Blue-green algae bloom remote sensing recognition system based on multi-source satellite and intelligent threshold correction

Abstract

The invention relates to a cyanobacterial bloom remote sensing recognition system based on a multi-source satellite and intelligent threshold correction, and the system comprises an intelligent image retrieval and screening module which is used for calling a multi-source image according to a daily frequency / weekly frequency period and generating a candidate set; the water body mask and preprocessing module is used for cloud removal, cutting, water body extraction and non-target plaque removal; the water bloom index and threshold value module is used for automatically calculating an ABDI or FAI index according to an image source and carrying out peak height proportion threshold value segmentation; the pattern spot extraction and clustering module is used for extracting algal bloom patches, merging pattern spots and removing small patches; and the standardized product generating and pushing module is used for generating and pushing grids, vectors, daily frequency brief reports and weekly frequency thematic reports. According to the invention, daily frequency / weekly frequency automatic and intelligent identification and achievement business output of cyanobacterial blooms can be realized, and the precision and continuity of remote sensing monitoring are significantly improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of environmental remote sensing monitoring and image processing, and particularly relates to a cyanobacterial bloom remote sensing identification system based on multi-source satellites and intelligent threshold correction. BACKGROUND

[0002] Cyanobacterial bloom is an important form of lake eutrophication, which can easily cause drinking water pollution, water ecological damage and fishery loss. Generally speaking, due to the influence of hydrological rhythm and complex meteorological conditions, the freshwater lake presents the typical characteristics of "lake in flood season and river in dry season", and the distribution of cyanobacterial bloom has the characteristics of "fragmentation and seasonal outbreak", with significant spatio-temporal variation and strong suddenness.

[0003] Although the existing satellite remote sensing based cyanobacterial bloom monitoring method has been applied to a certain extent, there are still the following deficiencies in the business monitoring of Poyang Lake: the automation of image retrieval and screening is insufficient, most methods rely on manual screening of satellite data, and there is a lack of batch automatic retrieval mechanism for daily and weekly periodic business; at the same time, the cloud screening mostly uses a single fixed threshold (such as 30%), which can easily cause data interruption in the cloudy season, and it is difficult to meet the monitoring needs of daily rapid response and weekly trend analysis. The data priority rule is missing, although some studies have tried to integrate multi-source satellite images, but the advantages and differences of different data sources and the adaptability of daily and weekly monitoring have not been fully considered. For example, Sentinel-2 has a 10m high resolution advantage in dry season, while Landsat-8 / 9 has a wide coverage advantage in flood season, but there is a lack of dynamic priority judgment mechanism for the flood and dry hydrological cycle of Poyang Lake and the daily / weekly monitoring needs.

[0004] At present, the threshold value is determined depending on the experience of the human, and the existing methods generally use fixed experience threshold or Otsu method to segment the bloom index histogram, but the Otsu method often fails to calculate in a large area, and cannot automatically locate the mutation point of "background peak end-water bloom start" by using the characteristics of "obvious background peak" in the index distribution, so the threshold value still needs expert manual intervention. The interference of small patches is not effectively removed, and there are a large number of small pit water bodies in the dry season, if not filtered specifically, the false bloom patches of non-target water bodies are easy to be included in the statistics, reducing the overall recognition accuracy. The result output lacks business adaptation, the existing methods mostly stay at the level of bloom distribution identification, and do not form standardized results such as daily frequency report and weekly special report, lack of automatic output mechanism of key attributes such as bloom patch latitude, longitude and area, and cannot meet the needs of daily rapid check and weekly comprehensive judgment of water environment management departments.

[0005] In summary, the existing satellite remote sensing cyanobacteria bloom monitoring method has deficiencies in data automatic retrieval, rich and lean period and day / week frequency differentiation processing, threshold intelligent correction, fine filtering of map patches, and business output of results. It is urgent to build a set of "retrieval-screening-identification-output" full-process automated monitoring technology that adapts to the characteristics of lakes, to realize efficient fusion of multi-source satellite data, intelligent extraction of cyanobacteria bloom, and business service of results, so as to improve the accuracy, stability and day / week frequency collaborative application value of remote sensing monitoring. SUMMARY

[0006] The purpose of the present application is to provide a cyanobacteria bloom remote sensing identification system based on multi-source satellites and intelligent threshold correction, to solve the problems of insufficient image retrieval automation, missing data priority, threshold dependence on manual experience, small patch interference, and insufficient business output of results in the prior art, and to realize the automation, intelligentization and day frequency / week frequency collaboration of cyanobacteria bloom remote sensing identification.

