A Watershed Water Ecological Health Classification Method Based on Multi-Source Remote Sensing and Spatial Big Data

Abstract

This invention discloses a watershed water ecological health classification method based on multi-source remote sensing and spatial big data, belonging to the field of data management technology. First, it acquires an image set of the target watershed water body boundary area from a multi-source remote sensing and spatial big data platform and generates a shallow water candidate pixel set. Then, it performs cloud expansion determination on the candidate pixel set, identifies pixels affected by cloud shading or expansion, and recovers their spectral and water body information. Next, it analyzes the spectral mixing degree of each pixel in the shallow water candidate pixel set, and optimizes the spectrum using an endmember-driven spectral decomposition method and correction strategy. Finally, it generates watershed-level water ecological health classification results. With the support of multi-source data, this method can effectively mitigate the interference of cloud shading and mixing effects on spectral information, thus providing a refined, reliable, and traceable analytical basis for watershed water ecological health classification, ensuring the spatiotemporal consistency of health classification and the ability to sensitively identify abnormal areas.

Description

Technical Field

[0001] This invention relates to the field of data management technology, and in particular to a watershed water ecological health classification method based on multi-source remote sensing and spatial big data. Background Technology

[0002] With the rapid development of remote sensing technology, geographic information systems (GIS), and spatial big data, watershed water ecological health classification based on multi-source remote sensing and spatial big data has gradually become an important technical direction for watershed ecological monitoring and water environment management. Traditional watershed water ecological assessment mainly relies on ground monitoring point data, which is limited by insufficient spatial coverage and poor timeliness, making it difficult to comprehensively reflect the spatiotemporal variation characteristics of water quality within the watershed. However, multi-source remote sensing data (including optical, SAR, thermal infrared, and hyperspectral imagery) can acquire the reflectance characteristics and dynamic information of water bodies in the watershed on a large-scale, continuous time-series scale, providing key support for identifying water quality distribution, monitoring ecological degradation, and analyzing the evolution of water health.

[0003] For example, CN116129287A discloses a method, device, and medium for early warning and identification of watershed water ecological risks based on remote sensing. The early warning method includes: acquiring a satellite remote sensing image of the watershed; performing grayscale processing on the satellite remote sensing image of the watershed to obtain a grayscale image; dividing the grayscale image into several pixels and comparing the difference in grayscale values ​​between two adjacent and touching pixels; when the difference in grayscale values ​​between two pixels is greater than a preset first threshold, the contact point between the two pixels is used as a marker point; and marking the area formed by connecting all the marker points as a risk warning area. This invention is based on satellite remote sensing images of the watershed, performs grayscale processing on the image, subdivides the image into pixels, determines marker points based on the difference in grayscale values ​​between two adjacent and touching pixels, and connects the marker points to accurately mark the risk warning area based on color changes.

[0004] For example, CN116310794B discloses a method for predicting river aquatic biodiversity based on remote sensing and environmental DNA. This method determines the first species abundance at each sampling point within the entire watershed based on the environmental DNA of each sampling point. It then determines patches corresponding to each sampling point based on the sampling point and the species activity range, and determines a first remote sensing index based on pre-acquired remote sensing vegetation indicators and their corresponding first species abundance within each patch. Based on the first remote sensing index and its corresponding first species abundance within each patch, a first preset model is trained. When the first preset model reaches a preset standard, it is determined as a species abundance model. The entire watershed is divided into multiple patches, and a second remote sensing index is determined within each patch. The second remote sensing index is determined by the remote sensing vegetation indicators corresponding to the first remote sensing index. The second remote sensing index within each patch is input into the species abundance model to determine the second species abundance within the entire watershed, which is used to determine the species diversity within the entire watershed.

[0005] The above-mentioned technology has at least the following technical problems:

[0006] Due to factors such as cloud, fog, aerosols, or changes in solar altitude angle, the water reflectance spectrum is prone to strong distortion or loss. This causes the differences in spatial resolution, temporal synchronization, and spectral response characteristics of multi-source remote sensing data to overlap, resulting in noise propagation and feature shift during multi-dimensional image fusion. This leads to discontinuity in local data and affects the calculation of overall watershed ecological indicators and spatial coherence. Summary of the Invention

[0007] To address the technical problems existing in the prior art, embodiments of the present invention provide a method for watershed water ecological health classification based on multi-source remote sensing and spatial big data. The technical solution is as follows, including:

[0008] Image sets of the water boundary region of the target watershed are obtained from a multi-source remote sensing and spatial big data platform, and a set of shallow water candidate pixels of the water boundary region of the target watershed is generated based on a unified grid.

[0009] Cloud expansion determination analysis is performed on the shallow water candidate pixel set to obtain the cloud expansion determination results of each pixel in the shallow water candidate pixel set. Based on the cloud expansion determination results of each pixel in the shallow water candidate pixel set, multi-source reconstruction is performed on each pixel.

[0010] The shallow water spectral mixture of each pixel in the shallow water candidate pixel set is analyzed to obtain the mixed value of the pixel spectrum of each pixel in each band. The pixel spectrum of each pixel in the shallow water candidate pixel set is subjected to spectral decomposition and correction processing, and the consistency of spectral decomposition and correction is checked to generate the water ecological health classification result.

[0011] The beneficial effects of the technical solutions provided by the embodiments of the present invention include at least the following:

[0012] (1) This invention proposes a watershed water ecological health classification method based on multi-source remote sensing and spatial big data. First, it obtains the image set of the water boundary area of ​​the target watershed from the multi-source remote sensing and spatial big data platform and generates a shallow water candidate pixel set. Then, it performs cloud expansion judgment on the candidate pixel set, identifies the pixels affected by cloud shading or expansion and restores their spectral and water information. Then, it analyzes the spectral mixing degree of each pixel in the shallow water candidate pixel set, and optimizes the spectrum using the endmember-driven spectral decomposition method and correction strategy. Finally, it generates watershed-level water ecological health classification results. With the support of multi-source data, it can effectively alleviate the interference of cloud shading and mixing effects on spectral information, thereby providing a fine, reliable and traceable analytical basis for watershed water ecological health classification, ensuring the spatiotemporal consistency of health classification and the ability to sensitively identify abnormal areas.

[0013] (2) This invention acquires multi-temporal images of the water boundary area of ​​the target watershed from a multi-source remote sensing and spatial big data platform, and generates a set of shallow water candidate pixels to provide basic pixel units for subsequent analysis. Based on this, by generating confidence factors for shallow water areas and combining cloud expansion probability risk scores obtained by a spatiotemporal cloud shadow detector, cloud expansion is determined for candidate pixels, identifying high-risk pixels affected by cloud occlusion or expansion. With the support of multi-source data, it can accurately identify disturbed pixels, providing targeted processing objects for subsequent multi-source reconstruction and spectral correction, thereby ensuring the reliability of the restored spectral and water body information. This helps to alleviate spectral confusion caused by cloud occlusion and data loss, improves the accuracy of shallow water pixel spectral and water quality parameters, and provides a refined, traceable, and spatiotemporally consistent analytical basis for watershed water ecological health classification.

