Method and system for monitoring and early warning of enteromorpha distribution based on stationary meteorological satellite

Abstract

This invention provides a method and system for dynamic monitoring and early warning of *Ulva prolifera* distribution based on geostationary meteorological satellites, belonging to the field of *Ulva prolifera* distribution dynamic monitoring technology. The method includes: acquiring standardized real-time monitoring data of the target sea area; extracting *Ulva prolifera* characteristic parameters to construct a *Ulva prolifera* identification feature set; constructing a deep learning-based *Ulva prolifera* identification model, inputting the *Ulva prolifera* identification feature set, and outputting suspected *Ulva prolifera* distribution areas; performing boundary purification and coordinate calibration on the suspected *Ulva prolifera* distribution areas, and calculating the current core indicators of *Ulva prolifera*; constructing a dynamic prediction model for *Ulva prolifera*, inputting the current core indicators of *Ulva prolifera* and real-time marine environmental factor data, and outputting the predicted results of dynamic changes in *Ulva prolifera* over time; and generating tiered monitoring and early warning information. This invention improves the accuracy of *Ulva prolifera* identification by constructing and optimizing the *Ulva prolifera* identification model; and achieves dynamic prediction and proactive control of *Ulva prolifera* through the dynamic prediction model and tiered early warning mechanism, reducing monitoring costs and contributing to marine ecological protection.

Description

Technical Field

[0001] This invention belongs to the field of dynamic monitoring technology for the distribution of Ulva prolifera, specifically a method and system for dynamic monitoring and early warning of Ulva prolifera distribution based on geostationary meteorological satellites. Background Technology

[0002] Ulva prolifera is a large green algae widely distributed in coastal waters worldwide. It is characterized by rapid reproduction and strong adaptability, and readily exhibits explosive growth under suitable temperature, light, and nutrient conditions. As a typical marine phytoplankton, Ulva prolifera itself does not produce toxins, but large-scale aggregations can have multiple impacts on the marine ecosystem and coastal economic activities. Ulva prolifera outbreaks in coastal waters have become a global marine ecological disaster. Its rapid reproduction and spread lead to water hypoxia, shading of sunlight, and destruction of marine habitats. It also affects port navigation, mariculture, and coastal tourism, causing significant economic losses and ecological damage.

[0003] Traditional methods for monitoring *Ulva prolifera* (seaweed) mainly rely on ship patrols, drone aerial photography, and manual observation. These methods suffer from limitations such as limited monitoring range, poor real-time performance, high labor costs, and discontinuous data acquisition, making it difficult to meet the needs of large-scale, high-precision dynamic monitoring. With the development of satellite remote sensing technology, multispectral remote sensing has become an important means of *Ulva prolifera* monitoring. However, existing technologies still have many shortcomings. Insufficient image preprocessing precision leads to data distortion due to atmospheric scattering, cloud cover, and geometric distortion, affecting subsequent identification accuracy. Furthermore, feature extraction is relatively simplistic, often relying solely on spectral features while ignoring spatial structural information such as texture, making it difficult to effectively distinguish *Ulva prolifera* from seawater, suspended sediment, and other background elements. Moreover, the lack of effective dynamic prediction and graded early warning mechanisms makes it impossible to predict the drift and aggregation trends of *Ulva prolifera* in advance, resulting in passive control efforts. Therefore, developing a *Ulva prolifera* monitoring technology system that combines high-precision identification, real-time dynamic prediction, and scientific early warning functions has become an urgent need in the field of marine ecological environment protection. Summary of the Invention

[0004] This invention provides a method and system for dynamic monitoring and early warning of the distribution of Ulva prolifera based on geostationary meteorological satellites, in order to overcome the deficiencies in the existing technology.

[0005] On the one hand, this invention provides a method for dynamic monitoring and early warning of *Ulva prolifera* distribution based on geostationary meteorological satellites, including:

[0006] S1. Real-time acquisition of geostationary meteorological satellite multispectral image data stream of the target sea area, and sequential radiometric correction, geometric correction and cloud removal and noise reduction preprocessing of each frame of image data to obtain standardized real-time image data.

[0007] S2. Extract the feature parameters of Ulva prolifera from standardized real-time image data, and construct a Ulva prolifera identification feature set based on the feature parameters.

[0008] S3. Construct a deep learning-based model for identifying seaweed, and use the whale optimization algorithm to optimize the hyperparameters of the model. Input the seaweed identification feature set and output the suspected distribution area of ​​seaweed and the seaweed identification confidence of each pixel in the area.

[0009] S4. Perform boundary purification and coordinate calibration on the suspected distribution area of ​​Ulva prolifera, and calculate the current core indicators of Ulva prolifera, including the coverage area, distribution density and coordinate location of the core area.

[0010] S5. Collect real-time marine environmental factor data, combine historical Ulva prolifera core indicator time series data with corresponding marine environmental factor time series data, construct a Ulva prolifera dynamic prediction model based on time series-environmental factor coupling, input the current Ulva prolifera core indicator and real-time marine environmental factor data, and output the prediction results of Ulva prolifera dynamic changes in the time series.

[0011] S6. Generate graded monitoring and early warning information based on the current core indicators of Ulva prolifera and the prediction results of its dynamic changes.

[0012] According to the method for dynamic monitoring and early warning of *Ulva prolifera* distribution based on geostationary meteorological satellites provided by this invention, the multispectral image data stream represents a sequence of target sea area image data collected in real time by a multispectral sensor carried by a geostationary meteorological satellite, containing multiple specific wavelength bands, including blue, green, red, and near-infrared bands. *Ulva prolifera* characteristic parameters include spectral reflectance features, vegetation index features, and texture features. Spectral reflectance features include blue band reflectance, green band reflectance, red band reflectance, and near-infrared band reflectance. Vegetation index features include normalized difference vegetation index (NDVI), enhanced vegetation index (EDI), and ratio vegetation index (RI). Texture features include contrast, correlation, energy, and entropy values. Marine environmental factor data include seawater temperature data, salinity data, ocean current velocity data, ocean current direction data, and light intensity data.

[0013] According to the method for dynamic monitoring and early warning of Ulva prolifera distribution based on geostationary meteorological satellites provided by the present invention, step S2, the process of extracting Ulva prolifera feature parameters and constructing Ulva prolifera identification feature set includes:

[0014] S21. Separate single-band images of blue, green, red and near-infrared bands from standardized real-time image data, calculate the spectral reflectance of each pixel in each single-band image, and obtain spectral reflectance features including blue band reflectance matrix, green band reflectance matrix, red band reflectance matrix and near-infrared band reflectance matrix.

[0015] S22. Based on spectral reflectance characteristics, the normalized vegetation index, enhanced vegetation index, and ratio vegetation index are calculated using the vegetation index calculation formula to form the corresponding vegetation index matrix, which serves as the vegetation index feature.

[0016] S23. The grayscale image of the standardized real-time image data is analyzed using the grayscale co-occurrence matrix method to calculate four texture features: contrast, correlation, energy, and entropy.

[0017] S24. Align the spectral reflectance features, vegetation index features, and texture features in dimensions, and concatenate the feature values ​​corresponding to the same pixel to form the feature vector of that pixel. The set of feature vectors of all pixels constitutes the Ulva prolifera recognition feature set.

