Classification mapping method of wetland by combining with the fusion of the cci and the aquatic vegetation isolation index

Abstract

The application discloses a wetland classification mapping method fusing a veining cap conversion and a hydrophyte separation index. The method of the application is not constructed for a single target ground object extraction method, but is faced with a complex scene of coexistence of multiple ecological components in a shallow wetland, a hierarchical and step-by-step comprehensive identification process is established, and the collaborative extraction of multiple ground objects such as terrestrial vegetation, hydrophyte, open water, submerged vegetation and algae bloom can be realized.

Description

Technical Field

[0001] This invention relates to the field of remote sensing information extraction and wetland ecological monitoring technology, and in particular to a wetland classification and mapping method that integrates tassel cap transformation and aquatic vegetation separation index. Background Technology

[0002] Shallow eutrophic lakes are ecologically dynamic aquatic systems, susceptible to the combined effects of climate change and human activities. The ecological evolution of these lakes manifests not only in changes in overall water area but also in the interconversion, spatial reorganization, and altered distribution patterns of various ecological components, including open water, aquatic vegetation, terrestrial vegetation, and algal blooms. For example, open water surfaces may shrink, accompanied by the expansion of aquatic or terrestrial vegetation; exposed mudflats may be transformed into farmland or other types of land cover under human disturbance; and changes in water level, hydrological processes, and trophic status can further lead to the substitution or fluctuation between submerged vegetation and algal blooms. These changes significantly impact habitat structure, ecosystem stability, and water quality.

[0003] For this type of wetland ecosystem, monitoring only the area change of a single land cover type is insufficient to fully reflect wetland ecological processes and their evolutionary patterns. Therefore, long-term, continuous, and synchronous monitoring of multiple ecological components has become a crucial technical requirement for wetland ecological management and water environment assessment. While traditional field surveys or manual visual interpretation methods can obtain land cover information with a certain degree of accuracy, they suffer from problems such as high workload, high implementation costs, and poor repeatability in large-scale, long-term monitoring tasks, making them unsuitable for long-term operational applications. Remote sensing technology, due to its advantages of large-scale, repeated observations and continuous time-series acquisition, has been widely used for the extraction and monitoring of objects such as water bodies, aquatic vegetation, and algal blooms.

[0004] Most existing wetland remote sensing extraction methods still focus on single-target enhancement, with common methods including the Normalized Difference Water Index (NDDI), Modified Normalized Difference Water Index (MDDI), Normalized Difference Vegetation Index (NDVE), and Phytoplankton Index (PDI). These indices are typically constructed based on the differences between a specific target and the background, thus performing well in binary or few-class tasks such as water body extraction, vegetation extraction, or algal bloom identification. However, in shallow wetland environments, multiple land cover types often coexist with complex boundary transitions, and significant spectral confusion exists between different ecological components. Relying solely on indices constructed for a single target often fails to accurately distinguish multiple components simultaneously. In particular, existing general-purpose indices typically lack sufficient discriminative power between aquatic vegetation and adjacent terrestrial vegetation, and between open water and submerged biological features, thus limiting the accuracy of multi-component comprehensive identification.

[0005] To reduce interference from non-target features, existing technologies employ several task-specific processing approaches. First, they limit the extraction of target features based on prior ecological zoning or preset spatial distribution ranges. Second, they construct specific indices, discriminant functions, or classification rules for specific objects of confusion, such as distinguishing between aquatic vegetation and water bodies, or between algal blooms and large aquatic plants. Third, they utilize phenological differences or temporal persistence among different features to suppress non-target interference through long-term image series. While these methods can improve the extraction of one or several types of targets in specific scenarios, most still focus on single targets or limited combinations of categories. They struggle to address the complex situations in shallow wetlands where multiple ecological components coexist, transition between each other, and exhibit hierarchical confusion, thus failing to establish a unified classification framework that is interpretable, feasible, and applicable in the long term.

[0006] Furthermore, while machine learning methods can handle complex classification boundaries by introducing more features and demonstrate strong capabilities in multi-class classification tasks, their classification performance typically depends heavily on the design of input features and the quality of representative training samples. In long-term monitoring scenarios of shallow wetlands, difficulties in obtaining training samples, significant temporal differences, and insufficient feature stability still limit the long-term widespread application of such methods. Therefore, current technologies still lack a comprehensive mapping method for the complex ecological components of shallow wetlands that can balance multi-class identification, temporal applications, and physical-ecological interpretability. Summary of the Invention

[0007] The purpose of this invention is to address the technical shortcomings of existing shallow wetland remote sensing classification methods, which are mainly designed for single targets and are difficult to adapt to the comprehensive identification of multiple ecological components and complex transitional zone conditions. This invention provides a wetland classification and mapping method that integrates tassel cap transformation and aquatic vegetation separation index.

[0008] The technical solution adopted to achieve the purpose of this invention is: A wetland classification and mapping method integrating tassel cap transformation and aquatic vegetation separation index includes the following steps: Step 1: Acquire remote sensing image data of the study area and preprocess the remote sensing image data; Step 2: Perform tassel transformation on the preprocessed remote sensing image data to obtain the greenness component and the humidity component; the greenness component is used for the initial separation of vegetation and non-vegetation, and the humidity component is used for further differentiation of land-related features and water-related features in non-vegetation areas.

