A large-scale remote sensing automatic extraction method for coastal tidal flat

Abstract

The application discloses a kind of large-scale remote sensing automation extraction methods of coastal tidal flat, belong to remote sensing tidal flat mapping technical field, comprising: based on cloud-removing time sequence remote sensing image per-pixel calculation turbidity water index and select contrast band accordingly, and adaptive mixed water index is calculated in combination with green light band;For each image determined by adaptive mixed water index, calculate its value range histogram, the adaptive mixed water index value corresponding to the derivative abrupt point in graph is regarded as adaptive water-land segmentation threshold;Improved normalized difference water index is calculated per scene, based on improved normalized difference water index and adaptive water-land segmentation threshold, the maximum water surface range is extracted, and based on each image and adaptive water-land segmentation threshold, tidal flat exposure frequency image is extracted, further extract preliminary tidal flat spatial distribution;After connectivity processing, tidal flat spatial distribution visualization result is generated.The application can realize the fast, accurate and stable extraction of tidal flat spatial distribution in large-scale and complex environment.

Description

Technical Field

[0001] This invention belongs to the field of remote sensing tidal flat mapping technology, specifically relating to a method for large-scale remote sensing automated extraction of coastal tidal flats. Background Technology

[0002] Tidal flats are areas periodically submerged by tides, submerged at high tide and exposed at low tide. They include mudflats, sandy beaches, and rocky beaches, and are an important component of the coastal zone. As biodiversity hotspots, tidal flats provide critical habitats for coastal organisms and play a vital role in windbreak and shoreline stabilization, resource supply, water purification, and carbon sequestration. However, as one of the world's most vulnerable ecosystems, tidal flats face the dual threats of natural factors (such as sea-level rise and climate change) and human activities (such as industrialization and coastal development). The continuous shrinkage of tidal flats has triggered a chain of environmental problems, including ecological degradation and biodiversity loss. Therefore, accurate monitoring of tidal flat extent is crucial for the protection and restoration of tidal flat wetlands and the sustainable development of the coastal zone.

[0003] Tidal flats are characterized by their wide range, rapid changes, and periodic inundation, making large-scale traditional field surveys difficult. Remote sensing technology, with its wide coverage, low cost, and real-time capabilities, has become an effective means of tidal flat monitoring. Currently, common tidal flat mapping schemes mainly include the following categories: methods based on tidal models, which are limited by the spatiotemporal resolution of satellite imagery and the differences in tidal levels across different regions, resulting in uncertainties in the observation results; classification methods based on machine learning, while offering high classification accuracy, rely on a large number of training samples, leading to high costs; and threshold-based methods, which are simple to implement and efficient, typically employing empirical thresholds or Otsu adaptive thresholds.

[0004] In recent years, research trends have gradually evolved towards multi-source remote sensing data fusion and automated threshold optimization to improve the robustness and accuracy of tidal flat identification. For example, patent document CN119048852A provides an automatic classification method and device for coastal ecosystems based on multi-source remote sensing fusion. By integrating Landsat and MODIS images, using the NSPI algorithm for cloud filtering, and introducing a regression-based spatiotemporal fusion model, it reconstructs fused images with low cloud cover and continuous time. Then, by combining spectral index curves, it achieves automated differentiation between tidal flats, mangroves, and salt marshes. Patent document CN116152655A provides a coastal wetland classification method, device, equipment, and storage medium. By synthesizing multi-temporal images of high tide, low tide, summer, and winter, it uses the enhanced mangrove vegetation index and the OTSU algorithm for initial classification. Then, based on tidal level characteristics and phenological characteristics, it automatically extracts random sample points and calculates classification thresholds, improving the objectivity and repeatability of classification.

[0005] While the aforementioned methods have made progress in multi-source fusion and automation, the empirical threshold is highly subjective, and the Otsu method's segmentation results are unstable when the gray-level histogram does not form a typical bimodal distribution. Therefore, how to achieve rapid and accurate extraction of tidal flat extent remains a research goal that urgently needs to be addressed in the field of remote sensing monitoring. Summary of the Invention

[0006] In view of the above, the purpose of this invention is to provide an automated remote sensing extraction method for large-scale coastal tidal flats. Based on the Google Earth Engine cloud platform, this method efficiently processes massive amounts of remote sensing images. By constructing an adaptive mixed water index that can automatically adapt to water turbidity, and combining it with an adaptive threshold segmentation method based on derivative mutation point detection, it overcomes the dependence of traditional methods on ideal bimodal histograms. Furthermore, by integrating time-series frequency analysis and maximum water surface range constraints, it achieves rapid, accurate, and stable extraction of the spatial distribution of tidal flats in large-scale and complex environments. This method is applicable to applications such as coastal resource monitoring, ecological assessment, and protection management.

