which requires a large amount of labor cost for data screening and sample training; or is based on frequency segmentation of big data, obtaining the tidal flat range through percentage threshold cutting, but ignoring data at extremely low tide levels, resulting in data underestimation; or adopts a scheme of constructing spectral indices, but there are still situations where the classification accuracy is not high and it cannot be used on a large scale. Therefore, in the context of the development of remote sensing big data, it is still difficult to achieve large-scale intelligent extraction of tidal flats, accurately obtain the spatial distribution range of tidal flats at the lowest tide level, and at the same time meet the generalization and robustness of the method. Summary of the Invention

[0004] The purpose of the present invention is to solve the deficiencies of current remote sensing technology in large-scale intelligent extraction of tidal flats. By providing a method for large-scale remote sensing intelligent extraction of tidal flats, through the construction of the minimum unit for tidal flat extraction, using a remote sensing feature dataset, a decision tree suitable for large-scale intelligent tidal flat extraction is constructed. Further, in the extraction results of long-term sequence data, the tidal flat areas within the minimum unit are sorted to achieve the synthesis of the lowest tide level data and the extraction of the maximum tidal flat spatial range, overcoming the current situation in traditional schemes such as large data processing volume of classification algorithms, inability of the frequency method to obtain the lowest tidal flat data images, interference from cloud data, and poor spatial generalization ability of specific indices, improving the accuracy and efficiency of remote sensing big data in intelligent extraction of tidal flats.

[0005] The specific technical solution to achieve the purpose of the present invention is as follows:

[0006] A method for large-scale remote sensing intelligent extraction of tidal flats, comprising the following steps:

[0007] Step 1: Sample Point Extraction

[0008] Based on the published literature related to coastal tidal flat mapping, high-resolution low-tide satellite data with a resolution better than one meter, and low-tide unmanned aerial vehicle mapping data, etc., two types of samples, namely tidal flat samples and non-tidal flat samples, are obtained. Each type of sample includes land cover type, coordinate information, and sample acquisition time. The land cover type within a 100-meter range of each sample point should be consistent, and the spatial geometric distance between different points should be greater than 150 meters;

[0009] Step 2: Shoreline Data Extraction

[0010] Through remote sensing data, the shoreline data is manually interactively drawn to separate land pixels and coastal zone pixels; the principle of the drawing is to ensure that one side of the shoreline is land elements including impervious surfaces, farmland, and aquaculture ponds, and the other side is coastal elements including coastal salt marsh vegetation, mangroves, tidal flats, and open waters;

[0011] Step 3: Determination of the Spatial Range for Tidal Flat Extraction

[0012] Based on the shoreline data drawn in Step 2, a 5-kilometer buffer zone is generated on the land side and a 10-kilometer buffer zone is generated on the ocean side, and several 20-kilometer × 20-kilometer grids that are closely connected and completely cover the buffer zones are generated; each grid is regarded as the smallest unit for tidal flat extraction;

[0013] Step 4: Construction of a Remote Sensing Feature Dataset

[0014] According to the buffer zone determined in Step 3, all surface reflectance products of remote sensing images with a cloud cover of less than 50% within three natural years that cover the buffer zone are screened, and a remote sensing feature dataset that matches each surface reflectance product is constructed. Specifically, it includes: (1) Extract the quality control bands of the remote sensing feature dataset, and pixels with cloud recognition results of "high / medium / low confidence" are all excluded, and pixels with cloud recognition results of "not detected" are regarded as available pixels; (2) Extract the data of each original band in the excluded remote sensing feature dataset, and calculate six features: NDVI, mNDWI, Max(R, NIR), DIFF(G, SWIRII), DIFF(NIR, G), DIFF(NIR+R,G+B); (3) Synthesize the original band data and the six calculated features to form a remote sensing feature dataset; the calculation methods of each feature dataset are as follows:

[0015] 1. NDVI (Normalized Difference Vegetation Index):

[0016] NDVI = (NIR - R) / (NIR + R);

[0017] 2. mNDWI (Modified Normalized Difference Water Index):

[0018] mNDWI = (G - SWIR1) / (G + SWIR1);

[0019] 3. Max(R, NIR) (Maximum Value):

[0020] Max(R, NIR) = max(R, NIR);

[0021] 4. DIFF(G, SWIRII) (Difference I):