[0009] Step 3: Separate vegetated areas from non-vegetated areas based on the greenness component; Step 4: In non-vegetated areas, the humidity component is used to separate the land area and the water area. Step 5: Within the vegetation area, considering that both aquatic and terrestrial vegetation have strong spectral responses and conventional vegetation indices are difficult to distinguish effectively, the aquatic vegetation separation index (AVSI) is used in conjunction with the phenological characteristics of the study area to separate the aquatic vegetation area and the terrestrial vegetation area. Step 6: Within the water body area, the aquatic vegetation separation index is used to separate the open water area from the submerged biological characteristic area. Step 7: Within the submerged biological characteristic area, the submerged vegetation area and the algal bloom area are separated using the time persistence feature. Step 8: Based on the region separation results, output the wetland classification and mapping results. In the above technical solution, step 1, the process of acquiring and preprocessing remote sensing image data, is as follows: acquiring multi-temporal and multispectral remote sensing image data of the study area, performing image quality control, cloud masking, and effective pixel screening quality control on the remote sensing image data, retaining images that meet the effective pixel ratio requirements; and using the quality control band to perform cloud masking to improve the data quality for subsequent classification.

[0010] In the above technical solution, the multi-temporal multispectral remote sensing image data is Landsat 8 surface reflectance image from June to September each year.

[0011] In the above technical solution, in step 3, the vegetation coverage in the image is initially extracted by utilizing the significant response differences of vegetation in the red light band and near-infrared band, combined with the greenness component. The initial extraction is mainly based on significant vegetation signals, which are dominated by emergent or high-canopy vegetation. Low-greenness backgrounds such as open water bodies, land, submerged vegetation, and algal blooms are retained for further discrimination in subsequent steps.

[0012] In the above technical solution, during the vigorous growth period of vegetation from July to September, an adaptive threshold method based on the distribution characteristics of greenness components is used to determine the current threshold and separate vegetated areas from non-vegetated areas; during the early growth stage in June, an empirically stable threshold is used to separate vegetated areas from non-vegetated areas. This step prioritizes the extraction of significant vegetation signals dominated by emergent or high-canopy vegetation.

[0013] In the above technical solution, in step 4, considering that the land-water boundary in shallow wetlands is usually not a sharp boundary, but a wide transition zone composed of saturated mudflats, shallowly flooded areas and mixed pixels, an adaptive threshold determination method based on boundary gradient constraints is used to separate the land area and the water area.

[0014] In the above technical solution, the specific process of the adaptive threshold determination method based on boundary gradient constraints is as follows: representative pure sample areas of two adjacent target classes are selected by visual interpretation; a class transition buffer is constructed according to the spatial contact relationship of the representative pure sample areas of the two target classes; the spatial gradient of the humidity component is calculated in the class transition buffer; the pixel with the largest local gradient is selected as the boundary candidate pixel; and the classification threshold in the current scene is determined according to the statistical characteristics of the humidity component of the boundary candidate pixel.

[0015] In the above technical solution, the expression for the aquatic vegetation separation index in step 5 is:

[0016] in, For near-infrared reflectivity, For green light band reflectivity, This index represents the reflectance in the red light band. By jointly incorporating near-infrared and green light responses and suppressing the red light reflectance enhancement effect, it enhances subtle spectral differences between aquatic and terrestrial vegetation while reducing the impact of brightness differences on classification results.

[0017] In the above technical solution, step 5 specifically includes the following steps: Step 5.1: Select the multispectral remote sensing image in June as the dominant time phase for emergent vegetation identification. Under this dominant time phase, the aquatic vegetation has formed a relatively obvious canopy, while the surrounding farmland vegetation has not yet reached its growth peak, thereby improving the separability between aquatic and terrestrial vegetation.

[0018] Step 5.2: Determine the threshold for the dominant phase of each year based on the spatial gradient of the aquatic vegetation separation index, and use the multi-year threshold statistics to determine the final segmentation threshold in order to extract the annual emergent vegetation range and separate the aquatic vegetation area from the terrestrial vegetation area.

[0019] In the above technical solution, in step 6, within the water body area, an adaptive threshold based on boundary gradient constraints is adopted. By selecting the transition region between open water and submerged biological features, the AVSI gradient is calculated and the scene threshold is determined to improve the ability to distinguish between open water and submerged biological features, thus differentiating between open water and submerged biological features.

[0020] In the above technical solution, the open water body includes clear water and turbid water, and its spectrum is mainly affected by water scattering and suspended matter; the characteristics of the submerged organisms are affected by the bio-optical effects caused by the biomass of aquatic vegetation or algae, and its spectrum is affected by the red light absorption characteristics relative to green light.