[0014] (3) This invention selects pixels that are determined to be high-risk due to cloud expansion from the shallow water candidate pixel set, uses the SAR data of these first pixels to restore their water body information, and obtains temporal residuals to help generate continuous and reliable pixel data. It can effectively repair the spectral and water body information of shallow water pixels affected by cloud shading or expansion, reduce the impact of cloud interference and data loss on the analysis results, and provide reliable input data for watershed water ecological health classification based on multi-source remote sensing and spatial big data.

[0015] (4) This invention obtains the mixed spectral values ​​of each pixel after completing the multi-source reconstruction of cloud expansion pixels, which helps to automatically identify pixels with high mixing degree and trigger the spectral decomposition and correction process to eliminate spectral distortion caused by differences in bottom sediment, water composition and observation conditions, and further generates neighborhood residuals and temporal residuals, and judges the consistency of the spectrum in the spatial and temporal dimensions. With the support of multi-source data, it can realize dynamic correction and quality assessment of shallow water spectral mixing effect, and ensure that the spectral data finally used for watershed water ecological health classification has higher physical authenticity and spatiotemporal stability, thereby improving the accuracy of water quality parameter inversion and health classification, and supporting fine-grained, continuous monitoring and scientific classification of watershed ecological status. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying 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.

[0017] Figure 1 This is a schematic diagram of the method provided in an embodiment of the present invention;

[0018] Figure 2This is a flowchart of shallow water candidate pixel generation and cloud expansion determination provided in an embodiment of the present invention;

[0019] Figure 3 This is a flowchart of multi-source reconstruction and spectral analysis provided in an embodiment of the present invention;

[0020] Figure 4 This is a flowchart of spectral decomposition, correction, and water ecological health classification provided in the embodiments of the present invention. Detailed Implementation

[0021] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0022] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0023] In the embodiments of this invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning. Similarly, the terms "of," "corresponding (relevant)," and "corresponding" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning.

[0024] In this embodiment of the invention, sometimes a subscript such as W1 may be written in a non-subscript form such as W1. When the difference is not emphasized, the meaning they express is the same.

[0025] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0026] like Figure 1 As shown, embodiments of the present invention provide a watershed water ecological health classification method based on multi-source remote sensing and spatial big data, including:

[0027] Image sets of the water boundary region of the target watershed are obtained from a multi-source remote sensing and spatial big data platform, and a set of shallow water candidate pixels of the water boundary region of the target watershed is generated based on a unified grid.

[0028] Cloud expansion determination analysis is performed on the shallow water candidate pixel set to obtain the cloud expansion determination results of each pixel in the shallow water candidate pixel set. Based on the cloud expansion determination results of each pixel in the shallow water candidate pixel set, multi-source reconstruction is performed on each pixel.

[0029] The shallow water spectral mixture of each pixel in the shallow water candidate pixel set is analyzed to obtain the mixed value of the pixel spectrum of each pixel in each band. The pixel spectrum of each pixel in the shallow water candidate pixel set is subjected to spectral decomposition and correction processing, and the consistency of spectral decomposition and correction is checked to generate the water ecological health classification result.

[0030] It should be noted that optical imagery includes (such as Sentinel-2, Landsat, and commercial high-resolution imagery), SAR imagery (such as Sentinel-1), aerial or UAV LiDAR / sonar bathymetry data, and existing ground water quality monitoring data. Each data point is recorded with complete metadata upon data entry, including sensor information, acquisition time, spatial resolution, band calibration parameters, processing chain history, and rapid quality check information such as effective pixel ratio, cloud cover, and geometric registration error. The shallow water candidate pixel set includes a pixel index, historical water persistence statistics, a reference image pointer for the most recent cloudless observation day, and water depth estimates or priority information for the candidate pixels.

[0031] It should be noted that cloud inflation refers to the misidentification of clouds and their surrounding translucent edges as larger cloud areas in remote sensing images by algorithms. This causes pixels that should belong to nearshore or shallow water areas to be incorrectly masked or abnormally brightened, thus reducing effective water body information. In particular, the spectral characteristics of high-reflectivity shallow water areas are "covered" or "diluted" by the bright pixels of clouds. This inflation effect is particularly pronounced in the visible and near-infrared bands, directly disrupting the reflectance curve of shallow water areas and resulting in a systematically lower inversion result for water quality parameters such as chlorophyll, suspended matter, and transparency. Secondly, the shallow water spectral mixing effect is more prominent when spatial resolution is limited or the bottom albedo is high. Because the optical properties of shallow water are affected by the underwater substrate, sediments, algae, and changes in water depth, the reflected signals received by the sensor are often a linear or nonlinear superposition of water signals and bottom sediment signals. This leads to significant differences in the spectral mixing degree between different pixels. When cloud expansion and spectral mixing occur simultaneously, the problem becomes even more complex: the mixing of the highly reflective portion of clouds with the reflectance of the shallow water bottom causes irregular shifts in the spectral representation of water bodies in multi-source images. This is especially problematic in multi-temporal data fusion or time-series analysis, where algorithms may misinterpret these shifts as genuine changes in water condition. For example, bright areas caused by cloud expansion may be misjudged as water shrinkage or improved water quality, while highly reflective shallow water pixels resulting from spectral mixing may be perceived as increased pollution or algal blooms, leading to systematic fluctuations and misleading interpretations of the ecological health index across the entire spatiotemporal series.