[0007] To achieve the above-mentioned purpose, the present application provides a cyanobacteria bloom remote sensing identification system based on multi-source satellites and intelligent threshold correction, comprising:

[0008] An intelligent image retrieval and screening module is used to retrieve multi-source satellite images according to a day frequency or week frequency cycle and select the optimal image according to priority;

[0009] A water body mask and preprocessing module is used to generate a water body mask based on a water body index and remove small patches through morphological operation, connectivity analysis and area pixel double threshold filtering;

[0010] A water bloom index and threshold module is used to calculate the ABDI or FAI index and automatically determine the water bloom extraction threshold value through the peak height ratio method;

[0011] A patch extraction and clustering module is used to extract water bloom patches and perform clustering and merging based on STRtree spatial index and boundary repair;

[0012] A standardized product generation and pushing module is used to add attributes to the water bloom patches and output and push the results in a standardized manner in the form of vectors, rasters and reports;

[0013] The cyanobacteria bloom remote sensing identification system identifies cyanobacteria bloom through the following steps:

[0014] Step S1, multi-source satellite image automatic retrieval and multi-level screening: according to a preset day frequency or week frequency cycle, automatically retrieve Sentinel-2 S2_SR images and Landsat-8 / 9 T1_L2 images of the target lake area, project the retrieved images uniformly, perform multi-level screening according to cloud cover, spatial coverage and image quality bands, and generate a candidate image set;

[0015] Step S2, Multi-source data priority determination and image selection: In the candidate image set, based on the hydrological characteristics of the lake during the high and low water seasons and the sensor resolution, Sentinel-2 images are preferentially selected for cyanobacterial bloom identification; if Sentinel-2 images are unavailable, Landsat-8 / 9 images with the lowest cloud cover and the time closest to the target monitoring time are selected, and the Landsat-8 / 9 images are resampled to 10m resolution to determine the images used for bloom identification;

[0016] Step S3, Water Mask Extraction and Preprocessing: Apply cloud and snow masks to the image, and generate an initial water mask using the improved Normalized Difference Water Index (MNDWI). The MNDWI index is calculated as follows: MNDWI = (ρ Green -ρ SWIR ) / ( ρ Green +ρ SWIR ), ρ Green ρ is the reflectivity in the green light band. SWIR To determine the reflectivity of the shortwave infrared band, regions with MNDWI > 0 are identified as water bodies. Threshold binarization, morphological opening and closing operations, connected component analysis, and area-pixel dual threshold filtering are then performed sequentially to remove isolated pits and noise patches, resulting in an optimized water body mask.

[0017] Step S4, Algal Bloom Index Calculation and Threshold Determination: Within the water body area, the ABDI index of cyanobacterial bloom is calculated for Sentinel-2 imagery. Since Landsat-8 / 9 imagery lacks a red-edge band, the ABDI index cannot be calculated. Instead, the FAI index of cyanobacterial bloom is calculated for Landsat-8 / 9 imagery. An algal bloom index histogram is constructed within the water body mask, and the threshold is determined using the peak height ratio method. The steps for determining the threshold using the peak height ratio method include: locating the peak height of the background peaks; using a preset ratio of the peak height as the judgment threshold; searching from the right side of the background peaks towards higher values, and the first bin boundary below the preset ratio threshold is the identification threshold.

[0018] Step S5, algal bloom patch extraction and clustering: The algal bloom region is segmented using a threshold and vector patches are generated. Patches below the preset area threshold are filtered out. Then, multiple rounds of iterative clustering and merging are performed based on the STRtree spatial index to merge adjacent patches or patches that are less than the preset spatial distance threshold. After merging, buffering and debuffering operations are performed to repair boundary defects, resulting in optimized algal bloom patches.