[0021] In the above technical solution, in step 7, submerged vegetation is typically rooted or attached to the substrate, and its spatial location is relatively stable during the growing season; algal blooms, on the other hand, are easily affected by wind fields and hydrodynamic conditions, and their spatial distribution exhibits strong drift and intermittency. Based on multi-temporal and multispectral remote sensing image data from June to September, the continuous occurrence of submerged biological feature pixels throughout the growing season is determined: if a pixel appears consistently and stably in multiple temporal phases, it is identified as submerged vegetation; if it appears only sporadically or its spatial location is unstable, it is identified as algal bloom. This step is only performed within the scope of features already identified as submerged biological features to reduce the risk of misjudgment in complex aquatic environments.

[0022] In the above technical solution, in step 8, the wetland classification and mapping results include at least terrestrial vegetation, emergent or floating aquatic vegetation, open water bodies, submerged vegetation, and algal blooms, so as to realize comprehensive mapping and long-term dynamic monitoring of multiple ecological components of shallow wetlands.

[0023] Another aspect of the present invention includes an electronic device comprising one or more processors and a memory; the memory is used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the wetland classification and mapping method described above.

[0024] Another aspect of the present invention includes a computer program product comprising computer-executable instructions which, when executed, implement the wetland classification and mapping method.

[0025] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention is not aimed at constructing and extracting a single target feature, but rather at the complex scenario of multiple ecological components coexisting in shallow wetlands. It establishes a hierarchical and step-by-step comprehensive identification process, which can achieve the collaborative extraction of multiple types of features such as terrestrial vegetation, aquatic vegetation, open water bodies, submerged vegetation, and algal blooms.

[0026] 2. This invention utilizes the greenness component and humidity component in the tassel transformation to undertake the tasks of initial vegetation classification and land-water separation, respectively, so that different classification steps are matched with the confusion relationship of corresponding land features, thereby improving the interpretability of the method.

[0027] 3. The Aquatic Vegetation Separation Index (AVSI) proposed in this invention integrates near-infrared, green, and red light band information, which can enhance the difference between aquatic and terrestrial vegetation and improve the separation ability between open water bodies and submerged biological characteristics.

[0028] 4. The present invention adopts an adaptive threshold strategy with boundary gradient constraints, which makes the threshold determination focus on the actual transition interface between adjacent categories, reducing the sensitivity of fixed empirical thresholds to different scenarios, seasons and background conditions.

[0029] 5. This invention combines the dominant phenological window in June with multi-temporal continuity information, which can not only improve the identification effect of emergent vegetation, but also further distinguish submerged vegetation from algal blooms, thereby improving the integrity and stability of the comprehensive mapping of complex ecological components in shallow wetlands.

[0030] 6. This invention is applicable to long-term operational monitoring based on time-series remote sensing images, and has good feasibility and application value. Attached Figure Description

[0031] Figure 1 This is a schematic flowchart of the wetland integrated mapping method of the present invention.

[0032] Figure 2 This is a schematic diagram of threshold determination based on boundary gradient constraints in this invention.

[0033] Figure 3 The diagram shows the spectral characteristics and threshold gradient distribution of typical land cover types. (a) shows the spectral characteristic curves of five land cover types, (b) shows the pixel frequency distribution and threshold determination of the greenness index, (c) shows the bar chart of the maximum pixel gradient of the humidity index, and (d) shows the bar chart of the maximum pixel gradient of the aquatic vegetation index.

[0034] Figure 4 The spatial distribution of major land features in Wuliangsuhai Lake in different months.

[0035] Figure 5 The images show the spatial distribution of multi-temporal remote sensing indices for 2015, 2016, 2020, and 2022, where (a) represents NDVI and (b) represents AVSI of this invention.

[0036] Figure 6 Among them, (a) is the AVSI of the present invention, and the greenness index, humidity index and brightness index used, and (b) is the box plot comparison results of the remote sensing index classification performance of the Normalized Difference Vegetation Index, the Improved Normalized Difference Water Index and the Aquatic Vegetation Separation Index.

[0037] Figure 7 The classification results of different land cover buffer zones are compared with the JRC-GSW data. Among them, (a) is the consistency verification result of the classification of the lake boundary area and emergent plant area based on JRC-GSW data, (b) is the Landsat 8 true color image of Wuliangsuhai area in August 2020, (c) is the Landsat 8 wetland classification result of Wuliangsuhai area in August 2020, and (d) is the Landsat 8 JRC data of Wuliangsuhai area in August 2020.

[0038] Figure 8This represents the variation of F1-score in the classification accuracy of various regions under long-term time series conditions.

[0039] Figure 9 The stability analysis of the classification results under threshold perturbation is shown in (a), where (b) shows the change of F1 value of each category under different greenness threshold perturbation, and (c) shows the change of F1 value of each category under different aquatic vegetation classification index threshold perturbation.

[0040] Figure 10 The results show the spatial distribution of submerged vegetation verification samples. (a) shows the distribution of sampling points from the field survey, (b) shows the spatial distribution of aquatic vegetation classification index compared with reference sampling points, and (c) shows the distribution of reference sampling points in the monitoring area of ​​submerged vegetation in Wuliangsuhai Lake based on publicly reported existing research in different years.