[0032] It should be noted that, in order to accurately identify and extract candidate pixels in nearshore and shallow water areas, a unified spatial grid system or a vector framework based on riverbank buffer zones needs to be established first. The specific process is as follows: First, based on high-precision DEM and river system vector data, the contours of the main channel, tributaries, and their connected shorelines are determined using a watershed hierarchical zoning model. A buffer zone of a certain width (e.g., adjustable from 50 to 300 meters) is generated using shoreline geometric features, serving as the initial spatial extent of potential nearshore and shallow water areas. On this basis, spatial resampling and geometric correction are performed on remote sensing images from different sources (e.g., Landsat, Sentinel, GF series, etc.) to ensure that their resolution, projection coordinates, and temporal matching accuracy meet the alignment requirements under the same standard grid. Subsequently, a unified grid (e.g., based on 1 / 3 arcsecond or 10-meter resolution) is used to divide the water body boundary region, projecting each pixel onto standard grid points, and identifying pixels belonging to the buffer zone as the candidate pixel set. To further improve the accuracy of water body identification for candidate pixels, the system uses the Normalized Difference Water Index (NDWI), the Modified Difference Water Index (MNDWI), and near-infrared reflectance thresholds for initial screening, eliminating mixed pixels of shadows, vegetation, or bare land. Simultaneously, pixel consistency is determined across multi-source data. For pixels exhibiting high spectral similarity but significant differences in classification labels across different time series or sensor images, statistical temporal stability and spectral morphological similarity are used to determine whether they belong to typical shallow water areas or the edge of dynamic water bodies. Next, the system estimates water depth gradient characteristics using a ground reflectance model and shoreline slope information, marking pixels with shallow water and abnormally high reflectance as potential spectral mixing sensitive areas. Finally, by fusing optical, radar, and topographic features into a multi-dimensional attribute space, a probability distribution map of candidate pixels in shallow water areas is established. Based on this distribution and confidence thresholds stored in the database, a set of highly reliable candidate pixels is selected. The shallow water candidate pixel set contains multiple pixels because it represents a group of potential shallow water areas extracted from optical images of the water boundary region of the target watershed based on a uniform grid. In remote sensing imagery, a pixel is the basic unit of the image; each pixel corresponds to a small area of ​​the ground with a certain spatial resolution (e.g., 10m × 10m) and carries multispectral information (e.g., reflectance values ​​in the visible and near-infrared bands). These spectral features can be used to distinguish land cover types such as water bodies, shallow water, or deep water.

[0033] like Figure 2 As shown, Figure 2The flowchart for shallow water candidate pixel generation and cloud expansion determination provided in this embodiment of the invention starts with acquiring multi-source remote sensing and spatial big data. A riverbank buffer is constructed based on the riverbank vector and DEM, generating a set of shallow water candidate pixels under a unified grid. By calculating the confidence factor of the shallow water area and the cloud probability expansion risk score, combined with a preset threshold, cloud expansion is determined. Pixels meeting the criteria are marked as the first pixel (pixels affected by clouds), laying the foundation for subsequent multi-source data reconstruction processing. The specific process for generating the set of shallow water candidate pixels under a unified grid is as follows:

[0034] A multi-source standardized image set of the target watershed within a specified time period is obtained. A riverbank buffer zone is constructed based on the riverbank vector and digital elevation model. The riverbank buffer zone is projected onto a unified grid, and each grid cell in the unified grid is filtered to obtain a set of shallow water candidate pixels for the water boundary region of the target watershed, including each shallow water candidate pixel.

[0035] It should be noted that the construction of the riverbank buffer zone based on the riverbank vector and digital elevation model specifically involves retrieving high-precision riverbank vector data of the target watershed and digital elevation model (DEM) data consistent with its spatial reference system from a multi-source spatial database. Spatial registration and coordinate system unification are then performed on both to ensure geographical consistency and spatial overlay accuracy. Subsequently, based on the spatial distribution characteristics of the main channel and tributaries, the system extracts the geometric attribute information of the riverbank, including line segment direction, curvature variation, river width estimation, and slope differences on both sides of the river. These parameters are then used to determine the initial setting range for the width of the riverbank buffer zone. The buffer zone width can be adaptively adjusted according to the river width, watershed topographic relief, water level variation, and bank slope. Specifically, a larger buffer distance is set in areas with gentler slopes or wider channels, while a narrower buffer zone is used in steeper areas or where the river channel narrows. The system uses the riverbank vector as the center line and generates equidistant buffer zone objects along the normal direction based on the aforementioned adaptive width. During generation, elevation constraints from the DEM are introduced to filter and correct the elevation gradient at the buffer zone boundaries, eliminating areas with abrupt elevation changes or steep terrain rises to prevent the buffer zones from extending into non-water or mountainous areas. The generated buffer zone objects undergo topology checks and overlap repair to ensure continuity and spatial closure of each buffer zone at river confluences or bifurcations. Finally, the system outputs a vector dataset containing the spatial geometry, width parameters, mean elevation, and slope distribution of the buffer zones, and establishes a spatial index for subsequent analysis modules such as shallow water candidate pixel selection, dynamic water body boundary identification, and nearshore water quality parameter estimation.

[0036] Specifically, cloud expansion determination analysis is performed on the shallow water candidate pixel set. The specific process is as follows: a multi-temporal historical image time window is preset. Within the multi-temporal historical image time window, the shallow water candidate pixel set of historical images of the target watershed water body boundary area is obtained. The average intersection ratio of cloud expansion and water body boundary and SAR reflectance are extracted. The number of times the intersection ratio of cloud expansion and water body boundary is higher than the threshold of the intersection ratio of cloud expansion and water body boundary stored in the database is counted and recorded as the cloud expansion event frequency. The average intersection ratio of cloud expansion and water body boundary, SAR reflectance and cloud expansion event frequency are normalized, reverse mapped and then fused to obtain the shallow water area confidence factor of each pixel in the shallow water candidate pixel set. The shallow water area confidence factor of each pixel in the shallow water candidate pixel set is used to quantify the reliability of the pixel belonging to the shallow water area.

[0037] It should be noted that the intersection of cloud expansion and water body boundaries refers to the spatial overlap between the cloud expansion area and the water body boundary area identified by cloud detection algorithms in remote sensing imagery. Because clouds exhibit dynamic changes in time-series imagery, when cloud shadows or cloud expansion phenomena approach or cover the water body edge, the water's reflectance characteristics may be partially masked or misidentified. Therefore, it is necessary to quantify this spatial overlap. The water body boundary refers to the outer edge of the water body region extracted using water body indices (such as NDWI, MNDWI) or water body masks; it represents the transition zone between water and land. This boundary reflects the spatial extent and outline of the water body distribution in the image. When the spatial positions of the cloud expansion area and the water body boundary overlap, an intersection is formed. The area ratio reflects the degree of interference of cloud expansion on the water body edge identification results, and is used to subsequently calculate the intersection ratio of cloud expansion and the water body boundary and assess the degree to which pixels are affected by clouds. Water body boundaries are the outlines or boundary regions extracted from the differences in reflectance or scattering characteristics between water and land in multi-source remote sensing data (such as optical images, SAR images, and thermal infrared data). They are typically obtained using water body indices (such as NDWI and MNDWI) or segmentation algorithms (such as thresholding and edge detection). In practical calculations, the spatial extent of the water body boundary is first determined by overlaying the water body identification results from optical and SAR images. Then, the cloud expansion range is determined by comparing the current temporal cloud detection results with the cloud shadow region, and the spatial overlap ratio between the cloud expansion range and the water body boundary is calculated.

[0038] Meanwhile, based on the trained spatiotemporal cloud shadow detector, the cloud expansion probability risk score of each pixel in the shallow water candidate pixel set is directly extracted from multi-source remote sensing and spatial big data images.