[0018] According to the method for dynamic monitoring and early warning of Ulva prolifera distribution based on geostationary meteorological satellites provided by the present invention, the process of constructing a Ulva prolifera identification model based on deep learning in step S3 includes:

[0019] S31. Collect historical geostationary meteorological satellite multispectral image data, corresponding measured distribution data of Ulva prolifera, and measured confidence labeling data. Preprocess the image data to obtain standardized historical image data. Divide the standardized historical image data, Ulva prolifera measured distribution data, and confidence labeling data into training set and validation set.

[0020] S32. Perform multi-scale feature extraction on standardized historical image data to obtain historical spectral reflectance features, historical vegetation index features, and historical texture features, and fuse them to construct a historical Ulva prolifera identification feature set.

[0021] S33. Construct a lightweight deep convolutional neural network model, introduce an attention mechanism to enhance the response of Ulva prolifera features, take the historical Ulva prolifera identification feature set as input, and take the Ulva prolifera distribution label and identification confidence as output.

[0022] S34. Train the model, verify the model performance through the validation set, retain the model parameters that meet the requirements of real-time performance and recognition accuracy, and obtain the seaweed recognition model.

[0023] According to the method for dynamic monitoring and early warning of Ulva prolifera distribution based on geostationary meteorological satellites provided by the present invention, in step S3, the process of optimizing the hyperparameters of the Ulva prolifera identification model using the whale optimization algorithm includes:

[0024] S35. Randomly generate a set of initial whale individual positions. Each whale individual represents a hyperparameter combination, which includes the convolution kernel size, learning rate decay coefficient, number of network layers, and attention mechanism weight coefficient.

[0025] S36. For each individual whale, calculate the model's comprehensive performance index on the validation set. The comprehensive performance index is a weighted fusion value of recognition accuracy and real-time processing efficiency, and is used as the fitness value.

[0026] S37. Update the location of each individual whale, including surrounding prey, bubble net attacks, and random searches.

[0027] S38. Repeat the position update process and calculate the fitness value of each individual whale after each iteration until the preset number of iterations is reached.

[0028] S39. Select the hyperparameter combination representing the location of the whale individual with the best fitness value as the optimal hyperparameter combination for the seaweed recognition model.

[0029] According to the method for dynamic monitoring and early warning of *Ulva prolifera* distribution based on geostationary meteorological satellites provided by the present invention, step S4, the process of calculating the current core indicators of *Ulva prolifera* includes:

[0030] S41. Set the confidence threshold for Ulva prolifera identification. The confidence level of areas suspected of being distributed by *Ulva prolifera* was less than [a certain value]. The pixels are used to obtain the preliminary distribution area of ​​the seaweed.

[0031] S42. Set the set of pixels in the initial distribution area of ​​*Ulva prolifera* as follows: ,in Represents the image coordinates of the i-th Ulva prolifera pixel. , Let be the x-coordinate of the image of the i-th pixel. Let be the ordinate of the image of the i-th pixel. This represents the total number of pixels in the seaweed plant.

[0032] S43. Transform image coordinates using a geographic coordinate transformation matrix. Convert to geographic coordinates ,in Let be the longitude of the i-th pixel. Let be the latitude of the i-th pixel.

[0033] S44. Perform connectivity analysis and boundary extraction on the calibrated seaweed pixels, and use a polynomial fitting method to smooth the boundaries.

[0034] S45. Solve for the fitting coefficients using the least squares method to obtain the smoothed boundary curve of the Ulva prolifera distribution area.

[0035] S46. Calculate the coverage area and distribution density of Ulva prolifera based on the boundary curve, and calculate the coordinates of the core area of ​​Ulva prolifera. The coordinates of the core area include the coordinates of the geometric center, the coordinates of the boundary vertex, and the coordinates of the key feature points.

[0036] According to the method for dynamic monitoring and early warning of Ulva prolifera distribution based on geostationary meteorological satellites provided by the present invention, the process of constructing a dynamic prediction model of Ulva prolifera based on time series-environmental factor coupling in step S5 includes:

[0037] S511. Collect time-series data of historical Ulva prolifera core indicators and corresponding marine environmental factors, and construct a coupled time-series dataset.

[0038] S512. Perform data preprocessing on the coupled time series dataset, including stationarity test, outlier removal, missing value imputation and normalization, to obtain standardized coupled time series data.

[0039] S513. Construct a dual-branch coupled prediction model, including a time-series feature extraction branch, an environmental factor influence modeling branch, and a prediction head network. The time-series feature extraction branch uses a long short-term memory network to extract the time-series dependency features of the core indicators of *Ulva prolifera*, and outputs a time-series feature vector. The environmental factor impact modeling branch uses a fully connected neural network to perform nonlinear mapping on marine environmental factor data, outputting an environmental impact feature vector. .

[0040] S514, Transfer the time series feature vector Environmental impact feature vector Perform fusion to construct a fused feature vector. The input is fed into the prediction head network to obtain the prediction results of the dynamic changes of Ulva prolifera. The prediction head network is used to perform nonlinear transformation and decision output on the fused feature vector.

[0041] S515. Divide the standardized coupled time series data into training and testing sets, train and validate the dual-branch coupled prediction model, retain the model parameters that meet the prediction accuracy requirements, and obtain the dynamic prediction model for Ulva prolifera.

[0042] According to the method for dynamic monitoring and early warning of Ulva prolifera distribution based on geostationary meteorological satellites provided by the present invention, step S5, the process of outputting the prediction results of Ulva prolifera dynamic changes over time includes:

[0043] S521, Set the future forecast time series as , The current core indicators of seaweed and real-time marine environmental factors are input into the seaweed dynamic prediction model.

[0044] S522. Time series calculated using the Ulva prolifera dynamic prediction model. The dynamic prediction parameters for *Ulva prolifera* at each time point include:

[0045] S523, Output the geometric center coordinates of the core region at each time point. , The drift path of the seaweed is obtained by connecting the geometric center coordinates of each time point.

[0046] S524. Calculate the rate of change of coverage area at each time point. and density change rate ,like and If it is determined to be a clustering trend, and The trend is determined to be spreading, where S is the coverage area of ​​*Ulva prolifera*. The distribution density of *Ulva prolifera* For the first Predicted values ​​of Ulva prolifera coverage area at specific time points. For the first Predicted distribution density of *Ulva prolifera* at specific time points.

[0047] S525. Output the coordinates of the boundary vertex of the distribution area of ​​*Ulva prolifera*, the coordinates of key feature points, and the coordinates of the geometric center of the core area at each time point, forming a complete set of location coordinates.

[0048] According to the method for dynamic monitoring and early warning of *Ulva prolifera* distribution based on geostationary meteorological satellites provided by the present invention, the process of generating graded monitoring and early warning information in step S6 includes:

[0049] S61. Set early warning thresholds for core indicators and dynamic changes of seaweed.

[0050] S62. Determine the warning level based on real-time core indicators and forecast results, including general warning, moderate warning and severe warning.

[0051] S63. Generate early warning information based on the early warning level determination results, including the current core indicator data of Ulva prolifera, real-time location distribution, prediction results of dynamic changes in future time series, early warning level, and response suggestions.

[0052] On the other hand, the present invention also provides a dynamic monitoring and early warning system for the distribution of *Ulva prolifera* based on geostationary meteorological satellites, comprising:

[0053] The real-time image preprocessing module is used to acquire geostationary meteorological satellite multispectral image data streams of the target sea area in real time, and perform radiometric correction, geometric correction and cloud removal and noise reduction preprocessing on each frame of image data to obtain standardized real-time image data.

[0054] The feature extraction module is used to extract the feature parameters of Ulva prolifera from standardized real-time image data and construct a feature set for Ulva prolifera recognition.