[0041] Figure 11 This is a schematic diagram showing the classification results of Wuliangsuhai Wetland. Detailed Implementation

[0042] The present invention will be further described in detail below with reference to specific embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0043] Example 1 like Figure 1 As shown, this embodiment uses the shallow wetland of Wuliangsuhai Lake as the application object to comprehensively identify multiple types of ecological components, including open water, land, terrestrial vegetation, aquatic vegetation, submerged vegetation, and algal blooms. This method is based on multi-temporal multispectral remote sensing images and follows a step-by-step classification approach from coarse to fine, progressively distinguishing different categories.

[0044] A wetland classification and mapping method integrating tassel cap transformation and aquatic vegetation separation index includes the following steps: Step 1: Acquire remote sensing image data of the study area and preprocess the remote sensing image data; Landsat 8 surface reflectance imagery was selected as the base remote sensing data. To ensure consistency in time series length, spatial resolution, and data quality, monthly average surface reflectance imagery for the period from 2013 to 2025 was acquired using the Google Earth Engine platform. Considering the long ice cover period of Wuliangsuhai Lake, the imagery selection period was limited to June to September each year to reduce the impact of ice and snow cover on wetland classification results. The Landsat surface reflectance products used have undergone atmospheric correction and are suitable for inland water remote sensing applications.

[0045] Furthermore, cloud masking was applied to the remote sensing image data using the QA_PIXEL quality control band, and the effective pixel ratio after cloud masking was calculated. Only images with an effective pixel ratio greater than 80% were retained for subsequent classification analysis. After screening, a total of 40 images were retained to construct a growing season time series dataset.

[0046] Step 2 involves performing a tasseled cap transformation on the preprocessed remote sensing image data to obtain greenness and humidity components. The tasseled cap transformation is an orthogonal linear transformation, its function being to compress the multispectral remote sensing image data into a small number of feature components related to the main physical properties of land features. Specifically, the brightness component characterizes the overall surface reflectance intensity; the greenness component enhances vegetation-related responses; and the humidity component characterizes differences in soil, water bodies, and moist background conditions. The greenness component is used for initial separation of vegetation and non-vegetation areas, while the humidity component is used for further delineation of land and water bodies within non-vegetation areas.

[0047] Step 3: Initial separation of vegetation (vegetation component) and non-vegetation (non-vegetation component) in the study area is performed using the greenness component. This is based on the significant contrast of the green vegetation canopy in the red and near-infrared bands. Although algal blooms and submerged vegetation also have photosynthetic activity, under shallow lake conditions, due to water decay and the influence of surface wetting background, their effective near-infrared response is usually weaker than that of emergent vegetation, and therefore they often appear as lower background values ​​in the greenness component space.

[0048] During the period from July to September each year, due to the full development of emergent vegetation canopy, the greenness component in the wetland area usually shows an approximately unimodal distribution, accompanied by a low-value tail composed of algal blooms, submerged vegetation, open water bodies and land.

[0049] Furthermore, the triangular thresholding method was used to determine the greenness component threshold in the images from July to September to separate vegetated and non-vegetated areas. During June each year, since emergent vegetation is still in the early growth stage and the background is relatively separable, a fixed greenness threshold of 0 was used to separate vegetation from non-vegetation.

[0050] This step prioritizes the extraction of signals from emergent or high-canopy vegetation, while preserving low-greenness backgrounds such as algal blooms, submerged vegetation, open water bodies, and land for further identification in subsequent steps.

[0051] Step 4: In non-vegetated areas, the humidity component is used to separate the land area and the water area. Furthermore, since there are usually obvious transition zones between different categories in shallow wetlands, and the category boundaries are greatly affected by season, light, water level, and background humidity, an adaptive threshold determination method based on boundary gradient constraints is adopted in multiple segmentation steps to improve cross-scene classification stability. Specifically, the steps include the following: Adjacent pure water and pure soil sample areas were selected through visual interpretation to construct a land-water transition buffer zone. Then, the spatial gradient of the humidity component was calculated within the buffer zone, and the pixel with the largest local gradient was selected as the boundary candidate pixel. A classification threshold was determined based on the statistical characteristics of the humidity component of the boundary candidate pixels, thereby improving the stability of land-water separation under different scenarios and reducing the omission of pixels in shallow water areas and transition interfaces, thus separating land and water areas. Specifically, the classification threshold is the mean of the humidity component of the boundary candidate pixels minus one standard deviation. The final humidity component threshold used is -0.00387. This threshold is slightly offset towards the land side, thereby ensuring the land-water separation effect while reducing the omission of pixels in shallow water areas and saturated interfaces into land areas. Figure 3 ).

[0052] This step is applicable to segmentation tasks where adjacent categories have a clear contact interface in the image space. When the two target classes do not contact each other or the boundary is too fragmented at the sensor's spatial resolution, the boundary candidate pixels may not accurately represent the actual transition interface. Therefore, this method is mainly applied to segmentation scenarios in Wuliangsuhai Lake where adjacent categories are continuously distributed.