[0039] It should be noted that the specific process of normalizing and inversely mapping the average intersection ratio of cloud expansion with water body boundaries, SAR reflectivity, and cloud expansion event frequency to align with shallow water reliability is as follows: First, the minimum and maximum values ​​of each indicator are calculated in the shallow water candidate pixel set. Then, based on this range, the original indicator values ​​of each pixel are linearly transformed proportionally to the interval between 0 and 1, so that the minimum value in the set corresponds to zero and the maximum value corresponds to one. This step completes the normalization process, allowing indicators with different dimensions and ranges to be compared uniformly. Next, inverse mapping is performed. For indicators such as the average intersection ratio of cloud expansion with water body boundaries and cloud expansion event frequency, which originally indicate more severe interference with higher values, a subtraction method is used for transformation, so that the higher normalized values ​​correspond to low interference and high shallow water reliability. SAR reflectivity, on the other hand, inherently indicates open water bodies with low values, meaning high shallow water reliability. Therefore, after normalization, the original direction can be directly retained or slightly adjusted as needed to maintain consistency with other indicators. To enhance sensitivity to extreme values, a nonlinear transformation, such as exponential decay or a sigmoid function, can be applied to the index values ​​after normalization or inverse mapping. This significantly reduces the confidence of pixels with severe cloud expansion or SAR anomalies, while increasing the confidence of pixels with lower interference. The resulting inverse mapping or adjustment values ​​of the three indices can then be used as their respective contributions for weighted or probabilistic fusion with other positive indices (such as blue-green band reflectance and spectral gradient) to generate a comprehensive shallow water area confidence factor. This quantifies the reliability of each candidate pixel belonging to a shallow water area. The entire process maintains consistency in index direction while considering the relative importance of different indices, achieving a unified evaluation of cloud expansion, SAR characteristics, and historical event frequency.

[0040] Specifically, the cloud expansion determination results for each pixel in the shallow water candidate pixel set are obtained through the following process:

[0041] Extract the shallow water region confidence factor of each pixel in the shallow water candidate pixel set and the cloud expansion probability risk score of each pixel in the shallow water candidate pixel set. If the shallow water region confidence factor of a pixel in the shallow water candidate pixel set is higher than or equal to the shallow water region confidence factor threshold stored in the database and the cloud expansion probability risk score of that pixel in the shallow water candidate pixel set is higher than the cloud expansion probability risk score stored in the database, then the cloud expansion determination result of that pixel in the shallow water candidate pixel set is marked as cloud expansion. Otherwise, it is not necessary to mark the cloud expansion determination result of that pixel in the shallow water candidate pixel set as cloud expansion. In this way, the cloud expansion determination result of each pixel in the shallow water candidate pixel set is obtained.

[0042] It should be noted that the pixel-level cloud probability and expansion risk scores extracted directly from multi-source remote sensing and spatial big data imagery based on the trained spatiotemporal cloud shadow detector first extracts a pre-generated set of shallow water candidate pixels from the database, and then inputs the multi-temporal, multi-source images corresponding to these pixels into the cloud shadow detector. The spatiotemporal cloud shadow detector utilizes multi-temporal optical feature changes, spectral curve characteristics, and optional SAR backscattering information to assess the probability of cloud or shadow interference for each pixel in the current temporal phase, and combines historical observation data and spatial neighborhood information to calculate the expansion risk score. The cloud probability represents the likelihood of a pixel being covered by clouds, while the expansion risk score reflects the magnitude of the risk that cloud shadows may expand or encroach on near-shore / shallow water pixels.

[0043] like Figure 3 As shown, Figure 3 The flowchart of multi-source reconstruction and spectral analysis provided in this embodiment of the invention restores water body information affected by clouds by calling SAR data and obtains historical cloudless image data for temporal residual analysis. Based on the comparison result of the residual and a threshold, it is determined whether to perform spatiotemporal reconstruction and fusion processing. Subsequently, the spectral mixing value is calculated, and based on the mixing degree threshold, it is determined whether spectral decomposition and correction are needed to complete the spectral quality optimization of contaminated pixels. The specific process of determining whether to perform spatiotemporal reconstruction and fusion processing based on the comparison result of the residual and the threshold is as follows: the cloud expansion determination result of each pixel in the shallow water candidate pixel set is extracted as the pixel in the shallow water candidate pixel set corresponding to the cloud expansion, and marked as the first pixel in the shallow water candidate pixel set.

[0044] The SAR data of the first pixel in the shallow water candidate pixel set is called to recover the water information of each first pixel.

[0045] It should be noted that after the shallow water candidate pixel set is generated, the system first filters out pixels identified as being under cloud expansion based on the judgment results of the cloud expansion detection module, and marks them as "the first pixel in the shallow water candidate pixel set". Then, the system calls the SAR image data corresponding to these pixels on the same or recent dates to restore the water information obscured by cloud expansion. Specifically, the system first completes the spatial registration and resolution unification of the SAR image and optical image, mapping the SAR backscattering intensity values ​​to the optical coordinate grid. Since SAR signals can penetrate clouds, areas with low backscattering values ​​usually correspond to open water bodies, and the system extracts pixels that may exist below the cloud area. Then, through spatial overlap analysis with historical water body boundaries, local mean smoothing, and edge consistency constraints, the confidence level of these potential water body areas is assessed, filtering out pixels with significant noise interference. Finally, the water body areas inferred from the SAR are used to replace the corresponding obscured areas in the optical image.

[0046] Specifically, multi-source reconstruction of each pixel is performed based on the cloud expansion determination results of each pixel in the shallow water candidate pixel set. This also includes: after recovering the water information of each first pixel by calling the SAR data of each first pixel in the shallow water candidate pixel set, historical cloudless image data of each first pixel under similar observation conditions are retrieved from the database, including the SAR reflectance of each first pixel, and the estimated value of the SAR reflectance of each first pixel after recovery is obtained. The difference between the estimated value of the SAR reflectance of each first pixel after recovery and the historical average SAR reflectance of each first pixel is calculated to obtain the temporal residual of each first pixel. The temporal residual of each first pixel is compared with the temporal residual threshold stored in the database. If the temporal residual of a certain first pixel is higher than or equal to the temporal residual threshold, spatiotemporal reconstruction and fusion are performed on the first pixel.

