Multi-source remote sensing data collaborative monitoring and intelligent interpretation and extraction system and method for lake aquatic vegetation
By combining multi-dimensional feature intelligent interpretation and a five-layer decision tree hierarchical stripping method with multi-source data and RTK UAV feedback optimization, the problem of distinguishing between cyanobacterial blooms and floating vegetation in lake aquatic vegetation extraction was solved, achieving high-precision and automated aquatic vegetation monitoring and extraction.
Patent Information
- Application Number
- CN202611077050.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2026-07-20
- Publication Date
- 2026-11-17
- Estimated Expiration
- 2046-07-20
AI Technical Summary
Existing remote sensing extraction technologies for lake aquatic vegetation cannot effectively distinguish between cyanobacterial blooms and floating vegetation. Mixed pixel signals at the water-vegetation interface are diluted. Fixed threshold segmentation methods result in missed extraction of edge vegetation and overly rigid boundaries, poor generalization ability, and a lack of systematic solutions for collaborative monitoring and intelligent interpretation of multi-source data.
A multi-source remote sensing data collaborative monitoring system was adopted. An enhanced cyanobacterial bloom detection index and an adaptive edge vegetation index were constructed through a multi-dimensional feature intelligent interpretation and extraction unit. Combined with a five-layer decision tree hierarchical stripping method, the system utilized domestic high-resolution satellites and Sentinel-2 multi-band registration, along with RTK UAV truth feedback optimization, to achieve adaptive extraction of aquatic vegetation.
It improves the accuracy and boundary definition of lake aquatic vegetation extraction, reduces misclassification and omission rates, enables adaptive monitoring of different lakes and seasons, and meets the needs of high-precision, continuous, dynamic monitoring.
Smart Images

Figure CN122598007B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the fields of remote sensing science and Earth observation, specifically to a system and method for collaborative monitoring and intelligent interpretation and extraction of multi-source remote sensing data for aquatic vegetation in lakes. Background Technology
[0002] Remote sensing science and Earth observation technology is a comprehensive technology that uses non-contact sensors to detect, acquire, and process electromagnetic wave information from the Earth's surface and atmospheric environment from a distance, in order to identify, analyze, and monitor its properties, state, and changes. This technology encompasses an integrated "air-space-ground" system, from satellites, space stations, high-altitude aircraft to low-altitude UAVs, achieving a multi-platform collaborative three-dimensional observation network that combines macro-level surveys with micro-level detailed surveys. It utilizes advanced multi-spectral and multi-modal sensors such as high spatial resolution optical cameras, hyperspectral imagers capable of identifying material composition, and synthetic aperture radar (SAR) with all-weather and all-time observation capabilities. It also employs intelligent data processing and interpretation technologies such as radiation and geometric correction, multi-source data fusion, and automated, high-precision information extraction methods driven by deep learning and large models. Furthermore, it integrates RS, GIS, and GNSS technologies, realizing a closed-loop chain from data acquisition, processing, and analysis to decision-making applications. The synergy of these technologies lays a solid foundation for the collaborative monitoring, intelligent interpretation, and extraction of multi-source remote sensing data for lake aquatic vegetation.
[0003] However, existing remote sensing extraction technologies for lake aquatic vegetation still face three major bottlenecks: First, the spectral characteristics of cyanobacterial blooms and floating vegetation are highly similar, and traditional spectral indices cannot effectively distinguish them, resulting in a large amount of cyanobacterial blooms being mixed into the floating vegetation extraction results, leading to a high misclassification rate. Second, the mixed pixel signals at the water-vegetation interface are diluted by the water background, and traditional fixed threshold segmentation methods are prone to missing edge vegetation and having overly hard boundaries, resulting in a large deviation between the extracted boundaries and the actual situation. Third, most methods use fixed threshold classification, which cannot adapt to the spectral differences of different lakes and different seasons, resulting in poor generalization ability. To address these issues, there is currently no systematic solution that can simultaneously solve the three major pain points of cyanobacterial spectral confusion, missing edge mixed pixels, and threshold adaptation.
[0004] Existing research only optimizes spectral indices or single classification thresholds, which cannot simultaneously overcome the three major coupled challenges of spectral confusion in algal blooms, pixel edge distortion in mixed water and land areas, and poor generalization of fixed thresholds. Furthermore, it lacks an integrated implementation architecture that combines domestic high-resolution satellites with multi-band registration and collaboration of Sentinel-2 and UAV ground truth closed-loop iteration. The related technical means are fragmented and have a low degree of automation, making it difficult to meet the needs of large-scale, high-precision, and time-series continuous dynamic monitoring of aquatic vegetation in lakes. Summary of the Invention
[0005] To address the aforementioned problems, this invention aims to provide a system and method for collaborative monitoring and intelligent interpretation and extraction of multi-source remote sensing data for aquatic vegetation in lakes.
[0006] To achieve the above objectives, this invention provides the following technical solution: a multi-source remote sensing data collaborative monitoring and intelligent interpretation and extraction system and method for lake aquatic vegetation, comprising a multi-source remote sensing data acquisition module, an integrated image preprocessing module, an intelligent interpretation and collaborative correction module, a UAV verification and closed-loop feedback optimization module, a coverage calculation and thematic product generation module, and a standardized data management and archiving module. The multi-source remote sensing data acquisition module is used for multi-source data acquisition; the integrated image preprocessing module is used for data standardization processing; the intelligent interpretation and collaborative correction module includes a multi-dimensional feature intelligent interpretation and extraction unit and a human-computer interaction collaborative correction unit; the multi-dimensional feature intelligent interpretation and extraction unit proposes an adaptive remote sensing interpretation algorithm for multi-dimensional features of aquatic vegetation to intelligently interpret and extract multi-dimensional features; the human-computer interaction collaborative correction unit is used for intelligent assisted visual interpretation; the UAV verification and closed-loop feedback optimization module is used for ground truth sample generation and accuracy evaluation; the coverage calculation and thematic product generation module is used for ecological indicator quantification and thematic result visualization; and the standardized data management and archiving module is used for full life cycle data management.