[0055] The real-time seaweed identification module is used to build a seaweed identification model based on deep learning. The hyperparameters of the seaweed identification model are optimized using the whale optimization algorithm. The input is the seaweed identification feature set, and the output is the real-time suspected distribution area of ​​seaweed and the identification confidence.

[0056] The core indicator calculation module is used to purify the boundaries and calibrate the coordinates of suspected Ulva prolifera distribution areas, and to calculate the coverage area, distribution density, and coordinate location of the core area of ​​Ulva prolifera.

[0057] The marine environmental data acquisition module is used to collect marine environmental factor data in real time, including seawater temperature, salinity, ocean current speed, ocean current direction, and light intensity.

[0058] The Ulva prolifera dynamic prediction module is used to combine historical time-series data and real-time data to build a dynamic prediction model for Ulva prolifera based on the coupling of time-series sequence and environmental factors, and output the drift path, aggregation and dispersion trend and real-time location coordinates of Ulva prolifera in the time series.

[0059] The graded early warning module is used to generate graded monitoring and early warning information based on real-time data and dynamic prediction results of the core indicators of seaweed.

[0060] This invention provides a method and system for dynamic monitoring and early warning of *Ulva prolifera* distribution based on geostationary meteorological satellites. By employing multispectral imagery data from geostationary meteorological satellites and combining it with refined preprocessing procedures such as radiometric correction, geometric correction, and cloud removal and noise reduction, the reliability of the image data is ensured. A comprehensive identification feature set is constructed by integrating spectral, vegetation index, and texture multidimensional features. Combined with a deep learning model optimized by the whale optimization algorithm, the accuracy of *Ulva prolifera* identification is improved, meeting real-time monitoring needs. A predictive model coupling time-series data and environmental factors is constructed, integrating historical data of core *Ulva prolifera* indicators with environmental factors such as seawater temperature and ocean currents, to accurately predict the drift path, aggregation trend, and distribution location of *Ulva prolifera*. A tiered early warning mechanism is established, combining core indicators and dynamic change thresholds to generate targeted response suggestions, providing scientific decision-making support for prevention and control work, and realizing a shift from passive disposal to proactive prevention and control. The fully automated processing process reduces manual intervention and lowers the manpower and equipment costs of monitoring. By predicting the spread direction and outbreak risk of *Ulva prolifera* in advance, the system guides the precise deployment of prevention and control resources, improves clearance efficiency, reduces the damage of *Ulva prolifera* to marine ecosystems and coastal industries, and contributes to marine ecological environmental protection and sustainable development. Attached Figure Description

[0061] The invention will now be further described with reference to the accompanying drawings.

[0062] Figure 1 This is a flowchart illustrating the dynamic monitoring and early warning method for the distribution of *Ulva prolifera* based on geostationary meteorological satellites in this invention.

[0063] Figure 2 This is a schematic diagram of the structure of the Ulva prolifera distribution dynamic monitoring and early warning system based on geostationary meteorological satellites in this invention. Detailed Implementation

[0064] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.

[0065] like Figures 1 to 2 As shown, the method and system for dynamic monitoring and early warning of Ulva prolifera distribution based on geostationary meteorological satellites provided in this embodiment of the invention can be implemented by a method for dynamic monitoring and early warning of Ulva prolifera distribution based on geostationary meteorological satellites, the method including:

[0066] S1. Real-time acquisition of geostationary meteorological satellite multispectral image data stream of the target sea area, and sequential radiometric correction, geometric correction and cloud removal and noise reduction preprocessing of each frame of image data to obtain standardized real-time image data.

[0067] Multispectral image data streams represent real-time image data sequences of target sea areas collected by multispectral sensors aboard geostationary meteorological satellites, encompassing multiple specific wavelength bands. These specific wavelength bands include the blue band (wavelength range approximately 450-520 nm), the green band (wavelength range approximately 530-600 nm), the red band (wavelength range approximately 630-690 nm), and the near-infrared band (wavelength range approximately 760-900 nm). The selection of these bands is based on the differences in spectral reflectance characteristics between *Ulva prolifera* and seawater. The blue and green bands reflect the clarity of seawater and the basic reflectance information of *Ulva prolifera*, the red band effectively distinguishes the absorption differences between *Ulva prolifera* and seawater, and the near-infrared band is sensitive to the vegetation characteristics of *Ulva prolifera*, providing crucial spectral evidence for its identification.

[0068] The preprocessing of each frame of image data includes:

[0069] An absolute radiometric correction method based on satellite sensor calibration coefficients is employed to eliminate the influence of sensor-specific response characteristics, atmospheric transmission effects, and variations in solar altitude angle on image radiometric values. This corrects radiometric distortions caused by atmospheric scattering and absorption, converting the image's digital quantization values ​​into true surface reflectance.

[0070] Based on high-precision geospatial reference data of the target sea area, a polynomial correction method is used to correct geometric distortion in the imagery. A sufficient number of evenly distributed ground control points are selected on the imagery; these control points must possess both clear image features and accurate geographic coordinates. A polynomial transformation model is constructed, and the model parameters are solved using the least squares method to achieve a precise mapping between image pixel coordinates and geographic coordinates. Bilinear interpolation is then used to resample the imagery, obtaining geometrically accurate image data.

[0071] To address potential cloud cover and sensor noise issues in geostationary meteorological satellite imagery, a multi-step cloud denoising strategy is employed. For cloud-covered areas, cloud pixels are initially identified using threshold segmentation based on the spectral characteristics of clouds. Morphological filtering and neighborhood analysis are then used to eliminate misidentified non-cloud pixels. Linear interpolation or texture synthesis based on surrounding valid pixels is then used to fill and repair cloud-covered areas. For sensor noise, a combination of median filtering and Gaussian filtering is used. Median filtering removes salt-and-pepper noise, while Gaussian filtering smooths Gaussian noise, preserving the detailed features of the seaweed (Ulva prolifera) to the greatest extent possible while eliminating noise and avoiding image distortion.

[0072] S2. Extract the feature parameters of Ulva prolifera from standardized real-time image data, and construct a Ulva prolifera identification feature set based on the feature parameters.

[0073] The characteristic parameters of *Ulva prolifera* include spectral reflectance, vegetation index, and texture. Spectral reflectance includes blue, green, red, and near-infrared reflectance. As an aquatic plant, *Ulva prolifera* exhibits typical vegetation spectral characteristics: low reflectance in the blue and red bands due to pigment absorption, a small reflectance peak in the green band, and high reflectance in the near-infrared band due to scattering from the internal structure of its leaves. This significantly differs from the spectral reflectance characteristics of seawater (high reflectance in the blue band, moderate reflectance in the green band, and low reflectance in the red and near-infrared bands).

[0074] Vegetation index features include the Normalized Difference Vegetation Index (NDVI), Enhanced Vegetation Index (EVI), and Ratio Vegetation Index (RVI), which are composite characteristic indicators constructed based on spectral reflectance. NDVI effectively suppresses the influence of atmospheric and soil background, enhancing the distinction between vegetation and non-vegetation areas. EVI, by introducing atmospheric correction coefficients and canopy background adjustment coefficients, reduces interference from atmospheric aerosols and soil background, and is more sensitive to areas with high vegetation cover. RVI amplifies the spectral differences between vegetation and non-vegetation areas by using the ratio of near-infrared to red band reflectance.