[0047] It is important to note that in water body monitoring, SAR reflectance (usually referring to the backscattering coefficient) is closely related to the roughness, dielectric constant, and surface condition of the water surface. The smoother the water surface, the weaker the backscattering and the lower the reflectance value; conversely, if the water surface is disturbed (such as by waves, increased suspended solids concentration, vegetation cover, siltation, or human interference), the reflectance value will increase significantly. Therefore, in shallow water areas or at the water-land interface, when the estimated value of the recovered SAR reflectance is significantly higher than the historical average SAR reflectance of that pixel, it often means that the surface condition of the current pixel has changed abnormally. This change is usually not due to natural time-series fluctuations, but may be caused by insufficient recovery due to cloud expansion obscuring the surface, spectral unmixing errors, inaccurate water body identification, or misreconstruction of non-water body features. Conversely, if the recovered SAR reflectance value is slightly lower than the historical average, it usually does not cause significant problems. This is because a low SAR value corresponds to a smoother or deeper water condition, which is more common under natural conditions, such as reduced scattering due to decreased wind speed, reduced sediment, or a temporary increase in water depth. Such declines typically reflect short-term changes in the water body or meteorological conditions and do not indicate a calculation error. Therefore, in determining temporal consistency, the system primarily detects and triggers spatiotemporal reconstruction for "excessively high" deviations, rather than overreacting to "excessively low" deviations. This avoids excessive corrections caused by natural water body fluctuations and contributes to the authenticity of water body spectra and SAR features in the temporal dimension and the continuity in the spatial dimension.

[0048] It should be noted that the estimated SAR reflectance of each first pixel after restoration is obtained by extracting the backscattering intensity value corresponding to the first pixel in SAR images from the same or recent dates, and then dynamically correcting it by combining the SAR characteristics of surrounding cloudless pixels with historical statistical characteristics. Specifically, the system first locates the pixel position in the SAR data corresponding to the cloud expansion area in the optical image, and extracts its backscattering intensity (usually σ). 0 or γ 0The system then normalizes the value using the average scattering characteristics of cloudless water pixels in the neighborhood to reduce bias caused by local noise and differences in incident angle. Next, the system maps the normalized SAR intensity to the optical reflectivity space and inverts it using the SAR-optical reflectivity regression relationship established in historical time series or a bivariate distribution model based on similar time phases to obtain the optical reflectivity estimate of each first pixel at the current time.

[0049] It should be noted that when performing spatiotemporal reconstruction and fusion on the first pixel, a mask to be restored is first generated within the image coordinates, and its spatial topology and neighborhood information are recorded. Then, recoverable methods are tried in order of priority: pixel-level replacement is first performed using historical cloudless optical slices from the same region, from similar dates, and with the closest current observation conditions. If suitable historical slices are insufficient, spatial interpolation and hole filling based on neighboring cloudless pixels are performed, using Thrust Laplacian / Markov random field interpolation combined with edge-preserving filtering (such as guided filtering or dual-domain preservation) to maintain the connectivity and texture consistency of water body boundaries. At the same time, the SAR water body mask is used to constrain the water / non-water category of the interpolation results. When spatial interpolation is still insufficient to restore details, example filling or patch-based texture synthesis is called to extract similar textures from the neighborhood to fill the gaps. After completion, Poisson fusion or multi-resolution fusion is performed on the entire restored image to eliminate stitching artifacts. If none of the above automatic methods can bring the residuals back to an acceptable range, the system will mark the area as requiring manual review and highlight it in the visualization interface. At the same time, the source, parameters, and residual distribution of the recovery attempt will be recorded in the metadata for auditing and subsequent model improvement.

[0050] It should be noted that in the monitoring of temporal changes or occlusion recovery in shallow water areas, the system establishes a historical reflectance (or backscattering intensity) time series for each spatial pixel. These reflectance values ​​are derived from the observation results of that pixel at multiple time points (multi-temporal SAR images). Through this time series data, the variation patterns of that pixel under normal conditions can be analyzed, such as the reflectance change trend caused by seasonal changes in water depth, wind and wave conditions, or sediment concentration.

[0051] Specifically, the mixed values ​​of the pixel spectra of each pixel in the shallow water candidate pixel set in each band are obtained. The specific process is as follows: after completing the multi-source reconstruction of each pixel, all the recovered first pixels are uniformly incorporated into the shallow water candidate pixel set according to their image coordinates and spatial topology information. The standard spectral values ​​and proportion coefficients of each pixel in the shallow water candidate pixel set in each band are obtained and normalized and fused to obtain the mixed values ​​of the pixel spectra of each pixel in the shallow water candidate pixel set in each band. The mixed values ​​of the pixel spectra of each pixel in the shallow water candidate pixel set in each band are used for the spectral purity and mixing degree of the pixel.

[0052] It should be noted that the specific analysis conditions for the mixed values ​​of the pixel spectra of each pixel in each band in the shallow water candidate pixel set are as follows:

[0053] ;

[0054] In the formula, R i,b f represents the mixture value of the pixel spectrum of the i-th pixel in the b-th band within the shallow water candidate pixel set. i,k E represents the proportion coefficient of the i-th pixel in the k-th endmember in the shallow water candidate pixel set. k,b Min represents the standard spectral value of the k-th endmember in the b-th band within the shallow water candidate pixel set. j,b The function represents the minimum reconstructed spectral value of all pixels in all bands within the shallow water candidate pixel set. j,b The function represents the maximum value of the reconstructed spectral values ​​of all pixels in all bands in the shallow water candidate pixel set, j represents the index number of each pixel, j=1,2,…,N represents the total number of pixel index numbers, b represents the index number of each band, b=1,2,…,B, where B is the total number of bands, represents the index number of each endmember, k=1,2,…,K, where K is the total number of endmembers.

[0055] It is important to note that a band is a sampling point in the spectral dimension used to record the reflectance characteristics of an endmember at different wavelengths. It reflects the proportion of different physical components (endmembers) that make up the pixel, rather than the distribution between bands. Endmembers (such as clear water, suspended matter, and sediment) represent physically different land cover types or water components. They constitute the true composition of a pixel in space. The significance of the proportion coefficient lies in characterizing the spatial mixing ratio of each endmember in a pixel, not the distribution ratio in the spectral dimension. The band is merely a dimension for measuring the spectral characteristics of these components.

[0056] It should be noted that by finding the minimum and maximum values ​​of the reconstructed spectral values ​​of all pixels in all bands in the shallow water area, the spectral mixture values ​​of each pixel in each band in the shallow water candidate pixel set can reflect the physical composition of the pixels, that is, the actual contribution ratio of endmembers such as water, suspended matter, and bottom sediment in the pixels. This provides a basis for water quality parameter estimation. Furthermore, by reconstructing the spectrum and normalizing it, the influence of spectral noise and observation errors can be reduced, and the brightness differences between different bands and environmental influences (such as solar altitude angle, water depth, turbidity, etc.) can be weakened. This maintains the proportional relationship between water and bottom sediment in the original pixels, and the contrast of the shallow water area spectrum is enhanced through standardization. This provides a stable and reliable spectral input for subsequent watershed water ecological health classification, making the entire analysis process more accurate and comparable under the fusion of multi-source remote sensing and spatiotemporal data.