[0007] Furthermore, the multi-source remote sensing data acquisition module is used for multi-source data acquisition, collaboratively acquiring multi-source heterogeneous data such as domestic high-resolution satellite imagery, international open satellite data, historical hydrological vectors, and high-precision verification images.
[0008] Furthermore, the integrated image preprocessing module is used for data standardization processing. Through a series of standardized preprocessing steps, including radiometric calibration, atmospheric correction, geometric fine correction, image fusion, and water body mask cropping, multi-source remote sensing images are processed to generate standardized data that can be intelligently interpreted.
[0009] Furthermore, the multidimensional feature intelligent interpretation and extraction unit proposes an adaptive remote sensing interpretation algorithm for multidimensional features of aquatic vegetation to intelligently interpret and extract multidimensional features, and realizes intelligent automatic extraction of aquatic vegetation based on multidimensional features and an adaptive decision tree model.
[0010] Furthermore, the adaptive remote sensing interpretation algorithm for multidimensional features of aquatic vegetation is as follows:
[0011] 1) Construction of multidimensional feature space: Improved normalized water index is calculated on the water area of the image to lock the water background. Normalized vegetation index is calculated to initially identify vegetation response. To solve the problem of high spectral confusion between cyanobacterial blooms and floating vegetation, a texture uniformity factor is introduced on the basis of traditional phytoplankton index for joint modulation to construct an enhanced cyanobacterial bloom detection index. To solve the problem of mixed pixels at the water-vegetation interface causing hard vegetation boundaries and missed detection, a new index that can adaptively enhance edge vegetation signals is proposed to construct an adaptive edge vegetation index. At the same time, texture statistics are extracted through gray-level co-occurrence matrix to form a "spectral-texture-geometry" multidimensional feature space.
[0012] 2) Five-layer decision tree layering: Based on the above features, a five-layer decision tree is constructed. The threshold of each layer is dynamically determined by the Otsu's method. The optimal segmentation surface is obtained by the improved normalized water body index through the Otsu's method, generating a complete water body mask. Within the obtained water body mask, emergent / wet vegetation masks are extracted by dynamic threshold determination of the adaptive edge vegetation index. After removing them from the water body mask, the remaining water body area is obtained. In the remaining water body area after removing emergent vegetation, the cyanobacterial bloom area is identified and removed by dynamic threshold determination of the cyanobacterial bloom detection index, thus obtaining a pure water body mask without bloom interference. In the water body mask without bloom interference, the floating vegetation mask is extracted by using the dynamic threshold of the adaptive edge vegetation index combined with the patch area constraint for noise filtering. The subset of the water body mask without bloom interference that meets the habitat conditions of water depth / transparency is obtained by determining the submerged vegetation sensitivity index.
[0013] 3) Post-processing and output: The vegetation masks output from the above layers are merged, and the final vegetation classification raster data is morphologically processed and vectorized to generate a preliminary aquatic vegetation spatial distribution vector layer containing category attributes. This layer is then delivered to the downstream human-machine collaborative correction stage to intelligently interpret and extract the multidimensional features of lake aquatic vegetation.
[0014] Furthermore, the human-computer interaction collaborative correction unit is used for intelligent assisted visual interpretation and accurate error correction, and introduces expert-experienced interpreters to overlay high-resolution fused images.
[0015] Furthermore, the UAV verification and closed-loop feedback optimization module is used for ground truth sample generation and accuracy evaluation. Based on the centimeter-level high-precision orthophotos acquired by the UAV equipped with RTK, ground truth samples are generated, the accuracy of the intelligent interpretation results is evaluated, and the evaluation results are fed back to the intelligent interpretation unit to drive its model parameters to adaptively optimize.
[0016] Furthermore, the coverage calculation and thematic product generation module is used for the quantification of ecological indicators and the visualization of thematic results. It performs spatiotemporal dimension processing on the final aquatic vegetation vector data that has been verified and corrected, and performs chart-based analysis and display of spatial patterns, type proportions, and temporal evolution patterns.
[0017] Furthermore, the standardized data management and archiving module is used for full lifecycle data management. According to the standard hierarchical directory, it performs structured archiving and management of the summary vectors, maximum water body vectors, thematic maps, and original image analysis results generated throughout the entire process.
[0018] Furthermore, a collaborative monitoring and intelligent interpretation and extraction method for multi-source remote sensing data of lake aquatic vegetation includes the following steps:
[0019] S1: Multi-source data acquisition and integrated preprocessing: Collaborative acquisition of domestic high-resolution satellite imagery, international open-source multispectral satellite data, historical hydrological vectors, and RTK UAV imagery, followed by radiometric calibration, atmospheric correction, geometric fine correction, and image fusion processing. The shortwave infrared bands required for spectral index calculation are provided by Sentinel-2, retaining the original 10m / 20m resolution for initial extraction of coarse-scale water bodies and cyanobacterial blooms, generating a coarse classification mask. After rigorous registration with the domestic Gaofen-2 imagery based on the same set of DOP / DLG control points and DEM geometric benchmarks, the coarse classification mask is spatially mapped to a 0.8m pixel grid, serving as a spatial constraint for high-resolution fine interpretation. Visible and near-infrared bands are fused to 0.8m resolution for pixel-by-pixel fine extraction of vegetation boundaries and types, fundamentally avoiding spectral distortion caused by SWIR band over-resolution resampling.
[0020] S2: Multidimensional feature space construction: Calculate the improved normalized water index MNDWI, normalized vegetation index NDVI, enhanced cyanobacterial bloom detection index ECDI, and adaptive edge vegetation index AEVI pixel by pixel, and extract texture statistics through gray-level co-occurrence matrix to construct a "spectral-texture-geometry" multidimensional feature space.