[0075] Texture features, including contrast, correlation, energy, and entropy, reflect the gray-level distribution patterns and spatial structure information of *Ulva prolifera* images. *Ulva prolifera* exhibits specific texture patterns in images (such as continuous sheet-like or banded textures), clearly distinguishing it from the uniform texture of seawater. Contrast reflects the degree of difference in image gray levels; due to its complex texture structure, the contrast of *Ulva prolifera* areas is usually higher than that of seawater areas. Correlation describes the degree of association between the gray-level values ​​of adjacent pixels in the image; the gray-level correlation between adjacent pixels is strong in *Ulva prolifera* areas. Energy reflects the uniformity and regularity of the image's gray-level distribution; seawater areas have higher energy, while *Ulva prolifera* areas have relatively lower energy. Entropy reflects the complexity of image information; due to its rich texture details, the entropy value of *Ulva prolifera* areas is higher than that of seawater areas.

[0076] The process of extracting the feature parameters of *Ulva prolifera* and constructing a feature set for *Ulva prolifera* recognition includes:

[0077] S21. Separate single-band images of blue, green, red and near-infrared bands from standardized real-time image data, calculate the spectral reflectance of each pixel in each single-band image, and obtain spectral reflectance features including blue band reflectance matrix, green band reflectance matrix, red band reflectance matrix and near-infrared band reflectance matrix.

[0078] S22. Based on spectral reflectance characteristics, the normalized vegetation index, enhanced vegetation index, and ratio vegetation index are calculated using the vegetation index calculation formula to form the corresponding vegetation index matrix, which serves as the vegetation index feature. The normalized vegetation index calculation formula is as follows:

[0079]

[0080] The formula for calculating the enhanced vegetation index is:

[0081]

[0082] The formula for calculating the ratio vegetation index is:

[0083]

[0084] In the formula, For near-infrared reflectivity, For red band reflectivity, For blue band reflectivity, This is the gain factor (default value is 2.5). , These are atmospheric correction factors (valued at 6.0 and 7.5 respectively). This is the canopy background adjustment factor (set to 1.0).

[0085] S23. The grayscale image of the standardized real-time image data is analyzed using the grayscale co-occurrence matrix method to calculate four texture features: contrast, correlation, energy, and entropy.

[0086] S24. Align the spectral reflectance features, vegetation index features, and texture features in dimensions, and concatenate the feature values ​​corresponding to the same pixel to form the feature vector of that pixel. The set of feature vectors of all pixels constitutes the Ulva prolifera recognition feature set.

[0087] S3. Construct a deep learning-based model for identifying seaweed, and use the whale optimization algorithm to optimize the hyperparameters of the model. Input the seaweed identification feature set and output the suspected distribution area of ​​seaweed and the seaweed identification confidence of each pixel in the area.

[0088] The process of building a deep learning-based model for identifying seaweed includes:

[0089] S31. Collect geostationary meteorological satellite multispectral image data of the target sea area from the past 5-10 years (covering scenes with different seasons and different outbreak levels of Ulva prolifera), corresponding measured distribution data of Ulva prolifera (actual distribution range and density data of Ulva prolifera obtained through ship patrol monitoring, UAV aerial photography monitoring, etc.), and measured confidence label data (confidence values ​​labeled according to the reliability of the measured data, ranging from 0-1). Perform radiometric correction, geometric correction, and cloud removal and noise reduction on the collected historical image data according to the aforementioned image preprocessing workflow to obtain standardized historical image data. Then, divide the standardized historical image data, the corresponding measured distribution data of Ulva prolifera, and the confidence label data into training and validation sets in a 7:3 ratio.

[0090] S32. Multi-scale feature extraction is performed on standardized historical image data. Historical spectral reflectance features, historical vegetation index features, and historical texture features are extracted using the same methods as real-time images. A multi-scale feature fusion strategy is introduced, performing convolution operations at 1×1, 3×3, and 5×5 scales on each feature to extract feature information at different scales. Features at different scales are then fused through feature concatenation to construct a more dimensional and representative historical Ulva prolifera identification feature set, thereby improving the model's ability to identify Ulva prolifera populations of different sizes.

[0091] S33. Construct a lightweight deep convolutional neural network model. The model adopts an Encoder-Decoder structure. The Encoder part consists of four convolutional blocks, each containing two 3×3 convolutional layers, one batch normalization layer, and one ReLU activation function layer. The convolutional layers use a stride of 2 to achieve downsampling, gradually extracting the deep features of the seaweed. The Decoder part consists of four deconvolutional blocks, each containing one 2×2 deconvolutional layer, one batch normalization layer, and one ReLU activation function layer. Deconvolution operations are used to upsample the feature maps, restoring the spatial resolution of the image. An attention mechanism (using a CBAM attention module combining channel attention and spatial attention) is introduced between the Encoder and Decoder. The channel attention module adaptively adjusts the weights of each feature channel to enhance the response to key channel features for seaweed identification, while the spatial attention module highlights the spatial features of the seaweed-containing area and suppresses interference from the background area. The model takes a historical set of Ulva prolifera identification features as input and outputs two parts: one part is the distribution identifier of Ulva prolifera (a binary classification result, where 0 represents a non-Ulva prolifera pixel and 1 represents a Ulva prolifera pixel), and the other part is the confidence score of Ulva prolifera identification (representing the probability that the model judges the pixel to be Ulva prolifera, ranging from 0 to 1).

[0092] S34. Train the model, verify the model performance through the validation set, retain the model parameters that meet the requirements of real-time performance and recognition accuracy, and obtain the seaweed recognition model.

[0093] The process of optimizing the hyperparameters of the *Ulva prolifera* recognition model using the whale optimization algorithm includes:

[0094] S35. Randomly generate a set of initial whale individual positions. Each whale individual represents a hyperparameter combination, which includes the convolution kernel size, learning rate decay coefficient, number of network layers, and attention mechanism weight coefficient.

[0095] S36. For each individual whale, calculate the model's comprehensive performance index on the validation set. The comprehensive performance index is a weighted fusion value of recognition accuracy and real-time processing efficiency, and is used as the fitness value.

[0096] S37. Update the position of each individual whale. The position update strategy includes three methods: surrounding the prey, bubble net attack, and random search. The update method is selected by random probability to balance the global search capability and local optimization capability of the algorithm.

[0097] When the random probability p < 0.5 and the convergence factor A < 1, the prey encirclement strategy is used to update the position of individual whales. The update formula is expressed as:

[0098]

[0099] in, This represents the position of the individual whale at the (t+1)th iteration (corresponding to a set of hyperparameter combinations). Let be the position of the individual whale at the t-th iteration. This represents the optimal position of a whale in the population at the t-th iteration (corresponding to the optimal combination of hyperparameters). For the coefficient vector, , , The convergence factor is initially 2, which decreases linearly to 0 with the number of iterations. and It is a random number between [0,1].

[0100] When the random probability p ≥ 0.5 and the convergence factor A < 1, the bubble net attack strategy is used to update the position of individual whales, simulating the bubble net foraging behavior of whales during hunting. The update formula is expressed as:

[0101]

[0102] in, The logarithmic spiral shape coefficient (with a value of 1, used to control the trajectory shape of the bubble net attack, causing individual whales to perform a spiral search around the optimal position, enhancing the refinement of the local search). A random number between [-1, 1] (used to adjust the direction of the spiral motion to make the search range more comprehensive).

[0103] When the convergence factor A≥1, a random search strategy is used to update the position of individual whales, simulating the behavior of whales randomly selecting prey, thus escaping local optima. The update formula is expressed as:

[0104]

[0105] in, The position of the whale individual randomly selected from the population at the t-th iteration (corresponding to a random set of hyperparameter combinations) is used to expand the search range by randomly selecting reference individuals, thus avoiding the algorithm from getting stuck in local optima.