[0057] It should be noted that endmembers are standard spectral templates representing pure land cover types, such as water bodies, vegetation, or substrate. Their spectral characteristics are typical and stable across all bands, serving as an ideal combination to describe pixel spectra. In this formula, the endmember represents the basic spectral template used to reconstruct the pixel spectrum. Each endmember has a fixed standard spectral value in each band, while a pixel is typically a mixture of spectra from multiple land cover types. By directly using endmembers and proportion coefficients, complex pixel spectra can be decomposed into linear combinations of different land cover components, thereby quantifying the mixing degree and spectral purity of the pixel. At the same time, endmembers provide a unified reference, allowing the spectra of all pixels to be normalized under the same standard, facilitating subsequent mixing degree scoring, spectral analysis, and health classification.

[0058] Specifically, the pixel spectra of each pixel in the shallow water candidate pixel set are subjected to spectral decomposition and correction processing. The specific process is as follows: extract the mixed value of the pixel spectrum of each pixel in the shallow water candidate pixel set in each band, and compare it with the mixed value threshold of the pixel spectrum stored in the database. If the mixed value of the pixel spectrum of a certain pixel in the shallow water candidate pixel set is higher than or equal to the mixed value threshold of the pixel spectrum, then the pixel spectrum of that pixel in the shallow water candidate pixel set is subjected to spectral decomposition and correction processing. Otherwise, it is not necessary to perform spectral decomposition and correction processing on the pixel spectrum of the pixels in the shallow water candidate pixel set.

[0059] It should be noted that the spectral decomposition and correction of the pixel spectrum in the shallow water candidate pixel set specifically involves using historical multi-temporal images and the spectra of cloudless neighboring pixels to perform spectral correction on the unmixed pixels. The specific steps are as follows: When performing spectral correction on high-mixing pixels in the shallow water candidate pixel set, the system first extracts the spectra of cloudless water pixels in the neighborhood of each target pixel and calculates the average reflectance of each band as a local reference spectrum. Subsequently, for each band of the target pixel, its actual observed spectrum and the average spectrum of its neighborhood are smoothed. Short-term anomalous fluctuations are reduced using methods such as weighted moving average or Gaussian filtering, while considering spatial continuity to ensure smooth and natural spectral changes between adjacent pixels. Next, the system retrieves multi-temporal images of the pixel under the same or similar historical observation conditions to form a historical spectral sequence. By comparing the current temporal spectrum with the historical spectral sequence, the system can identify anomalous bands or values ​​that deviate from the historical pattern and perform retrospective correction when necessary, bringing the spectrum closer to the historical mean or trend line, thereby maintaining the temporal consistency and spatial continuity of the spectrum. During the correction process, the weights of each band will be dynamically adjusted. For anomalous bands that deviate significantly from the neighborhood or historical patterns, their impact on the overall correction will be reduced to avoid excessive interference from single-point anomalies in the estimation of water features.

[0060] like Figure 4 As shown, Figure 4The flowchart for spectral decomposition, correction, and water ecological health classification provided in this embodiment of the invention performs a spatiotemporal consistency check by calculating neighborhood residuals and historical time-series residuals. Based on the check results, it determines whether to construct a multidimensional feature matrix or perform manual verification. The feature matrix is ​​constructed by integrating the corrected spectral data, water component estimates, water depth information, and geospatial topology. A machine learning model is used to generate ecological health classification results, ultimately outputting a classification map, statistical reports, and a warning list. Specifically, the spatiotemporal consistency check by calculating neighborhood residuals and historical time-series residuals is a consistency check of spectral decomposition and correction. The specific process involves obtaining the values ​​of each pixel in the corrected shallow water candidate pixel set at each wavelength. The spectral reflectance value of the segment, the average spectral reflectance value of cloudless water pixels in the neighborhood of the pixel stored in the database, and the average spectral value of each pixel in the same band in historical multi-temporal images are used to obtain the neighborhood residual of each pixel in the corrected shallow water candidate pixel set by subtracting the neighborhood average spectral reflectance value of each band in the current temporal phase. At the same time, the historical mean or trend line value of the corrected spectral reflectance value of the pixel is subtracted to obtain the historical temporal residual of each pixel in the corrected shallow water candidate pixel set. The spatiotemporal consistency of the corrected spectrum is judged based on the neighborhood residual of each pixel in the corrected shallow water candidate pixel set and the historical temporal residual of each pixel in the corrected shallow water candidate pixel set.

[0061] It should be noted that subtracting the historical mean or trend line value from the corrected spectral reflectance value of a pixel refers to the difference calculation performed on each pixel in each band of the shallow water candidate pixel set. That is, within each band range of multispectral or hyperspectral data (e.g., blue, green, red, near-infrared bands), the corrected spectral reflectance of the current time phase is compared with the average spectral reflectance of the same band in historical multi-temporal images. The resulting historical temporal residual reflects the degree of deviation of each band in the time dimension and is used to assess the stability and consistency of the spectral changes of the pixel at different time scales. In the subsequent spatiotemporal consistency judgment, it is not only based on the residual of a single band, but also by comprehensively analyzing the joint distribution characteristics of the historical temporal residuals and neighborhood residuals of multiple bands. The overall consistency index of the residuals of each band is calculated to determine whether the spectral correction result of the pixel conforms to the spatiotemporal continuity law. If the residuals of most bands are within the normal range and the trend is consistent, the spectral correction is considered reliable; if there are significant anomalies in some bands, it is necessary to readjust the endmember ratio or re-perform spectral decomposition correction to ensure that the corrected spectrum is consistent with the neighborhood and historical time series in multiple band dimensions.

[0062] It should be noted that the spatiotemporal consistency of the corrected spectrum is determined based on the neighborhood residuals of each pixel in the corrected shallow water candidate pixel set and the historical time-series residuals of each pixel in the corrected shallow water candidate pixel set. Specifically, if the residual of a pixel in its spatial neighborhood is within a reasonable range and the historical time-series residual does not exceed a preset threshold, then the corrected spectrum of that pixel is considered to be spatiotemporally consistent. If any residual exceeds the corresponding threshold stored in the database, then the spectrum is considered to have spatiotemporal inconsistency, and the pixel needs to be iteratively corrected and included in the manual review process to ensure that the spectral characteristics of the entire shallow water candidate pixel set are reliable and continuous.

[0063] It should be noted that after smoothing and historical backtracking correction, the system performs a consistency check on the entire correction area to ensure that the corrected pixel spectrum is consistent with the spectrum of neighboring cloudless pixels and matches the historical spectral pattern, while preserving the spatial details and variation trends of the original spectrum. Finally, the corrected spectrum is updated in the shallow water candidate pixel set, providing reliable input for subsequent estimation of water quality parameters such as water depth, chlorophyll, and suspended matter, while ensuring the continuity and consistency of the spectral characteristics of the entire watershed.

[0064] Specifically, the process of generating water ecological health classification results is as follows: integrating and correcting spectral data, estimated values ​​of water body components in each pixel, water depth information, and geospatial topology data to construct a multi-dimensional water ecological feature matrix for the watershed; combining historical water quality and ecological monitoring data to establish a health indicator database; and then integrating the data and inputting it into a statistical analysis and machine learning model to generate water ecological health classification results.