[0053] Furthermore, by Figure 2 The threshold determination process based on boundary gradient constraints is as follows: First, for the land and water regions to be separated, diagnostic variables that can well characterize the differences between the land and water regions and have obvious variation characteristics in their boundary areas are selected, and the spatial distribution map of the diagnostic variables is calculated. Then, representative clean pixels of the land and water regions are obtained based on visual interpretation, and clean feature 1 pixels (masks) corresponding to the land region and clean feature 2 pixels (masks) corresponding to the water region are constructed respectively. Then, according to the spatial positional relationship between clean feature 1 pixels and clean feature 2 pixels, a transition buffer is constructed in the contact area between the two.

[0054] Based on this, the spatial gradient of the diagnostic variables is calculated only within the transition buffer, and the pixel with the largest local gradient is extracted as the boundary candidate pixel. Finally, the values ​​of the diagnostic variables corresponding to the boundary candidate pixels are statistically analyzed, and the segmentation threshold for the current scene is determined accordingly. Since this method limits the threshold extraction to the actual transition interface between the two types of land features, it can reduce the influence of irrelevant background pixels, global noise, and outliers on the threshold determination result, thereby improving the accuracy and robustness of the threshold determination.

[0055] like Figure 3As shown, in step 5, within the vegetation area, considering the problem that both aquatic and terrestrial vegetation have strong spectral responses and are difficult to distinguish effectively using conventional vegetation indices, the Aquatic Vegetation Separation Index (AVSI) is used in conjunction with phenological characteristics within the study area to separate the aquatic vegetation area and the terrestrial vegetation area; step 5 specifically includes the following steps: Step 5.1, considering the phenological conditions of Wuliangsuhai Lake, according to the field survey results from late May to June 2025, the emergent and floating aquatic vegetation of Wuliangsuhai Lake had formed a relatively dense canopy at this time, while the main crops around the study area had not yet reached the peak of canopy growth, and some farmland was still in a state of waterlogging or sparse vegetation. Therefore, the separability between aquatic and terrestrial vegetation was most obvious in June. Figure 4 Therefore, the multispectral remote sensing imagery of June was selected as the dominant time phase for emergent vegetation identification. Under this dominant time phase, the aquatic vegetation has formed a relatively obvious canopy, while the vegetation in the surrounding farmland has not yet reached its growth peak, thereby improving the separability between aquatic and terrestrial vegetation.

[0056] Step 5.2: For each year's dominant phase, a threshold is determined based on the spatial gradient of the aquatic vegetation separation index. The final segmentation threshold is then determined using multi-year threshold statistics to extract the annual emergent vegetation range and separate the aquatic and terrestrial vegetation regions. The final threshold is 0.32346 (…). Figure 3 The extent of emergent and floating aquatic vegetation extracted from June imagery using this threshold was used as a proxy for the extent occupied in that year. Furthermore, the expression for the aquatic vegetation separation index is: ; in, For near-infrared reflectivity, For green light band reflectivity, This index represents the reflectance in the red light band. By jointly incorporating near-infrared and green light responses and suppressing the red light reflectance enhancement effect, it enhances subtle spectral differences between aquatic and terrestrial vegetation while reducing the impact of brightness differences on classification results.

[0057] like Figure 5 As shown, in step 6, within the water body region (water body characteristic component), the open water body region (open water region) and the submerged biological characteristic region are separated using the aquatic vegetation separation index; wherein, the open water body includes clear water and turbid water, and its spectrum is mainly affected by water scattering and suspended matter; the submerged biological characteristics are affected by the bio-optical effects caused by the biomass of aquatic vegetation or algae, and its spectrum is affected by the red light absorption characteristics relative to green light.

[0058] The specific process is as follows: AVSI (Ambient Air Quality Index) is used as the key discriminant feature between open water and submerged biological features. First, open water and submerged biological feature regions are selected visually, and pure endmember pixels are removed, retaining only the uncertain transition zone between the two types of regions. Then, the AVSI spatial gradient is calculated within the transition zone, and the pixel with the largest local gradient is selected as the boundary candidate pixel. Finally, the segmentation threshold between open water and submerged biological features is determined based on the AVSI statistics of the boundary candidate pixels, separating the open water and submerged biological feature regions. Specifically, the threshold is the mean of the AVSI of the boundary candidate pixels minus one standard deviation. In this embodiment, the final AVSI threshold used is 0.10738 (…). Figure 3 This threshold is slightly offset towards the open water side, which helps to preserve the weaker submerged vegetation signal under shallow or turbid conditions.

[0059] like Figure 4 As shown, in step 7, the submerged vegetation area and the algal bloom area are separated within the submerged biological characteristic area using the time persistence feature.