[0106] S38. Set the preset number of iterations to 50 rounds, repeat the position update process, calculate the fitness value of each whale individual after each iteration, and update the position of the best whale individual in the population.

[0107] S39. When the preset number of iterations is reached, the iteration is stopped, and the hyperparameter combination represented by the position of the whale individual with the best fitness value in the population is selected as the optimal hyperparameter combination for the seaweed recognition model. The optimal hyperparameter combination is substituted into the seaweed recognition model, and the model is retrained (training iterations are 100 rounds) to obtain the seaweed recognition model.

[0108] S4. Perform boundary purification and coordinate calibration on the suspected distribution area of ​​Ulva prolifera, and calculate the current core indicators of Ulva prolifera, including the coverage area, distribution density and coordinate location of the core area.

[0109] The process of calculating the current core indicators of *Ulva prolifera* includes:

[0110] S41. Based on statistical analysis of a large amount of measured data, set a confidence threshold for Ulva prolifera identification. (This threshold can be dynamically adjusted based on the distribution characteristics of *Ulva prolifera* in the target sea area and the required monitoring accuracy. When the *Ulva prolifera* density in the target sea area is low, the threshold can be appropriately lowered to 0.6; when the requirement for extremely high identification accuracy is required, the threshold can be increased to 0.8.) All pixels within the suspected *Ulva prolifera* distribution area are traversed, and pixels with an identification confidence score lower than [a certain value] are removed. For pixels, retain those with a confidence level greater than or equal to 1. The pixels are used to obtain the preliminary distribution area of ​​the seaweed.

[0111] S42. Set the set of pixels in the initial distribution area of ​​*Ulva prolifera* as follows: ,in Represents the image coordinates of the i-th Ulva prolifera pixel. , Let be the x-coordinate of the image of the i-th pixel. Let be the ordinate of the image of the i-th pixel. This represents the total number of pixels in the seaweed plant.

[0112] S43. Transform image coordinates using a geographic coordinate transformation matrix. Convert to geographic coordinates ,in Let be the longitude of the i-th pixel. Given the latitude of the i-th pixel, perform coordinate calibration.

[0113] S44. Perform connectivity analysis on the calibrated *Ulva prolifera* pixels. Using the 8-neighborhood connectivity criterion, adjacent *Ulva prolifera* pixels are divided into the same connected region, and isolated small connected regions with fewer than 5 pixels are removed. The chain code tracing algorithm is used to extract the boundary pixels of each connected region to obtain the initial boundary curve. Since the initial boundary curve has jagged fluctuations, a polynomial fitting method is used to smooth the boundary. The fitting formula is:

[0114]

[0115] in, The longitude of the boundary point, The latitude of the boundary point These are the fitting coefficients. The order of the fitted polynomial is given.

[0116] S45. The least squares method is used to solve for the fitting coefficients. The objective function is constructed with the goal of minimizing the sum of squared residuals between the actual latitude and the fitted latitude of the boundary points. The optimal fitting coefficients are obtained by solving for the extreme values ​​of the objective function, and then the smoothed boundary curve of the Ulva prolifera distribution area is obtained.

[0117] S46. Calculate the coverage area of ​​*Ulva prolifera* based on the boundary curve. The formula is:

[0118]

[0119] in, The geographical area enclosed by the boundary curve. For longitude micro-element, For latitude micro-elements, This is the area correction factor for geographic coordinates. The trapezoidal integral method is used to solve for the specific value of the area covered by *Ulva prolifera*.

[0120] Calculate the distribution density of Ulva prolifera. The formula is:

[0121]

[0122] in, The total number of valid pixels within the distribution area of ​​*Ulva prolifera* (confidence level greater than or equal to) (number of pixels) This represents the area covered by *Ulva prolifera*.

[0123] Calculate the coordinates of the core region of *Ulva prolifera*. The core region coordinates include the geometric center coordinates, boundary vertex coordinates, and key feature point coordinates. The geometric center coordinates are:

[0124]

[0125] In the formula, Represents the coordinates of the geometric center. This represents the longitude of the i-th pixel. Let represent the latitude of the i-th pixel, and N represent the total number of valid pixels within the distribution area of ​​the seaweed.

[0126] Extract the coordinates of the four vertices corresponding to the maximum and minimum longitude and the maximum and minimum latitude values ​​on the boundary curve. , , , Used as the coordinates of the boundary vertices.

[0127] Extract the pixel coordinates corresponding to the maximum curvature on the boundary curve, and use them as the key feature point coordinates of the *Ulva prolifera* distribution. The curvature calculation formula is:

[0128]

[0129] in, The first derivative of the boundary curve. Let be the second derivative of the boundary curve. Calculate the curvature value at each point on the boundary curve, and identify points whose curvature values ​​are greater than a preset curvature threshold (set according to the smoothness of the boundary curve) as key feature points.

[0130] S5. Collect real-time marine environmental factor data, including seawater temperature data, salinity data, ocean current velocity data, ocean current direction data, and light intensity data. Combine historical time-series data of Ulva prolifera core indicators with corresponding time-series data of marine environmental factors to construct a dynamic prediction model of Ulva prolifera based on time-series-environmental factor coupling. Input the current Ulva prolifera core indicators and real-time marine environmental factor data, and output the prediction results of Ulva prolifera dynamic changes in the time series.

[0131] The process of constructing a dynamic prediction model for *Ulva prolifera* based on the coupling of time series and environmental factors includes:

[0132] S511. Collect historical time-series data of key indicators of *Ulva prolifera* over the past 5-10 years (including time-series sequences of coverage area, distribution density, and geometric center coordinates of the core area), as well as corresponding time-series data of marine environmental factors (including time-series sequences of seawater temperature, salinity, ocean current velocity, ocean current direction, and light intensity). Correlate and match the *Ulva prolifera* key indicator data with the marine environmental factor data at the same time point to construct a coupled time-series dataset. The time granularity of the time-series dataset is a preset time unit.

[0133] S512. Perform data preprocessing on the coupled time series dataset, including stationarity test, outlier removal, missing value imputation and normalization, to obtain standardized coupled time series data.

[0134] S513. Construct a dual-branch coupled prediction model, including a temporal feature extraction branch, an environmental factor influence modeling branch, and a prediction head network. The temporal feature extraction branch uses a Long Short-Term Memory (LSTM) network to extract the temporal dependency features of the core indicators of *Ulva prolifera*. The LSTM network contains three hidden layers, each with 128 neurons. The input is a 24-dimensional temporal sequence of the core indicators of *Ulva prolifera* (i.e., the core indicator data of *Ulva prolifera* over the past 24 hours). Through the gating mechanism (input gate, forget gate, output gate) of the LSTM network, long-term dependencies in the temporal data are adaptively captured, and the output is a 64-dimensional temporal feature vector. This vector comprehensively represents the historical trends and temporal patterns of the core indicators of *Ulva prolifera*.

[0135] The environmental factor impact modeling branch employs a fully connected neural network (FNN) to perform nonlinear mapping on marine environmental factor data. The FNN network contains two hidden layers: the first layer has 64 neurons, and the second layer has 32 neurons. The ReLU activation function is used for both layers. The input consists of marine environmental factor data from the current moment and the past 23 hours (a total of 24 time points, with each time point's environmental factor vector having a dimension of 5). Through the nonlinear transformation of the fully connected layers, the environmental factor data is mapped into a 32-dimensional environmental impact feature vector. This vector quantifies the combined effects of marine environmental factors on the growth and drift of Ulva prolifera.