[0007] To achieve the above-mentioned objectives, the present invention provides the following technical solution:

[0008] In a first aspect, the present invention provides a method for large-scale automated remote sensing extraction of coastal tidal flats, comprising the following steps:

[0009] Based on the de-clouded temporal remote sensing images of the study area, the turbidity index is calculated pixel by pixel and the contrast band is selected accordingly. The adaptive mixed water index is calculated based on the contrast band and the green light band.

[0010] For each image determined by the adaptive mixed water index, its value range histogram is calculated and smoothed before the first derivative is obtained. The adaptive mixed water index value corresponding to the derivative mutation point is used as the adaptive land-water segmentation threshold for this image.

[0011] For each scene of cloud-free time-series remote sensing image, the improved normalized differential water index is calculated. The maximum water surface range is extracted based on the improved normalized differential water index and the adaptive land-water segmentation threshold. The tidal flat exposure frequency image is extracted based on each scene image and the adaptive land-water segmentation threshold. The preliminary spatial distribution of the tidal flat is extracted by combining the tidal flat exposure frequency image and the maximum water surface range.

[0012] After connectivity processing, the preliminary spatial distribution of tidal flats is used to generate a visualization of the spatial distribution of tidal flats.

[0013] Preferably, the construction of cloud-free time-series remote sensing imagery includes:

[0014] Sentinel-2 surface reflectance images covering the study area were selected from the GEE cloud platform, and preliminary cloud masking was performed based on the quality control bands of the images.

[0015] Further, by associating the Cloud Score+S2_HARMONIZED dataset with its cloud detection quality assurance band, residual clouds and cloud shadows are finely removed.

[0016] The number of effective pixels in the declouded images is counted scene by scene. Images with fewer than a preset threshold of effective pixels are removed to construct high-quality declouded time-series remote sensing images.

[0017] Preferably, the step of calculating the turbidity index per pixel and selecting a contrast band accordingly, and calculating an adaptive mixed water index based on the contrast band and the green light band, includes:

[0018] The turbidity index TWI is calculated pixel by pixel as TWI = RED - SWIR1, where RED is the reflectance in the red band and SWIR1 is the reflectance in the first band of shortwave infrared, based on a preset turbidity threshold. th To classify pixels, when TWI ≥ TWI th For turbid water pixels, when TWI < TWI th At that time, the pixels were for land and non-turbid water bodies;

[0019] For pixels in turbid water, SWIR1 was selected as the reflectance B of the contrast band. A For pixels on land and in non-turbid water, the reflectance NIR in the near-infrared band is selected as the reflectance B in the contrast band. A Furthermore, the adaptive mixed water index ACWI is calculated for different pixel categories: ACWI = (GREEN - B) / ( ... A ) / (GREEN+B A ), where GREEN is the reflectivity of the green light band.

[0020] Preferably, the step of using the adaptive mixed water index value corresponding to the derivative abrupt change point as the adaptive land-water segmentation threshold of the scene image includes:

[0021] Locate the first significant peak point within the interval where the adaptive mixed water index value is greater than zero in the histogram. Search along the direction of decreasing adaptive mixed water index value from this peak point, and determine the adaptive mixed water index value corresponding to the first position where the first derivative turns from negative to positive as the adaptive land-water segmentation threshold for the current image.

[0022] Preferably, the step of calculating the improved normalized differential water index scene-by-scene for de-clouded temporal remote sensing images, and extracting the maximum water surface range based on the improved normalized differential water index and the adaptive land-water segmentation threshold, includes:

[0023] For cloud-free time-series remote sensing images, the improved normalized differential water index mNDWI is calculated scene by scene: mNDWI = (GREEN-SWIR1) / (GREEN+SWIR1), where GREEN is the reflectance of the green band and SWIR1 is the reflectance of the first band of shortwave infrared. The maximum value of mNDWI for all time phases is synthesized pixel by pixel to generate the mNDWI maximum value image.

[0024] An adaptive water-land segmentation threshold is applied to the mNDWI maximum image for water-land binary segmentation. Regions with pixel values ​​greater than or equal to the adaptive water-land segmentation threshold are classified as water bodies, while regions with pixel values ​​less than the adaptive water-land segmentation threshold are classified as non-water bodies. The segmented water body regions are converted into vectors and fused to obtain a complete and continuous maximum water surface range.

[0025] Preferably, the step of extracting tidal flat exposure frequency images based on each scene image and an adaptive land-water segmentation threshold includes:

[0026] For each image determined by the adaptive mixed water index, water-land binary segmentation is performed using its corresponding adaptive water-land segmentation threshold to generate a temporal water-land mask image.

[0027] The number of times the temporal water-land mask image was identified as a non-water body in all valid observations was counted pixel by pixel. The ratio of this number to the total number of valid observations was calculated as the tidal flat exposure frequency. The tidal flat exposure frequency was then assigned back to the corresponding pixel position to generate a tidal flat exposure frequency image.