[0060] Furthermore, since submerged vegetation is typically rooted or attached to the substrate, it maintains a relatively stable spatial location during the growing season, while algal blooms are significantly affected by wind and hydrodynamic forces, resulting in a highly mobile and intermittent spatial distribution. Therefore, this embodiment utilizes multi-temporal duration characteristics to distinguish between the two. Submerged vegetation is typically rooted or attached to the substrate, maintaining a relatively stable spatial location during the growing season; algal blooms, on the other hand, are easily affected by wind and hydrodynamic conditions, exhibiting a highly mobile and intermittent spatial distribution. Based on multi-temporal, multispectral remote sensing image data from June to September, the continuous occurrence of submerged biomarker pixels throughout the growing season is assessed: if a pixel appears consistently and stably across multiple temporal phases, it is identified as submerged vegetation; if it appears only sporadically or has an unstable spatial location, it is identified as algal bloom. This step is only performed within the scope of pixels already identified as submerged biomarkers to reduce the risk of misjudgment in complex aquatic environments.

[0061] like Figure 11 As shown, in step 8, based on the regional separation results, the wetland classification and mapping results are output, resulting in algal bloom areas and submerged plant areas. The wetland classification and mapping results include at least terrestrial vegetation, emergent or floating aquatic vegetation, open water bodies, submerged vegetation, and algal blooms, thereby achieving comprehensive mapping and long-term dynamic monitoring of multiple ecological components of shallow wetlands.

[0062] like Figure 5As shown, compared with the traditional Normalized Difference Vegetation Index (NDVI), the Aquatic Vegetation Separation Index (AVSI) proposed in this invention is more suitable for identifying floating emergent vegetation in lake wetland environments. Its advantages mainly stem from its enhanced targeting of the spectral characteristics of aquatic vegetation and complex background conditions. The traditional NDVI mainly utilizes the difference between the near-infrared band and the red band to characterize vegetation growth status, making it more suitable for monitoring terrestrial vegetation. In lake wetland environments, both floating emergent vegetation and shoreline terrestrial vegetation typically exhibit high near-infrared reflectance and low red reflectance, thus easily showing similar index values, leading to blurred classification boundaries. Furthermore, this index is quite sensitive to changes in background brightness, and its stability further decreases under the influence of wet soil, water-mixed pixels, shadows, and different substrate conditions.

[0063] Meanwhile, the aquatic vegetation separation index proposed in this invention further incorporates green light band information and suppresses high red light reflectance characteristics. Because floating emergent vegetation is affected by the aquatic environment, it typically exhibits reflectance characteristics in the green light band that differ from terrestrial vegetation. Furthermore, the aquatic background modulates the red and near-infrared bands. Therefore, this invention enhances the synergistic response of near-infrared and green light bands to aquatic vegetation through normalized combination, while simultaneously reducing red light background interference, thereby effectively improving the spectral separability between floating emergent vegetation and terrestrial vegetation.

[0064] To further verify the performance and applicability of the method of this invention, the Normalized Difference Vegetation Index (NDVI), the Modified Normalized Difference Water Index (MNDWI), and the aquatic vegetation separation index proposed in this invention were selected for comparative analysis. The results are as follows: Figure 6As shown in the figure, the results indicate that while the traditional normalized vegetation index (NVI) can enhance vegetation information, significant overlap exists between different categories due to the high responses of both terrestrial and floating emergent vegetation, leading to misclassification, especially in complex water-land transition zones. The improved normalized differential water body index (NDDI) can better enhance water body information, but its ability to distinguish between different aquatic vegetation types is limited, and some overlap still exists in areas covered by emergent vegetation. In contrast, the aquatic vegetation separation index proposed in this invention exhibits better separation characteristics between different categories. Open water bodies generally show lower index values, floating emergent vegetation has the highest index response, terrestrial vegetation is in the middle range, and submerged vegetation shows a relatively independent low value range, significantly reducing the overlap between categories. This demonstrates that this invention can effectively enhance the spectral differences between aquatic and terrestrial vegetation while reducing the interference of background brightness, water body mixed pixels, and complex wetland environments on the classification results. Furthermore, the box plot results show that the index proposed in this invention has a low degree of dispersion within the corresponding category and a reduced number of outliers, indicating that it has good robustness and long-term stability, and is more suitable for dynamic monitoring of aquatic vegetation in lakes and wetlands under long-term sequence conditions.

[0065] like Figure 7 As shown, the effectiveness of the method of the present invention was verified by comparing the extraction results with JRC-GSW water body products. The verification area mainly included the land-water transition zone and the area where open water and floating emergent vegetation intersect, and F1-score was used as the evaluation index. The results show that the method of the present invention has high consistency in the land-water boundary region. The F1-score is consistently higher than 0.80 for most periods and exceeds 0.90 for some periods, indicating that the present invention can effectively identify water body boundaries in complex shoreline environments and has good temporal stability. Figure 7 In the transitional zone between floating and emergent vegetation, although consistency fluctuated at certain times, analysis showed that this difference mainly stemmed from the differences in category definitions between the method of this invention and the JRC-GSW product. The JRC-GSW product primarily represents open water bodies, while this invention can further distinguish between different aquatic ecological types such as harmful algal blooms, submerged vegetation, and floating emergent vegetation, thus exhibiting higher thematic resolution in complex transitional zones. Compared to traditional water index methods, the classification framework proposed in this invention can effectively reduce errors caused by mixed pixels and vegetation interference in complex lake and wetland environments, improving the accuracy of water body and aquatic vegetation identification, and demonstrating good stability and application value.