[0136] S514. The prediction head network is constructed using a fully connected neural network, containing two hidden layers (64 neurons in the first layer and 32 neurons in the second layer) and one output layer. The temporal feature vector... Environmental impact feature vector The features are concatenated to construct a fusion feature vector with a dimension of 96. The fused feature vector is input into the prediction head network. Through nonlinear transformation of the hidden layer and linear transformation of the output layer, the fused feature vector is mapped to a preset target for predicting the dynamic changes of *Ulva prolifera*. The output of the prediction head network is the dynamic prediction result of *Ulva prolifera* for the next 12 hours (1-hour time granularity), including the geometric center coordinates of the core area, the rate of change of coverage area, and the rate of change of density at each time point. The mapping relationship formula is expressed as:

[0137]

[0138] in, For the future Dynamic prediction results of Ulva prolifera over a time unit (including drift path node coordinates, coverage area change rate, and density change rate). To predict the time step, For predicting head networks, The time-series feature extraction branch extracts core indicator data of *Ulva prolifera* from the first t time units. The extraction results The branch for modeling the impact of environmental factors uses marine environmental factor data from the first t time units. The mapping result.

[0139] S515. Divide the standardized coupled time series data into training and testing sets, train and validate the dual-branch coupled prediction model, retain the model parameters that meet the prediction accuracy requirements, and obtain the dynamic prediction model for Ulva prolifera.

[0140] The process of outputting the prediction results of the dynamic changes of Ulva prolifera on the time series includes:

[0141] S521, Set the future forecast time series as , The current core indicators of seaweed and real-time marine environmental factors are input into the seaweed dynamic prediction model.

[0142] S522. Obtain the time series through model calculation. The dynamic prediction parameters for *Ulva prolifera* at each time point include:

[0143] S523, Output the geometric center coordinates of the core region at each time point. , Connecting the coordinate points yields the drift path of the seaweed.

[0144] S524. Calculate the rate of change of coverage area at each time point. and density change rate ,like and This is judged as a clustering trend, indicating that the distribution range of *Ulva prolifera* is shrinking but the density is increasing, potentially forming a high-density *Ulva prolifera* aggregation area. If... and If both the coverage area change rate and density change rate are greater than 0, it is considered an expansion and aggregation trend, indicating that the distribution range of Ulva prolifera is expanding while its density is decreasing, and it may spread to a wider sea area. If both are less than 0, it is considered a contraction and dispersion trend, indicating that the distribution range of Ulva prolifera is shrinking and its density is decreasing, and its growth is inhibited.

[0145] S525. Output the coordinates of the boundary vertex of the distribution area of ​​*Ulva prolifera*, the coordinates of key feature points, and the coordinates of the geometric center of the core area at each time point, forming a complete set of location coordinates.

[0146] Where S represents the coverage area of ​​the seaweed. The distribution density of *Ulva prolifera* For the first Predicted values ​​of Ulva prolifera coverage area at specific time points. For the first Predicted distribution density of *Ulva prolifera* at specific time points.

[0147] S6. Based on the current core indicators of *Ulva prolifera* and the predicted results of its dynamic changes, generate tiered monitoring and early warning information. The process includes:

[0148] S61. Based on the marine ecological carrying capacity, Ulva prolifera control capabilities, and historical Ulva prolifera outbreak disaster data of the target sea area, set early warning thresholds for core Ulva prolifera indicators and early warning thresholds for dynamic changes:

[0149] In this example, the coverage area threshold The density threshold is set based on the key protected area of ​​the target sea area (such as within a 10-kilometer radius of a port, or the entire aquaculture area). Based on the degree of impact of Ulva prolifera on the marine ecological environment, the value was set at 500 pixels per square kilometer (that is, when the number of effective Ulva prolifera pixels per square kilometer of sea area reaches 500, it will have a significant impact on marine water quality and the living environment of marine organisms).

[0150] Drift speed threshold Based on the protection response time setting for key sea areas, a value of 0.5 m / s is used (when the drift speed of *Ulva prolifera* exceeds this threshold, emergency prevention and control measures must be initiated within a short period of time). Diffusion / aggregation rate threshold. The value is set to 0.2 based on the speed of change of the *Ulva prolifera* (i.e., when the rate of change in coverage area or density exceeds 20% per hour, it indicates that the *Ulva prolifera* is in a state of rapid change and requires close monitoring). A trend level classification coefficient is also set. and It is used to distinguish between moderate and rapid changing trends.

[0151] S62. The warning level is determined based on real-time core indicators and forecast results. The warning level is divided into three levels: general warning, moderate warning, and severe warning. The specific determination rules are as follows:

[0152] General early warnings include real-time core indicators (coverage area) ,density None of the values ​​exceeded the threshold, but a moderate diffusion / aggregation trend is predicted in the future (i.e., 0.1 ≤ change rate ≤ 0.2). This warning level indicates that the current distribution of Ulva prolifera has a relatively small impact on the marine ecological environment, but there is a slow trend of change in Ulva prolifera in the future. It is necessary to increase the frequency of monitoring and closely monitor the dynamic changes of Ulva prolifera.

[0153] A moderate alert is determined if any of the following conditions are met:

[0154] One of the real-time core metrics exceeds the threshold ( or (1), the other item did not exceed the threshold.

[0155] It is predicted that there will be a rapid diffusion / aggregation trend in the future (the rate of change is greater than 1%). That is, the rate of change is >0.2).

[0156] Predicted future drift speed of seaweed will exceed (Drift speed > 0.5 m / s).

[0157] This level of warning indicates that the current seaweed bloom has already impacted the local marine ecological environment, or that the seaweed bloom will change rapidly in the future. It is necessary to activate the emergency monitoring plan and organize personnel and equipment to prepare for prevention and control work.

[0158] A severe warning is issued if any of the following conditions are met:

[0159] Both of the real-time core metrics exceeded the threshold ( and Furthermore, it is predicted that there will be a rapid diffusion / aggregation trend in the future (change rate > 0.2) and the duration will exceed 4 hours.

[0160] The predicted drift path will affect key sea areas (for example, the area where the predicted distribution of seaweed overlaps with the area of ​​ports, aquaculture areas, and tourist attractions exceeds 30% of the total area of ​​key sea areas).

[0161] This level of warning indicates that *Ulva prolifera* has already had a serious impact on the marine ecological environment, or is about to pose a threat to key sea areas. It is necessary to immediately activate the emergency response plan and take comprehensive prevention and control measures such as physical removal, chemical control, and biological control to minimize the losses caused by *Ulva prolifera*.

[0162] S63. Generate early warning information based on the early warning level determination results. The early warning information includes the current core indicator data of Ulva prolifera, real-time location distribution, prediction results of dynamic changes in future time series, early warning level, and response suggestions.