[0022] DIFF(G, SWIRII) = G - SWIRII;

[0023] 5. DIFF(NIR, G) (Difference II):

[0024] DIFF(NIR, G) = NIR - G;

[0025] 6. DIFF(NIR + R, G + B) (Difference III):

[0026] DIFF(NIR + R, G + B) = (NIR + R) - (G + B);

[0027] where G, R, NIR, SWIRI, and SWIRII are the corresponding values of the green light band, red light band, near-infrared band, first short-wave infrared band, and second short-wave infrared band, respectively;

[0028] Step 5: Construction of Classification Decision Tree

[0029] Based on the two types of samples containing ground object type information and coordinate information obtained in Step 1, in the remote sensing feature dataset constructed in Step 4, obtain the frequency distribution diagrams of the two types of samples in each feature dataset, use the recursive method to determine the threshold, calculate the misclassification rate under different thresholds, and record the misclassification rate with the lowest value as the optimal segmentation threshold; the construction of the classification decision tree for tidal flat pixel recognition includes the following three components:

[0030] 1. Determination of High Tidal Flat Area

[0031] NDVI > T1 ∧ Max(R, NIR) ≥ T4 ∧ DIFF(NIR, G) ≥ 0 ∧ DIFF(NIR + R, G + B) ≥ 0;

[0032] 2. Determination of Medium Tidal Flat Area

[0033] T1 ≥ NDVI > T2 ∧ mNDWI < T5 ∧ Max(R, NIR) ≥ T4 ∧ DIFF(NIR, G) ≥0 ∧ DIFF(NIR + R, G + B) ≥ 0;

[0034] 3. Low-tide flat area determination

[0035] T2 ≥ NDVI > T3 ∧ mNDWI < T6 ∧ Max(R, NIR) ≥ T4 ∧ T7 > DIFF(NIR,G) ≥ T8 ∧ T9 > DIFF(NIR + R, G + B) ≥ 0;

[0036] Where T1 - T9 are the optimal segmentation thresholds of each feature in the remote sensing feature dataset; all pixels with the recognition result of high-tide flat area, medium-tide flat area or low-tide flat area are recorded as tidal flat pixels;

[0037] Step 6: Lowest tide level data synthesis

[0038] Using the classification decision tree obtained in Step 5, extract tidal flat pixels from all remote sensing image products in Step 4 and their matching remote sensing feature datasets to obtain tidal flat pixels and non-tidal flat pixels; within the overlapping area of the minimum extraction unit and buffer of tidal flat extraction in Step 3, calculate the proportion of tidal flat pixels in all pixels, denoted as the proportion of tidal flat pixels;

[0039] Extract the top 5% of remote sensing image products with the highest proportion of tidal flat pixels and their matching remote sensing feature datasets, and calculate the lowest tide level synthesis data through averaging; check whether there are spatial missing data. If there are missing data, extract remote sensing image products with a tidal flat pixel proportion of 5% - 25% and their matching remote sensing feature datasets, and perform averaging to fill in the missing part. The filled data is used as the lowest tide level data synthesis product; if there is no vacancy, it is directly used as the lowest tide level data synthesis product;

[0040] Step 7: Tidal flat spatial range extraction

[0041] Based on the lowest tide level data synthesized in Step 6, use the classification decision tree in Step 5 to identify tidal flat pixels and obtain the tidal flat spatial distribution product;

[0042] Step 8: Data post-processing and statistical mapping

[0043] Set the projection for the tidal flat spatial distribution product obtained in Step 7, merge patches with an area less than 2 hectares into the nearest patch, calculate the total area of the tidal flat and the area of the tidal flat within each minimum extraction unit; use mapping software for symbolization processing to obtain the visualized tidal flat extraction product.

[0044] Compared with the prior art, the present invention has the following advantages:

[0045] (1) With the support of remote sensing big data, the present invention utilizes sufficient time series data to achieve the intelligent extraction of the large-scale tidal flat range.

[0046] (2) By constructing the minimum unit for tidal flat extraction, the present invention extracts the tidal flat at the lowest tide level, avoiding the mapping error caused by the large tidal level difference in a single scene image in the traditional method.