[0021] S3: Five-layer adaptive decision tree classification: The threshold is dynamically determined layer by layer by the maximum inter-class variance method. The water body mask is extracted in sequence: emergent / wet vegetation mask is extracted, cyanobacterial blooms are removed, floating vegetation mask is extracted, and submerged vegetation mask is extracted. After merging, morphological processing and vectorization are performed to output a preliminary aquatic vegetation distribution vector layer.
[0022] S4: Human-computer interactive collaborative correction: The initial vector overlay of the high-resolution fused image is then corrected by experts based on the judgment rule library, including polygon deletion, shoreline truncation, boundary expansion, and manual delineation, and the final vector result is output.
[0023] S5: Multi-source verification and closed-loop optimization: Emergent and floating vegetation are verified by constructing ground truth samples based on centimeter-level orthophotos from RTK UAVs; submerged vegetation is verified by obtaining underwater ground truth using shipborne underwater cameras combined with quadrat surveys; after completing the accuracy evaluation, the misclassified pixel features are fed back to the interpretation unit, and three core parameters, namely ECDI texture entropy weight coefficient, AEVI local variance adjustment coefficient, and SVSI segmentation threshold d, are optimized in a targeted manner to form a self-optimizing closed loop of "extraction-verification-parameter iteration-re-extraction";
[0024] S6: Coverage Calculation, Thematic Mapping and Standardized Archiving: Calculate the area and coverage of various vegetation types, generate spatial distribution thematic maps and type proportion charts, and archive them in a structured manner according to standard hierarchical directory.
[0025] Compared with existing technologies, this invention possesses three progressive core innovations, resulting in significant technological advantages:
[0026] 1. This invention innovatively introduces a texture homogeneity factor to modulate the phytoplankton index to obtain the ECDI index, breaking away from the traditional approach of relying solely on a single spectral band to distinguish between cyanobacterial blooms and floating vegetation. Traditional FAI and NDVI only utilize spectral information, resulting in high spectral overlap and blurred distinction between the two types of land cover. This invention integrates spectral and texture dimensions, utilizing the inherent characteristics of homogeneous texture in cyanobacterial blooms and mottled texture in floating vegetation to construct a composite discriminant factor. This amplifies the numerical difference between the two targets from a physical mechanism perspective, fundamentally reducing the misclassification rate of floating vegetation. This composite index design concept has not been publicly disclosed in existing literature on aquatic vegetation remote sensing extraction, demonstrating outstanding novelty.
[0027] 2. A unique adaptive edge vegetation index (AEVI) is developed, coupling NDVI vegetation signals, MNDWI water background, and dynamic weighting based on local pixel spectral heterogeneity to solve the problem of signal attenuation in mixed pixels at the water-land interface. Existing technologies use fixed coefficient indices, which cannot adaptively adjust the gain according to the proportion of local water and vegetation. This invention utilizes the local NDVI standard deviation to adjust the enhancement weight in real time, automatically compensating for the diluted vegetation signal in edge mixed pixels. It eliminates the need for manual segmentation of boundary correction parameters, unlike traditional passive solutions that rely on manual edge trimming. This achieves fully automated edge vegetation enhancement and extraction, representing a novel approach to mixed pixel restoration.
[0028] 3. A five-layer progressive maximum inter-class variance adaptive decision tree layering and stripping framework is proposed, coupled with a "space-air-ground" collaborative closed-loop optimization system, forming a complete automated monitoring solution. Traditional classification methods mostly use single-layer fixed thresholds and binary classification models, which cannot separate the five types of targets—water bodies, emergent vegetation, cyanobacterial blooms, floating vegetation, and submerged vegetation—layer by layer. This invention removes interfering ground features layer by layer, and dynamically solves the optimal segmentation threshold for each layer using the Otsu method, eliminating the need for manual adjustment of threshold parameters for different lakes and seasons. Simultaneously, it is equipped with a domestic high-resolution imaging system + Sentinel-2 multi-source band precise registration and fusion, and an RTK UAV truth feedback iteration mechanism, forming a fully integrated system of "data fusion - intelligent interpretation - field verification - parameter self-optimization." No similar complete collaborative architecture has been found in existing technologies, and the overall technical solution has significant innovation and engineering practical value. Attached Figure Description
[0029] The invention will be further illustrated with reference to the accompanying drawings, but the embodiments in the drawings do not constitute any limitation on the invention. For those skilled in the art, other drawings can be obtained based on the following drawings without any creative effort.
[0030] Appendix Figure 1 This is a schematic diagram of the structure of the present invention. Detailed Implementation
[0031] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0032] A collaborative monitoring and intelligent interpretation and extraction system and method for multi-source remote sensing data of lake aquatic vegetation is proposed. This system includes a multi-source remote sensing data acquisition module, an integrated image preprocessing module, an intelligent interpretation and collaborative correction module, a UAV verification and closed-loop feedback optimization module, a coverage calculation and thematic product generation module, and a standardized data management and archiving module. The multi-source remote sensing data acquisition module is used for multi-source data acquisition; the integrated image preprocessing module is used for data standardization; the intelligent interpretation and collaborative correction module includes a multi-dimensional feature intelligent interpretation and extraction unit and a human-computer interaction collaborative correction unit. The multi-dimensional feature intelligent interpretation and extraction unit proposes an adaptive remote sensing interpretation algorithm for multi-dimensional features of aquatic vegetation to intelligently interpret and extract multi-dimensional features; the human-computer interaction collaborative correction unit is used for intelligent assisted visual interpretation; the UAV verification and closed-loop feedback optimization module is used for ground truth sample generation and accuracy evaluation; the coverage calculation and thematic product generation module is used for ecological indicator quantification and thematic result visualization; and the standardized data management and archiving module is used for full life-cycle data management.
[0033] Preferably, the multi-source remote sensing data acquisition module is used for multi-source data acquisition, and collaboratively acquires multi-source heterogeneous data such as domestic high-resolution satellite imagery, international open satellite data, historical hydrological vectors, and high-precision verification images.