[0019] Step S6, Standardized Product Generation and Push Steps: Calculate the area and centroid coordinates of the algal bloom patches, and overlay administrative division and lake code data to generate algal bloom products with attributes such as latitude and longitude, area, administrative division, and lake code; output and push them to the terminal in the form of vector, raster, daily briefing and weekly report.

[0020] Preferably, in step S1, the multi-level screening criteria include: the first priority is cloud cover <30%; if no image meets the criteria, the priority is relaxed to 60%, and all images must meet the requirement of spatial coverage >50%; finally, clouds, shadows and invalid pixels are removed using the quality bands inherent in the image.

[0021] Preferably, in step S3, the area threshold is a preset parameter, which is set according to the lake size, image resolution, and monitoring requirements.

[0022] Preferably, in step S4, the formula for calculating the ABDI index is: ABDI=[ρ RE2 -ρ Red -(ρ NIRn -ρ Red )×(λ RE2 -λ Red ) / (λ NIRn -λ Red )]-[ρ Red -0.5×ρ Green ], where ρ Red ρ Green ρ NIRn、 ρ RE2 The reflectances of the red band, green band, narrow near-infrared band, and second red-edge band are respectively λ. Red , λ NIRn With λ RE2 These are the center wavelengths of the red light band, the narrow near-infrared band, and the second red-edge band, respectively; the formula for calculating the FAI index is: FAI = ρ NIR -ρ NIR ′ , ρ NIR ′ =ρ Red +(ρ SWIR – ρ Red ) ×(λ NIR -λ Red ) / (λ SWIR -λ Red ), ρ Red , ρ NIR , ρ SWIR Reflectance in the red, near-infrared, and short-wave infrared bands, respectively, λ Red , λ NIR , λ SWIR These are the center wavelengths of the red light, near-infrared, and short-wave infrared bands, respectively.

[0023] Preferably, in step S4, the ratio parameter of the peak height ratio method is preset according to the distribution characteristics of the algal bloom index, and the number of histogram bins is adjusted according to the image quality and the size of the water body.

[0024] Preferably, in the clustering and merging process of step S5, the spatial distance threshold is an adjustable parameter and is gradually adjusted and reduced during the iteration process.

[0025] Based on the above technical solution, the advantages of the present invention are:

[0026] The remote sensing identification system for cyanobacterial blooms of this invention supports automated daily / weekly image retrieval and screening, significantly reducing manual intervention and improving monitoring continuity. It establishes a priority mechanism prioritizing Sentinel-2 and supplementing it with Landsat-8 / 9, balancing high resolution and wide coverage. Threshold extraction is performed using the peak-height ratio method, enabling adaptive threshold correction. Combining area, morphological operations, and spatial clustering, it effectively eliminates non-target small patches. The system outputs standardized results in "raster + vector + daily briefing + weekly report," facilitating application in various operational scenarios. Attached Figure Description

[0027] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings:

[0028] Figure 1 This is a structural diagram of the module of the present invention;

[0029] Figure 2 This is an overall flowchart of the method of the present invention;

[0030] Figure 3 This is a schematic diagram of the water body pattern before and after filtration in an embodiment of the present invention;

[0031] Figure 4 This is a schematic diagram of the algal bloom pattern before and after filtration in an embodiment of the present invention;

[0032] Figure 5 This is a spatial distribution map of algal blooms in an embodiment of the present invention. Detailed Implementation

[0033] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0034] This invention provides a remote sensing identification system for cyanobacterial blooms based on multi-source satellite imagery and intelligent threshold correction. The system includes an intelligent image retrieval and filtering module, a water body masking and preprocessing module, a bloom index and thresholding module, a patch extraction and clustering module, and a standardized product generation and delivery module. The system comprises automated multi-source satellite image retrieval and multi-level filtering; setting data priorities based on hydrological characteristics; generating and preprocessing water body masks; calculating the ABDI or FAI bloom index within the water body and automatically determining the threshold using the peak-to-height ratio method; clustering and merging extracted bloom patches and removing small patches based on spatial indexing; and outputting standardized bloom products with attributes such as area, latitude and longitude, and administrative division.

[0035] This invention enables automated and intelligent identification of cyanobacterial blooms on a daily / weekly basis and allows for the operational output of the results, significantly improving the accuracy and continuity of remote sensing monitoring.