[0047] (3) By constructing a classification decision tree, the present invention improves the accuracy of traditional tidal flat classification, further removes the interference of pixels such as clouds and haze, and synthesizes the lowest tide level data based on the sorting thinking, avoiding the phenomenon that the extremely low tide level is missed in the traditional method using the frequency idea. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 is a flow chart of the present invention;

[0049] Figure 2 is a schematic diagram of the time series Landsat 8 original image dataset in the embodiment of the present invention;

[0050] Figure 3 is a schematic diagram of the proportion of pixels classified as tidal flats by the classification decision tree in all images of three years in the embodiment of the present invention;

[0051] Figure 4 is a schematic diagram of the lowest tide level data of the Yangtze River Estuary region synthesized in the embodiment of the present invention;

[0052] Figure 5 is a schematic diagram of the extraction result of some minimum extraction units in the embodiment of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0053] In order to clearly express the technical solution, purpose and advantages of the present invention, the following will explain the present invention in combination with the drawings in the embodiments of the present invention, so as to enable an ordinary practitioner in this field to implement the present invention. The following embodiments are examples of the present invention and are only used for explanation. Obviously, without departing from the principle of the present invention, any non-creative modification, improvement and equivalent solutions, as well as other technical solutions that do not deviate from the scope of the present invention, should be included in the protection scope of the present invention.

[0054] Embodiment: Refer to Figure 1In this embodiment, a large-scale remote sensing intelligent extraction method for tidal flats is constructed by constructing the smallest unit for tidal flat extraction and using the remote sensing feature data set to construct a decision tree suitable for large-scale intelligent tidal flat extraction. Furthermore, in the extraction results of long time series data, the tidal flat area in the smallest unit is sorted to achieve the synthesis of the lowest tide data and the extraction of the largest tidal flat spatial range. The main steps are as follows:

[0055] Step 1: Sample point extraction, including: obtaining the point information of tidal flat samples and non-tidal flat samples based on existing literature containing coastal tidal flats or vegetation maps, high-resolution low-tide satellite data better than one meter, low-tide drone mapping data, etc. The point should ensure that the type of land features within 100 meters is consistent, and the spatial geometric distance between different points should be greater than 150 meters.

[0056] Step 2: Extract shoreline data, including: drawing shoreline data manually and interactively through remote sensing data, separating land pixels and coastal pixels. The drawing principle is to ensure that one side of the shoreline is land elements (such as impervious surfaces, farmland, aquaculture ponds, etc.), and the other side is coastal elements (such as coastal salt marsh vegetation, mangroves, tidal flats, open waters, etc.).

[0057] Step 3: Determine the spatial scope of tidal flat extraction, including: generating a buffer zone of 5 km on the land side and 10 km on the ocean side through the shoreline data drawn in step 2, and generating several 20 km × 20 km grids that are closely connected and completely cover the above buffer zones. Each grid is regarded as the smallest unit for tidal flat extraction. In this embodiment, the spatial scope of tidal flat extraction is set in the Yangtze River Estuary area, starting from Nantong City, Jiangsu Province in the north and Jiaxing City, Zhejiang Province in the south.

[0058] Step 4: Construct a remote sensing feature dataset, including: based on the buffer zone determined in step 3, screen all remote sensing image surface reflectance products with cloud cover less than 50% in the three natural years covering the buffer zone, and construct a matching remote sensing feature dataset for each surface reflectance product. The construction method is as follows: (1) extract the quality control band of the remote sensing feature dataset, and remove the three types of pixels with high, medium and low cloud identification results; (2) extract the data of each original band from the removed remote sensing feature dataset, and calculate a total of six features: NDVI, mNDWI, Max(R, NIR), DIFF(G, SWIRII), DIFF(NIR, G), and DIFF(NIR+R,G+B); (3) synthesize the original band data and the calculated six features to form a remote sensing feature dataset. The calculation method of each feature dataset is as follows:

[0059] 1. NDVI (Normalized Difference Vegetation Index):

[0060] NDVI = (NIR - R) / (NIR + R);

[0061] 2. mNDWI (Modified Normalized Difference Water Index):

[0062] mNDWI = (G - SWIR1) / (G + SWIR1);

[0063] 3. Max(R, NIR) (Maximum Value):

[0064] Max(R, NIR) = max(R, NIR);

[0065] 4. DIFF(G, SWIRII) (Difference One):

[0066] DIFF(G, SWIRII) = G – SWIRII;