[0028] Preferably, the step of extracting the preliminary spatial distribution of tidal flats by combining tidal flat exposure frequency images and maximum water surface area includes:

[0029] The areas with tidal flat exposure frequencies higher than a preset threshold are extracted from the tidal flat exposure frequency images as potential tidal flat areas. The potential tidal flat areas are then masked using the maximum water surface area to exclude stable land areas that are not submerged at high tide, thus obtaining the preliminary spatial distribution of tidal flats.

[0030] Secondly, embodiments of the present invention also provide a large-scale remote sensing automated extraction device for coastal tidal flats, which is implemented using the above-mentioned large-scale remote sensing automated extraction method for coastal tidal flats, including: an index calculation module, a threshold determination module, a tidal flat extraction module, and a result optimization module;

[0031] The index calculation module is used to calculate the turbidity index pixel by pixel based on the de-clouded time-series remote sensing image of the study area and select the contrast band accordingly. Based on the contrast band and the green light band, the adaptive mixed water index is calculated.

[0032] The threshold determination module is used to calculate the value range histogram of each image determined by the adaptive mixed water index, and after smoothing, obtain the first derivative. The adaptive mixed water index value corresponding to the derivative mutation point is used as the adaptive land-water segmentation threshold of the current image.

[0033] The tidal flat extraction module is used to calculate the improved normalized differential water index for each scene of the cloud-free time-series remote sensing image, extract the maximum water surface range based on the improved normalized differential water index and the adaptive water-land segmentation threshold, extract the tidal flat exposure frequency image based on each scene image and the adaptive water-land segmentation threshold, and extract the preliminary tidal flat spatial distribution by combining the tidal flat exposure frequency image and the maximum water surface range.

[0034] The result optimization module is used to generate a visualization result of the spatial distribution of tidal flats after the preliminary spatial distribution of tidal flats has undergone connectivity processing.

[0035] Thirdly, embodiments of the present invention also provide an electronic device, including a memory and one or more processors, wherein the memory is used to store a computer program, and the processor is used to implement the above-described method for large-scale remote sensing automated extraction of coastal tidal flats when the computer program is executed.

[0036] Fourthly, embodiments of the present invention also provide a computer-readable storage medium storing a computer program, which, when executed by a computer, implements the above-described method for large-scale remote sensing automated extraction of coastal tidal flats.

[0037] Compared with the prior art, the beneficial effects of the present invention include at least the following:

[0038] (1) This invention proposes an adaptive mixed water body index, which combines the turbidity index and turbidity threshold to dynamically select the near-infrared or short-wave infrared band, effectively distinguishing between clear and turbid water bodies, thereby achieving high-precision integrated extraction of various water bodies under complex turbid environments, and significantly improving the applicability and accuracy of the water body index.

[0039] (2) This invention innovatively proposes an adaptive water-land segmentation threshold extraction method based on derivative mutation point detection. By analyzing the change of the first derivative of the exponential histogram, the starting boundary of the statistical distribution of water pixels is identified, which effectively overcomes the dependence of traditional methods on ideal bimodal histograms and solves the problem of threshold instability caused by image radiation differences and land feature mixing, significantly enhancing the robustness of tidal flat water-land boundary identification.

[0040] (3) This invention constructs a complete technical chain from exponential calculation and adaptive threshold segmentation to information extraction and optimization: by synthesizing the maximum time-series maximum value, the maximum water surface range is obtained, effectively suppressing the interference of linear features such as cross-sea bridges and island-connecting dikes on the connectivity of tidal flats; by calculating the frequency of tidal flat exposure to quantify the submergence dynamics of pixels, and combining it with the maximum water surface range for masking, stable identification and accurate extraction of periodically exposed tidal flat pixels are achieved. This method is highly systematic and has clear physical meaning, and is suitable for large-scale, long-term series dynamic monitoring and change analysis of tidal flats under different coastal tidal environments. Attached Figure Description

[0041] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0042] Figure 1 A flowchart illustrating the automated remote sensing extraction method for large-scale coastal tidal flats provided in this embodiment of the invention;

[0043] Figure 2 This is a schematic diagram of the filtered Yellow River Estuary time-series Sentinel-2 image dataset in an embodiment of the present invention;

[0044] Figure 3 This is a comparison of the performance of NDWI, mNDWI, and ACWI in the Yellow River Estuary in the embodiments of the present invention;

[0045] Figure 4 This refers to the composite image with the maximum mNDWI value in this embodiment of the invention.

[0046] Figure 5 Here is an example of an NDWI image histogram in an embodiment of the present invention ( Figure 3 area);

[0047] Figure 6 Example of mNDWI image histogram in an embodiment of the present invention ( Figure 3 area);

[0048] Figure 7 This is an example of the ACWI image histogram and its thresholding results in an embodiment of the present invention. Figure 3 area);

[0049] Figure 8 This refers to the maximum water surface area in this embodiment of the invention;

[0050] Figure 9 This is a diagram showing the frequency of tidal flat exposure in an embodiment of the present invention;

[0051] Figure 10 The results of tidal flat extraction in this embodiment of the invention;

[0052] Figure 11 This is a schematic diagram of the structure of the large-scale remote sensing automated extraction device for coastal tidal flats provided in an embodiment of the present invention. Detailed Implementation

[0053] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and do not limit the scope of protection of this invention.