[0163] In summary, this embodiment provides a dynamic monitoring and early warning method for *Ulva prolifera* distribution based on geostationary meteorological satellites. By employing multispectral imagery data from geostationary meteorological satellites, combined with refined preprocessing procedures such as radiometric correction, geometric correction, and cloud removal and noise reduction, the reliability of the image data is ensured. A comprehensive identification feature set is constructed by integrating spectral, vegetation index, and texture multidimensional features. This, coupled with a deep learning model optimized by the whale optimization algorithm, improves the accuracy of *Ulva prolifera* identification, meeting real-time monitoring requirements. A time-series-environmental factor coupled prediction model is built, integrating historical data of core *Ulva prolifera* indicators with environmental factors such as seawater temperature and ocean currents, to accurately predict the drift path, aggregation trend, and distribution location of *Ulva prolifera*. A tiered early warning mechanism is established, combining core indicators and dynamic change thresholds to generate targeted response suggestions, providing scientific decision-making support for prevention and control work, and realizing a shift from passive handling to proactive prevention and control. The fully automated processing process reduces manual intervention and lowers the manpower and equipment costs of monitoring. By predicting the spread direction and outbreak risk of *Ulva prolifera* in advance, the method guides the precise deployment of prevention and control resources, improves removal efficiency, reduces the damage of *Ulva prolifera* to marine ecosystems and coastal industries, and contributes to marine ecological environmental protection and sustainable development.

[0164] Based on the same general inventive concept, this invention also protects a dynamic monitoring and early warning system for the distribution of *Ulva prolifera* based on geostationary meteorological satellites. The dynamic monitoring and early warning system for the distribution of *Ulva prolifera* based on geostationary meteorological satellites provided by this invention will be described below. The dynamic monitoring and early warning system for the distribution of *Ulva prolifera* based on geostationary meteorological satellites described below can be referred to in correspondence with the dynamic monitoring and early warning method and system for the distribution of *Ulva prolifera* based on geostationary meteorological satellites described above.

[0165] The Ulva prolifera distribution dynamic monitoring and early warning system based on geostationary meteorological satellites includes a real-time image preprocessing module, a feature extraction module, a real-time Ulva prolifera identification module, a core indicator calculation module, a marine environmental data acquisition module, a Ulva prolifera dynamic prediction module, and a graded early warning module.

[0166] The real-time image preprocessing module is used to acquire geostationary meteorological satellite multispectral image data streams of the target sea area in real time, and to perform radiometric correction, geometric correction and cloud removal and noise reduction preprocessing on each frame of image data to obtain standardized real-time image data.

[0167] The feature extraction module is used to extract the feature parameters of Ulva prolifera from standardized real-time image data and construct a feature set for Ulva prolifera recognition.

[0168] The real-time seaweed identification module is used to build a seaweed identification model based on deep learning. The hyperparameters of the seaweed identification model are optimized using the whale optimization algorithm. The input is the seaweed identification feature set, and the output is the real-time suspected distribution area of ​​seaweed and the identification confidence.

[0169] The core indicator calculation module is used to purify the boundaries and calibrate the coordinates of suspected Ulva prolifera distribution areas, and to calculate the coverage area, distribution density, and coordinate location of the core area of ​​Ulva prolifera.

[0170] The marine environmental data acquisition module is used to collect marine environmental factor data in real time, including seawater temperature, salinity, ocean current speed, ocean current direction, and light intensity.

[0171] The Ulva prolifera dynamic prediction module is used to combine historical time-series data and real-time data to build a Ulva prolifera dynamic prediction model based on the coupling of time series and environmental factors, and outputs the drift path, aggregation and dispersion trend and real-time location coordinates of Ulva prolifera in the time series.

[0172] The graded early warning module is used to generate graded monitoring and early warning information based on real-time data and dynamic prediction results of the core indicators of seaweed.

[0173] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0174] 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 the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for dynamic monitoring and early warning of *Ulva prolifera* distribution based on geostationary meteorological satellites, characterized in that, include: S1. Obtain real-time monitoring data of geostationary meteorological satellites in the target sea area, extract the characteristic parameters of seaweed, and construct a seaweed identification feature set; S2. Construct a deep learning-based seaweed identification model, optimize the hyperparameters of the seaweed identification model using the whale optimization algorithm, input the seaweed identification feature set, and output the suspected distribution area of ​​seaweed and the seaweed identification confidence of each pixel in the area. S3. Perform boundary purification and coordinate calibration on the suspected distribution area of ​​Ulva prolifera, and calculate the current core indicators of Ulva prolifera, including the coverage area, distribution density and coordinate location of the core area. S4. Collect real-time marine environmental factor data, combine historical Ulva prolifera core indicator time series data with corresponding marine environmental factor time series data, construct a Ulva prolifera dynamic prediction model based on time series-environmental factor coupling, input the current Ulva prolifera core indicator and real-time marine environmental factor data, and output the prediction results of Ulva prolifera dynamic changes in the time series. S5. Based on the current core indicators of Ulva prolifera and the prediction results of Ulva prolifera dynamic changes, generate graded monitoring and early warning information.

2. The method for dynamic monitoring and early warning of *Ulva prolifera* distribution based on geostationary meteorological satellites according to claim 1, characterized in that, The *Ulva prolifera* characteristic parameters include spectral reflectance features, vegetation index features, and texture features. The spectral reflectance features include blue band reflectance, green band reflectance, red band reflectance, and near-infrared band reflectance. The vegetation index features include normalized difference vegetation index, enhanced vegetation index, and ratio vegetation index. The texture features include contrast, correlation, energy, and entropy. The marine environmental factor data include seawater temperature data, salinity data, ocean current velocity data, ocean current direction data, and light intensity data.

3. The method for dynamic monitoring and early warning of *Ulva prolifera* distribution based on geostationary meteorological satellites according to claim 1, characterized in that, In step S1, the process of extracting the feature parameters of *Ulva prolifera* and constructing the *Ulva prolifera* identification feature set includes: S11. Separate single-band images of blue, green, red and near-infrared bands from standardized real-time monitoring data, calculate the spectral reflectance of each pixel in each single-band image, and obtain spectral reflectance features including blue band reflectance matrix, green band reflectance matrix, red band reflectance matrix and near-infrared band reflectance matrix. S12. Based on the spectral reflectance characteristics, the normalized vegetation index, enhanced vegetation index and ratio vegetation index are calculated respectively using the vegetation index calculation formula to form the corresponding vegetation index matrix, which serves as the vegetation index feature. S13. The grayscale co-occurrence matrix method is used to perform texture analysis on the grayscale images of standardized real-time monitoring data, and four texture features, namely contrast, correlation, energy and entropy, are calculated. S14. Align the spectral reflectance feature, the vegetation index feature and the texture feature in dimensions, and concatenate the feature values ​​corresponding to the same pixel to form the feature vector of the pixel. The set of feature vectors of all pixels constitutes the Ulva prolifera recognition feature set.

4. The method for dynamic monitoring and early warning of *Ulva prolifera* distribution based on geostationary meteorological satellites according to claim 1, characterized in that, In step S2, the process of constructing a deep learning-based model for identifying seaweed includes: S21. Collect historical geostationary meteorological satellite multispectral image data, corresponding measured distribution data of Ulva prolifera, and measured confidence labeling data. Preprocess the image data to obtain standardized historical image data. Divide the standardized historical image data, Ulva prolifera measured distribution data, and confidence labeling data into training set and validation set. S22. Multi-scale feature extraction is performed on standardized historical image data to obtain historical spectral reflectance features, historical vegetation index features and historical texture features, which are then fused to construct a historical Ulva prolifera identification feature set. S23. Construct a lightweight deep convolutional neural network model, introduce an attention mechanism to enhance the feature response of Ulva prolifera, take the historical Ulva prolifera identification feature set as input, and take the Ulva prolifera distribution label and identification confidence as output. S24. Train the model, verify the model performance through the validation set, retain the model parameters that meet the requirements of real-time performance and recognition accuracy, and obtain the seaweed recognition model.