[0067] 5. DIFF(NIR, G) (Difference Two):

[0068] DIFF(NIR, G) = NIR - G;

[0069] 6. DIFF(NIR + R, G + B) (Difference Three):

[0070] DIFF(NIR + R, G + B) = (NIR + R) - (G + B);

[0071] Where G, R, NIR, SWIRI, and SWIRII are the corresponding values of the green light band, red light band, near-infrared band, shortwave infrared first band, and shortwave infrared second band respectively. In this embodiment, the time is set from 2019 to 2021, and the selected image is Landsat 8 OLI data. The schematic diagram of the original image dataset is as Figure 2 . Where G, R, NIR, SWIRI, and SWIRII respectively correspond to the 3rd band (wavelength range: 0.525–0.600 µm), 4th band (wavelength range: 0.630–0.680 µm), 5th band (wavelength range: 0.845–0.885 µm), 6th band (wavelength range: 1.560–1.660 µm), and 7th band (wavelength range: 2.100–2.300 µm) of Landsat 8 OLI data.

[0072] Step 5: Construction of classification decision tree, including: based on the tidal flat samples and non-tidal flat samples obtained in Step 1, in the remote sensing feature dataset constructed in Step 4, obtain the frequency distribution maps of the two types of samples in each feature dataset, use the recursive method to determine the threshold, calculate the misclassification rate at different thresholds, and record the one with the lowest misclassification rate as the optimal segmentation threshold. The construction of the classification decision tree for tidal flat pixel recognition includes three components:

[0073] 1. Determination of high-tide tidal flat area

[0074] NDVI ≥ T1 ∧ Max(R, NIR) ≥ T4 ∧ DIFF(NIR, G) ≥ 0 ∧ DIFF(NIR + R,G + B) ≥ 0;

[0075] 2. Determination of medium-tide tidal flat area

[0076] T1 ≥ NDVI > T2 ∧ mNDWI < T5 ∧ Max(R, NIR) ≥ T4 ∧ DIFF(NIR, G) ≥0 ∧ DIFF(NIR + R, G + B) ≥ 0;

[0077] 3. Determination of low-tide tidal flat area

[0078] T2 ≥ NDVI > T3 ∧ mNDWI < T6 ∧ Max(R, NIR) ≥ T4 ∧ T7 > DIFF(NIR,G) ≥ T8 ∧ T9 > DIFF(NIR + R, G + B) ≥ 0;

[0079] where T1 - T9 are the optimal segmentation thresholds of each feature in the remote sensing feature dataset. Denote all the pixels in the high-tide tidal flat area, medium-tide tidal flat area and low-tide tidal flat area as tidal flat pixels.

[0080] In this embodiment, the T1 - T9 thresholds are set to 0.1, 0.05, -0.2, 0.08, 0.5, 0.8, 0.03, -0.03, 0.1 according to the distribution of the two types of samples in Step 1 in each feature dataset of Landsat 8 data in Step 4.

[0081] Step 6: Synthesis of lowest tide level data, including: using the classification decision tree obtained in Step 5, in all the remote sensing image products in Step 4 and their matching remote sensing feature datasets, extract tidal flat pixels to obtain tidal flat pixels and non-tidal flat pixels. Calculate the proportion of tidal flat pixels in all pixels within the overlapping area of the minimum unit and buffer area for tidal flat extraction in Step 3, and denote it as the proportion of tidal flat pixels such as Figure 3 . Figure 3The pixel extraction result of the classification decision tree for this embodiment is the pixel of the tidal flat. Among all remote sensing images in three years, the proportion of pixels judged as tidal flat pixels is represented by a black-and-white transitional color, where the proportion ranges from 0 - 1. That is, white indicates that the proportion of this pixel judged as a tidal flat in all remote sensing images in three years is 0%, and black indicates that the proportion of this pixel judged as a tidal flat in all remote sensing images in three years is 100%.