[0034] Preferably, the integrated image preprocessing module is used for data standardization processing. Through a series of standardization preprocessing steps, such as radiometric calibration, atmospheric correction, geometric fine correction (with the registration error controlled within 0.5 pixels), image fusion, and water body mask cropping, standardized data that can be intelligently interpreted is generated.
[0035] Preferably, the multidimensional feature intelligent interpretation and extraction unit proposes an adaptive remote sensing interpretation algorithm for multidimensional features of aquatic vegetation to intelligently interpret and extract multidimensional features, and realizes intelligent automatic extraction of aquatic vegetation based on multidimensional features and an adaptive decision tree model.
[0036] Specifically, the adaptive remote sensing interpretation algorithm for multidimensional features of aquatic vegetation is as follows: First, deep feature mining is performed on the standard water body image after integrated preprocessing to construct a multidimensional feature space that integrates innovative spectral indices, texture, geometry, and phenological information. This includes spectral feature calculation and texture feature extraction. An improved normalized water body index is calculated to lock in the water body background. A series of spectral indices are calculated to capture the spectral response characteristics of different land cover species. The improved normalized water body index is then used to... ,in, The surface reflectance of the image in the green band. The surface reflectance of the image in the shortwave infrared band is used to accurately separate water bodies from land. The normalized difference vegetation index is... ,in, The surface reflectance of the image in the near-infrared band. The surface reflectance of the image in the red band is used for preliminary identification of vegetation areas. To address the issue of high spectral confusion between cyanobacterial blooms and floating vegetation, a texture uniformity factor is introduced for joint modulation based on the traditional phytoplankton index, constructing an enhanced cyanobacterial bloom detection index. The texture uniformity factor is obtained by normalized texture entropy transformation based on the gray-level co-occurrence matrix, and its value is [value missing]. ,in, To normalize to The texture entropy value of the interval; the more uniform the texture, the higher the value of this factor. Therefore, the enhanced cyanobacterial bloom detection index is calculated as follows: ,in, It is the phytoplankton index, and ,in, The near-infrared reflectance is the atmospherically corrected surface reflectance. The near-infrared baseline reflectance is obtained by linear interpolation of reflectance in the red and short-wave infrared bands, and... ,in, This represents the surface reflectance in the red band after atmospheric correction. This represents the surface reflectance in the shortwave infrared band after atmospheric correction. It is the center wavelength of the near-infrared band. It is the center wavelength of the red light band. It is the center wavelength of the shortwave infrared band. For calculation based on the gray-level co-occurrence matrix, normalized to The texture entropy value of the interval is calculated, and the texture features are all calculated based on a 3×3 window, a stride of 1, and 8 gray levels. To address the problem of overly hard vegetation boundaries and missed extraction caused by mixed pixels at the water-vegetation interface, a novel index that can adaptively enhance edge vegetation signals is proposed. The adaptive edge vegetation index is: ,in, For adaptive weighting coefficients, and ,in, For empirical adjustment, the value range is 0.8~1.5, typical value. , All pixels within a local window centered on the current pixel. The standard deviation of the values is used to quantify local spectral heterogeneity for spectral feature calculation. Simultaneously, texture statistics are extracted from the image using the gray-level co-occurrence matrix to distinguish between complex emergent vegetation and homogeneous phytoplankton. The second moment of the angle is calculated as follows: ,in, and The grayscale level of the image. The gray-level co-occurrence matrix is the first... line, number The element values in this column are used to measure the uniformity of texture. Blue-green algae bloom patches exhibit uniform grayscale and high ASM values; floating / emergent vegetation has mottled textures and low ASM values, serving as a core texture indicator for distinguishing homogeneous blooms from heterogeneous vegetation. Contrast is calculated as follows: This is used to measure texture contrast and roughness. Emergent vegetation canopies have large undulations and high contrast; open water surfaces have smooth textures and low contrast, used to identify vegetation canopy structural features. Correlation is calculated as follows: ,in, is the mean of all elements in the gray-level co-occurrence matrix. The variance of all elements in the gray-level co-occurrence matrix measures the degree of linear correlation between gray levels in the row and column directions, and can distinguish the spatial distribution patterns of strip-shaped aquatic vegetation from those of clustered algal blooms. Entropy is calculated as follows: ,in, This is a logarithmic operation used to measure texture complexity; the more complex the texture, the higher the entropy value. It forms the basis for calculating the texture uniformity factor in the ECDI index and is used to modulate the FAI to distinguish between algal blooms and floating vegetation. This is then used for texture feature extraction.