[0066] like Figure 8As shown, to verify the classification reliability and interannual stability of the method of the present invention, one representative image from each year during the period of 2019-2024 was selected for visual interpretation verification. Manual sample interpretation was performed for the main land cover categories, with a focus on selecting category transition areas, boundary areas, and areas with complex spectral characteristics as verification samples to enhance the evaluation ability of classification performance under complex mixed pixel conditions. Sample points with interpretation discrepancies were removed to ensure the reliability of the verification results. Figure 8 The verification results show that the method of the present invention exhibits high pixel-level classification accuracy in most years, with good overall separation effect between different categories. Specifically, the classification results for July 2020, September 2021, and September 2022 show high overall consistency, with each category achieving high recognition accuracy. In contrast, there is some degree of category confusion in July 2019, June 2023, and June 2024, with errors mainly concentrated in the transition zone between water and land, the area between terrestrial vegetation and bare land, and the area where water and floating emergent vegetation intersect. Overall, the method of the present invention can maintain high classification stability under long-term remote sensing image conditions, indicating its good robustness and long-term applicability.

[0067] like Figure 9 As shown, to verify the stability of the method of the present invention to threshold changes, sensitivity analyses were conducted on the greenness threshold, humidity threshold, and aquatic vegetation separation index threshold. Based on the baseline threshold, perturbations of ±10%, ±20%, and ±30% were applied, while other thresholds remained unchanged. The classification results were evaluated using a long-term validation dataset to analyze the impact of different threshold changes on the classification results. Figure 9 The results show that the method of the present invention has good threshold robustness overall. The humidity threshold has the least impact on the classification results; even within a large perturbation range, the accuracy of each category remains relatively stable, indicating high stability in the water-land distinction process. In contrast, the greenness threshold and the aquatic vegetation separation index threshold have relatively significant impacts on vegetation categories. Changes in the greenness threshold mainly affect the pixel allocation between terrestrial vegetation and floating emergent vegetation, but the overall change is small, and the classification results remain generally stable. The aquatic vegetation separation index threshold has a more significant impact on vegetation classification, mainly affecting the separation process between terrestrial vegetation and floating emergent vegetation, while having a smaller impact on water and land categories. When the threshold decreases, the two types of vegetation are more easily confused; when the threshold increases, the vegetation separation effect is enhanced, indicating that this index is an important parameter affecting fine-grained vegetation classification. Overall, even under large threshold perturbation conditions, the method of the present invention can maintain high classification stability, indicating that the proposed classification framework has good robustness and adaptability.

[0068] like Figure 10As shown, submerged vegetation in shallow lake environments is easily affected by water depth, turbidity, bottom sediment background, and mixed pixels, making it difficult to obtain stable and reliable verification results solely through visual interpretation of remote sensing images. Therefore, this invention uses independent reference data to verify the classification results of submerged vegetation, including submerged vegetation samples obtained from field surveys and publicly reported submerged vegetation monitoring areas. Specifically, using "Wuliangsuhai" and "submerged vegetation" as keywords, relevant research literature was retrieved from Chinese and foreign academic databases such as CNKI, Web of Science, and Google Scholar, collecting publicly reported information on the survey locations, monitoring areas, and distribution of submerged vegetation. For monitoring areas with clearly identifiable spatial locations, georegistration and location correction were performed using remote sensing images, geographic coordinates, and ground feature features to construct a submerged vegetation reference sample dataset. Figure 10 (c) The dataset was then spatially compared with the submerged vegetation results extracted in this invention to verify the reliability of the submerged vegetation extraction results. The verification results show that the spatial distribution of submerged vegetation extracted in this invention has a high degree of consistency with the independent reference data. The field survey samples were mainly distributed in the central and southern parts of the lake area, and generally fell within the submerged vegetation range extracted in this invention. The submerged vegetation areas reported in previous studies also maintain a good spatial correspondence with the results of this invention, especially showing high consistency in the central part of the lake area and the main submerged vegetation zones, indicating that this invention can effectively identify the distribution characteristics of submerged vegetation in lakes. Figure 10 Furthermore, after subjecting the submerged vegetation classification thresholds to a ±30% perturbation test, the classification accuracy remained largely stable, indicating that the submerged vegetation extraction results are not sensitive to threshold changes and possess good stability and robustness. Based on the verification results from independent reference data and threshold sensitivity analysis, the classification framework proposed in this invention can stably and effectively identify submerged vegetation in complex lake environments, and has good reliability and applicability.