[0065] It should be noted that after completing the spectral decomposition and correction of each pixel in the shallow water candidate pixel set, and performing spatiotemporal consistency verification through neighborhood residuals and historical time-series residuals, the system first extracts and integrates multi-source information for each pixel. This includes corrected spectral reflectance values ​​for each band, estimated water components obtained based on endmember unmixing (such as chlorophyll a concentration, suspended matter concentration, pigment index, etc.), water depth information obtained through optical or SAR inversion, and geospatial information such as the pixel's spatial coordinates and watershed topology. For the estimation of chlorophyll a concentration, the system first uses the spectral characteristics representing typical water components in the endmember library to perform linear or variable endmember unmixing, obtaining the proportion coefficient of each pixel on each endmember. One or more endmembers are highly correlated with chlorophyll a concentration. Then, these endmember proportion coefficients are used as input, combined with the corrected spectral reflectance, and mapped to chlorophyll a concentration using a spectral-water quality empirical formula. For estimating suspended solids concentration, the system first obtains endmember proportion coefficients representing suspended solids through endmember unmixing. These endmembers typically correspond to turbid water or sediment mixtures. These proportion coefficients are then combined with the calibrated spectrum, and a physics-based water model is used to calculate continuous values ​​of suspended solids concentration. The physics model involves the nonlinear relationship between water reflectance and suspended solids concentration, and uses spectral response characteristics to invert suspended solids content. The data-driven model learns the mapping relationship between spectral characteristics and suspended solids concentration using historical multi-temporal data, ensuring the model's applicability under different lighting, meteorological, and observation conditions. Regarding the estimation of pigment indices, the system first identifies characteristic bands reflecting water pigment content (such as combinations of blue, green, and red light). Subsequently, through a spectral-water quality relationship model, these spectral contributions are mapped to pigment index values. The key to this process lies in the accuracy of the endmember proportion coefficients and the precision of spectral calibration, ensuring that the pigment index reflects the true distribution of pigments in the water while maintaining consistency and spatial continuity over time. These parameters are used to construct multi-dimensional feature vectors at the watershed unit or pixel level. Each vector contains spectral features, chemical and physical parameters, water depth, and spatial relationship indicators, forming a multi-dimensional aquatic ecological feature matrix for the entire watershed. Historical water quality monitoring data and an ecological assessment index library are also incorporated as references for model calibration and training. Based on this, the system performs data preprocessing on the feature matrix, including missing value imputation, outlier removal, normalization, and feature selection, to ensure the consistency and comparability of the input data. Subsequently, machine learning models are used to model and predict the feature matrix. Supervised learning methods such as Random Forest, Gradient Boosting, Multilayer Perceptron (MLP), or lightweight convolutional neural networks (1D-CNN / Temporal CNN) can be selected. The input is the multi-source feature vector for each pixel or watershed unit, and the output is the corresponding ecological health level or continuous health index.The model is trained using historical health grading samples to learn the mapping relationship between water body characteristics, spectral information, and ecological health levels. Simultaneously, optimization methods such as cross-validation and grid search are used to adjust model parameters, ensuring the model's generalization ability across different watershed conditions and seasonal variations. During the prediction process, the multidimensional feature vector of each pixel is calculated by the trained model to obtain the probability distribution or continuous score of the ecological health level. Uncertainty indices, such as prediction confidence intervals or entropy values, are also calculated to mark potential risk or abnormal areas. Subsequently, the system performs spatial smoothing or topological constraints on the pixel-level prediction results based on spatial neighborhood and hydrological connectivity, ensuring the consistency of health levels within the same water body or river segment. Finally, the health levels of each pixel or watershed unit are summarized to generate a watershed-level ecological health grading map. Statistical indicators for each grade (such as area proportion, water coverage, and potential risk points) are also output. The model inputs, outputs, training parameters, and grading results are recorded in the database, achieving full-process data traceability. The principle behind this process is to integrate multi-source spectral and water quality information, and use machine learning models to capture the nonlinear correlation between spectral features, chemical and physical indicators, spatial topology, and ecological health, thereby achieving dynamic, accurate assessment and hierarchical management of the watershed's water ecological health status.

[0066] It should also be understood that the memory in the embodiments of the present invention can be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. The non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of random access memory (RAM) are available, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate synchronous DRAM (DDR SDRAM), enhanced synchronous DRAM (ESDRAM), synchronous linked DRAM (SLDRAM), and direct rambus RAM (DR RAM).

[0067] The above embodiments can be implemented, in whole or in part, by software, hardware (such as circuits), firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. A computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the flow or function according to the embodiments of the present invention is generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. Computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means. A computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. Available media can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media. Semiconductor media can be solid-state drives.

[0068] It should be understood that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. A and B can be singular or plural. Additionally, the character " / " in this article generally indicates an "or" relationship between the preceding and following related objects, but it can also represent an "and / or" relationship. Please refer to the context for a more accurate understanding.

[0069] In this invention, "at least one" means one or more, and "more than one" means two or more. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of a single item or a plurality of items. For example, at least one of a, b, or c can represent: a, b, c, ab, ac, bc, or abc, where a, b, and c can be a single item or multiple items.

[0070] It should be understood that, in various embodiments of the present invention, the order of the above-mentioned process numbers does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0071] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0072] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the devices, apparatuses, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0073] In the embodiments provided by this invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0074] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0075] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0076] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0077] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A watershed water ecological health classification method based on multi-source remote sensing and spatial big data, characterized in that, The method includes: Image sets of the water boundary region of the target watershed are obtained from a multi-source remote sensing and spatial big data platform, and a set of shallow water candidate pixels of the water boundary region of the target watershed is generated based on a unified grid. Cloud expansion determination analysis is performed on the shallow water candidate pixel set to obtain the cloud expansion determination results of each pixel in the shallow water candidate pixel set. Based on the cloud expansion determination results of each pixel in the shallow water candidate pixel set, multi-source reconstruction is performed on each pixel. The shallow water spectral mixture of each pixel in the shallow water candidate pixel set is analyzed to obtain the mixed value of the pixel spectrum of each pixel in each band. The pixel spectrum of each pixel in the shallow water candidate pixel set is subjected to spectral decomposition and correction processing, and the consistency of spectral decomposition and correction is checked to generate the water ecological health classification result. The cloud expansion determination analysis is performed on the candidate pixel set in shallow water. The specific process is as follows: A multi-temporal historical image time window is preset. Within the multi-temporal historical image time window, a shallow water candidate pixel set of historical images of the water body boundary area of ​​the target watershed is obtained. The average intersection ratio of cloud expansion and water body boundary and SAR reflectance are extracted. The number of times the intersection ratio of cloud expansion and water body boundary is higher than the threshold of the intersection ratio of cloud expansion and water body boundary stored in the database is counted and recorded as the cloud expansion event frequency. The average intersection ratio of cloud expansion and water body boundary, SAR reflectance and cloud expansion event frequency are normalized, reverse mapped and then fused to obtain the shallow water area confidence factor of each pixel in the shallow water candidate pixel set. The shallow water area confidence factor of each pixel in the shallow water candidate pixel set is used to quantify the reliability of the pixel belonging to the shallow water area. Meanwhile, based on the trained spatiotemporal cloud shadow detector, the cloud expansion probability risk score of each pixel in the shallow water candidate pixel set is directly extracted from multi-source remote sensing and spatial big data images.