[0054] The inventive concept of this invention is as follows: Addressing the shortcomings of existing technologies, such as the insufficient adaptability of single water body indices to complex turbid water environments and the poor stability caused by the reliance on ideal bimodal histograms for water-land separation thresholds, this invention provides a large-scale remote sensing automated extraction method for coastal tidal flats. By constructing an adaptive mixed water body index (ACWI), the optimal contrast band is dynamically selected based on pixel turbidity, achieving high-precision integrated extraction of clear and turbid water bodies. Furthermore, an adaptive threshold selection method based on derivative mutation point detection is proposed to overcome the dependence on the bimodal shape of the histogram, achieving robust water-land separation in complex scenarios. Finally, by integrating temporal frequency analysis and maximum water surface area constraints, a complete technical chain is constructed, ultimately achieving high-precision, fully automated, and robust extraction of large-scale coastal tidal flats.

[0055] like Figure 1 As shown in the embodiment, a method for large-scale remote sensing automated extraction of coastal tidal flats is provided, including the following steps:

[0056] Step 1: Construct cloud-free time-series remote sensing images.

[0057] In this embodiment, Sentinel-2 surface reflectance imagery data covering the study area and with cloud cover below 60% within one year from the GEE (Google Earth Engine) cloud platform were selected and then declouded. The specific process is as follows:

[0058] (1) Cloud masking based on QA band:

[0059] Based on the image quality control band (QA band), opaque clouds and cirrus clouds are removed through bit operations and masking algorithms.

[0060] (2) Cloud-optimized processing integrating Cloud Score+ processor:

[0061] By further linking the Cloud Score+S2_HARMONIZED dataset and leveraging its cloud detection quality assurance band (CS band), with a discrimination threshold of 0.4, residual clouds and cloud shadows are finely removed, significantly improving the proportion of usable pixels in the image.

[0062] (3) Valid pixel count screening:

[0063] To ensure the reliability of subsequent index calculations and tidal flat frequency statistics, the effective pixel count of each image after cloud removal was counted, and images with fewer than 50,000 pixels were removed. This ensured that the selected images had sufficient spatial coverage and information integrity, and were constructed as follows: Figure 2 The image dataset shown is a time-series Sentinel-2 image dataset of the Yellow River Estuary.

[0064] Step 2: Construct an adaptive mixed water index.

[0065] To address the significant spatial variability of turbidity in estuarine and coastal waters, this invention proposes an Adaptive Combined Water Index (ACWI). This method dynamically selects the optimal contrast band based on a turbidity threshold, thereby improving the overall extraction accuracy of both turbid and clear water. The specific process is as follows:

[0066] (1) Calculation of turbidity index:

[0067] First, calculate the Turbid Water Index (TWI) pixel-by-pixel:

[0068] TWI = RED-SWIR1,

[0069] Where RED is the reflectance in the red light band, and SWIR1 is the reflectance in the first band of shortwave infrared. Based on a preset turbidity threshold TWI... th =0.107, classifying pixels as turbid water bodies (TWI ≥ TWI) th ) and terrestrial and non-turbid water bodies (TWI) <TWI th ).

[0070] (2) Adaptive band selection:

[0071] Based on the turbidity determination results, the reflectance band B for constructing the water index is dynamically selected. A :

[0072] ,

[0073] For pixels in turbid water, SWIR1 was selected as the contrast band reflectance B. AFor pixels on land and in non-turbid water, the reflectance NIR in the near-infrared band is selected as the contrast reflectance B. A .

[0074] (3) Adaptive Mixed Water Index Calculation:

[0075] The adaptive mixed water index is calculated based on the green light band and an adaptively selected contrast band:

[0076] ACWI=(GREEN-B A ) / (GREEN+B A ),

[0077] Wherein, GREEN represents the reflectivity of the green light band.

[0078] like Figure 3 The image shows a comparison of the performance of NDWI, mNDWI, and ACWI in the Yellow River Estuary. It can be seen that in turbid water areas, NDWI is insufficient in distinguishing between turbid water and tidal flats, with significantly lower NDWI values ​​in turbid water, especially in areas with high suspended sediment concentrations such as the Yellow River Estuary. mNDWI struggles to effectively distinguish between wet tidal flats and water bodies, with significantly higher mNDWI values ​​in wet tidal flats, similar to water bodies, leading to omissions in the tidal flat area. ACWI, on the other hand, is more stable, effectively separating turbid water and compensating for omissions in wet tidal flats, maintaining the integrity of the tidal flat area. This step utilizes the high absorption characteristics of the NIR band in clear water areas to maintain the accuracy of shallow water and tidal flat identification; in turbid water areas, switching to the SWIR1 band suppresses high reflection interference from suspended sediment, significantly enhancing the water-land contrast. This achieves high-precision integrated extraction of all water bodies in complex turbid environments using a single-body index, automatically adapting to different turbidity conditions without manual intervention or regional segmentation.