[0082] Extract the top 5% of remote sensing image products with the highest proportion of tidal flat pixels and their corresponding remote sensing feature datasets, and calculate the lowest tide level synthetic data through averaging. Check whether there are spatial missing values in the data. If there are missing values, extract the remote sensing image products with a tidal flat pixel proportion of 5% - 25% and their corresponding remote sensing feature datasets, and perform averaging calculation to fill the missing part. The data after filling is used as the lowest tide level data synthesis product; if there are no vacancies, it is directly used as the lowest tide level data synthesis product. In this embodiment, the synthesized lowest tide level data for the Yangtze Estuary area is as Figure 4 . Figure 4 The left figure, Image A, is a schematic diagram of the top 5% of remote sensing image products with the highest proportion of tidal flat pixels. The white part is a schematic of the data missing pixels, which are filled by the remote sensing image products with a tidal flat pixel proportion of 5% - 25%, that is Figure 4 The middle picture, Image B, is the top 5% of the above-mentioned remote sensing image products with the highest proportion of tidal flat pixels and the lowest tide level data for the Yangtze Estuary area jointly synthesized by the remote sensing image products with a tidal flat pixel proportion of 5% - 25%, that is Figure 4 The right figure, Max Tidal Flats Image.

[0083] Step 7: Extract the tidal flat spatial range. Based on the lowest tide level data synthesized in Step 6, use the classification decision tree in Step 5 to identify the tidal flat pixels and obtain the tidal flat spatial distribution product. The schematic diagram of the extraction result of some minimum units is as Figure 5 . Figure 5 Successively display several groups of grayscale diagrams of remote sensing images cut into minimum units and the corresponding binary diagrams of the tidal flat spatial distribution products. Among them, the grayscale diagram is the real remote sensing image, and the white part of the binary diagram is the area identified as tidal flat pixels after recognition, and the black part is the area identified as non-tidal flat pixels after recognition.

[0084] Step 8: Data post-processing and statistical mapping. Set a suitable projection for the tidal flat spatial distribution product obtained in Step 7, merge the patches with an area less than 2 hectares into the nearest patch, and calculate the total area of the tidal flat and the area of the tidal flat within each minimum extraction unit. Use mapping software for symbolization processing to obtain a visualized tidal flat extraction product.

[0085] Based on the above steps, the present invention can utilize remote sensing big data, adopt the concept of cutting the smallest unit of tidal flats, combine the classification decision tree and the land pixel sorting method to obtain the lowest tide level data, and realize the intelligent extraction of the tidal flat area with the largest range, overcoming the limitations of the existing methods such as low classification accuracy, omission of extremely low tide levels, and poor generalization performance, and forming a set of efficient, simple, reproducible, and robust remote sensing tidal flat intelligent extraction solutions.

[0086] As described above, it is only a specific implementation manner of the present invention. Any variations, improvements, equivalent solutions obtained without creative labor, and other technical solutions that do not deviate from the scope of the present invention shall fall within the protection scope of the present invention.

Claims

1. A large-scale remote sensing intelligent extraction method for tidal flats, characterized in that It includes the following steps: Step 1: Sample point extraction Based on the published literature on coastal tidal flat mapping, high-resolution low-tide satellite data with a resolution better than one meter, and low-tide unmanned aerial vehicle mapping data, two types of samples, namely tidal flat samples and non-tidal flat samples, are obtained. Each type of sample includes land cover type, coordinate information, and sample acquisition time. The land cover type within 100 meters of each sample point is ensured to be consistent, and the spatial geometric distance between different points is at least 150 meters; Step 2: Shoreline data extraction Through remote sensing data, the shoreline data is manually interactively drawn to separate land pixels and coastal zone pixels; the drawing ensures that on one side of the shoreline are land elements including impervious surfaces, farmland, and aquaculture ponds, and on the other side are coastal elements including coastal salt marsh vegetation, mangroves, tidal flats, and open water; Step 3: Determination of the spatial range for tidal flat extraction Based on the shoreline data drawn in Step 2, a 5-kilometer buffer zone is generated on the land side and a 10-kilometer buffer zone is generated on the ocean side, and several 20-kilometer × 20-kilometer grids that are closely connected and completely cover the buffer zones are generated; each grid is regarded as the smallest unit for tidal flat extraction; Step 4: Construction of a remote sensing feature dataset According to the buffer zones determined in Step 3, surface reflectance products of remote sensing images with a cloud cover of less than 50% within three natural years that cover the buffer zones are screened, and a matching remote sensing feature dataset is constructed for each scene of surface reflectance product. Specifically, it includes: (1) Extract the quality control bands of the remote sensing feature dataset, and pixels with high-confidence, medium-confidence, or low-confidence cloud identification results are all excluded, and pixels with an undetected cloud identification result are regarded as available pixels; (2) Extract the data of each original band in the excluded remote sensing feature dataset, and calculate six features: NDVI, mNDWI, Max(R, NIR), DIFF(G, SWIRII), DIFF(NIR, G), DIFF(NIR + R, G + B); (3) Synthesize the original band data and the six calculated features to form a remote sensing feature dataset; the calculation methods of each feature dataset are as follows:

1. Normalized Difference Vegetation Index NDVI: NDVI = (NIR - R) / (NIR + R); 2. Modified Normalized Difference Water Index mNDWI: mNDWI = (G - SWIRI) / (G + SWIRI); 3. Maximum value Max(R, NIR): Max(R, NIR) = max(R, NIR); 4. Difference one DIFF(G, SWIRII): DIFF(G, SWIRII) = G - SWIRII; 5. Difference two DIFF(NIR, G): DIFF(NIR, G) = NIR - G; 6. Difference three DIFF(NIR + R, G + B): DIFF(NIR + R, G + B) = (NIR + R) - (G + B); Where G, R, NIR, SWIRI, and SWIRII are the values corresponding to the green light band, red light band, near-infrared band, first short-wave infrared band, and second short-wave infrared band, respectively; Step 5: Construction of classification decision tree Based on the two types of samples obtained in Step 1, in the remote sensing feature dataset constructed in Step 4, obtain the frequency distribution diagrams of the two types of samples in each feature dataset. Use the recursive method to determine the threshold, calculate the misclassification rate under different thresholds, and record the threshold with the lowest misclassification rate as the optimal segmentation threshold. The construction of the classification decision tree for tidal flat pixel recognition includes the following three components:

1. Determination of high tidal flat area NDVI > T1 ∧ Max(R, NIR) ≥ T4 ∧ DIFF(NIR, G) ≥ 0 ∧ DIFF(NIR + R, G +B) ≥ 0; 2. Determination of medium tidal flat area T1 ≥ NDVI > T2 ∧ mNDWI < T5 ∧ Max(R, NIR) ≥ T4 ∧ DIFF(NIR, G) ≥ 0∧ DIFF(NIR + R, G + B) ≥ 0; 3. Determination of low tidal flat area T2 ≥ NDVI > T3 ∧ mNDWI < T6 ∧ Max(R, NIR) ≥ T4 ∧ T7 > DIFF(NIR, G)≥ T8 ∧ T9 > DIFF(NIR + R, G + B) ≥ 0; Where T1 - T9 are the optimal segmentation thresholds of each feature in the remote sensing feature dataset; all pixels identified as high tidal flat area or medium tidal flat area or low tidal flat area are recorded as tidal flat pixels; Step 6: Synthesis of lowest tide level data Using the classification decision tree obtained in Step 5, extract tidal flat pixels in all remote sensing image products in Step 4 and their corresponding remote sensing feature datasets to obtain tidal flat pixels and non-tidal flat pixels; within the overlapping area of the minimum unit and buffer for tidal flat extraction in Step 3, calculate the proportion of tidal flat pixels in all pixels, denoted as the proportion of tidal flat pixels; Extract the top 5% of the remote sensing image products with the highest proportion of tidal flat pixels and their corresponding remote sensing feature datasets, and calculate the lowest tide level synthesis data through averaging; check whether there are spatial missing values in the data. If there are missing values, extract the remote sensing image products with a tidal flat pixel proportion of 5% - 25% and their corresponding remote sensing feature datasets, and perform averaging to fill the missing part. The filled data is used as the lowest tide level data synthesis product; if there are no missing values, it is directly used as the lowest tide level data synthesis product; Step 7: Extraction of tidal flat spatial range Based on the lowest tide level data synthesized in Step 6, use the classification decision tree in Step 5 to identify tidal flat pixels and obtain the tidal flat spatial distribution product; Step 8: Data post-processing and statistical mapping Set the projection for the intertidal flat spatial distribution product obtained in step 7, merge the patches with an area less than 2 hectares into the nearest patch, calculate the total area of the intertidal flat and the area of the intertidal flat within each minimum extraction unit; perform symbolization using mapping software to obtain a visualized intertidal flat extraction product.

Images