[0036] like Figure 1 As shown, the system of the present invention includes the following modules: an intelligent image retrieval and filtering module: used to retrieve multi-source images according to daily / weekly cycles and generate candidate sets; a water body masking and preprocessing module: used for cloud removal, cropping, water body extraction, and removal of non-target patches; an algal bloom index and threshold module: used to automatically calculate the ABDI or FAI index and peak height ratio threshold segmentation based on the image source; a patch extraction and clustering module: used to extract algal bloom patches, merge patches, and remove small patches; and a standardized product generation and push module: used to generate and push raster, vector, daily briefing, and weekly thematic reports.

[0037] Furthermore, such as Figure 2 As shown, the cyanobacterial bloom remote sensing identification system performs cyanobacterial bloom remote sensing identification through the following steps:

[0038] Step S1: Automated retrieval and multi-level filtering of multi-source satellite imagery

[0039] Sentinel-2 S2_SR and Landsat-8 / 9 T1_L2 images within the Poyang Lake area are automatically retrieved according to a preset daily / weekly frequency cycle, with a unified projection of WGS84 / UTM49N. A multi-level screening strategy is adopted: images with cloud cover <30% are prioritized, and if no image meets the condition, the threshold is relaxed to 60%; the spatial coverage of the image must be greater than 50%; clouds, shadows, and invalid pixels are removed by combining cloud masking and quality bands to generate a candidate image set.

[0040] Step S2: Prioritization of Multi-Source Data and Image Selection

[0041] Among the candidate images, priorities are set based on the characteristics of Poyang Lake during its high / low water seasons and the advantages of the sensors: if Sentinel-2 and Landsat-8 / 9 images exist simultaneously, Sentinel-2 (10m resolution) is selected first for algal bloom identification; when Sentinel-2 is unavailable, Landsat-8 / 9 images with the lowest cloud cover and the closest time are selected and resampled to 10m resolution. This priority mechanism enables continuous output of sliding time series with daily updates and weekly optimization.

[0042] Step S3: Water Mask Extraction and Pretreatment

[0043] Cloud and snow masks were applied to the selected images (Sentinel-2 used cloud probability data with a threshold of <60%, and Landsat used the QA band); water bodies were extracted using the improved MNDWI index (MNDWI>0 was considered water bodies), and then isolated pits and noise patches were removed by threshold binarization, morphological opening and closing operations, connected component analysis, and area-pixel dual threshold filtering to obtain an optimized water body mask.

[0044] Step S4: Algal bloom index calculation and threshold determination

[0045] Within the water body, the ABDI index was calculated using Sentinel-2 imagery, and the FAI index was calculated using Landsat-8 / 9 imagery. An algal bloom index histogram was constructed within the water body mask, and the threshold was determined using the peak-height ratio method. This involved: ① locating the peak values ​​of the background peaks; ② setting a counting threshold according to a certain proportion of the peak height; ③ searching from the right side of the background peaks towards higher values, with the first bin boundary below the proportional threshold serving as the algal bloom identification threshold.

[0046] Step S5: Extraction and clustering of algal bloom patches

[0047] The threshold is used to segment the algal bloom region to generate vector patches, and patches below the preset area threshold are filtered out. Based on the STRtree spatial index, multiple rounds of iterative clustering and merging are performed to merge adjacent or neighboring patches according to the decreasing spatial distance threshold. After merging, the boundary defects are repaired through buffering / debugging operations to obtain optimized algal bloom patches.

[0048] Step S6: Standardized Product Generation and Push

[0049] The area and centroid coordinates of algal bloom patches are calculated, and administrative divisions and lake codes are superimposed to generate algal bloom products containing attributes such as latitude and longitude, area, administrative division, and lake codes. These products are output and pushed to the terminal in the form of vector, raster, daily briefing and weekly report.

[0050] Compared with existing technologies, this invention has at least the following advantages: it supports automated daily / weekly retrieval and screening of multi-source images, significantly reducing manual intervention and improving monitoring continuity; it establishes a priority mechanism with Sentinel-2 as the priority and Landsat-8 / 9 as a supplement, taking into account the advantages of high resolution and wide coverage; it uses the peak height ratio method for threshold extraction to achieve adaptive threshold correction; it combines area, morphological operations, and spatial clustering to effectively remove non-target small patches; and it outputs standardized results of "raster + vector + daily briefing + weekly report", which is convenient for application in different operational scenarios.