[0079] Step 3: Determine the adaptive water-land segmentation threshold.

[0080] Traditional thresholding methods based on histogram morphology (such as the Otsu algorithm) typically rely on histograms exhibiting a clear bimodal characteristic. However, in actual remote sensing images, especially in areas affected by complex land cover mixtures, turbid water bodies, and tidal flat transition zones, water index histograms often exhibit non-ideal bimodal morphology, unimodal skewness, or peak overlap, leading to deviations or even failures in threshold determination using traditional methods.

[0081] To overcome the above problems, this invention proposes an adaptive threshold selection method based on derivative mutation point detection. This method does not rely on the bimodal assumption but instead captures the initial boundary where water pixels begin to accumulate significantly in the statistical distribution, thus achieving robust land-water segmentation. The specific process is as follows:

[0082] (1) Constructing an exponential histogram:

[0083] In the GEE platform, taking the ACWI image as an example, it is converted into an array, and the ee.Reducer.histogram function in GEE is called to calculate the frequency distribution of ACWI values, obtaining a histogram object. The abscissa is the ACWI value (-1 to 1), and the ordinate is the pixel frequency within the corresponding ACWI value interval.

[0084] (2)Smoothing the histogram curve:

[0085] The original histogram is smoothed using a moving average method with a window size of 3 to suppress the minor fluctuations caused by fewer pixels or noise.

[0086] (3)Identifying the peak positions:

[0087] Traverse the histogram sequence to identify all peak points n that satisfy G[n - 1]<G[n]>G[n + 1], where G[n] represents the frequency value of the histogram at position n.

[0088] (4)Calculating the first derivative:

[0089] Calculate the first derivative D(f) of the smoothed histogram curve to quantify the frequency change trend and identify the growth and decay stages of the distribution.

[0090] (5)Removing interfering peaks

[0091] Calculate the global maximum derivative value D max , for each peak point n, extract the maximum derivative value D max_n of its left region (ACWI < n part). If D max_n <N * D max (where N is an empirical coefficient, usually taken as 0.01 - 0.1 and can be adjusted according to the regional complexity), then it is considered that this peak is caused by statistical noise or minor features and is removed, only retaining the significant peak body.

[0092] (6)Extracting the water body peak and determining the segmentation threshold

[0093] As Figure 7 shown, in the interval where the abscissa of the histogram (ACWI) is greater than 0 (usually corresponding to the water body pixel aggregation area), locate the first significant peak n closest to 0 in the abscissa. Figure 7 (the coordinate of n in Figure 7 is (0.66, 223)). Starting from this peak point n, search along the decreasing direction of the abscissa (ACWI) (to the left) to find the first position where the first derivative changes from negative to positive. The ACWI value t

[0094] corresponding to this point (0.125 in Figure 5 The example of an NDWI image histogram shown, and as shown Figure 6 The mNDWI image histogram examples shown all correspond to Figure 3 In terms of regions, it can be seen that the NDWI image histogram has a large number of water-land mixed pixels, with unclear transitions and numerous and chaotic peaks; the mNDWI image histogram shows a more obvious water-land transition, with very obvious water peaks, tending towards a single-peak distribution. Figure 3 The exponential effect of the mNDWI histogram is that the water peaks include wet tidal flats, which will significantly affect the identification of tidal flats; the ACWI image histogram combines the advantages of both, with obvious transitions between water and land, which facilitates thresholding.

[0095] This step effectively overcomes the dependence of traditional segmentation methods on ideal bimodal histograms. By detecting the initial boundary of the statistical distribution of water pixels, it achieves robust and adaptive threshold selection for complex histogram shapes such as unimodal, skewed, and overlapping peaks. In complex coastal environments, this method can automatically identify statistical abrupt changes in the water-land transition, significantly improving the accuracy and stability of water extraction in turbid waters, tidal flats, and shaded areas. Furthermore, it is suitable for large-scale, long-term batch automated processing of images on the GEE platform, providing a reliable, efficient, and highly adaptable foundation for water-land segmentation in tidal flat extraction.

[0096] Step 4: Extract the maximum water surface area.

[0097] To extract the maximum water surface area under high tide conditions and effectively suppress interference from inland water surfaces in non-tidal areas, this invention employs a maximum value synthesis method based on the improved normalized difference water index (mNDWI). mNDWI, utilizing a combination of green light and shortwave infrared bands, has been proven to have good distinguishing ability between land and water boundaries under high tide conditions. The specific process is as follows:

[0098] (1) Calculate the temporal mNDWI image set:

[0099] For the time-series images selected in step one, calculate mNDWI scene by scene:

[0100] mNDWI=(GREEN-SWIR1) / (GREEN+SWIR1),

[0101] The calculation results are organized in the original time order to obtain a time-series mNDWI image set.