[0037] Then, based on the above features, a five-layer decision tree is constructed, and non-target features are stripped away layer by layer to achieve fine classification of aquatic vegetation. The threshold for each layer is dynamically determined using the maximum inter-class variance method, and the water body extent is extracted as follows: ,in, The segmentation threshold to be evaluated. To be at the threshold The inter-class variance between the foreground and background classes. For pixel values less than or equal to the threshold The proportion of pixels, For pixel values greater than the threshold The proportion of pixels, For pixel values less than or equal to the threshold The average grayscale value of all pixels. For pixel values greater than the threshold The average grayscale value of all pixels is used to extract the complete water mask from the image. Within the obtained water body mask range Internal application Index, determined by criteria Extraction of emergent / hygrophytic vegetation cover ,in, To determine the dynamic segmentation threshold for emergent vegetation extraction, in the remaining water body area after removing emergent vegetation, the following applies: Index, determined by criteria Identify and remove areas of cyanobacterial blooms, among which... A dynamic segmentation threshold is used to remove cyanobacterial blooms, thereby obtaining a pure water mask free from bloom interference. Reused in a pure water mask An index is used, combined with geometric area constraints to filter out noise, and the decision conditions are met. Extracting floating vegetation mask ,in, The dynamic segmentation threshold for extracting floating vegetation. The area of the connected pixel patch, The minimum area threshold for determining a valid vegetation patch, typically 5 pixels (approximately 3.2 m² in a 0.8 m resolution image), is used to filter out salt-and-pepper noise. Within a subset of water bodies meeting specific habitat conditions (such as suitable water depth and transparency), the submerged vegetation sensitivity index is used for extraction. The determination criteria are as follows: ,in, The dynamic segmentation threshold for submerged vegetation is automatically solved for SVSI images using the Otsu's method, which is consistent with the threshold solution methods for the other four layers. The sensitivity index for submerged vegetation, and ,in, The surface reflectance is in the green light band. Because of the water depth, For water transparency, the above water depth and transparency values are typical habitat constraints for shallow urban lakes. In practical applications, they can be dynamically adjusted according to lake type, trophic status, and season. The system supports dynamic input of hydrological monitoring data interpolation, historical inversion results, and field measurements. It also outputs submerged vegetation masking. Subsequently, after completing the five-layer decision-making process, the vegetation masks output from each layer are merged, i.e. The final vegetation classification raster data undergoes morphological processing (using 3×3 circular structuring elements for opening operations to remove small patches and closing operations to fill holes) and vectorization to generate a preliminary aquatic vegetation spatial distribution vector layer containing category attributes (emergent, floating, and submerged). This layer is then delivered to the downstream human-machine collaborative correction stage. The aquatic vegetation multidimensional feature adaptive remote sensing interpretation algorithm first proposes an enhanced cyanobacterial bloom detection index. This index is jointly modulated by introducing a texture uniformity factor on top of the traditional phytoplankton index to distinguish between spectrally similar cyanobacterial blooms and floating algae blooms. Aquatic vegetation is analyzed, and then an adaptive edge vegetation index is proposed. By coupling vegetation indices, water body indices, and local spectral heterogeneity, adaptive enhancement of vegetation signals in mixed pixels at the water-vegetation interface is achieved, thereby alleviating the problems of overly hard boundaries and missed extraction. The above values are only used to demonstrate the calculation logic of each spectral index. All layered segmentation thresholds in this invention are dynamically solved based on real image data of the current period using the Otsu's method, without the need for manually preset fixed reflectance or threshold parameters. The algorithm can adapt to remote sensing images of different lakes, different seasons, and different water body optical conditions. Finally, based on these intelligently interpreted multidimensional features, a five-layer decision tree with threshold adaptation is used to achieve fine and automatic interpretation and extraction of emergent, floating, and submerged vegetation, thereby intelligently interpreting and extracting the multidimensional features of lake aquatic vegetation.
[0038] Preferably, the human-computer interaction collaborative correction unit is used for intelligent assisted visual interpretation and accurate error correction. It introduces expert-experienced interpreters to overlay high-resolution fused images, performs operations according to a preset judgment rule library, and provides fine correction tools such as polygon deletion, shoreline truncation, boundary expansion, and manual delineation to ensure that the boundaries and categories of the extracted results conform to the actual ecological status.
[0039] Preferably, the UAV verification and closed-loop feedback optimization module is used for ground truth sample generation and accuracy evaluation. Based on the centimeter-level high-precision orthophotos acquired by the UAV equipped with RTK, ground truth samples are generated, the accuracy of the intelligent interpretation results is evaluated, and the evaluation results are fed back to the intelligent interpretation unit to drive its model parameters to adaptively optimize, forming a technical closed loop of "extraction-verification-optimization".
[0040] Preferably, the coverage calculation and thematic product generation module is used for ecological indicator quantification and thematic result visualization. It performs spatiotemporal dimension processing on the final aquatic vegetation vector data that has been verified and corrected, and transforms the dry raster / vector data into intuitive decision support products. It generates thematic maps of the spatial distribution of aquatic vegetation for single months / multi-months / years, and performs chart-based analysis and display of spatial patterns, type proportions and time-series evolution.
[0041] Preferably, the standardized data management and archiving module is used for full lifecycle data management. According to the standard hierarchical directory, the summary vectors, maximum water body vectors, thematic maps and original image analysis results generated throughout the process are structured and archived and managed to ensure the traceability of data and the reproducibility of results.
[0042] The proposed system and method for collaborative monitoring and intelligent interpretation and extraction of multi-source remote sensing data for lake aquatic vegetation, using Xinglong Lake in Chengdu, Sichuan Province in July 2025 as the monitoring target, aims to obtain the fine spatial distribution and coverage of emergent, floating, and submerged vegetation. The specific implementation process is as follows:
[0043] 1) Multi-source data acquisition and integrated preprocessing: This involves the collaborative acquisition of multi-source heterogeneous data, including high-resolution images from the domestic Gaofen-2 satellite, multispectral data from Sentinel-2, historical hydrological vectors, and high-precision verification images from UAVs equipped with RTK. The integrated image preprocessing module performs a series of standardized preprocessing steps, including radiometric calibration, atmospheric correction, geometric fine correction, and image fusion, generating standardized data products with a spatial resolution of 0.8 meters that are suitable for subsequent intelligent interpretation. The shortwave infrared bands required for MNDWI and FAI coarse classification are provided by Sentinel-2 products. The original 10m / 20m resolution was retained to perform preliminary extraction of coarse-scale water bodies and cyanobacterial blooms, generating a coarse classification mask. After rigorous registration with the domestic Gaofen-2 image based on the same set of digital orthophoto map / digital line map control points and digital elevation model geometric benchmarks, the coarse classification mask was spatially mapped to a 0.8m pixel grid as a spatial constraint for high-resolution fine interpretation. The visible-near-infrared bands were fused to a 0.8m resolution for pixel-by-pixel fine extraction of vegetation boundaries and types, avoiding spectral distortion caused by SWIR super-resolution resampling.