2. The watershed water ecological health classification method based on multi-source remote sensing and spatial big data according to claim 1, characterized in that, The specific process of generating a shallow water candidate pixel set for the target watershed boundary region based on a unified grid is as follows: A multi-source standardized image set of the target watershed within a specified time period is obtained. A riverbank buffer zone is constructed based on the riverbank vector and digital elevation model. The riverbank buffer zone is projected onto a unified grid, and each grid cell in the unified grid is filtered to obtain a set of shallow water candidate pixels for the water boundary region of the target watershed, including each shallow water candidate pixel.

3. The watershed water ecological health classification method based on multi-source remote sensing and spatial big data according to claim 1, characterized in that, The specific process for obtaining the cloud expansion determination results for each pixel in the shallow water candidate pixel set is as follows: Extract the shallow water region confidence factor of each pixel in the shallow water candidate pixel set and the cloud expansion probability risk score of each pixel in the shallow water candidate pixel set. If the shallow water region confidence factor of a pixel in the shallow water candidate pixel set is higher than or equal to the shallow water region confidence factor threshold stored in the database and the cloud expansion probability risk score of that pixel in the shallow water candidate pixel set is higher than the cloud expansion probability risk score stored in the database, then the cloud expansion determination result of that pixel in the shallow water candidate pixel set is marked as cloud expansion. Otherwise, it is not necessary to mark the cloud expansion determination result of that pixel in the shallow water candidate pixel set as cloud expansion. In this way, the cloud expansion determination result of each pixel in the shallow water candidate pixel set is obtained.

4. The watershed water ecological health classification method based on multi-source remote sensing and spatial big data according to claim 3, characterized in that, The process of multi-source reconstruction of each pixel based on the cloud expansion determination results of each pixel in the shallow water candidate pixel set is as follows: The cloud expansion determination result of each pixel in the shallow water candidate pixel set is extracted as each pixel in the shallow water candidate pixel set corresponding to cloud expansion, and marked as the first pixel in the shallow water candidate pixel set. The SAR data of the first pixel in the shallow water candidate pixel set is called to recover the water information of each first pixel.

5. The watershed water ecological health classification method based on multi-source remote sensing and spatial big data according to claim 4, characterized in that, The step of performing multi-source reconstruction on each pixel based on the cloud expansion determination results of each pixel in the shallow water candidate pixel set also includes: After recovering the water information of each first pixel by calling the SAR data of each first pixel in the shallow water candidate pixel set, historical cloudless image data under similar observation conditions of each first pixel are retrieved from the database, including the SAR reflectance of each first pixel. The estimated SAR reflectance of each first pixel after recovery is obtained. The difference between the estimated SAR reflectance of each first pixel after recovery and the historical average SAR reflectance of each first pixel is calculated to obtain the temporal residual of each first pixel. The temporal residual of each first pixel is compared with the temporal residual threshold stored in the database. If the temporal residual of a certain first pixel is higher than or equal to the temporal residual threshold, spatiotemporal reconstruction and fusion are performed on the first pixel.

6. The watershed water ecological health classification method based on multi-source remote sensing and spatial big data according to claim 1, characterized in that, The specific process for obtaining the mixed values ​​of the pixel spectra of each pixel in each band in the shallow water candidate pixel set is as follows: After completing the multi-source reconstruction of each pixel, all the recovered first pixels are uniformly incorporated into the shallow water candidate pixel set according to their image coordinates and spatial topology information. The standard spectral values ​​and proportion coefficients of each pixel in the shallow water candidate pixel set in each band are obtained and normalized and fused to obtain the mixed value of the pixel spectrum of each pixel in the shallow water candidate pixel set in each band. The mixed value of the pixel spectrum of each pixel in the shallow water candidate pixel set in each band is used for the spectral purity and mixing degree of the pixel.

7. The watershed water ecological health classification method based on multi-source remote sensing and spatial big data according to claim 1, characterized in that, The specific process of performing spectral decomposition and correction on the pixel spectra of each pixel in the shallow water candidate pixel set is as follows: Extract the mixed values ​​of the pixel spectra of each pixel in each band in the shallow water candidate pixel set, and compare them with the mixed value threshold of the pixel spectra stored in the database. If the mixed value of the pixel spectrum of a certain pixel in the shallow water candidate pixel set is higher than or equal to the mixed value threshold of the pixel spectrum, then perform spectral decomposition and correction processing on the pixel spectrum of that pixel in the shallow water candidate pixel set; otherwise, it is not necessary to perform spectral decomposition and correction processing on the pixel spectrum of the pixels in the shallow water candidate pixel set.

8. The watershed water ecological health classification method based on multi-source remote sensing and spatial big data according to claim 1, characterized in that, The specific process for verifying the consistency between spectral decomposition and correction is as follows: The spectral reflectance values ​​of each pixel in the corrected shallow water candidate pixel set in each band are obtained, along with the average spectral reflectance values ​​of cloudless water pixels in the neighborhood of that pixel stored in the database, and the average spectral values ​​of each pixel in the same band in historical multi-temporal images. The neighborhood average spectral reflectance value is subtracted from the corrected spectral reflectance value of each band in the current temporal phase to obtain the neighborhood residual of each pixel in the corrected shallow water candidate pixel set. At the same time, the historical mean or trend line value is subtracted from the corrected spectral reflectance value of that pixel to obtain the historical temporal residual of each pixel in the corrected shallow water candidate pixel set. The spatiotemporal consistency of the corrected spectrum is determined based on the neighborhood residuals and the historical temporal residuals of each pixel in the corrected shallow water candidate pixel set.

9. The watershed water ecological health classification method based on multi-source remote sensing and spatial big data according to claim 1, characterized in that, The specific process for generating the water ecological health classification results is as follows: By integrating and correcting spectral data, estimated water components for each pixel, water depth information, and geospatial topology data, a multidimensional water ecological feature matrix for the watershed is constructed. A health indicator database is established by combining historical water quality and ecological monitoring data. The data is then fused and input into statistical analysis and machine learning models to generate water ecological health classification results.

Images