[0069] The above description is only a preferred embodiment of the present invention. It should be noted that, for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A wetland classification and mapping method integrating tassel cap transformation and aquatic vegetation separation index, characterized in that, Includes the following steps: Step 1: Acquire remote sensing image data of the study area and preprocess the remote sensing image data; Step 2: Perform tassel transformation on the preprocessed remote sensing image data to obtain the greenness component and the humidity component; the greenness component is used for the initial separation of vegetation and non-vegetation, and the humidity component is used for further differentiation of land-related features and water-related features in non-vegetation areas. Step 3: Separate vegetated areas from non-vegetated areas based on the greenness component; Step 4: In non-vegetated areas, the humidity component is used to separate the land area and the water area. Step 5: Within the vegetation area, considering that both aquatic and terrestrial vegetation have strong spectral responses and conventional vegetation indices are difficult to distinguish effectively, the aquatic vegetation separation index is used in conjunction with the phenological characteristics of the study area to separate the aquatic vegetation area and the terrestrial vegetation area. Step 6: Within the water body area, the aquatic vegetation separation index is used to separate the open water area from the submerged biological characteristic area. Step 7: Within the submerged biological characteristic area, the submerged vegetation area and the algal bloom area are separated using the time persistence feature. Step 8: Based on the regional separation results, output the wetland classification and mapping results.

2. The wetland classification and mapping method according to claim 1, characterized in that, In step 1, the process of acquiring and preprocessing remote sensing image data is as follows: acquire multi-temporal and multispectral remote sensing image data of the study area, perform image quality control, cloud masking, and effective pixel screening quality control on the remote sensing image data, retain images that meet the effective pixel ratio requirements, and perform cloud masking using the quality control band.

3. The wetland classification and mapping method according to claim 1, characterized in that, The multi-temporal multispectral remote sensing image data is Landsat 8 surface reflectance image. Preferably, during the vigorous growth period of vegetation, an adaptive threshold method based on the distribution characteristics of greenness components is used to determine the current threshold and separate vegetated areas from non-vegetated areas; in the early growth stage, an empirical stable threshold is used to separate vegetated areas from non-vegetated areas.

4. The wetland classification and mapping method according to claim 1, characterized in that, In step 3, the vegetation coverage in the image is initially extracted by utilizing the significant response differences of vegetation in the red light band and near-infrared band, combined with the greenness component. The initial extraction is mainly based on significant vegetation signals dominated by emergent or high-canopy vegetation.

5. The wetland classification and mapping method according to claim 1, characterized in that, In step 4, considering that the land-water boundary in shallow wetlands is usually not a sharp boundary, but a wide transition zone composed of saturated mudflats, shallowly flooded areas and mixed pixels, an adaptive threshold determination method based on boundary gradient constraints is used to separate the land area and the water area. Preferably, the adaptive threshold determination method based on boundary gradient constraints specifically involves: selecting representative pure sample areas of two adjacent target classes through visual interpretation; constructing a class transition buffer based on the spatial contact relationship between the representative pure sample areas of the two target classes; calculating the spatial gradient of the humidity component within the class transition buffer; selecting the pixel with the largest local gradient as the boundary candidate pixel; and determining the classification threshold for the current scene based on the statistical characteristics of the humidity component of the boundary candidate pixel.

6. The wetland classification and mapping method according to claim 1, characterized in that, Step 5 specifically includes the following steps: Step 5.1: Select the multispectral remote sensing imagery from June as the dominant time phase for emergent vegetation identification; Step 5.2: Determine the threshold for the dominant phase of each year based on the spatial gradient of the aquatic vegetation separation index, and use the multi-year threshold statistics to determine the final segmentation threshold in order to extract the annual emergent vegetation range and separate the aquatic vegetation area from the terrestrial vegetation area. Preferably, the expression for the aquatic vegetation separation index is: ； in, For near-infrared reflectivity, For green light band reflectivity, This refers to the reflectivity in the red light band.

7. The wetland classification and mapping method according to claim 1, characterized in that, In step 6, within the water body area, the adaptive threshold based on boundary gradient constraints is used. By selecting the transition region between open water and submerged biological features, the AVSI gradient is calculated and the scene threshold is determined to improve the ability to distinguish between open water and submerged biological features. Preferably, the open water body includes clear water and turbid water, and its spectrum is mainly affected by water scattering and suspended matter; the characteristics of the submerged organisms are affected by the bio-optical effects caused by the biomass of aquatic vegetation or algae, and its spectrum is affected by the red light absorption characteristics relative to green light.

8. The wetland classification and mapping method according to claim 1, characterized in that, In step 7, submerged vegetation is usually rooted or attached to the substrate and its spatial position is relatively stable during the growing season; algal blooms, on the other hand, are easily affected by wind and hydrodynamic conditions, and their spatial distribution has strong drift and intermittency. Based on multi-temporal and multispectral remote sensing image data, the continuous occurrence of submerged biological feature pixels throughout the growing season is judged: if a pixel appears continuously and stably in multiple temporal phases, it is determined to be submerged vegetation; if it only appears sporadically or its spatial position is unstable, it is determined to be algal bloom.

9. An electronic device, characterized in that, It includes one or more processors and a memory; the memory is used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the wetland classification and mapping method as described in any one of claims 1 to 8.

10. A computer program product, characterized in that, The aforementioned computer program product includes computer-executable instructions that, when executed, implement the wetland classification and mapping method as described in any one of claims 1 to 8.

Images