[0044] 2) Multidimensional feature calculation: Calculation is performed pixel-by-pixel on the preprocessed image. Taking the measured reflectance feature values of four typical land features—open water, cyanobacterial blooms, floating vegetation, and emergent vegetation—in the July 2025 image of Xinglong Lake as an example, the values are: Open water ( , , , ), Cyanobacterial blooms ( , , , ), floating vegetation ( , , , emergent vegetation , , , According to the formula The normalized water indexes for open water bodies, cyanobacterial blooms, floating vegetation, and emergent vegetation were +0.311, +0.451, +0.236, and -0.124, respectively. This indicates that pure water bodies and bloom areas... The values are positive (+0.31 to +0.45), while emergent vegetation falls into the negative value zone. Based on this, the water body and terrestrial vegetation can be initially separated. Then, according to the formula... The normalized vegetation indices (NDIs) for the four pixel classes were calculated to be -0.385, +0.200, +0.575, and +0.736, respectively. It can be seen that only through... When cyanobacterial blooms and floating vegetation both show positive values and overlapping bands, then according to the formula... , and Calculate the enhanced cyanobacterial bloom detection index of cyanobacterial blooms and floating vegetation, and the cyanobacterial bloom... , , floating vegetation , , The detection indices for enhanced cyanobacterial blooms and floating vegetation-induced cyanobacterial blooms were calculated to be 0.0183 and 0.0164, respectively. The dynamic segmentation threshold for cyanobacterial bloom removal was... Then it satisfies The region was identified as a cyanobacterial bloom and removed. This effectively separates cyanobacterial blooms with highly similar spectra from floating vegetation. Then, a mixed pixel at the water-emergent vegetation boundary was taken. The reflectance of this mixed pixel is... , , , According to the formula The adaptive edge vegetation index is calculated to be 0.851. If directly used... Perform threshold segmentation (set the threshold to 0.35), and the mixed pixel... These pixels should have been extracted—but in practice, due to atmospheric scattering / edge blurring, the actual observed boundary pixels are mixed. It is often diluted to between 0.20 and 0.30, resulting in missed extraction. pass An additional gain of +0.15 to +0.25 is added at this position to raise the equivalent segmentation surface, ensuring that edge pixels that should belong to vegetation stably cross the dynamic segmentation threshold for emergent vegetation extraction. This alleviates the problems of overly rigid boundaries and missed points.
[0045] 3) Five-layer adaptive decision tree classification: for Maximize the inter-class variance of the image Regarding the image in this example, Values are concentrated in the range of +0.18 to +0.55 for water pixels and in the range of -0.40 to -0.10 for land / vegetation pixels. After traversal, the peak position of the inter-class variance is extracted to obtain the complete water mask. Subsequently, emergent water extraction, algal bloom removal, floating water extraction, and submerged water extraction were performed layer by layer. Water depth and transparency data were obtained from synchronous observation data from three hydrological monitoring stations around Xinglong Lake. The inverse distance weighted interpolation method was used to obtain the water depth and transparency raster of the entire lake. The three vegetation masks were merged and morphological processing and vectorization were performed by opening / closing operations of 3×3 circular structural elements to output a preliminary aquatic vegetation distribution vector layer.
[0046] 4) Human-computer interactive collaborative correction: The initial vector overlay of the 0.8m fused image is processed, and the interpreter operates according to the rule base: polygon deletion is performed on the pink flocculent area in the center of the lake (falsely identified as an algal bloom), shoreline truncation is performed on the terrestrial shrubs and grasses along the shoreline, and so on. The rigid boundary of the transition zone is extended outward by 2-3 pixels towards the water side for supplementation, and the final vector result is output.
[0047] 5) UAV RTK Validation and Closed-Loop Feedback Optimization: 500 validation sample points were randomly deployed for emergent and floating vegetation based on the orthophoto data from an RTK UAV; underwater ground truth samples for submerged vegetation were simultaneously obtained using shipborne underwater cameras combined with in-situ quadrat surveys. The confusion matrix yielded the following results: overall accuracy OA=92.4%, Kappa=0.89, emergent PA=95.1%, floating PA=91.8%, and submerged PA=88.3%. Misclassified features were fed back to the interpretation unit, and three core parameters—ECDI texture entropy weight coefficient, AEVI local variance adjustment coefficient k, and SVSI segmentation threshold d—were specifically optimized to form a self-optimizing closed loop of "extraction-validation-parameter iteration-re-extraction". To verify the technical effectiveness of this invention, the same image was compared and extracted using a traditional NDVI combined with a fixed threshold decision tree method. The results showed that the overall accuracy of the traditional method was 78.2%, with Kappa=0.71. Among them, the accuracy of floating vegetation users was only 67.5% (due to the large amount of cyanobacterial blooms mixed in), and the omission rate of edge vegetation reached 22.3%. In contrast, the overall accuracy of the method of this invention was improved by 14.2 percentage points, the accuracy of floating vegetation users was improved to 90.6%, and the omission rate of edge vegetation was reduced to 4.7%, which significantly improved the classification accuracy and boundary accuracy.
[0048] 6) Coverage calculation, thematic mapping and standardized archiving: Final statistics: emergent area 3.81ha / floating area 2.18ha / submerged area 1.27ha, effective water area of the lake extracted by remote sensing in July is 25.06ha, coverage is 28.0% (emergent area 15.2% / floating area 8.7% / submerged area 5.1%). The system generates a thematic map of spatial distribution and a type percentage chart for the month, and archives the original image, intermediate feature layer, final vector, thematic map and accuracy report together according to the "Xinglong Lake / 2025-07 / " hierarchical directory.