[0102] (2) Synthesize the maximum mNDWI image:

[0103] The temporal maximum value synthesis method is used to extract the maximum mNDWI value from all temporal phases of the temporal mNDWI image set pixel by pixel, generating an mNDWI maximum value image, such as... Figure 4As shown, this image effectively highlights water pixels while suppressing noise caused by tidal changes, cloud shadows, and transient surface variations.

[0104] (3) Binary segmentation of land and water:

[0105] The adaptive land-water segmentation threshold determined in step three is applied to the mNDWI maximum value image and binarized to obtain the land-water segmentation result. Regions with pixel values ​​higher than the threshold are classified as water bodies, while regions with values ​​lower than the threshold are classified as non-water bodies.

[0106] (4) Extract the maximum water surface area:

[0107] To ensure the integrity of the water surface area, especially in situations where linear features such as cross-sea bridges and island-connecting barriers might missegment the water area, the water area is marked using the `point` function in Geographic Object Modeling (GEE). The `filterBounds` function in GEE then filters out all water polygons intersecting with these marked points. These filtered polygons are then merged to obtain the complete and continuous vector data of the maximum water surface area. Figure 8 As shown.

[0108] Step 5: Calculate the frequency of tidal flat exposure.

[0109] To quantify the dynamic inundation and exposure characteristics of tidal flats, this invention constructs a land-water mask based on time-series water index images and calculates the tidal flat exposure frequency pixel by pixel. The specific process is as follows:

[0110] Temporal generation of land and water masks:

[0111] For the image set for which the index has been calculated in step two, each image in the ACWI is assigned an adaptive land-water segmentation threshold t according to the method described in step three, and then subjected to land-water binary segmentation to construct a mask band. Water pixels are assigned a value of 0, and non-water pixels are assigned a value of 1.

[0112] ,

[0113] The mask band is added as a new band to the original image using the addBand function in GEE, forming an image set containing temporal information on land and water conditions.

[0114] Calculation of tidal flat exposure frequency:

[0115] Based on the above temporal masking results, the frequency of tidal flat exposure is calculated pixel by pixel, defined as the proportion of times a pixel is identified as a "non-water body" in the total number of valid observations. The calculation formula is as follows:

[0116] ,

[0117] in, The total number of observations. Indicates the first pixels in the scene image The state of flooding ( It means to be submerged. This indicates that the signal was not submerged, i.e., 1 and 0 in the mask band. Indicates the first pixels in the scene image Observation status ( Indicates that it was observed. (Indicating unobserved), the TFF value ranges from 0 to 1, with higher values ​​indicating a more stable exposure state of the pixel over time. The tidal flat exposure frequency is then assigned back to the corresponding pixel location to generate a tidal flat exposure frequency image, such as... Figure 9 As shown.

[0118] Step 6: Extract the spatial extent of the tidal flats.

[0119] Based on the tidal flat exposure frequency image obtained in step five, pixels with a frequency value greater than 0.05 are extracted as potential tidal flat areas. This frequency threshold of 0.05 effectively eliminates misjudgments caused by short-term submersion or occasional exposure, preserving long-term stable tidal flat areas. Subsequently, using the maximum water surface area extracted in step four, the aforementioned potential tidal flat areas are masked to exclude stable land areas not submerged at high tide, ensuring that the final extracted tidal flats only include periodically exposed tidal flat areas. Through these steps, a preliminary spatial distribution of tidal flats is obtained.

[0120] Step 7: Data post-processing.

[0121] The connectivity processing of the tidal flat distribution data obtained in step six is ​​performed. In GEE, the `connectedPixelCount` function is called to label and count connected components in the binary tidal flat image, removing isolated small patches with fewer than 50 connected pixels to eliminate fragmented patches caused by image noise or short-term interference, thus improving the continuity of the tidal flat patches. Small cavities formed within the tidal flat patches due to tidal channels, localized water accumulation, or residual sediment are further filled to maintain the integrity of the tidal flat landform. This yields the tidal flat extent result as follows: Figure 10 As shown in the figure. Based on the processed tidal flat vector data, the area of ​​each tidal flat patch is calculated, and the area is visualized and rendered according to area level or spatial distribution to generate a thematic map of tidal flat spatial distribution that can be used for further analysis and application.

[0122] In summary, the present invention provides a method for large-scale automated remote sensing extraction of coastal tidal flats. Utilizing the Google Earth Engine cloud platform and time-series Sentinel-2 imagery, it achieves high-precision, fully automated, and robust extraction of coastal tidal flats in complex turbid environments through a complete technical process combining adaptive mixed water index (ACWI) construction, robust threshold extraction based on derivative mutation points, and time-series frequency analysis with spatial constraints. The method is simple and efficient, overcoming the dependence of existing technologies on ideal histograms, significantly improving the automation level and result stability of tidal flat monitoring, and is suitable for large-scale, long-term operational applications of dynamic tidal flat monitoring.