[0051] To further illustrate the cyanobacterial bloom remote sensing identification system based on multi-source satellites and intelligent threshold correction of the present invention, the following description uses the automatic identification of cyanobacterial blooms in Poyang Lake in August 2025 as an example.

[0052] This embodiment uses the high-water season of Poyang Lake (August 26 to September 1, 2025) as the research period. Utilizing Sentinel-2 S2_SR and Landsat-8 / 9 T1_L2 multi-source satellite imagery, it demonstrates the implementation process of the automatic cyanobacterial bloom identification system based on multi-source satellite imagery and intelligent threshold correction. The specific steps are as follows:

[0053] Step 1: Automated Retrieval and Multi-Level Filtering of Multi-Source Satellite Imagery

[0054] In this embodiment, the system automatically retrieves the Sentinel-2 S2_SR image and Landsat-8 / 9 T1_L2 image of the Poyang Lake area from August 26 to September 1, 2025, according to the weekly cycle (7 days), and projects them uniformly into the WGS84 / UTM49N coordinate system.

[0055] A multi-level screening strategy was adopted: the first priority was to screen images with cloud cover of less than 30%, and two Sentinel-2A images from August 26 and August 28 were retrieved. After merging, the spatial coverage of the Poyang Lake area was 90%. Combining the image quality bands (cloud probability data of Sentinel-2 and QA bands of Landsat-8 / 9), clouds, shadows and invalid pixels were removed to generate a candidate image set (only the Sentinel-2A image in this period met the conditions).

[0056] Step 2: Prioritizing Multi-Source Data and Selecting Images

[0057] Sentinel-2 imagery exists among the candidate images, and Poyang Lake has entered its dry season ahead of schedule. Based on its 10m spatial resolution advantage, this embodiment prioritizes Sentinel-2A imagery and does not call Landsat imagery, proceeding to the subsequent processing flow.

[0058] Step 3: Water Mask Extraction and Pretreatment

[0059] Cloud and snow masking: Using Sentinel-2 cloud probability data, a threshold of less than 60% is set to mask the cloud and snow cover on the image to remove cloud-covered areas.

[0060] Water body extraction: Using the improved MNDWI index (MNDWI=(B3-B11) / (B3+B11)), areas with MNDWI greater than 0 were identified as water bodies, generating an initial water body mask, and 3108 water body patches were detected.

[0061] Non-target water body removal: Through morphological opening and closing operations (structural element is 3×3 pixels), connected component analysis, and combined with area and pixel dual thresholds (area threshold set to 5000 square meters, pixel threshold set to 500), non-target pits and noise patches were removed, resulting in an optimized water body mask (473 water body patches). The results are as follows: Figure 3 As shown.

[0062] Step 4: Identification and Optimization of Cyanobacterial Blooms

[0063] Algal bloom index calculation: Within the optimized water mask area, the ABDI index is calculated for Sentinel-2A imagery. ABDI = [ρ RE2 -ρ Red -(ρ NIRn -ρ Red )×(λ RE2 -λ Red ) / (λ NIRn -λ Red )]-[ρ Red -0.5×ρ Green ], where ρ Red ρ Green ρ NIRn、 ρ RE2 The reflectances of the red band, green band, narrow near-infrared band, and second red-edge band are respectively λ. Red , λ NIRn With λ RE2 These are the center wavelengths of the red light band, the narrow near-infrared band, and the second red edge band, respectively.