[0049] This invention relates to a system and method for collaborative monitoring and intelligent interpretation and extraction of multi-source remote sensing data for lake aquatic vegetation. It aims to achieve automated, high-precision monitoring and intelligent interpretation and extraction of lake aquatic vegetation by collaboratively fusing multi-source remote sensing data and based on multi-dimensional features and an adaptive decision model. The system integrates a multi-source remote sensing data acquisition module, an integrated image preprocessing module, an intelligent interpretation and collaborative correction module, a UAV verification and closed-loop feedback optimization module, a coverage calculation and thematic product generation module, and a standardized data management and archiving module. It provides a system and method for collaborative monitoring and intelligent interpretation and extraction of multi-source remote sensing data for lake aquatic vegetation, and proposes an adaptive remote sensing interpretation algorithm for multi-dimensional features of aquatic vegetation to intelligently interpret and extract these features. The innovation of this invention lies in the fact that the adaptive remote sensing interpretation algorithm for multi-dimensional features of aquatic vegetation first proposes an enhanced cyanobacterial bloom detection index. This index is achieved by introducing a texture uniformity factor for joint modulation on the basis of the traditional phytoplankton index, in order to distinguish between cyanobacterial blooms with highly similar spectra and those of other algae. This paper proposes an adaptive edge vegetation index for floating vegetation. By coupling the vegetation index, water body index, and local spectral heterogeneity, it aims to adaptively enhance the vegetation signal in mixed pixels at the water-vegetation interface, thereby alleviating the problems of overly hard boundaries and missed extractions. Finally, based on these intelligently interpreted multidimensional features, a five-layer decision tree with threshold adaptation is used to achieve fine and automatic interpretation and extraction of emergent, floating, and submerged vegetation. This enables intelligent interpretation and extraction of multidimensional features of lake aquatic vegetation, effectively improving the working effect of multi-source remote sensing data collaborative monitoring and intelligent interpretation and extraction systems and methods for lake aquatic vegetation. It provides more intelligent and accurate technical support for multi-source remote sensing data collaborative monitoring and intelligent interpretation and extraction systems and methods for lake aquatic vegetation. Furthermore, this invention relates to remote sensing science and Earth observation technology, providing an efficient and convenient method and system for collaborative monitoring and intelligent interpretation and extraction of multi-source remote sensing data for lake aquatic vegetation, contributing significant application value to remote sensing science and Earth observation.
[0050] Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A multi-source remote sensing data collaborative monitoring and intelligent interpretation extraction system for lake aquatic vegetation, characterized in that, It includes a multi-source remote sensing data acquisition module, an integrated dual-scale image preprocessing module, a multi-dimensional feature intelligent interpretation and collaborative correction module, an UAV truth verification and closed-loop feedback optimization module, a coverage calculation and thematic product generation module, and a standardized full lifecycle data management and archiving module. The multi-source remote sensing data acquisition module is used for collaborative acquisition of heterogeneous data from multiple sources in air, space, and ground, and simultaneously acquires high-resolution visible-near-infrared satellite images, short-wave infrared multispectral satellite images, historical hydrological vector data of lakes, and high-precision verification images from RTK UAVs, thereby constructing a multi-scale, multi-modal observation data source. The integrated dual-scale image preprocessing module is used to perform multi-source image dual-scale fusion and spatial constraint preprocessing: it fuses the visible-near-infrared bands to a high spatial resolution for subsequent pixel-by-pixel fine ground feature interpretation; it retains the original lower resolution of the short-wave infrared band to complete the initial extraction of water bodies and cyanobacterial blooms at the coarse scale, and after strict spatial registration, it generates a coarse classification mask and maps it to the high-resolution image as an interpretation spatial constraint, thereby avoiding spectral distortion caused by over-resolution resampling of the short-wave infrared band from the root, and outputting a standardized preprocessed image with a unified spatiotemporal reference; The multi-dimensional feature intelligent interpretation and collaborative correction module includes a multi-dimensional feature intelligent interpretation and extraction unit and a human-computer interaction collaborative correction unit. The multidimensional feature intelligent interpretation and extraction unit incorporates an adaptive remote sensing interpretation algorithm for multidimensional features of aquatic vegetation, and constructs a multidimensional feature space that integrates water body index, vegetation index, enhanced cyanobacterial bloom detection index, adaptive edge vegetation index, and gray-level co-occurrence matrix texture statistics. The enhanced cyanobacterial bloom detection index is obtained by jointly modulating the phytoplankton index and the texture uniformity factor, and the texture difference between cyanobacterial blooms and floating vegetation is used to widen the discrimination boundary between the two types of land cover. Adaptive edge vegetation index coupled with vegetation spectral response, water background features and local spectral heterogeneity to adaptively enhance vegetation signal intensity in mixed pixels at the water-land interface; The algorithm adopts a five-layer progressive adaptive decision tree layer-by-layer stripping framework. The segmentation threshold of each layer is dynamically and automatically solved by the maximum inter-class variance method: the initial water body mask extraction, emergent / wet vegetation extraction and water body boundary correction are completed in sequence, cyanobacterial bloom identification and removal, floating vegetation extraction, and submerged vegetation extraction are completed; finally, the multi-layer vegetation masks are merged and the preliminary vegetation vector layer is output after morphological filtering. The human-computer interaction collaborative correction unit is used to overlay high-resolution fused images and complete the fine manual correction of vegetation boundaries. The UAV ground truth verification and closed-loop feedback optimization module is used to construct ground truth samples based on high-precision orthophotos of RTK UAVs and underwater field surveys, and to evaluate the accuracy of intelligent interpretation results; and to feed back the feature patterns of misclassified pixels to the interpretation unit, and to optimize three core algorithm parameters in a targeted manner: cyanobacteria detection weight, edge enhancement coefficient, and submerged vegetation segmentation threshold, forming a self-optimizing technology closed loop. The coverage calculation and thematic product generation module is used to perform coverage quantification statistics on the final verified and corrected aquatic vegetation data, and generate thematic chart products on the spatial distribution pattern, type proportion and time series evolution of vegetation. The standardized full lifecycle data management and archiving module is used to perform structured archiving management of the original images, intermediate feature layers, vector results, thematic products and accuracy reports throughout the entire process, according to a standard hierarchical directory. The adaptive edge vegetation index characterizes spectral heterogeneity using the standard deviation of the vegetation index within a local window and dynamically adjusts the edge enhancement weight. The higher the degree of water-land mixing in the region where the pixel is located, the greater the enhancement of the vegetation signal, thereby compensating for the vegetation response diluted by the water background in the mixed pixels.