[0123] Based on the same inventive concept, such as Figure 11 As shown, this embodiment of the invention also provides a large-scale remote sensing automated extraction device 110 for coastal tidal flats, including: an index calculation module 111, a threshold determination module 112, a tidal flat extraction module 113, and a result optimization module 114.

[0124] The index calculation module 111 is used to calculate the turbidity index pixel by pixel based on the de-clouded time-series remote sensing image of the study area and select the contrast band accordingly. Based on the contrast band and the green light band, the adaptive mixed water index is calculated.

[0125] The threshold determination module 112 is used to calculate the value range histogram of each image determined by the adaptive mixed water index, and after smoothing, obtain the first derivative. The adaptive mixed water index value corresponding to the derivative mutation point is used as the adaptive land-water segmentation threshold of the current image.

[0126] The tidal flat extraction module 113 is used to calculate the improved normalized differential water index for each scene of the declouded time-series remote sensing image, extract the maximum water surface range based on the improved normalized differential water index and the adaptive land-water segmentation threshold, extract the tidal flat exposure frequency image based on each scene image and the adaptive land-water segmentation threshold, and extract the preliminary tidal flat spatial distribution by combining the tidal flat exposure frequency image and the maximum water surface range.

[0127] The result optimization module 114 is used to generate a visualization result of the spatial distribution of tidal flats after connectivity processing of the preliminary spatial distribution.

[0128] Based on the same inventive concept, embodiments of the present invention also provide an electronic device, including a memory and one or more processors, wherein the memory is used to store a computer program, and the processor is used to implement the above-described method for large-scale remote sensing automated extraction of coastal tidal flats when the computer program is executed.

[0129] Based on the same inventive concept, embodiments of the present invention also provide a computer-readable storage medium storing a computer program, which, when executed by a computer, implements the above-described method for large-scale remote sensing automated extraction of coastal tidal flats.

[0130] It should be noted that the large-scale remote sensing automated extraction device, electronic equipment, and computer-readable storage medium provided in the above embodiments all belong to the same inventive concept as the large-scale remote sensing automated extraction method for coastal tidal flats. For details of its implementation process, please refer to the embodiments of the large-scale remote sensing automated extraction method for coastal tidal flats, which will not be repeated here.

[0131] The specific embodiments described above illustrate the technical solution and beneficial effects of the present invention in detail. It should be understood that the above description is only the most preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, additions, and equivalent substitutions made within the scope of the principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for large-scale automated remote sensing extraction of coastal tidal flats, characterized in that, Includes the following steps: Based on de-clouded temporal remote sensing images of the study area, the turbidity index is calculated pixel-by-pixel, and a contrast band is selected accordingly. An adaptive mixed water index is then calculated based on the contrast band and the green band, including: calculating the turbidity index TWI = RED - SWIR1 pixel-by-pixel, where RED is the reflectance of the red band and SWIR1 is the reflectance of the first shortwave infrared band, and a preset turbidity threshold is used for TWI. th To classify pixels, when TWI ≥ TWI th For turbid water pixels, when TWI < TWI th The time frame includes pixels representing land and non-turbid water bodies; for pixels representing turbid water bodies, SWIR1 is selected as the reflectance B of the contrast band. A For pixels on land and in non-turbid water, the reflectance NIR in the near-infrared band is selected as the reflectance B in the contrast band. A Furthermore, the adaptive mixed water index ACWI is calculated for different pixel categories: ACWI = (GREEN - B) / ( ... A ) / (GREEN+B A ), where GREEN is the reflectivity in the green light band; For each image determined by the adaptive mixed water index, its value range histogram is calculated and smoothed before the first derivative is obtained. The adaptive mixed water index value corresponding to the derivative mutation point is used as the adaptive land-water segmentation threshold for this image. For each scene of cloud-free time-series remote sensing image, the improved normalized differential water index is calculated. The maximum water surface range is extracted based on the improved normalized differential water index and the adaptive land-water segmentation threshold. The tidal flat exposure frequency image is extracted based on each scene image and the adaptive land-water segmentation threshold. The preliminary spatial distribution of the tidal flat is extracted by combining the tidal flat exposure frequency image and the maximum water surface range. After connectivity processing, the preliminary spatial distribution of tidal flats is used to generate a visualization of the spatial distribution of tidal flats.

2. The method for large-scale automated remote sensing extraction of coastal tidal flats according to claim 1, characterized in that, The construction of cloud-free time-series remote sensing images includes: Sentinel-2 surface reflectance images covering the study area were selected from the GEE cloud platform, and preliminary cloud masking was performed based on the quality control bands of the images. Further, by associating the Cloud Score+S2_HARMONIZED dataset with its cloud detection quality assurance band, residual clouds and cloud shadows are finely removed. The number of effective pixels in the declouded images is counted scene by scene. Images with fewer than a preset threshold of effective pixels are removed to construct high-quality declouded time-series remote sensing images.