[0064] Peak height ratio method for threshold determination: Construct an ABDI index histogram, locate the peak height of the background peak as -0.0687 (count is 1056285), set the ratio parameter to 10% (default 10%, threshold is adjustable in the range of 5%-15%), and the threshold count is 105628.50; search from the right side of the background peak towards the higher value direction, the first bin boundary with a count less than 105628.50 is ABDI=-0.0163 (in actual application, -0.016 is used), and this is used as the algal bloom identification threshold;

[0065] Algal bloom patch optimization: Algal bloom regions were segmented using a threshold method to generate vector patches, identifying 1105 initial algal bloom patches. Four rounds of iterative clustering and merging were performed based on the STRtree spatial index. The initial spatial distance threshold was set to 50m, and the threshold decreased by a factor of 0.8 after each iteration (50m → 40m → 32m → 25.6m). Adjacent patches and patches with a distance less than the current threshold were merged. After merging, boundary defects were repaired through buffering (2m buffer distance) and anti-buffering operations, and patches with an area <1000㎡ were filtered out. Noise was removed, resulting in 209 optimized algal bloom patches covering a total area of ​​6.35km², mainly distributed in lakes surrounding Poyang Lake such as Zaohu Lake, Nanhu Lake, Fangxi Lake, and branches of the Ganjiang River. Results are as follows: Figure 4 As shown.

[0066] Step 5: Standardized Product Generation and Push

[0067] Output vector files (Shapefile format) and raster files (TIF format): The vector file contains the unique ID of the algal bloom patch, the monitoring date (20250826_20250901), the satellite type (Sentinel-2A), the area (m²), the latitude and longitude of the center point, the administrative division (involving districts under the jurisdiction of cities such as Nanchang and Jiujiang), the lake code and other attributes.

[0068] Generate standardized products: weekly thematic reports, including image overviews and spatial distribution maps of algal blooms, such as... Figure 5 As shown, the time series analysis of the area of ​​water blooms in key rivers is performed, and the data is automatically pushed to the designated server or business terminal of the water environment management department according to the preset interface.

[0069] Step 6: Effect Verification and Accuracy Evaluation

[0070] The accuracy of the results in this embodiment was evaluated by generating a truth map of algal blooms through manual interpretation. The results show that the overall accuracy of the method described in this embodiment is 92.4%, and the recall rate is 89.7%; the overall accuracy of the comparison method (fixed threshold method) is 81.2%. These results verify the applicability and effectiveness of the intelligent threshold correction strategy in this embodiment.

[0071] The method of this invention was validated using multi-temporal Sentinel-2 and Landsat-8 / 9 imagery from the Poyang Lake region in 2024-2025, demonstrating stable identification of cyanobacterial blooms under multi-source satellite data conditions. Compared to existing identification methods relying on a single data source or fixed thresholds, this invention effectively avoids instability in identification results caused by differences in image quality or complex backgrounds by introducing multi-source image priority filtering and intelligent threshold correction mechanisms. The system achieves automated image scheduling, batch index calculation, and adaptive threshold correction processing, possessing good versatility and scalability, and can meet the application requirements of large-scale, multi-temporal automated monitoring of cyanobacterial blooms.

[0072] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them; although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications can still be made to the specific implementation of the present invention or equivalent substitutions can be made to some technical features without departing from the spirit of the technical solutions of the present invention, and all such modifications and substitutions should be covered within the scope of the technical solutions claimed in the present invention.