5. The method for dynamic monitoring and early warning of *Ulva prolifera* distribution based on geostationary meteorological satellites according to claim 1, characterized in that, In step S2, the process of optimizing the hyperparameters of the *Ulva prolifera* recognition model using the whale optimization algorithm includes: S25. Randomly generate a set of initial whale individual positions, each whale individual representing a hyperparameter combination, the hyperparameter combination including convolution kernel size, learning rate decay coefficient, number of network layers and attention mechanism weight coefficient; S26. For each individual whale, calculate the comprehensive performance index of the model on the validation set. The comprehensive performance index is a weighted fusion value of recognition accuracy and real-time processing efficiency, and use the comprehensive performance index as the fitness value. S27. Update the location of each individual whale, including surrounding prey, bubble net attacks, and random searches; S28. Repeat the position update process and calculate the fitness value of each whale individual after each iteration until the preset number of iterations is reached. S29. Select the hyperparameter combination representing the location of the whale individual with the best fitness value as the optimal hyperparameter combination for the seaweed recognition model.

6. The method for dynamic monitoring and early warning of *Ulva prolifera* distribution based on geostationary meteorological satellites according to claim 1, characterized in that, In step S3, the process of calculating the current core indicators of *Ulva prolifera* includes: S31. Set the confidence threshold for Ulva prolifera identification. The confidence level of areas suspected of being distributed by *Ulva prolifera* was less than [a certain value]. The pixels are used to obtain the preliminary distribution area of ​​the seaweed; S32. Set the set of pixels in the initial distribution area of ​​*Ulva prolifera* as follows: ,in Represents the image coordinates of the i-th Ulva prolifera pixel. , Let be the x-coordinate of the image of the i-th pixel. Let be the ordinate of the image of the i-th pixel. This represents the total number of pixels in the seaweed plant; S33. Transform image coordinates using a geographic coordinate transformation matrix. Convert to geographic coordinates ,in Let be the longitude of the i-th pixel. Let i be the latitude of the i-th pixel; S34. Perform connectivity analysis and boundary extraction on the calibrated seaweed pixels, and use a polynomial fitting method to smooth the boundaries. S35. Solve for the fitting coefficients using the least squares method to obtain the smoothed boundary curve of the Ulva prolifera distribution area; S36. Calculate the coverage area and distribution density of Ulva prolifera based on the boundary curve, and calculate the coordinate position of the core area of ​​Ulva prolifera, including the coordinates of the geometric center, the coordinates of the boundary vertex, and the coordinates of the key feature points.

7. The method for dynamic monitoring and early warning of *Ulva prolifera* distribution based on geostationary meteorological satellites according to claim 1, characterized in that, In step S4, the process of constructing a dynamic prediction model for *Ulva prolifera* based on the coupling of time series and environmental factors includes: S411. Collect historical time-series data of key indicators of Ulva lactuca and corresponding marine environmental factors, and construct a coupled time-series dataset. S412. Perform data preprocessing on the coupled time series dataset, including stationarity test, outlier removal, missing value imputation and normalization, to obtain standardized coupled time series data. S413. Construct a dual-branch coupled prediction model, including a time-series feature extraction branch, an environmental factor influence modeling branch, and a prediction head network. The time-series feature extraction branch uses a long short-term memory network to extract the time-series dependency features of the core indicators of *Ulva prolifera*, and outputs a time-series feature vector. The environmental factor impact modeling branch uses a fully connected neural network to perform nonlinear mapping on marine environmental factor data, outputting an environmental impact feature vector. ; S414, Transfer the time series feature vector Environmental impact feature vector Perform fusion to construct a fused feature vector. The data is input into the prediction head network to obtain the prediction result of the dynamic change of Ulva prolifera. The prediction head network is used to perform nonlinear transformation and decision output on the fused feature vector. S415. Divide the standardized coupled time series data into training and testing sets, train and validate the dual-branch coupled prediction model, retain the model parameters that meet the prediction accuracy requirements, and obtain the dynamic prediction model for Ulva prolifera.

8. The method for dynamic monitoring and early warning of *Ulva prolifera* distribution based on geostationary meteorological satellites according to claim 1, characterized in that, In step S4, the process of outputting the predicted results of the dynamic changes of Ulva prolifera over time includes: S421. Set the future forecast time series as... , The current core indicators of seaweed and real-time marine environmental factors are input into the seaweed dynamic prediction model; S422. Time series data were calculated using the Ulva prolifera dynamic prediction model. The dynamic prediction parameters for *Ulva prolifera* at each time point include: S423, Output the geometric center coordinates of the core region at each time point. , The drift path of the seaweed is obtained by connecting the geometric center coordinates of each time node; S424. Calculate the rate of change of coverage area at each time point. and density change rate ,like and If it is determined to be a clustering trend, and The trend is determined to be spreading, where S is the coverage area of ​​*Ulva prolifera*. The distribution density of *Ulva prolifera* For the first Predicted values ​​of Ulva prolifera coverage area at specific time points. For the first Predicted distribution density of *Ulva prolifera* at specific time points; S425. Output the coordinates of the boundary vertex of the distribution area of ​​*Ulva prolifera*, the coordinates of key feature points, and the coordinates of the geometric center of the core area at each time point, forming a complete set of location coordinates.

9. The method for dynamic monitoring and early warning of *Ulva prolifera* distribution based on geostationary meteorological satellites according to claim 1, characterized in that, In step S5, the process of generating graded monitoring and early warning information includes: S51. Set early warning thresholds for core indicators and dynamic changes of *Ulva prolifera*. S52. Determine the warning level based on real-time core indicators and forecast results, including general warning, moderate warning and severe warning; S53. Generate early warning information based on the early warning level determination results, including the current core indicator data of Ulva prolifera, real-time location distribution, prediction results of dynamic changes in future time series, early warning level, and response suggestions.

10. A dynamic monitoring and early warning system for the distribution of *Ulva prolifera* based on geostationary meteorological satellites, employing the dynamic monitoring and early warning method for the distribution of *Ulva prolifera* based on geostationary meteorological satellites as described in any one of claims 1 to 9, characterized in that... The system includes: The real-time image preprocessing module is used to acquire geostationary meteorological satellite multispectral image data streams of the target sea area in real time, and perform radiometric correction, geometric correction and cloud removal and noise reduction preprocessing on each frame of image data to obtain standardized real-time monitoring data. The feature extraction module is used to extract Ulva prolifera feature parameters from standardized real-time monitoring data and construct a Ulva prolifera identification feature set; The real-time seaweed identification module is used to build a seaweed identification model based on deep learning. The hyperparameters of the seaweed identification model are optimized using the whale optimization algorithm. The input is the seaweed identification feature set, and the output is the real-time suspected distribution area of ​​seaweed and the identification confidence. The core indicator calculation module is used to purify the boundaries and calibrate the coordinates of suspected Ulva prolifera distribution areas, and to calculate the coverage area, distribution density and coordinate location of the core area of ​​Ulva prolifera. The marine environmental data acquisition module is used to collect marine environmental factor data in real time, including seawater temperature, salinity, ocean current speed, ocean current direction and light intensity data; The Ulva prolifera dynamic prediction module is used to combine historical time series data and real-time data to build a Ulva prolifera dynamic prediction model based on the coupling of time series and environmental factors, and output the drift path, aggregation and dispersion trend and real-time location coordinates of Ulva prolifera in the time series. The graded early warning module is used to generate graded monitoring and early warning information based on real-time data and dynamic prediction results of the core indicators of seaweed.

Images