3. The method for large-scale automated remote sensing extraction of coastal tidal flats according to claim 1, characterized in that, The step of using the adaptive mixed water index value corresponding to the derivative abrupt change point as the adaptive land-water segmentation threshold for the current scene image includes: Locate the first significant peak point within the interval where the adaptive mixed water index value is greater than zero in the histogram. Search along the direction of decreasing adaptive mixed water index value from this peak point, and determine the adaptive mixed water index value corresponding to the first position where the first derivative turns from negative to positive as the adaptive land-water segmentation threshold for the current image.

4. The method for large-scale automated remote sensing extraction of coastal tidal flats according to claim 1, characterized in that, The step of calculating the improved normalized differential water index scene-by-scene for de-clouded temporal remote sensing images, and extracting the maximum water surface range based on the improved normalized differential water index and adaptive land-water segmentation threshold, includes: For cloud-free time-series remote sensing images, the improved normalized differential water index mNDWI is calculated scene by scene: mNDWI = (GREEN-SWIR1) / (GREEN+SWIR1), where GREEN is the reflectance of the green band and SWIR1 is the reflectance of the first band of shortwave infrared. The maximum value of mNDWI for all time phases is synthesized pixel by pixel to generate the mNDWI maximum value image. An adaptive water-land segmentation threshold is applied to the mNDWI maximum image for water-land binary segmentation. Regions with pixel values ​​greater than or equal to the adaptive water-land segmentation threshold are classified as water bodies, while regions with pixel values ​​less than the adaptive water-land segmentation threshold are classified as non-water bodies. The segmented water body regions are converted into vectors and fused to obtain a complete and continuous maximum water surface range.

5. The method for large-scale automated remote sensing extraction of coastal tidal flats according to claim 1, characterized in that, The extraction of tidal flat exposure frequency images based on each scene image and adaptive land-water segmentation threshold includes: For each image determined by the adaptive mixed water index, water-land binary segmentation is performed using its corresponding adaptive water-land segmentation threshold to generate a temporal water-land mask image. The number of times the temporal water-land mask image was identified as a non-water body in all valid observations was counted pixel by pixel. The ratio of this number to the total number of valid observations was calculated as the tidal flat exposure frequency. The tidal flat exposure frequency was then assigned back to the corresponding pixel position to generate a tidal flat exposure frequency image.

6. The method for large-scale automated remote sensing extraction of coastal tidal flats according to claim 1, characterized in that, The preliminary spatial distribution of tidal flats is extracted by combining tidal flat exposure frequency images and maximum water surface area, including: The areas with tidal flat exposure frequencies higher than a preset threshold are extracted from the tidal flat exposure frequency images as potential tidal flat areas. The potential tidal flat areas are then masked using the maximum water surface area to exclude stable land areas that are not submerged at high tide, thus obtaining the preliminary spatial distribution of tidal flats.

7. A large-scale remote sensing automated extraction device for coastal tidal flats, implemented using the large-scale remote sensing automated extraction method for coastal tidal flats as described in any one of claims 1 to 6, characterized in that, include: The module includes an index calculation module, a threshold determination module, a tidal flat extraction module, and a result optimization module. The index calculation module is used to calculate the turbidity index pixel by pixel based on the de-clouded time-series remote sensing image of the study area and select the contrast band accordingly. Based on the contrast band and the green light band, the adaptive mixed water index is calculated. The threshold determination module is used to calculate the value range histogram of each image determined by the adaptive mixed water index, and after smoothing, obtain the first derivative. The adaptive mixed water index value corresponding to the derivative mutation point is used as the adaptive land-water segmentation threshold of the current image. The tidal flat extraction module is used to calculate the improved normalized differential water index for each scene of the cloud-free time-series remote sensing image, extract the maximum water surface range based on the improved normalized differential water index and the adaptive water-land segmentation threshold, extract the tidal flat exposure frequency image based on each scene image and the adaptive water-land segmentation threshold, and extract the preliminary tidal flat spatial distribution by combining the tidal flat exposure frequency image and the maximum water surface range. The result optimization module is used to generate a visualization result of the spatial distribution of tidal flats after the preliminary spatial distribution of tidal flats has undergone connectivity processing.

8. An electronic device comprising a memory and one or more processors, the memory for storing a computer program, characterized in that, The processor is used to implement the large-scale remote sensing automated extraction method for coastal tidal flats as described in any one of claims 1 to 6 when executing a computer program.

9. A computer-readable storage medium storing a computer program thereon, characterized in that, When the computer program is executed by a computer, it implements the large-scale remote sensing automated extraction method for coastal tidal flats as described in any one of claims 1 to 6.

Images