Claims

1. A remote sensing identification system for cyanobacterial blooms based on multi-source satellites and intelligent threshold correction, characterized in that: include: The intelligent image retrieval and filtering module is used to retrieve multi-source satellite images according to daily or weekly frequency and select the best image based on priority. The water mask and preprocessing module is used to generate a water mask based on the water index and remove small patches through morphological operations, connectivity analysis and area-pixel dual threshold filtering. The algal bloom index and threshold module is used to calculate the ABDI or FAI index and automatically determine the algal bloom extraction threshold using the peak height ratio method. The patch extraction and clustering module is used to extract algal bloom patches and perform clustering, merging, and boundary repair based on the STRtree spatial index. The standardized product generation and push module is used to add attributes to algal bloom patches and output and push them in a standardized manner as vector, raster and report formats. The cyanobacterial bloom remote sensing identification system performs cyanobacterial bloom remote sensing identification through the following steps: Step S1, Automated retrieval and multi-level screening of multi-source satellite images: According to the preset daily or weekly frequency cycle, the Sentinel-2 S2_SR image and Landsat-8 / 9 T1_L2 image of the target lake area are automatically retrieved, the retrieved images are uniformly projected, and multi-level screening is carried out based on cloud cover, spatial coverage and image quality bands to generate a candidate image set. Step S2, Multi-source data priority determination and image selection: In the candidate image set, based on the hydrological characteristics of the lake during the high and low water seasons and the sensor resolution, Sentinel-2 images are preferentially selected for cyanobacterial bloom identification; if Sentinel-2 images are unavailable, Landsat-8 / 9 images with the lowest cloud cover and the time closest to the target monitoring time are selected, and the Landsat-8 / 9 images are resampled to 10m resolution to determine the images used for bloom identification; Step S3, Water Mask Extraction and Preprocessing: Apply cloud and snow masks to the image, use the improved Normalized Difference Water Index (MNDWI) to generate an initial water mask, identify areas with MNDWI > 0 as water bodies, and sequentially perform threshold binarization, morphological opening and closing operations, connected component analysis, and area-pixel dual threshold filtering to remove isolated pits and noise patches, thus obtaining an optimized water mask. Step S4, Algal Bloom Index Calculation and Threshold Determination: Within the water body area, calculate the ABDI index for Sentinel-2 imagery and the FAI index for Landsat-8 / 9 imagery, and construct an algal bloom index histogram within the water body mask. Determine the threshold using the peak height ratio method. The steps for determining the threshold using the peak height ratio method include: locating the peak height of the background peaks; using a preset ratio of the peak height as the judgment threshold; searching from the right side of the background peaks towards higher values, and the first bin boundary below the preset ratio threshold is the identification threshold. Step S5, algal bloom patch extraction and clustering: The algal bloom region is segmented using a threshold and vector patches are generated. Patches below the preset area threshold are filtered out. Then, multiple rounds of iterative clustering and merging are performed based on the STRtree spatial index to merge adjacent patches or patches that are less than the preset spatial distance threshold. After merging, buffering and debuffering operations are performed to repair boundary defects, resulting in optimized algal bloom patches. Step S6, Standardized Product Generation and Push Steps: Calculate the area and centroid coordinates of the algal bloom patches, and overlay administrative division and lake code data to generate algal bloom products with attributes such as latitude and longitude, area, administrative division, and lake code; output and push them to the terminal in the form of vector, raster, daily briefing and weekly report.

2. The remote sensing identification system for cyanobacterial blooms according to claim 1, characterized in that: In step S1, the multi-level screening criteria include: the first priority is cloud cover <30%; if no image meets the criteria, the priority is relaxed to 60%, and all images must have a spatial coverage >50%; finally, clouds, shadows and invalid pixels are removed using the image's built-in quality bands.

3. The remote sensing identification system for cyanobacterial blooms according to claim 1, characterized in that: In step S3, the area threshold is a preset parameter, which is set according to the lake size, image resolution and monitoring requirements.

4. The remote sensing identification system for cyanobacterial blooms according to claim 1, characterized in that: In step S4, the formula for calculating the ABDI index is: ABDI=[p RE2 -r Red -(r NIRn -r Red )×(λ RE2 -l Red ) / (λ NIRn -l Red )]-[p Red -0.5×ρ Green ]; Where, ρ Red ρ Green ρ NIRn、 ρ RE2 The reflectances of the red band, green band, narrow near-infrared band, and second red-edge band are respectively λ. Red , λ NIRn With λ RE2 These are the center wavelengths of the red light band, the narrow near-infrared band, and the second red edge band, respectively. The formula for calculating the FAI index is as follows: FAI=p NIR -r NIR ′ ,r NIR ′ =p Red +(r SWIR – r Red ) ×(λ NIR -l Red ) / (λ SWIR -l Red ); Where, ρ Red , ρ NIR , ρ SWIR Reflectance in the red, near-infrared, and short-wave infrared bands, respectively, λ Red , λ NIR , λ SWIR These are the center wavelengths of the red light, near-infrared, and short-wave infrared bands, respectively.

5. The remote sensing identification system for cyanobacterial blooms according to claim 1, characterized in that: In step S4, the ratio parameter of the peak height ratio method is preset according to the distribution characteristics of the algal bloom index, and the number of histogram bins is adjusted according to the image quality and the size of the water body.

6. The remote sensing identification system for cyanobacterial blooms according to claim 1, characterized in that: In the clustering and merging process in step S5, the spatial distance threshold is an adjustable parameter and is gradually adjusted and reduced during the iteration process.

Images