2. The lake-oriented aquatic vegetation multi-source remote sensing data collaborative monitoring and intelligent interpretation extraction system according to claim 1, characterized in that, The multi-source remote sensing data acquisition module simultaneously acquires visible-near-infrared images from domestic high-resolution satellites, short-wave infrared multispectral data from Sentinel series satellites, historical hydrological vector data of lakes, and high-precision verification images from UAVs equipped with real-time dynamic positioning.
3. The lake-oriented aquatic vegetation multi-source remote sensing data collaborative monitoring and intelligent interpretation extraction system according to claim 1, characterized in that, The integrated dual-scale image preprocessing module first performs radiometric calibration, atmospheric correction, and geometric fine correction on the multi-source images in sequence, and then performs dual-scale fusion and coarse classification mask spatial mapping; the registration error of geometric fine correction is controlled at the sub-pixel level.
4. The lake-oriented aquatic vegetation multi-source remote sensing data collaborative monitoring and intelligent interpretation extraction system according to claim 1, characterized in that, The multidimensional feature space includes the improved normalized water index, the normalized vegetation index, the enhanced cyanobacterial bloom detection index, the adaptive edge vegetation index, and the gray-level co-occurrence matrix texture statistics, forming a multidimensional feature combination of spectrum, texture, and geometry.
5. The lake-oriented aquatic vegetation multi-source remote sensing data collaborative monitoring and intelligent interpretation extraction system according to claim 1, characterized in that, In the enhanced cyanobacterial bloom detection index, the texture uniformity factor is obtained by transforming the normalized texture entropy based on the gray-level co-occurrence matrix. The higher the texture uniformity of the ground features, the stronger the modulation gain, thereby amplifying the numerical difference between homogeneous cyanobacterial blooms and heterogeneous floating vegetation.
6. The lake-oriented aquatic vegetation multi-source remote sensing data collaborative monitoring and intelligent interpretation extraction system according to claim 1, characterized in that, The layer-by-layer stripping logic of the five-layer progressive adaptive decision tree is as follows: the first layer extracts the initial water body mask of the entire region; the second layer extracts emergent / wet vegetation in the water-land interface zone and corrects the water body boundary; the third layer identifies and removes cyanobacterial blooms to obtain a pure water body mask; the fourth layer extracts floating vegetation in the pure water body by combining patch area constraints; and the fifth layer extracts submerged vegetation in habitat areas with suitable water depth and transparency.
7. The lake-oriented aquatic vegetation multi-source remote sensing data collaborative monitoring and intelligent interpretation extraction system according to claim 1, characterized in that, The UAV ground truth verification and closed-loop feedback optimization module constructs ground truth samples using UAV orthophotos for emergent and floating vegetation, and constructs ground truth samples using shipborne underwater cameras combined with in-situ quadrat surveys for submerged vegetation. After completing the accuracy evaluation through the confusion matrix, the feature patterns of misclassified pixels are fed back to the interpretation unit to drive the adaptive iterative optimization of algorithm parameters.
8. The lake-oriented aquatic vegetation multi-source remote sensing data collaborative monitoring and intelligent interpretation extraction system according to claim 1, characterized in that, The human-computer interaction collaborative correction unit provides four types of refined correction tools: polygon deletion, shoreline truncation, boundary expansion, and manual delineation. It supports experts in correcting vegetation extraction boundaries and category determination based on image texture features. The coverage calculation and thematic product generation module statistically analyzes the area and coverage of various aquatic vegetation types and generates spatial distribution thematic maps, type proportion statistics charts, and time series evolution analysis charts.
9. A method for collaborative monitoring and intelligent interpretation and extraction of multi-source remote sensing data for lake aquatic vegetation, characterized in that, The system described in any one of claims 1-8 is implemented by comprising the following steps: S1 Multi-Source Data Collaborative Acquisition and Dual-Scale Preprocessing: Simultaneously acquire high-resolution visible-near-infrared satellite images, short-wave infrared multispectral satellite images, historical hydrological vectors, and high-precision verification images; perform high-resolution fusion of visible-near-infrared bands for fine interpretation, retain the original resolution of short-wave infrared bands to complete the initial extraction of coarse-scale water bodies and cyanobacteria, and after spatial registration, map the coarse classification mask onto the high-resolution image as the spatial constraint for interpretation. S2 Multidimensional Feature Space Construction: Calculate water index, vegetation index, enhanced cyanobacterial bloom detection index, and adaptive edge vegetation index per pixel, extract texture statistics, and construct a multidimensional feature space; S3 five-layer progressive adaptive decision tree classification: The maximum inter-class variance method is used to dynamically solve the segmentation threshold layer by layer. The water body, emergent vegetation, cyanobacterial blooms are removed, floating vegetation is extracted and submerged vegetation is extracted in sequence according to the layer-by-layer stripping logic. After merging, the preliminary vegetation vector is output after morphological processing. S4 Human-Computer Interaction Collaborative Correction: The initial vector overlay of high-resolution images is used for refined boundary and category correction; S5 Truth Verification and Closed-Loop Optimization: Based on high-precision verification data, truth samples are constructed. After completing the accuracy evaluation, misclassification features are fed back to the interpretation algorithm to iteratively optimize the core parameters and form a self-optimizing closed loop. S6 Coverage Statistics, Thematic Mapping and Standardized Archiving: Calculate the coverage of various vegetation types, generate thematic products and complete the entire data archiving process according to specifications.
Citation Information
Patent Citations
MODIS (moderate resolution imaging spectroradiometer) satellite synchronous monitoring method for cyanobacterial bloom and aquatic vegetation in eutrophic lakes
CN106315856A
Enteromorpha green tide remote sensing monitoring method, system, equipment, medium and program product
CN118533763A
